Download the Full Ebook and Access More Features - ebooknice.

com

(Ebook) Earth Observation Data Analytics Using

Machine and Deep Learning: Modern Tools,
Applications and Challenges by Sanjay Garg, Swati
Jain, Nitant Dube, Nebu Varghese (Editors) ISBN
9781839536175, 1839536179
https://ebooknice.com/product/earth-observation-data-
analytics-using-machine-and-deep-learning-modern-tools-
applications-and-challenges-50558476

OR CLICK HERE

DOWLOAD EBOOK

Download more ebook instantly today at https://ebooknice.com

 Instant digital products (PDF, ePub, MOBI) ready for you
Download now and discover formats that fit your needs...

Start reading on any device today!

(Ebook) Biota Grow 2C gather 2C cook by Loucas, Jason;

Viles, James ISBN 9781459699816, 9781743365571,
9781925268492, 1459699815, 1743365578, 1925268497
https://ebooknice.com/product/biota-grow-2c-gather-2c-cook-6661374

ebooknice.com

(Ebook) Matematik 5000+ Kurs 2c Lärobok by Lena

Alfredsson, Hans Heikne, Sanna Bodemyr ISBN 9789127456600,
9127456609
https://ebooknice.com/product/matematik-5000-kurs-2c-larobok-23848312

ebooknice.com

(Ebook) SAT II Success MATH 1C and 2C 2002 (Peterson's SAT

II Success) by Peterson's ISBN 9780768906677, 0768906679

https://ebooknice.com/product/sat-ii-success-
math-1c-and-2c-2002-peterson-s-sat-ii-success-1722018

ebooknice.com

(Ebook) Master SAT II Math 1c and 2c 4th ed (Arco Master

the SAT Subject Test: Math Levels 1 & 2) by Arco ISBN
9780768923049, 0768923042
https://ebooknice.com/product/master-sat-ii-math-1c-and-2c-4th-ed-
arco-master-the-sat-subject-test-math-levels-1-2-2326094

ebooknice.com
(Ebook) Cambridge IGCSE and O Level History Workbook 2C -
Depth Study: the United States, 1919-41 2nd Edition by
Benjamin Harrison ISBN 9781398375147, 9781398375048,
1398375144, 1398375047
https://ebooknice.com/product/cambridge-igcse-and-o-level-history-
workbook-2c-depth-study-the-united-states-1919-41-2nd-edition-53538044

ebooknice.com

(Ebook) Advanced Data Analytics Using Python: With Machine

Learning, Deep Learning and NLP Examples by Mukhopadhyay,
Sayan ISBN 9781484234495, 1484234499
https://ebooknice.com/product/advanced-data-analytics-using-python-
with-machine-learning-deep-learning-and-nlp-examples-55670484

ebooknice.com

(Ebook) Machine Learning and Big Data Analytics Paradigms:

Analysis, Applications and Challenges by Aboul Ella
Hassanien, Ashraf Darwish ISBN 9783030593377, 3030593371
https://ebooknice.com/product/machine-learning-and-big-data-analytics-
paradigms-analysis-applications-and-challenges-22035322

ebooknice.com

(Ebook) Deep Learning for Data Analytics: Foundations,

Biomedical Applications, and Challenges by Himansu Das;
Chittaranjan Pradhan; Nilanjan Dey ISBN 9780128197646,
0128197641
https://ebooknice.com/product/deep-learning-for-data-analytics-
foundations-biomedical-applications-and-challenges-11152452

ebooknice.com

(Ebook) Demystifying Big Data, Machine Learning, and Deep

Learning for Healthcare Analytics by Pradeep N Sandeep
Kautish Sheng-Lung Peng ISBN 9780128216330, 0128216336
https://ebooknice.com/product/demystifying-big-data-machine-learning-
and-deep-learning-for-healthcare-analytics-33381074

ebooknice.com
IET COMPUTING SERIES 56

Earth Observation Data

Analytics Using Machine
and Deep Learning
Other volumes in this series:
Volume 1 Knowledge Discovery and Data Mining M.A. Bramer (Editor)
Volume 3 Troubled IT Projects: Prevention and turnaround J.M. Smith
Volume 4 UML for Systems Engineering: Watching the wheels, 2nd Edition J. Holt
Volume 5 Intelligent Distributed Video Surveillance Systems S.A. Velastin and P.
Remagnino (Editors)
Volume 6 Trusted Computing C. Mitchell (Editor)
Volume 7 SysML for Systems Engineering J. Holt and S. Perry
Volume 8 Modelling Enterprise Architectures J. Holt and S. Perry
Volume 9 Model-Based Requirements Engineering J. Holt, S. Perry and M. Bownsword
Volume 13 Trusted Platform Modules: Why, when and how to use them A. Segall
Volume 14 Foundations for Model-based Systems Engineering: From patterns to
models J. Holt, S. Perry and M. Bownsword
Volume 15 Big Data and Software Defined Networks J. Taheri (Editor)
Volume 18 Modeling and Simulation of Complex Communication M.A. Niazi (Editor)
Volume 20 SysML for Systems Engineering: A model-based approach, 3rd Edition J.
Holt and S. Perry
Volume 22 Virtual Reality and Light Field Immersive Video Technologies for Real-
World Applications G. Lafruit and M. Tehrani
Volume 23 Data as Infrastructure for Smart Cities L. Suzuki and A. Finkelstein
Volume 24 Ultrascale Computing Systems J. Carretero, E. Jeannot, and A. Zomaya
Volume 25 Big Data-Enabled Internet of Things M. Khan, S. Khan, and A. Zomaya
(Editors)
Volume 26 Handbook of Mathematical Models for Languages and Computation A.
Meduna, P. Horáček, and M. Tomko
Volume 29 Blockchains for Network Security: Principles, technologies and
applications H. Huang, L. Wang, Y. Wu, and K.R. Choo (Editors)
Volume 30 Trustworthy Autonomic Computing T. Eze
Volume 32 Network Classification for Traffic Management: Anomaly detection,
feature selection, clustering and classification Z. Tari, A. Fahad, A. Almalawi,
and X. Yi
Volume 33 Edge Computing: Models, technologies and applications J. Taheri and S.
Deng (Editors)
Volume 34 AI for Emerging Verticals: Human-robot computing, sensing and
networking M.Z. Shakir and N. Ramzan (Editors)
Volume 35 Big Data Recommender Systems Vol 1 & 2 O. Khalid, S.U. Khan, and A.Y.
Zomaya (Editors)
Volume 37 Handbook of Big Data Analytics Vol 1 & 2 V. Ravi and A.K. Cherukuri (Editors)
Volume 39 ReRAM-based Machine Learning H.Y., L. Ni, and S.M.P. Dinakarrao
Volume 40 E-learning Methodologies: Fundamentals, technologies and applications
M. Goyal, R. Krishnamurthi, and D. Yadav (Editors)
Volume 44 Demystifying Graph Data Science: Graph algorithms, analytics methods,
platforms, databases, and use cases P. Raj, C. Surianarayanan, K. Seerangan,
and G. Ghinea (Editors)
Volume 44 Streaming Analytics: Concepts, architectures, platforms, use cases and
applications P. Raj, A. Kumar, V. Garcı́a Dı́az, and N. Muthuraman (Editors)
Volume 54 Intelligent Network Design Driven by Big Data Analytics, IoT, AI and Cloud
Computing S. Kumar, G. Mapp, and K. Cergiz (Editors)
Volume 57 AIoT Technologies and Applications for Smart Environments M. Alazab, M.
Gupta, and S. Ahmed (Editors)
Volume 60 Intelligent Multimedia Technologies for Financial Risk Management:
Trends, tools and applications K. Sood, S. Grima, B. Rawal, B. Balusamy,
E. Özen, and G.G.G. Gan (Editors)
Volume 115 Ground Penetrating Radar: Improving sensing and imaging through
numerical modelling X.L. Travassos, M. F. Pantoja, and N. Ida
Earth Observation Data
Analytics Using Machine
and Deep Learning
Modern tools, applications and challenges

Edited by
Sanjay Garg, Swati Jain, Nitant Dube and
Nebu Varghese

The Institution of Engineering and Technology

Published by The Institution of Engineering and Technology, London, United Kingdom
The Institution of Engineering and Technology is registered as a Charity in England &
Wales (no. 211014) and Scotland (no. SC038698).
† The Institution of Engineering and Technology 2023
First published 2023

This publication is copyright under the Berne Convention and the Universal Copyright
Convention. All rights reserved. Apart from any fair dealing for the purposes of research
or private study, or criticism or review, as permitted under the Copyright, Designs and
Patents Act 1988, this publication may be reproduced, stored or transmitted, in any
form or by any means, only with the prior permission in writing of the publishers, or in
the case of reprographic reproduction in accordance with the terms of licences issued
by the Copyright Licensing Agency. Enquiries concerning reproduction outside those
terms should be sent to the publisher at the undermentioned address:

The Institution of Engineering and Technology

Futures Place
Kings Way, Stevenage
Hertfordshire SG1 2UA, United Kingdom
www.theiet.org

While the authors and publisher believe that the information and guidance given in this
work are correct, all parties must rely upon their own skill and judgement when making
use of them. Neither the author nor publisher assumes any liability to anyone for any
loss or damage caused by any error or omission in the work, whether such an error or
omission is the result of negligence or any other cause. Any and all such liability is
disclaimed.
The moral rights of the author to be identified as author of this work have been
asserted by him in accordance with the Copyright, Designs and Patents Act 1988.

British Library Cataloguing in Publication Data

A catalogue record for this product is available from the British Library

ISBN 978-1-83953-617-5 (hardback)

ISBN 978-1-83953-618-2 (PDF)

Typeset in India by MPS Limited

Printed in the UK by CPI Group (UK) Ltd, Croydon

Cover Image: Digital Earth, conceptual illustration / EDUARD MUZHEVSKYI /

SCIENCE PHOTO LIBRARY via Getty Images
 Contents

About the editors xiii

Foreword xv

1 Introduction 1
Preeti Kathiria, Swati Jain, Kimee Joshi and Nebu Varghese
1.1 Earth observation data 1
1.1.1 Organization 2
1.2 Categories of EO data 2
1.2.1 Passive imaging system 3
1.2.2 Active imaging system 5
1.3 Need of data analytics in EO data 6
1.4 Data analytics methodology 6
1.4.1 Machine learning 7
1.4.2 Deep learning 8
1.5 Data visualization techniques 8
1.5.1 Cartogram map 8
1.5.2 Heat map 9
1.5.3 Choropleth map 9
1.6 Types of inferences from data analytics (application areas) 9
1.6.1 Agriculture 9
1.6.2 Forestry 9
1.6.3 Land cover classification 10
1.6.4 Flooding 12
1.6.5 Maritime 12
1.6.6 Defence and security 13
1.6.7 Wetland 13
1.7 Conclusion 13
References 14

Part I: Clustering and classification of Earth observation data 19

2 Deep learning method for crop classification using remote
sensing data 21
Kavita Bhosle and Vijaya Musande
2.1 Sources of remote sensing data collection 21
2.2 Tools for processing remote sensing data 22
vi Earth observation data analytics using machine and deep learning

2.3 Crop classification using remote sensing data 22

2.3.1 Methods for crop classification 23
2.3.2 Case study 24
2.4 Performance evaluation 27
2.5 Conclusion 29
References 30

3 Using optical images to demarcate fields in L band SAR images for

effective deep learning based crop classification and crop cover
estimation 33
Kimee Joshi, Madhuri Bhavsar, Zunnun Narmawala and Swati Jain
3.1 Introduction 33
3.1.1 Motivation 34
3.1.2 Research contribution 34
3.1.3 Organization 35
3.2 Related work 35
3.3 Proposed methodology 35
3.3.1 SAR image pre-processing and decomposition 38
3.3.2 Edge detection & field extraction 38
3.3.3 Classification using deep learning 39
3.4 Study area 40
3.5 Experimental setting 42
3.5.1 Dataset 1 42
3.5.2 Dataset 2 43
3.6 Experimental result and analysis 43
3.7 Conclusion 47
References 47

4 Leveraging twin networks for land use land cover classification 51

Pranshav Gajjar, Manav Garg, Pooja Shah, Vijay Ukani and Anup Das
4.1 Introduction 51
4.2 Related literature 52
4.3 Methodology 53
4.3.1 Dataset 54
4.3.2 Siamese network 55
4.3.3 Encoders 56
4.4 Results and discussion 58
4.5 Conclusion and future work 61
References 62

5 Exploiting artificial immune networks for enhancing RS image

classification 67
Poonam S. Tiwari, Hina Pande and Shrushti S. Jadawala
5.1 Introduction 68
 Contents vii

5.1.1 The immune system 69

5.1.2 Classification based on the AIS 71
5.2 Data used and study area 71
5.3 Experimental approach 72
5.3.1 Initialization 73
5.3.2 Randomly choose an antigen 74
5.3.3 Select the n highest affinity 74
5.3.4 Clone the n selected Ab’s 74
5.3.5 Allow each Ab’s in clone set 74
5.3.6 Calculate the affinity aff * j 75
5.3.7 Select the highest affinity 75
5.3.8 Decide 75
5.3.9 Replace 76
5.3.10 A stopping criterion 76
5.4 Result 77
5.5 Conclusion 78
References 78

6 Detection and segmentation of aircrafts in UAV images with

a deep learning-based approach 81
Hina Pande, Poonam Seth Tiwari, Parul Dhingra and Shefali Agarwal
6.1 Introduction 81
6.2 Background 83
6.2.1 Digital images and spatial resolution 84
6.2.2 Neural networks 84
6.2.3 CNNs 85
6.3 Methodology 86
6.3.1 Dataset 86
6.3.2 Object detection 87
6.3.3 Semantic segmentation 90
6.4 Model training and results 92
6.4.1 Object detection 92
6.4.2 Semantic segmentation 94
6.5 Conclusions and discussion 96
References 97

Part II: Rare event detection using Earth Observation data 99

7 A transfer learning approach for hurricane damage assessment
using satellite imagery 101
Jayesh Soni, Nagarajan Prabakar and Himanshu Upadhyay
7.1 Introduction 102
7.2 Literature review 102
7.3 Image processing techniques 103
viii Earth observation data analytics using machine and deep learning

7.3.1 Statistical-based algorithms 103

7.3.2 Learning-based algorithms 103
7.4 Transfer learning 106
7.4.1 AlexNet 107
7.5 Implementation 108
7.6 Conclusion 111
References 111

8 Wildfires, volcanoes and climate change monitoring from satellite

images using deep neural networks 115
Yash Kumar Shrivastava and Kavita Jhajharia
8.1 Introduction 117
8.2 Background and related work 121
8.3 Modern DL methods 122
8.3.1 U-Net 123
8.3.2 AlexNet 123
8.3.3 Inception-v3 123
8.3.4 Other neural networks 123
8.4 Benefits of using this approach 124
8.5 Long-term climate change monitoring using DL methods 125
8.6 Other applications of this approach 127
8.7 Possible problems 129
8.8 Conclusion 129
References 130

9 A comparative study on torrential slide shortcoming zones and

causative factors using machine learning techniques: a case study
of an Indian state 133
G. Bhargavi and J. Arunnehru
9.1 Introduction 133
9.2 Discussions on landslide influencing factors 135
9.3 Materials and methods 137
9.4 Dataset collections 137
9.5 Rainfall characteristics in Kerala 137
9.6 Landslide impacted earthquake 140
9.7 Anthropogenic activities 141
9.8 Machine learning techniques for landslide study using
satellite images 142
9.8.1 Highlights of machine learning techniques in
satellite images 142
9.9 Emergency rescue and mitigation 144
9.10 Conclusion 144
References 144
 Contents ix

10 Machine learning paradigm for predicting reservoir property:

an exploratory analysis 149
Saikia Pallabi, Deepankar Nankani and Rashmi Dutta Baruah
10.1 Introduction 149
10.2 Geo-scientific data sources for reservoir characterization 151
10.2.1 Seismic survey 151
10.2.2 Well logging 152
10.3 Research issues and objectives 152
10.4 Description of the case study 153
10.4.1 Geological background of the survey area 154
10.5 ML for reservoir characterization: the proposed approach 154
10.5.1 Well tie 155
10.5.2 Seismic signal reconstruction 155
10.5.3 Smoothing of well log 156
10.5.4 Seismic attributes selection 157
10.5.5 Outlier removal 159
10.6 Experimental results and analysis 160
10.6.1 Statistical data analysis 160
10.6.2 Results and analysis of ML modeling 163
10.6.3 Performance comparison of shallow vs. DNN model 167
10.7 Discussion and future prospects 169
10.8 Conclusion 170
Acknowledgment 170
References 170

Part III: Tools and technologies for Earth Observation data 175
11 The application of R software in water science 177
Nasrin Fathollahzadeh Attar and Mohammad Taghi Sattari
11.1 Introduction 177
11.1.1 What is hydrology? 178
11.1.2 What is computational hydrology? 178
11.1.3 What is hydroinformatics? 178
11.1.4 Free, open-source software (FOSS) 179
11.1.5 What is GitHub? 179
11.2 Material and methods 179
11.2.1 What is R? What is an integrated development
environment (IDE)? 179
11.2.2 What are R packages? 180
11.2.3 What are cheatsheets? 180
11.2.4 What are R communities? 180
11.2.5 What is RPubs? 181
11.2.6 What are popular conferences in R? 181
11.2.7 What is joss (open source software)? 182
x Earth observation data analytics using machine and deep learning

11.2.8 What is R studio cloud? 182

11.2.9 What is R application in hydrology? 182
11.2.10 What are hydrological packages? 183
11.2.11 Workflow of R in hydrology 183
11.2.12 Data for hydrology? How to retrieve datasets? 184
11.2.13 Preprocessing retrieved hydrological data (data tidying) 184
11.2.14 Different hydrology model types? 184
11.2.15 Hydrologic time series analysis tools in R? 189
11.2.16 Hydrological ML application tools in R? 190
11.2.17 Remote sensing tools in R 190
11.3 Conclusion and future prospects 192
References 192

12 Geospatial big data analysis using neural networks 201

Hemi Patel, Jai Prakash Verma, Sapan Mankad, Sanjay Garg,
Pankaj Bodani and Ghansham Sangar
12.1 Introduction 201
12.1.1 Geospatial data 202
12.1.2 Big data analysis 202
12.1.3 Fog computing 202
12.1.4 Neural network 202
12.1.5 Contribution 203
12.2 Related works 204
12.2.1 Big data analysis on geospatial data 205
12.2.2 Data processing techniques in fog environment 206
12.3 Proposed work 206
12.4 Methodology and concepts 207
12.4.1 Data pre-processing on fog environment 207
12.4.2 Prediction on cloud environment using ANN 207
12.5 Results and discussion 208
12.6 Conclusion 210
References 210

13 Software framework for spatiotemporal data analysis and

mining of earth observation data 213
KP Agrawal, Pruthvish Rajput, Shashikant Sharma and Ruchi Sharma
13.1 Introduction 213
13.1.1 Visualization 214
13.1.2 Multidimensional analysis 214
13.1.3 Data mining 214
13.2 Related work 214
13.3 Challenges 215
13.4 The ST-DAME 216
13.4.1 Conceptual architecture of the framework 216
13.4.2 Proposed framework 217
 Contents xi

13.4.3 ST-DAME in action 220

13.5 Result 222
13.5.1 Automated system 223
13.5.2 Customized system 224
13.6 Conclusion 225
References 225

14 Conclusion 227
Sanjay Garg, Kimee Joshi and Nebu Varghese
14.1 Excerpts from various chapters 227
14.2 Issues and challenges 229
14.2.1 Collecting meaningful and real-time data 229
14.2.2 Data storage 230
14.2.3 Resolution; quality promotion 230
14.2.4 Budget limitations 230
14.2.5 Standardization 231
14.2.6 Lack of ground truth data 231
14.2.7 Processing and analysis 231
References 231

Index 233
This page intentionally left blank
 About the editors

Sanjay Garg is a professor in the Department of Computer Science and

Engineering at the Jaypee University of Engineering and Technology, India. His
research interests include data science, algorithms and pattern recognition. He has
over 30 years of academic and research experience. He has completed six funded
research projects sponsored by ISRO under the RESPOND scheme and GUJCOST
as principal investigator in the field of earth observation data analytics and image
processing. He has supervised 10 doctoral dissertations in the same field. He is a
fellow of the Institution of Engineers (India), a senior member of IEEE and a
Senior member of ACM.

Swati Jain is an associate professor in the Department of Computer Science and

Engineering at the Institute of Technology, Nirma University, India. She works in
the areas of machine learning, data analytics, and deep learning. She is currently
working on four funded research projects, two of them under BRNS Department of
Atomic Energy and two by ISRO (Indian Space Research Organization). She has
completed one funded research project under ISRO-RESPOND as a CO-PI. She is
actively working on establishing a Center of Excellence in-Data Science in asso-
ciation with Binghamton University, USA. She received her PhD degree from
Nirma University in the areas of machine learning and image processing.

Nitant Dube is the group director of MOSDAC Research at the joint Space
Applications Centre (SAC) and ISRO, India. His research fields include satellite
image processing, big data analytics, AI/ML and its applications for Earth obser-
vation data, geo-intelligence and web-based processing. He is involved in the
design and development of software for meteorological and oceanographic appli-
cations. He has been responsible for the design and development of data products
and information processing systems for Indian remote sensing satellites and has
contributed towards the development and operationalization of data processing
systems at Indian and International ground stations. He is an ISRO nominated
member for the CEOS Working Group on Information System and Services
(WGISS). He holds a PhD degree in Computer Science from Nirma University,
Ahmedabad (GJ), India.

Nebu Varghese is an assistant manager (GIS) in the Land and Municipal Service at
Dholera Industrial City Development Limited (DICDL), India. He works in the
areas of GIS systems analysis and prepare design for new GIS methodologies, land
xiv Earth observation data analytics using machine and deep learning

use mapping, land cover mapping, urban land use analysis, spatial data manage-
ment, satellite image processing and analysis, machine learning and deep learning.
Currently, he is working on regional and city-level planning projects, where he
employs the most cutting-edge technologies for building information model (BIM)
to GIS Integration with micro-level asset information of all Infrastructure in city
development. He has been involved in various government DST, ISRO, and IIMA
funded projects and was also involved in the innovation hub Malawi project with
the Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ), India. He is a
member of ISPRS. He holds a master’s degree in remote sensing & GIS from Sam
Higginbottom University of Agriculture, Technology & Sciences (SHUATS),
Prayagraj (UP), India.
 Foreword

This compilation of various research outcomes in the form of a book is an excellent

collection of well-written extensive papers on various important and current topics
of application of modern computational technologies in the field of image proces-
sing of remote sensing data. These are outcomes of their own studies and research
on various topics. The topics covered range from Geospatial Big Data Analysis
Using Neural Networks Deep Learning methods for Crop classification, Transfer
Learning Approach for Hurricane Damage Assessment, Wildfires, Volcanoes and
Climate Change, Monitoring Using Deep Neural Networks, Exploiting Artificial
Immune Networks for Enhancing Remote Sensing Image Classification, Mining of
Earth Observation Data, and Software Framework for Spatiotemporal Data
Analysis. The authors have taken a lot of care in reviewing an extensive list of
literature on the topics covered. All these topics are of current intensive research.
Usually, there is a strong urge to bring about drastic changes that are urgently
required to move the world toward a sustainable and resilient course. To attain
these goals, fast computing methods and sophisticated algorithms for pattern
recognition are need of the hour. Various machine learning and deep learning
algorithms have these potentials which are needed to be exploited. Understanding
the scope and implications of this global challenge will require global approaches.
Only global information that is easily accessible from affordable sources, such as
satellite images and other widely available sources, can help us achieve sustainable
goals through their use and ensure their universality. Original and precise methods
for addressing the indicators linked to sustainable development goals are the need
of the hour.
A plethora of Earth-observing satellites gathers enormous amounts of data
from various sources. But today, the challenges in effectively utilizing such data lie
in data access technologies in various data formats, data visualization, and various
data processing methodologies, computational speeds. If the solutions to these
issues can be found, the scientific communities would have faster access to much
better data to support decisions making on weather, climate forecasting and their
impacts, including high-impact weather events, droughts, flooding, wildfires,
ocean/coastal ecosystems, air quality, and many more.
This book is a comprehensive resource that brings together the most recent
research and techniques to cover the theory and practice of Earth Observation data
analytics. The editors and authors have long experience in their fields and have put
their knowledge and expertise in various fields together in seeking solutions to
various problems in earth observation systems.
xvi Earth observation data analytics using machine and deep learning

I strongly recommend this book to all researchers and students interested in

these fields. The chapters in the book will certainly go a long way in enhancing
their knowledge and expertise in the files of their study. Certainly, not least, this
book will be a valuable and timely addition to both their personal collections as
well as their intuitional libraries.

Dr M.B. Potdar
Former Scientist (Indian Space Research Organization)
ISRO/Space Applications Centre, Govt. of India
Former Project Director (Bhaskaracharya National Institute for
Space Applications and Geo-informatics) BISAG, Govt. of Gujarat.
 Chapter 1
Introduction
Preeti Kathiria1, Swati Jain1, Kimee Joshi1 and
Nebu Varghese2

1.1 Earth observation data

Sustainable development and climate change are problems that require immediate
solutions, and both are critical to humanity’s present and future well-being [1,2].
Also, human activities are increasing strain on natural resources, which has a global
impact on the environment. Continuous and ongoing monitoring is required to
analyze, comprehend, and minimize these environmental changes [3]. The United
Nations (UN) seeks a sustainable development model for this generation and future
generations, as well as shared prosperity for people and the planet, through the
promotion of its Sustainable Development Agenda and the United Nations
Framework Convention on Climate Change (UNFCCC) [4,5]. The UN has defined
a set of 17 sustainable development goals (SDGs) as a plan of action to reach peace
and prosperity for all people on our planet by 2030. Several benchmarks and
indicators for each of the 17 goals are used to measure, track, and report the
development of every nation. The global framework established by the UN is
designed around 169 targets, and 232 indicators, 71 (42%) of these targets and 30
(13%) of the indicators can be measured directly or indirectly by Earth observation
(EO) [6–8].
EO plays an essential role in advancing many of the SDGs. Addressing sci-
entific issues like global warming and climate change, ecological change, and
reduction effects of habitat and biodiversity deterioration and producing statistics
and indicators that allow the quantification of SD. The UN report has shown the
viability of using EO data to produce official statistics, including SDGs statistics
like agricultural, urban, and land planning, or food security indicators [9]. Data on
the state of the atmosphere [10], oceans [11], crops [12], forests [13], climate [14],
natural disasters [15], natural resources [16], urbanization [17], biodiversity [18],
and human conditions [19] can be provided by EO. The SDGs that benefit from all
EO indicators are zero hunger (SDG 2), clean water and sanitation (SDG 6),

1
Institute of Technology, Nirma University, India
2
Dholera Industrial City Development Limited, India
2 Earth observation data analytics using machine and deep learning

climate action (SDG 13), life below water (SDG 14), and partnership for the goals
(SDG 17). EOs from satellites and airborne and in situ sensors provide accurate and
reliable information on the state of the atmosphere, oceans, coasts, rivers, soils,
crops, forests, ecosystems, natural resources, ice, snow, and building infrastructure
it changes over time. These observations are directly or indirectly required for all
governmental functions, all economic sectors, and nearly all daily societal activities
[7]. EO satellites make real-time observations of the land, ocean, atmosphere,
cryosphere, and carbon cycle from space, which continuously relay this data to the
ground.

1.1.1 Organization
This paper presents EO data along with various applications. The rest of the paper is
organized as follows: Section 1.2 discusses the categories of the EO data, Section 1.3
describes the need of data analytics, Section 1.4 describes the data analytics metho-
dology, Section 1.5 shows a data visualization techniques, Section 1.6 presents the
application areas, and concluding remarks are in Section 1.7.

1.2 Categories of EO data

This section proposes various types of EO imagery. The process of collecting

observations of the Earth’s surface and atmosphere using remote sensing tools is
known as Earth Observation. The captured data is typically in the form of digital
images [20]. The primary differences are based on the sensing device used to
capture the image (passive or active) and the wavelength of the electromagnetic
spectrum used for the observation. They are broadly classified into two categories
as shown in Figure 1.1: passive imaging system and active imaging system.

Earth
observation data

Passive imaging Active imaging

system system

Optical image Thermal image Lidar Radar

Panchromatic Multispectral RAR SAR

Hyperspectral

Figure 1.1 Overview of an EO data

 Introduction 3

1.2.1 Passive imaging system

Starting with the passive imaging system, passive EO sensors either measure the
thermal IR or microwave radiation emitted from the Earth or measure solar energy
reflected by the Earth’s surface in the visible to the middle infrared region of the
electromagnetic spectrum [21]. Passive sensors that are frequently used in EO will
be introduced in the subsections in the following categories.

1.2.1.1 Optical image

Optical sensors detect and record radiation in the visible, near-infrared, and short-
wave wavelengths for either:
1. One channel (panchromatic)
2. Several channels (multispectral)
3. Numerous channel (hyperspectral)

Panchromatic
Panchromatic scanners typically capture electromagnetic radiation (EMR) in a
single band that includes all wavelengths from the visible to infrared spectrum. A
grey-scale image, which makes pixels with lower image values appear dark and
those with higher values appear bright, is the most common way to display pan-
chromatic data. Concerning optical imagery, panchromatic channels typically
record low values for water and dense vegetation and high values for urban and
bare areas [21]. Panchromatic sensors produce images with higher spatial resolu-
tion than multispectral scanners [20,21].

Multispectral
The creation of “natural colour” images using measurements from three visible
spectrum bands is a typical example of a multispectral image (narrow bands cen-
tered around the blue, green, and red wavelengths) [20]. As the name suggests,
multispectral scanners (MSS) are a specific form of remote sensing equipment that
detects and digitally records radiation in various, defined wavelength areas of the
visible and infrared parts of the electromagnetic spectrum [21]. Multi-spectral
instruments typically have to collect energy on larger spatial extents to “fill” the
imaging detector, resulting in a lower resolution than for panchromatic images
because the range of wavelengths contributing to the radiation energy detected by
the sensor is reduced.
Hyperspectral
Hyperspectral scanners collect image data for hundreds of spectral channels.
Instead of assigning primary colours (red, green, blue) to each pixel, hyperspectral
imaging (HSI) analyzes a broad spectrum of light. To provide more information on
what is imaged, the light striking each pixel is broken down into many different
spectral bands [22].
In Figure 1.2, the images a and b handled by [23,24] are shown here as an
example of a panchromatic image obtained from SPOT satellite with 10 m resolution
and a multispectral image obtained from plants cope with 3.7 m resolution,
4 Earth observation data analytics using machine and deep learning

(a) (b) (c)

Figure 1.2 (a) Panchromatic image. (b) Multispectral image. (c) Hyperspectral image.

Figure 1.3 Thermal infrared image

respectively. In the same figure, image c shows the picture of a portion of Horseshoe
Bay Village in Xiong’ with 0.5 m spatial resolution reported by the Institute of
Remote Sensing and Digital Earth of the Chinese Academy of Sciences and the
Shanghai Institute of Technical Physics of the Chinese Academy of Sciences [25].

1.2.1.2 Thermal image

Thermal infrared radiometers detect energy emitted from the Earth’s surface with
wavelengths ranging from 3 to 15 m [21]. Since the infrared spectrum is not visible to
the human eye, thermal infrared imaging is also known as “non-visible” imaging [26].
With high thermal and spatial resolutions, thermal imaging maps any object’s surface
temperature [27]. The image shown in Figure 1.3 is downloaded from [28,29].
 Introduction 5

1.2.2 Active imaging system

Active remote sensing devices work on the principle of transmitting energy, either
as pulses or as a continuous signal, towards a specific target, then measuring the
energy returned from the target. Radar and lidar are two different subcategories of
active imaging sensors [21].

1.2.2.1 Lidar
Light detection and ranging (lidar) is a technique that uses a transmitted laser pulse
to detect the presence of a target and measures the distance to the target based on
the time and intensity of any reflected return pulse. Lidar is an active remote sen-
sing technology that operates in the ultraviolet to near-infrared wavelength range.
Lidar systems are available on a variety of sensing platforms, including satellite,
airborne, and ground-based systems [21].

1.2.2.2 Radar
RADAR is an acronym for RAdio Detection And Ranging, that is, using actively
transmitted radio waves to detect objects and determine their position or ‘range’. For
EO, the X, C, and L bands are the most commonly used [21]. Compared to infrared and
optical sensing devices, the primary goal of radar is to detect distant targets under
adverse weather conditions and determine their distance, range, and precision. The
radar has a transmitter that serves as an illumination source for target placement. It
generally operates in the microwave region of the electromagnetic spectrum, which is
measured in Hertz [30]. The electrical wave vibrations in the transmitted radar signals
can be constrained to a single plane, that is, perpendicular to the wave propagation
direction (rather than vibrating in all directions perpendicular to that of propagation).
The term “polarization” refers to this filtering procedure. In imaging radar, there are
two orthogonal polarization modes known as horizontal (H) and vertical (V), which are
normally transmitted individually [21]. Two types of radar-based systems are com-
monly used for microwave imaging on aircraft and satellite platforms:
1. Side-Looking Airborne Radar (SLAR) or Real Aperture Radar (RAR)
2. Synthetic Aperture Radar (SAR)
SLAR stands for synthetic aperture radar (Side Looking Airborne Radar). The
illumination of both Real Aperture Radar and Synthetic Aperture Radar is typically
perpendicular to the flight path, making them side-looking systems. The along-
track resolution, also known as the azimuth direction, distinguishes one system
from another. Actual Aperture Radars’ Azimuth resolution is based on the antenna
beam width, which is proportional to the distance between the radar and the target
(slant range). Using a series of signals that have been stored in the system memory,
synthetic aperture radar uses signal processing to create an aperture that is hundreds
of times more significant than the actual antenna [31].
SLAR was the first active sensor to produce terrain imagery from back-
scattered microwave radiation. An antenna is mounted beneath the platform in
SLAR to produce a fan beam (wide vertically and narrow horizontally) pointing to
the platform’s side [32].
6 Earth observation data analytics using machine and deep learning

Figure 1.4 SAR image

Synthetic aperture radars were developed to overcome the limitations of real

aperture radars. For good Azimuth resolution that is independent of the slant range
to the target, these systems combine small antennae with relatively long wave-
lengths [31]. The image shown in Figure 1.4 is downloaded from [33] Copernicus
Open Access Hub.

1.3 Need of data analytics in EO data

EO data analytics has been regarded as a major challenge since the dawn of time.
The amount of data available today is exponentially increasing, but it is only par-
tially structured and harmonized. Understanding the scientific, socioeconomic, and
environmental benefits of earth observation data analytics has become critical for
businesses and users alike [34,35]. The primary goal of data analytics is to uncover
hidden patterns, unknown correlations, and other useful information from large
amounts of heterogeneous data to aid in Earth science research [36].

1.4 Data analytics methodology

Data analysis is the process of transforming raw data into something useful and
comprehensive for a specific purpose. Following preprocessing, the primary goal of
data analytics is to elucidate obscured patterns, unidentified correlations, and other
pertinent information from a sizable volume of heterogeneous data to support Earth
science research [36,37].
 Introduction 7

Table 1.1 Data analytics method

Analytics type Methods

Machine learning Classification [18], Clustering [12], Regression [38],
Dimension reduction [39]
Deep learning Classification [40,41], Object detection [42],
Image segmentation [42]

Table 1.2 Examples of EO satellite applications

Field Main findings Method References

Agriculture The method was successful in mapping Random Forest [43]
the dynamics of crop types in com-
plex landscapes with small field
sizes.
Combining MODIS time series data Random Forest [44]
with machine learning algorithms
allows for the detection of successive
crops.
Land cover Classify RISAT-1 dataset over the CNN [45]
Mumbai region for land Cover clas-
sification.
Maritime The approach was used to detect and Random Forest [46]
map terrestrial oil contamination.
Developed an Oil Spill Detection CNN [47]
framework based on a deep learning
algorithm.
Develop an algorithm for automatic Improved-YOLOv3 [48]
ship detection from SAR and Optical
ship dataset.

To identify the ambiguous and complex relationships between variables and to

better comprehend the geographic distribution and frequency distribution of big
Earth data, traditional statistical methods, which are frequently predicated on spe-
cific assumptions, are frequently used [36]. In terms of non-linear relationship
understanding, machine learning methods generally outperform traditional statis-
tical methods [36]. The involved methods can be categorized as machine learning
and deep learning (Table 1.1).

1.4.1 Machine learning

Machine learning is a branch of computational algorithms that is constantly
developing and aims to mimic human intelligence by learning from the environ-
ment [49]. Machine learning is broadly classified into three subdomains: supervised
learning, unsupervised learning, and reinforcement learning. To summarize,
supervised learning necessitates training on labeled data with inputs and desired
8 Earth observation data analytics using machine and deep learning

outputs. Unsupervised learning, as opposed to supervised learning, does not require

labeled training data, and the environment only provides inputs with no desired
targets. Reinforcement learning allows for learning based on feedback from inter-
actions with the outside world [50].
Machine learning classifiers including Random Forest, Support Vector
Machines, and Maximum likelihood classifiers can produce the probability of an
observation belonging to a specific class of Earth process, such as land use and land
cover classification [51], and crop classification [52,53].

1.4.2 Deep learning

Deep learning techniques, which emerged from machine learning, have unique
capabilities for extracting and presenting features from Earth data at various,
detailed levels. In the classification and segmentation of Earth data tasks, these
features and characteristics are crucial. Deep learning has demonstrated excellent
performance in a variety of fields, including natural language processing, computer
vision, recommendation systems, and others, thanks to its more potent expression
and parameter optimization capabilities [54–56]. For example, the deep convolu-
tional neural networks (CNNs), ResUnet, and DeepUNet can perform satisfying
results in Earthquake-triggered Landslide detection [41,57] classified the different
types of the crop from multi-temporal data. Marmanis et al. [40] classified different
types of the scene using CNN classifier. Beyond image classification, objects can
be detected and segmented using deep learning techniques [42,58].

1.5 Data visualization techniques

A graphic representation of information and data is referred to as data visualization
[59]. The following is a list of different types of map layers or map types that are
mostly used to visualize data [60,61]. Figure 1.5 shows different types of map
[60,61] to visualize the data.

1.5.1 Cartogram map

This choropleth map variant combines a map and a chart. It entails taking a land area
map of a geographical region and segmenting it so that the sizes and/or distances are

(a) (b) (c)

Figure 1.5 (a) Cartogram map [62]. (b) Heat map [63]. (c) Choropleth map [64].
 Introduction 9

proportional to the values of the variable being measured. Then, to correspond to its
corresponding value, each segment is assigned a different colour or shade. As a
result, the data is more directly related to the land area to which it refers [60,61].

1.5.2 Heat map

Heat map uses colors or shades to represent various values or value ranges. A heat map
differs from a choropleth map in that its colors do not correspond to geographic bound-
aries [60]. Instead of presenting these values and ranges as discrete cells constrained by
arbitrary geographic or political boundaries, it presents them as a continuous spectrum.
A heat map can help you more clearly to see patterns of high (“hot spots”) and
low concentrations of a variable in this way. However, since algorithms are fre-
quently used to transform discrete data points into a continuous spectrum, this can
compromise accuracy.

1.5.3 Choropleth map

One more common kind of map is a choropleth map. It is created by first dividing the
area to be mapped into sections, such as by political or geographical boundaries, and
then filling each section with a different color or shade. Referring to [61], every shade
or color represents a different value or range of values that a variable can have [60].

1.6 Types of inferences from data analytics (application

areas)
Figure 1.6 shows the different application areas of Earth Observation data such as
Agriculture, Forestry, Flooding, Land Cover Classification, Maritime, Wetland,
Defence and Security.

1.6.1 Agriculture
For planning and decision-making objectives such as distribution and storage of
food grains, governmental policies, pricing, procurement, and food security, among
others, it is necessary to have access to crop statistics. The Ministry of Agriculture
and Farmers’ Welfare successfully employs modern satellite remote sensing tech-
nology in such decision-making [65].
Satellite imagery has the potential to increase revenue generation for agri-
cultural applications by providing information related to crop type, crop insurance
damage assessment, production management techniques, fertilizer application
requirements, yield estimations, re-growth monitoring, illicit crop monitoring, pest
and invasive species monitoring, and irrigation requirements and application,
monitoring agri-environmental measures (such as acreage) to inform subsidy allo-
cations, field boundary management, crop health mapping, field scale mapping, and
storm damage assessment [66] (see Figure 1.7).

1.6.2 Forestry
Over 1.6 billion people rely on forests for their livelihoods and their sources of
food, medicine, and fuel. Forests cover 31% of the total area of the planet. Provide
10 Earth observation data analytics using machine and deep learning

Figure 1.6 Applications of EO

more information for Obtaining information on forest acreage, stand density, spe-
cies composition, age, and condition to be recognized as a For management pur-
poses, a single unit) surveying, assessing, and keeping track of forest health,
Updating forest management plans: tree cutting, delineation, and tracking of par-
cels, estimating biomass, assessing plant health, and plantation surveillance,
Estimating damage from fire, storms, and other extreme weather, Conservation
area planning and protection, “Conducting fuel analysis and locating the locations
where the Fire risk is high, Deforestation mapping, Monitoring of forest con-
servation and regrowth initiatives [66] (see Figure 1.8).

1.6.3 Land cover classification

Land can be classified, allowing us to learn about its various uses and track its
evolution over time. This is useful for a variety of applications, including mon-
itoring mining and resource extraction, measuring deforestation, land cadastre, and
planning regulations [66].
 Introduction 11

Agriculture

Yield
estimation Crop type

Crop health
monitoring
Precision Regrowth
farming monitoring

Illicit crop
Crop pests monitoring

Figure 1.7 Application areas of agriculture

Burn scar mapping

Fire damage estimation. Risk map,
accurate mapping of burn scar

Forestry Illegal deforestation

Deforestation assessment

Forest stock mapping

Degradation, reforestation,
afforestation, illegal practices

Figure 1.8 Application areas of forestry

12 Earth observation data analytics using machine and deep learning

1.6.4 Flooding
Monitoring soil moisture and water levels, and how they fluctuate over time pro-
vide a solid indicator of how likely it is that flood and drought threats may occur.
Combining high fidelity interferometric synthetic aperture radar (SAR) measure-
ments with ground truth measurements, S-band SAR, medium resolution optical
images, and digital elevation modeling can be used to achieve this. Combining this
with accurate weather forecasts to determine the likelihood and anticipated amount
of rain enables the detection of potential flood and drought concerns [66].

1.6.5 Maritime
Oceans comprise 96.5% of the water on Earth and makeup 70% of its surface. Ship
tracking data from bathymetry is used to create nautical charts and measure beach erosion,
subsidence, and sea levels. Detection of iceberg threats on shipping routes, protecting the
environment in the ocean, oil spill monitoring, detection of an illegal oil discharge,
detection of unlicensed fishing vessels, port surveillance, detection of incoming hostile
objects in maritime piracy, and marine environmental protection [66] (see Figure 1.9).

1.6.5.1 Ship detection

Radar and optical imagery are both useful for tracking maritime activity and detecting
ships. These EO systems’ adaptability will enable the detection, type-classification, and
movement monitoring of ships, boats, and other vessels. In addition to finding ships, it
is also possible to determine other details like speed, heading, and, depending on
resolution, the general class of ship. Law enforcement, including enforcing regulations
governing fishing activities, environmental protection, search and rescue, ship traffic
monitoring, as well as customs and excise activities like preventing illegal smuggling
activities, all make use of ship detection information [66].

Ship detection
Detection of illegal ships, vessels
detection, fisheries monitoring, shipping
traffic

Oil spill monitoring

Detection of oil seepage, new oil
Maritime fields

Hydrology
Flood extent, inland waterways, water
levels, ice floe, iceberg movements,
glacier and lake monitoring

Figure 1.9 Application areas of Maritime

 Introduction 13

1.6.5.2 Oil spill monitoring

In the oceans across the world, many million tonnes of oil are spilled annually. One
of the major ocean pollutants is an oil spill, which poses a serious threat to the
marine environment [47]. Most maritime and coastal nations require the ability to
detect and monitor oil slicks on the sea’s surface to enforce maritime pollution
laws, identify offenders, support cleanup and control efforts, detect oil spills from
far-off pipelines, and detect oil seepage from ocean floors, which could indicate the
presence of new oil fields. SAR photographs can identify oil slicks because of the
way they differ from oil-free regions in appearance [66].

1.6.5.3 Hydrology
A hydrological survey is essential for understanding the coastal zones and inland
waterways of a territory. Timely and reliable assessments of available water
resources via satellites and models provide critical input for developing strategies
and water management in the country [67]. Coastal and tidal zones that change
frequently due to meteorological influences can be of concern. Using satellite
technology, information such as shallow water depths, mudflat topology, and the
presence or absence of outflow or sediments can be derived. Satellite imagery
provides information on changing bathymetry in coastal zones, which is especially
useful around ports and busy shipping areas. The surveys that can be conducted
from space provide a consistent and accurate overview [66].

1.6.6 Defence and security

Imaging capability aids nations seeking to better understand and police their bor-
ders, coastlines, and assets. Monitoring the flow of people and goods into and out of
a country helps policymakers make better decisions.

1.6.7 Wetland
Wetlands are the most productive ecosystems, with a diverse range of flora and
fauna. Wetland conservation is therefore critical for preserving biological diversity.
Wetlands are currently under stress due to biological diversity loss, deteriorating
water quality, sedimentation and shrinkage in the area, infestation by unwanted
weeds, and other factors. Remote sensing data is the primary source of information
for monitoring and mapping large areas, such as wetland extent, distribution, and
wetland types such as freshwater, peat swamps, and non-forested peatlands, among
others [68].

1.7 Conclusion
In this paper, we have presented detailed information on EO data and various
types of EO data. Then, we have listed various data analytics methods used to
represent the data into meaningful manner. Then, we have listed some data
visualization techniques. Finally, this paper presented types of inferences from
data analytics.
14 Earth observation data analytics using machine and deep learning

References
[1] Steffen W, Richardson K, Rockström J, et al. Planetary boundaries: guiding
human development on a changing planet. Science, 2015;347(6223):1259855;
2015.
[2] Biermann F, Bai X, Bondre N, et al. Down to earth: contextualizing the
Anthropocene. Global Environmental Change. 2016;39:341–350.
[3] Giuliani G, Camara G, Killough B, et al. Earth observation open science:
enhancing reproducible science using data cubes. MDPI. 2019;4:147.
[4] Transforming Our World: The 2030 Agenda for Sustainable Development,
2015. Available from: https://undocs. org/en/A/RES/70/71.
[5] Paris Agreement, 2015. Available from: https://unfccc.int/process-and-
meetings/the-paris-agreement/theparis- agreement.
[6] Persello C, Wegner JD, Hänsch R, et al. Deep learning and earth observation
to support the sustainable development goals: current approaches, open
challenges, and future opportunities. IEEE Geoscience and Remote Sensing
Magazine. 2022;10(2):172–200.
[7] Earth Observations in Support of the 2030 Agenda for Sustainable
Development. Available from: https://www.earthobservations.org/docu-
ments/publications/201703_geo_eo_for_2030_agenda.pdf.
[8] Estoque RC. A review of the sustainability concept and the state of SDG
monitoring using remote sensing. Remote Sensing. 2020;12(11):1770.
[9] Ferreira B, Iten M, and Silva RG. Monitoring sustainable development by
means of earth observation data and machine learning: a review.
Environmental Sciences Europe. 2020;32(1):1–17.
[10] Boyte SP, Wylie BK, Howard DM, et al. Estimating carbon and showing
impacts of drought using satellite data in regression-tree models.
International Journal of Remote Sensing. 2018;39(2):374–398.
[11] Poursanidis D, Topouzelis K, and Chrysoulakis N. Mapping coastal marine
habitats and delineating the deep limits of the Neptune’s seagrass meadows
using very high resolution Earth observation data. International Journal of
Remote Sensing. 2018;39(23):8670–8687.
[12] Reza MN, Na IS, Baek SW, et al. Rice yield estimation based on K-means
clustering with graph-cut segmentation using low-altitude UAV images.
Biosystems Engineering. 2019;177:109–121.
[13] Puletti N, Chianucci F, Castaldi C, et al. Use of Sentinel-2 for forest clas-
sification in Mediterranean environments. Annals of Silvicultural Research.
2018;42(1):32–38.
[14] Sathiaraj D, Huang X, and Chen J. Predicting climate types for the
Continental United States using unsupervised clustering techniques.
Environmetrics. 2019;30(4):e2524.
[15] Lizundia-Loiola J, Otón G, Ramo R, et al. A spatio-temporal active-fire
clustering approach for global burned area mapping at 250 m from MODIS
data. Remote Sensing of Environment. 2020;236:111493.
 Introduction 15

[16] Sharma B, Kumar M, Denis DM, et al. Appraisal of river water quality using
open-access earth observation data set: a study of river Ganga at Allahabad
(India). Sustainable Water Resources Management. 2019;5(2):755–765.
[17] Firozjaei MK, Sedighi A, Argany M, et al. A geographical direction-based
approach for capturing the local variation of urban expansion in the appli-
cation of CA-Markov model. Cities. 2019;93:120–135.
[18] Wang L, Dong Q, Yang L, et al. Crop classification based on a novel feature
filtering and enhancement method. Remote Sensing. 2019;11(4):455.
[19] Foody GM, Ling F, Boyd DS, et al. Earth observation and machine learning
to meet Sustainable Development Goal 8.7: mapping sites associated with
slavery from space. Remote Sensing. 2019;11(3):266.
[20] Newcomers Earth Observation Guide. Available from: https://business.esa.
int/newcomers-earth-observation-guide.
[21] Earth observation: Data, processing and applications – CRC for spatial ...
Australia and New Zealand CRC for Spatial Information. Available from:
https://www.crcsi.com.au/assets/Consultancy-Reports-and-Case-Studies/Earth-
Observation-reports-updated-Feb-2019/Vol1-Introduction-low-res-0.8MB.pdf.
[22] Vasefi F, MacKinnon N, and Farkas D. Hyperspectral and multispectral
imaging in dermatology. In: Imaging in Dermatology. New York, NY:
Elsevier; 2016. p. 187–201.
[23] Principles of Remote Sensing and Processing. Available from: https://crisp.
nus.edu.sg/ research/tutorial/opt_int.htm.
[24] Satellite imagery and archive, 2021. Available from: https://www.planet.
com/products/planet-imagery/.
[25] Wei L, Wang K, Lu Q, et al. Crops fine classification in airborne hyper-
spectral imagery based on multi-feature fusion and deep learning. Remote
Sensing. 2021;13(15):2917.
[26] What is Thermal Infrared Imaging, 2022. Available from: https://movitherm.
com/knowledgebase/thermal-infrared-imaging.
[27] Teena M and Manickavasagan A. Thermal infrared imaging. In: Imaging with
Electromagnetic Spectrum. New York, NY: Springer; 2014. p. 147–173.
[28] OpenSnow. Infrared Satellite – Explained. Available from: https://open-
snow.com/news/post/infrared-satellite-explained.
[29] Rap Real-Time Weather. Available from: https://weather.rap.ucar.edu/.
[30] Agarwal T. Radar – Basics, Types, Working, Range Equation & Its
Applications, 2020. Available from: https://www.elprocus.com/radar-basics-
types-and-applications/.
[31] Radar Course 2. Available from: https://earth.esa.int/eogateway/missions/
ers/radar-courses/radar-course-2#: :text=Real
[32] Remote Sensing GIS Applications 3(2+1). Available from: http://ecourse-
sonline.iasri.res.in/mod/page/view.php?id=2068.
[33] Sentinel Datset. Available from: https://scihub.copernicus.eu/dhus/#/home.
[34] Narain A. Data Analytics to Shape the Future of Earth Observation, 2019.
Available from: https://www.geospatialworld.net/blogs/future-of-earth-
observation/.
16 Earth observation data analytics using machine and deep learning

[35] Pesaresi M, Syrris V, and Julea A. A new method for earth observation data
analytics based on symbolic machine learning. Remote Sensing. 2016;8
(5):399.
[36] Yang C, Yu M, Li Y, et al. Big Earth data analytics: a survey. Big Earth
Data. 2019;3(2):83–107.
[37] Big Earth Data Observation. Available from: https://ati.ec.europa.eu/sites/
default/files/2020-06/Big%20Data
[38] Haase D, Jänicke C, and Wellmann T. Front and back yard green analysis
with subpixel vegetation fractions from earth observation data in a city.
Landscape and Urban Planning. 2019;182:44–54.
[39] Dogan T and Uysal KA. The impact of feature selection on urban land cover
classification. International Journal of Intelligent Systems and Applications
in Engineering. 2018;6:59–64.
[40] Marmanis D, Datcu M, Esch T, et al. Deep learning earth observation clas-
sification using ImageNet pretrained networks. IEEE Geoscience and
Remote Sensing Letters. 2015;13(1):105–109.
[41] Yi Y and Zhang W. A new deep-learning-based approach for earthquake-
triggered landslide detection from single-temporal RapidEye satellite
imagery. IEEE Journal of Selected Topics in Applied Earth Observations
and Remote Sensing. 2020;13:6166–6176.
[42] Hoeser T and Kuenzer C. Object detection and image segmentation with
deep learning on earth observation data: a review – Part I: evolution and
recent trends. Remote Sensing. 2020;12(10):1667.
[43] Som-ard J, Immitzer M, Vuolo F, et al. Mapping of crop types in 1989, 1999,
2009 and 2019 to assess major land cover trends of the Udon Thani Province,
Thailand. Computers and Electronics in Agriculture. 2022;198:107083.
[44] Kuchler PC, Simões M, Ferraz R, et al. Monitoring complex integrated
crop–livestock systems at regional scale in Brazil: a big Earth Observation
data approach. Remote Sensing. 2022;14(7):1648.
[45] Memon N, Parikh H, Patel SB, et al. Automatic land cover classification of
multi-resolution dualpol data using convolutional neural network (CNN).
Remote Sensing Applications: Society and Environment. 2021;22:100491.
[46] Löw F, Stieglitz K, and Diemar O. Terrestrial oil spill mapping using
satellite earth observation and machine learning: a case study in South
Sudan. Journal of Environmental Management. 2021;298:113424.
[47] Seydi ST, Hasanlou M, Amani M, et al. Oil spill detection based on multi-
scale multidimensional residual CNN for optical remote sensing imagery.
IEEE Journal of Selected Topics in Applied Earth Observations and Remote
Sensing. 2021;14:10941–10952.
[48] Hong Z, Yang T, Tong X, et al. Multi-scale ship detection from SAR and
optical imagery via a more accurate YOLOv3. IEEE Journal of Selected
Topics in Applied Earth Observations and Remote Sensing. 2021;14:6083–
6101.
[49] El Naqa I and Murphy MJ. What is machine learning? In: Machine Learning
in Radiation Oncology. New York, NY: Springer; 2015. p. 3–11.
 Introduction 17

[50] Qiu J, Wu Q, Ding G, et al. A survey of machine learning for big data
processing. EURASIP Journal on Advances in Signal Processing. 2016;2016
(1):1–16.
[51] Picoli MCA, Camara G, Sanches I, et al. Big earth observation time series
analysis for monitoring Brazilian agriculture. ISPRS Journal of
Photogrammetry and Remote Sensing. 2018;145:328–339.
[52] Löw F, Michel U, Dech S, et al. Impact of feature selection on the accuracy
and spatial uncertainty of per-field crop classification using support vector
machines. ISPRS Journal of Photogrammetry and Remote Sensing.
2013;85:102–119.
[53] Foerster S, Kaden K, Foerster M, et al. Crop type mapping using spectral–
temporal profiles and phenological information. Computers and Electronics
in Agriculture. 2012;89:30–40.
[54] Collobert R and Weston J. A unified architecture for natural language
processing: deep neural networks with multitask learning. In:
Proceedings of the 25th International Conference on Machine Learning,
2008. p. 160–167.
[55] Krizhevsky A, Sutskever I, and Hinton GE. Imagenet classification with
deep convolutional neural networks. Communications of the ACM. 2017;60
(6):84–90.
[56] Schmidhuber J. Deep learning in neural networks: an overview. Neural
Networks. 2015;61:85–117.
[57] Zhong L, Hu L, and Zhou H. Deep learning based multi-temporal crop
classification. Remote Sensing of Environment. 2019;221:430–443.
[58] Hoeser T, Bachofer F, and Kuenzer C. Object detection and image seg-
mentation with deep learning on Earth observation data: a review. Part II:
applications. Remote Sensing. 2020;12(18):3053.
[59] Goyal A. Must known data visualization techniques for Data Science.
Analytics Vidhya. 2021; Available from: https://www.analyticsvidhya.com/
blog/2021/06/must-known-data-visualization-techniques-for-data-science/.
[60] Lesley. 7 techniques to visualize geospatial data – Atlan: Humans of data,
2019. Available from: https://humansofdata.atlan.com/2016/10/7-techni-
ques-to-visualize-geospatial-data/.
[61] 12 Methods for Visualizing Geospatial Data on a Map. Available from:
https://www.safegraph.com/guides/visualizing-geospatial-data.
[62] GISGeography. Cartogram Maps: Data Visualization with Exaggeration,
2022. Available from: https://gisgeography.com/cartogram-maps/.
[63] Geothermal Heat Map. Available from: https://serc.carleton.edu/details/
images/46195.html.
[64] Rice S. Healthcare Scorecard Shows Little Progress Among U.S. States.
Modern Healthcare. 2014; Available from: https://www.modernhealthcare.
com/article/20140430/NEWS/304309964/healthcare-scorecard-shows-little-
progress-among-u-s-states.
[65] Space Applications Center, Indian Space Research Organization, Government
of India. Available from: https://vedas.sac.gov.in/en/.
18 Earth observation data analytics using machine and deep learning

[66] Applications of Earth Observation. Available from: https://www.ukspace.

org/wp-content/uploads/2019/05/The-many-uses-of-Earth-Observation-data.
pdf.
[67] Hydrosphere. Available from: https://www.sac.gov.in/Vyom/hydrosphere.
jsp.
[68] Wetlands. Available from: https://www.isro.gov.in/earth-observation/
wetlands.
Part I
Clustering and classification of Earth
Observation data
This page intentionally left blank
 Chapter 2
Deep learning method for crop classification
using remote sensing data
Kavita Bhosle1 and Vijaya Musande2

The Red Edge Position (REP) index plays an important role in agricultural remote
sensing applications. A wavelength from 350 nm to 990 nm is the common range
for green. In this chapter, we have focused mainly on crop classification using the
deep learning method. We have presented a study of crop classification using deep
learning methods on hyperspectral remote sensing data. Deep learning is the
evolved form of artificial neural network (ANN). It is based on a biological concept
that deals with the network of neurons in a brain. To solve problems regarding crop
classification, many machine-learning methods are used by researchers. Traditional
machine-learning algorithms, including support vector machine, decision tree-
based, and random forest, work on structured data only. Remote sensing data is
unstructured data. Hence more computational overheads are needed to organize the
unstructured data into structured ones. One of the most adaptable state-of-the-art
approaches for feature extraction and classification of unstructured and structured
data is deep learning. Thus we have focused on deep learning convolutional neural
network (CNN) for feature extraction and classification of crops.

2.1 Sources of remote sensing data collection

There are multiple sources available for the collection of remote sensing data.
Sensors are mounted on either satellites or aircraft. Each data has its own spectral
and spatial resolution. Spectral resolution defines intervals between two con-
secutive wavelengths. The finer the spectral resolution, the narrower the wave-
length range for a particular channel or band. There are two types of data collection
techniques used: active and passive. In passive, reflected sunlight is used to mea-
sure radiation. These types of sensors can work in the presence of sunlight. Active
sensors are not dependent on the Sun’s electromagnetic rays. These sensors use
their electromagnetic energy and incident on the earth’s surface. Reflected energy
is collected by active sensors.

1
MIT, India
2
Jawaharlal Nehru Engineering College, MGM University, India
22 Earth observation data analytics using machine and deep learning

Spectral reflectance can also be acquired with an ASD Spectro radiometer that
provides measurements in the spectral range starting from 350 nm to 2,500 nm with
3 nm spectral resolution and a 1 nm sampling step. These experiments can be
carried out in the field or laboratory. Obtained data can be viewed and exported in
ASCII file using View Spec Pro 6.2.

2.2 Tools for processing remote sensing data

Quantum GIS (QGIS) is available at https://qgis.org/en/site/forusers/download.html,
System for Automated Geoscientific Analyses (SAGA) is available at https://saga-gis.
sourceforge.io/en/index.html, Geographic Resources Analysis Support System
(GRASS) is available at https://grass.osgeo.org/, The Integrated Land and Water
Information System (ILWIS) is available at https://www.itc.nl/ilwis/, etc. are the tools
available for processing remote sensing data. Radiometric, geometry and atmospheric
correction is required in GIS. In the process of remote sensing, the atmosphere gives
radiation and is mixed with earth radiation. Atmospheric correction is required to
remove bands created by atmospheric error. Hyperspectral data has many numbers of
bands or dimensions. Selection techniques can be used for dimensionality reduction
[1]. Principal component analysis (PCA) and linear discriminant analysis (LDA) are
transformation techniques that play a significant role in the dimensionality reduction of
hyperspectral images [2]. Strong spectral correlation in hyperspectral photographs
results in redundant information. Therefore, it is necessary to minimise these photo-
graphs’ dimensions. Techniques for selection or transformation can be used to reduce
dimensionality. PCA is a transformation method that is important for reducing the
dimensionality of hyperspectral pictures. From hundreds of hyperspectral bands,
interesting bands have been extracted using kernel PCA.
Hyperspectral data requires dimensionality reduction since it contains a large
number of bands. Prior to applying a machine-learning algorithm to the data, the
dimensionality-reduction approach PCA has been used to convert a high-dimensional
dataset into a smaller-dimensional subspace. There is a tonne of information in the
thousands of bands of hyperspectral data. Using the Eigenvalues and the
Eigenvectors, PCA generates valuable and significant principal components.
As a segment for the experiment, the region of interest (ROI) has been
extracted. For ROI, there is empirical data. In order to obtain useful components
and reduce dimensionality, PCA has been performed. Using these elements and
ground truth, patches are produced. After that, the training and testing datasets are
created from these patches. This data has been subjected to a deep-learning CNN
model. Various CNN parameters have been changed, and the model has been
assessed using precision, recall, f1 score and test accuracy.

2.3 Crop classification using remote sensing data

Monospectral, multispectral, hyperspectral and many more types of remote sensing
data are available. Monospectral has only one band, whereas multispectral has
 Deep learning method for crop classification using remote sensing data 23

10–12 bands which are not contiguous [3]. Hyperspectral data has hundreds of
contiguous spectral bands. Hyperspectral imaging is the collection of hundreds of
continuous connecting spectral bands which can be employed to represent each
pixel [4].

2.3.1 Methods for crop classification

Conventional methods for crop classification are neural networks, random forest
method, support vector machines and many more [5]. Deep learning is the evolved
form of the ANN [6]. It is based on a biological concept that deals with the network
of neurons in a brain. To solve problems regarding crop classification, many
machine-learning methods are used by researchers. Traditional machine-learning
algorithms, including support vector machine, decision tree-based, and random
forest, work on structured data only. Remote sensing data is unstructured data [7].
Hence, more computational overheads are needed to organize the unstructured data
into structured ones. One of the most adaptable states-of-the-art approaches for
feature extraction and classification of unstructured and structured data is deep
learning [8]. Thus, we have presented a study using deep learning CNN for feature
extraction and classification of crops [9]. EO-1 hyperspectral images provide an
appropriate spectral and spatial resolution. This helps to classify the crops much
better than the conventional methods. EO-1 Hyperion data is available at https://
www.usgs.gov/centers/eros/science/earth-observing-1-eo-1. EO-1 Hyperion has
given 242 bands data of 10 nm spectral resolution and 30 m spatial resolution. Only
155 bands are received after performing atmospheric correction. The study area of
(49  69) consists of 3,381 pixels [10]. The advanced version of an ANN is deep
learning. It is based on a biological idea that pertains to the brain’s network of
neurons. Numerous machine-learning techniques are employed by researchers to
address crop classification issues. Support vector machines, decision trees and
random forests are examples of traditional machine-learning algorithms that only
function with structured input. Data from remote sensing is unstructured.
Therefore, converting unstructured data to structured data requires higher com-
puting overhead. Deep learning is one of the most flexible modern methods for
feature extraction and categorization of unstructured and structured data. We,
therefore, concentrated on deep learning CNN for feature extraction and crop
categorization.
The REP index plays an important role in agricultural remote sensing appli-
cations. A wavelength from 350 nm to 990 nm is the common range for green. REP
is different for each crop [11,12]. This value can be used for discrimination of crops
as well as conditions of crops. Crop conditions may be healthy or diseased crops
[13]. The REP index can be calculated using many methods like linear extrapola-
tion method, maximum first derivative and language interpolation. REP value can
be used for crop classification [14,15].
Experiments to assess plant health have been conducted using healthy and
diseased samples of mulberry, cotton and sugarcane plants. For locating REP, four
techniques have been investigated. The simplest method is the maximum
24 Earth observation data analytics using machine and deep learning

derivative. The easiest way to determine REP is linear interpolation, which takes
into account both the maximum and lowest shoulders of the chlorophyll reflectance
curve. For all crops and environmental conditions, the third linear extrapolation has
taken into account the fixed four spots on the chlorophyll reflectance curve. The
two approaches mentioned above are typically utilised when there are two peaks in
the first derivative and a nitrogen-affected REP.
A narrowband reflectance capability that responds to variations in chlorophyll
content is called the REP index. Increasing chlorophyll concentration broadens the
absorption characteristic and shifts the red edge to longer wavelengths. Red edge
location refers to the wavelength with the steepest slope between 690 and 740 nm.
Green vegetation typically has a wavelength between 700 and 730 nm.

2.3.2 Case study

Space-borne hyperspectral remote sensing data (EO-1 Hyperion) dataset has been
used in the present study. This data consists of 0m spatial and 10 nm spectral
resolution. The study area is selected from the Aurangabad District region in
Maharashtra, India. Data have gathered by conducting a campaign in the month of
December 2015. The selected region is segmented using the ENVI tool. This seg-
ment of ROI consists of fields of sugarcane, cotton, and mulberry plantations and
consists of rock as shown in Figure 2.1. This figure has been referred from the
research paper [10].
Labelling of ROI data corresponding to individual pixels which have been
labelled using collected ground truth. The final labelled training and testing dataset
has been formed using this process. Deep learning CNN has been used for feature
extraction and classification [16]. CNN consists of alternate convolutional and max
pooling layer followed by a fully connected neural network [10] as shown in
Figure 2.2.

10

20

30

40

0 10 20 30 40 50 60

Figure 2.1 Study area

 Deep learning method for crop classification using remote sensing data 25

Flattening Output
Layer
Cotton
PCA
Mulberry

Sugarcane

Rock

Hyperspectral
Image Convolutional
Convolutional Layer2 Fully
Layer1 Connected

Figure 2.2 Architecture of CNN

Data have been gathered via campaigns. The chosen terrain includes of rock as
well as sugarcane, cotton, mulberry and a limited quantity of other crop fields with
different plant heights between 0.7 and 1.5 m. Using gathered ground truth data as
well as the latitude and longitude of each pixel, the labelling of hyperspectral data
corresponding to individual pixels has been completed. This method resulted in the
creation of the final labelled training and testing dataset. Experts who obtained data
in the field manually marked various crop varieties. The extracted and labelled
hyperspectral data corresponding to each pixel were then created using the com-
bined ground truth data. The final labelled dataset used for training and testing was
created through this approach. The correctly labelled dataset for crop classification
was created by running a data-gathering effort.
Python experiments have been carried out utilising the Keras library functions
and the TensorFlow environment.

2.3.2.1 Convolutional layers

Consider d  d input hyperspectral segment given input to convolutional layer. We
have given p  p filter w; the output of the convolutional layer will be of the size
(d p + 1)  (d p + 1) [17,18]. In (2.1), after summing up the contributions,
weighted by the filter components from the previous layer cells, adding bias term
and then applying the activation function [19].
Using the softmax function, it has been connected to a dense layer that is
entirely connected. The activation functions of an ANN determine whether or not a
neuron should fire and whether the information it contains is meaningful. The
hidden layers have been implemented using rectified linear units (ReLUs). A rec-
tified linear unit outputs  when  is positive and 0 when  is negative. CNN only
has one channel and uses 2  2 and 3  3 filter sizes. Because more than two
classes or crops need to be detected, the CNN output layer uses the softmax func-
tion. Each unit’s corresponding output, between 0 and 1, has been provided.
Following are the steps that CNN took to train the model: The activation
functions ReLU and 3  3 filter size have been employed, and the learning rate has
26 Earth observation data analytics using machine and deep learning

been held constant at 0.0001. The batch size has been held constant at 32. CNN has
two convolutional layers in its proposed design, which are followed by fully con-
nected layers and output layers, as shown in Figure 2.2.

2.3.2.2 Max pooling layers

The max-pooling layers are k  k box and output a single value which is optimized
for that section. For example, if the input layer is d  d layer, then the output will
be computed as a (d/k)  (d/k) layer, k  k block is compact to presently a single
value through the max function. ReLu and Tanh are the activation function [20].
The output of the max pooling layer has to flatten and then can be given to a fully
connected layer with an activation function [21]. The softmax function is used in
this case in the output layer as class labels are not binary. In this case study, output
layer consists of four nodes of classes, mulberry, cotton, sugarcane and rocks.

2.3.2.3 Convolutional autoencoder

Stack autoencoder and convolutional autoencoder are similar terms (SAE). In many
unsupervised applications, dimensionality reduction is primarily achieved via
autoencoders and neural networks. Additionally, they are employed in image
retouching, image enhancement and image colourisation. Deep learning feature
extraction has been carried out using a multiscale convolutional autoencoder. To
reduce noise and improve the image in this investigation, we used a convolutional
autoencoder. It involves a two-step technique. Encoding and decoding are the first
and second steps, respectively. The input is transformed through encoding into a
latent state space with fewer dimensions. Utilising upsampling techniques, the
decoding process attempts to reconstitute the input. The input of size n  n is
transformed into (n m + 1)  (n m + 1) by the convolutional layer.
Latent state space is minimised by using numerous convolutional layers
autoencoders. The decoder then employs a convolutional layer and several
upsampling layers to obtain the image’s original size. To produce a magnified
image, upsampling repeats the data’s rows and columns by size [0] and size [1],
respectively. CNN models have been contrasted with deep neural network (DNN)
models and convolutional autoencoder models.

2.3.2.4 Deep neural network

A multilayered feed-forward perceptron, which consists of multiple layers of neu-
rons coupled to one another, is one type of neural network that is frequently
employed. In several sectors, remote sensing has been successfully implemented
using ANNs. Deep neural networks are multilayer perceptron (MLP) systems with
deep hidden layers. The complexity of a deep neural network rises with the number
of hidden layers.
In this study, there are four dense layers that make up deep NN. Thirty nodes
and a ReLU activation function are present in each layer. The Deep NN classifier
can be thought of as a parallel computing system made up of a very large number of
basic inter connected processors.
 Deep learning method for crop classification using remote sensing data 27

2.4 Performance evaluation

Deep learning CNN model has been implemented on this data. The model has been
evaluated using precision, recall, F1 score and test accuracy. About 75% of the
samples have been used for training, and 25% have been used for testing. CNN has
been compared with two different supervised classification methods, deep neural
network and convolutional autoencoder [22]. It has been observed that Adam, with
ReLU activation function and filter size of 2  2, has given better classification
accuracy compared to the other methods [23,24]. About 70% data has been used for
training the model, and 30% data has been used for testing purposes.
PCA and CNN experiments are carried out on the Python Tensorflow envir-
onment. The PCA receives as its input unstructured hyperspectral pictures. As these
major components have provided more information, the experiment will only take
into account the first 30 components.
Effective implementation of the predictive models has been possible for crop
classification using deep learning. Deep learning CNN extracts features and then
classified the crops [25–27]. As shown in Table 2.1, the overall accuracy of CNN is
88  2.43; convolutional autoencoder is 85  1.57. DNN has given 78  1.15
accuracy. REP also plays an important role in accuracy [28].
Table 2.2 shows the confusion matrix for the CNN model for four classes;
Cotton, Mulberry, Sugarcane and Rock have a number of samples as 269, 113, 128
and 345, respectively. Overall accuracy is calculated using a confusion matrix. The
confusion matrix shows actual and predicted values for four classes. Using the
following formulae, precision, recall, F1 score and overall accuracy have been
calculated. It has been observed that accuracy using the CNN method is good
compared to convolutional autoencoder and deep NN method. As there are more

Table 2.1 Overall accuracy comparison of CNN with other

methods

% Overall accuracy
CNN Convolutional autoencoder Deep NN
88  2.43 85  1.57 78  1.15

Table 2.2 Confusion matrix for CNN method

Predicted values
Actual values 0 (Cotton) 1 (Mulberry) 2 (Sugarcane) 3 (Rock)
0 (Cotton) TP 0 E 01 E 02 E 03
1 (Mulberry) E 10 TP 1 E 12 E 13
2 (Sugarcane) E 20 E 21 TP 2 E 23
3 (Rock) E 30 E 31 E 32 TP 3
28 Earth observation data analytics using machine and deep learning

continuous spectral bands in the hyperspectral dataset, the signature of each crop is
more obvious. This method of crop classification works well. Because it effectively
handles unstructured data, deep-learning CNN has been chosen. It can auto-
matically extract the information needed for crop detection or categorization. To
obtain an optimum CNN model, a fine-tuning strategy has been applied by
adjusting the values of various parameters.
DNNs and convolutional autoencoder, two distinct supervised classification
techniques, are contrasted with optimised CNN. The model’s performance can be
enhanced by employing precisely calibrated parameters. Data for training and
testing have been created through the use of significant field research. The samples
were taken using a random sampling strategy. In Aurangabad, cotton is regarded as
a significant winter crop. Sugarcane, mulberries and a few other minor crops are
additional important crops. There are rocks all across the study area as well:
Precision 0 = (TP 0) / (TP 0 + E 10 + E 20 + E 30)
Recall 0 = (TP 0) / (TP 0 + E 01 + E 02 + E 03)
F1 Score = (2 * precision * recall) / (precision + recall)
Overall accuracy = number of correctly classified samples/number of test samples
Overall accuracy = (TP 0 + TP 1 + TP 2 + TP 3)/(total test record)
Three techniques: convolutional autoencoder, DNN, and optimized CNN have been
used and compared. The improved CNN method has been found to produce
superior classification outcomes compared to the other approaches. Through the
use of flattened one-dimensional patches, convolutional autoencoders and DNNs
can also extract deep features via deep learning, albeit the performance may be
significantly lowered. Unsupervised learning is used by the convolutional auto-
encoder to acquire feature knowledge. As a result, it cannot properly process the
information on the label.
Our goal was to create a crop classification system that had already been
trained. This study concentrated on the efficient application of crop forecast pre-
dictive models. It also looks at how well the various predictive algorithms can
classify data. The overall classification accuracy on the dataset can be significantly
increased by the optimised CNN. It also functions with a minimal number of
training samples, as was seen with the dataset for the research area.
The current study may be viewed as the initial stage in creating a crop cate-
gorization model that has already been trained. Additionally, it demonstrates that
CNN’s use of the Adam optimiser enhances efficiency even with sparse input. To
improve accuracy on other crops, it is possible to build and test deep learning-based
systems for remote sensing data classification in the future.
The signature of each crop is more distinct in the hyperspectral data set since it
has more bands. Crop identification can be aided by this hyperspectral data feature.
Typically, the confusion matrix is used to test classification accuracy. The out-
comes of crop classification are compared to and examined using the confusion
matrix, which is made up of the actual pixel classification results. The proportion of
correctly classified pixels to all pixels is used to measure classification accuracy
Random documents with unrelated
content Scribd suggests to you:
Photographs of the Cotton Blossom. A book of pressed wild flowers.
Old newspaper clippings.
Julie lingered. Steve crossed the gangplank, turned, beckoned
with his head. Julie lingered. An unspoken question in her eyes.
Andy flushed and scratched the mutton-chop whiskers this side
and that. “Well, you know how she is, Julie. She don’t mean no
harm. But she didn’t let on to Magnolia just what time you were
going. Told her to-morrow, likely. Women folks are funny, that way.
She don’t mean no harm.”
“That’s all right,” said Julie; picked up the valises, was at Steve’s
side. Together the two toiled painfully up the steep river bank, Steve
turning to aid her as best he could. They reached the top of the
levee. They stood a moment, breathless; then turned and trudged
down the dusty Southern country road, the setting sun in their faces.
Julie’s slight figure was bent under the weight of the burden she
carried. You saw Steve’s fine blond head turned toward her, tender,
concerned, encouraging.
Suddenly from the upper deck that fronted Magnolia’s room and
Parthy’s came the sound of screams, a scuffle, a smart slap, feet
clattering pell-mell down the narrow wooden balcony stairs. A wild
little figure in a torn white frock, its face scratched and tear-stained,
its great eyes ablaze in the white face, flew past Andy, across the
gangplank, up the levee, down the road. Behind her, belated and
panting, came Parthy. Her hand on her heart, her bosom heaving,
she leaned against the inadequate support offered by Andy’s right
arm, threatening momentarily to topple him, by her own dead
weight, into the river.
“To think that I should live to see the day when—my own child—
she slapped me—her mother! I saw them out of the window, so I
told her to straighten her bureau drawers—a sight! All of a sudden
she heard that woman’s voice, low as it was, and she to the window.
When she saw her going she makes for the door. I caught her on the
steps, but she was like a wildcat, and raised her hand against me—
her own mother—and tore away, with me holding this in my hand.”
She held out a fragment of torn white stuff. “Raised her hand
against her own——”
Andy grinned. “Good for her.”
“What say, Andy Hawks!”
But Andy refused to answer. His gaze followed the flying little
figure silhouetted against the evening sky at the top of the high river
bank. The slim sagging figure of the woman and the broad-
shouldered figure of the man trudged down the road ahead. The
child’s voice could be heard high and clear, with a note of hysteria in
it. “Julie! Julie! Wait for me! I want to say good-bye! Julie!”
The slender woman in the black dress turned and made as
though to start back and then, with a kind of crazy fear in her pace,
began to run away from the pursuing little figure—away from
something that she had not the courage to face. And when she saw
this Magnolia ran on yet a little while, faltering, and then she
stopped and buried her head in her hands and sobbed. The woman
glanced over her shoulder, fearfully. And at what she saw she
dropped her bags and bundles in the road and started back toward
her, running fleetly in spite of her long ruffled awkward skirts; and
she held out her arms long before they were able to reach her. And
when finally they came together, the woman dropped on her knees
in the dust of the road and gathered the weeping child to her and
held her close, so that as you saw them sharply outlined against the
sunset the black of the woman’s dress and the white of the child’s
frock were as one.
 VIII

M
agnolia, at fifteen, was a gangling gawky child whose
eyes were too big for her face and whose legs were too long
for her skirts. She looked, in fact, all legs, eyes, and elbows.
It was a constant race between her knees and her skirt hems. Parthy
was for ever lengthening frocks. Frequently Magnolia, looking down
at herself, was surprised, like Alice in Wonderland after she had
eaten the magic currant cake, to discover how far away from her
head her feet were. Being possessed of a natural creamy pallor
which her mother mistook for lack of red corpuscles, she was dosed
into chronic biliousness on cod liver oil, cream, eggs, and butter, all
of which she loathed. Then suddenly, at sixteen, legs, elbows, and
eyes assumed their natural proportions. Overnight, seemingly, she
emerged from adolescence a rather amazing looking young creature
with a high broad forehead, a wide mobile mouth, great dark liquid
eyes, and a most lovely speaking voice which nobody noticed. Her
dress was transformed, with Cinderella-like celerity, from the
pinafore to the bustle variety. She was not a beauty. She was, in
fact, considered rather plain by the unnoticing. Being hipless and
almost boyishly flat of bust in a day when the female form was a
thing not only of curves but of loops, she was driven by her mother
into wearing all sorts of pads and ruffled corset covers and
contrivances which somehow failed to conceal the slimness of the
frame beneath. She was, even at sixteen, what might be termed
distinguished-looking. Merely by standing tall, pale, dark-haired, next
to Elly, that plump and pretty ingénue was transformed into a dumpy
and rather dough-faced blonde in whose countenance selfishness
and dissatisfaction were beginning to etch telltale lines.
She had been now almost seven years on the show boat. These
seven years had spread a tapestry of life and colour before her eyes.
Broad rivers flowing to the sea. Little towns perched high on the
river banks or cowering flat and fearful, at the mercy of the waters
that often crept like hungry and devouring monsters, stealthily over
the levee and into the valley below. Singing Negroes. Fighting
whites. Spawning Negroes. A life fantastic, bizarre, peaceful, rowdy,
prim, eventful, calm. On the rivers anything might happen and
everything did. She saw convict chain gangs working on the roads.
Grisly nightmarish figures of striped horror, manacled leg to leg. At
night you heard them singing plantation songs in the fitful glare of
their camp fires in the woods; simple songs full of hope. Didn’t My
Lord Deliver Daniel? they sang. Swing Low Sweet Chariot, Comin’ for
to Carry Me Home. In the Louisiana bayou country she saw the
Negroes perform that weird religious rite known as a ring shout,
semi-savage, hysterical, mesmerizing.
Iowa, Illinois, and Missouri small-town housewives came to be
Magnolia’s friends, and even Parthy’s. The coming of the show boat
was the one flash of blazing colour in the drab routine of their
existence. To them Schultzy was the John Drew of the rivers, Elly the
Lillian Russell. You saw them scudding down the placid tree-shaded
streets in their morning ginghams and calicoes, their bits of silver
clasped in their work-seamed hands, or knotted into the corner of a
handkerchief. Fifty cents for two seats at to-night’s show.
“How are you, Mis’ Hawks? . . . And the little girl? . . . My! Look
at the way she’s shot up in a year’s time! Well, you can’t call her
little girl any more. . . . I brought you a glass of my homemade
damson preserve. I take cup of sugar to cup of juice. Real rich, but it
is good if I do say so. . . . I told Will I was coming to the show every
night you were here, and he could like it or lump it. I been saving
out of the housekeeping money.”
They brought vast chocolate cakes; batches of cookies; jugs of
home-brewed grape wine; loaves of fresh bread; jars of strained
honey; stiff tight bunches of garden flowers. Offerings on the shrine
of Art.
Periodically Parthy threatened to give up this roving life and take
Magnolia with her. She held this as a weapon over Andy’s head when
he crossed her will, or displeased her. Immediately boarding schools,
convents, and seminaries yawned for Magnolia.
Perhaps Parthy was right. “What kind of a life is this for a child!”
she demanded. And later, “A fine kind of a way for a young lady to
be living—slopping up and down these rivers, seeing nothing but
loafers and gamblers and niggers and worse. What about her
Future?” Future, as she pronounced it, was spelled with a capital F
and was a thin disguise for the word husband.
“Future’ll take care of itself,” Andy assured her, blithely.
“If that isn’t just like a man!”
It was inevitable that Magnolia should, sooner or later, find
herself through force of circumstance treading the boards as an
actress in the Cotton Blossom Floating Theatre company. Not only
that, she found herself playing ingénue leads. She had been thrown
in as a stop-gap following Elly’s defection, and had become, quite
without previous planning, a permanent member of the troupe.
Strangely enough, she developed an enormous following, though
she lacked that saccharine quality which river towns had come to
expect in their show-boat ingénues. True, her long legs were a little
lanky beneath the short skirts of the woodman’s pure daughter, but
what she lacked in one extremity she made up in another. They got
full measure when they looked at her eyes, and her voice made the
small-town housewives weep. Yet when their husbands nudged
them, saying, “What you sniffling about?” they could only reply, “I
don’t know.” And no more did they.
Elly was twenty-eight when she deserted Schultzy for a gambler
from Mobile. For three years she had been restless, fault-finding,
dissatisfied. Each autumn she would announce to Captain Andy her
intention to forsake the rivers and bestow her talents ashore. During
the winter she would try to get an engagement through the Chicago
booking offices contrary to the custom of show-boat actors whose
habit it was to hibernate in the winter on the savings of a long and
economical summer. But the Chicago field was sparse and uncertain.
She never had the courage or the imagination to go as far as New
York. April would find her back on the Cotton Blossom. Between her
and Schultzy the bickerings and the quarrels became more and more
frequent. She openly defied Schultzy as he directed rehearsals. She
refused to follow his suggestions though he had a real sense of
direction. Everything she knew he had taught her. She invariably
misread a line and had to be coached in it, word by word; inflection;
business; everything.
Yet now, when Schultzy said, “No! Listen. You been kidnapped
and smuggled on board this rich fella’s yacht, see. And he thinks he’s
got you in his power. He goes to grab you. You’re here, see. Then
you point toward the door back of him, see, like you saw something
there scared the life out of you. He turns around and you grab the
gun off the table, see, and cover him, and there’s your big speech.
So and so and so and so and so and so and so and so——” the ad
lib. directions that have held since the day of Shakespeare.
Elly would deliberately defy him. Others in the company—new
members—began to take their cue from her.
She complained about her wardrobe; refused to interest herself
in it, though she had been an indefatigable needlewoman. Now,
instead of sewing, you saw her looking moodily out across the river,
her hands idle, her brows black. An unintelligent and unresourceful
woman turned moody and thoughtful must come to mischief, for
within herself she finds no solace.
At Mobile, then, she was gone. It was, they all knew, the black-
moustached gambler who had been following the show boat down
the river since they played Paducah, Kentucky. Elly had had dozens
of admirers in her show-boat career; had received much attention
from Southern gallants, gamblers, loafers, adventurers—all the
romantic beaux of the river towns of the ’80s. Her attitude toward
them had been puritanical to the point of sniffiness, though she had
enjoyed their homage and always displayed any amorous missives or
gifts that came her way.
True to the melodramatic tradition of her environment, she left a
note for Schultzy, written in a flourishing Spencerian hand that made
up, in part, for the spelling. She was gone. He need not try to follow
her or find her or bring her back. She was going to star at the head
of her own company and play Camille and even Juliet. He had
promised her. She was good and sick and tired of this everlasting
flopping up and down the rivers. She wouldn’t go back to it, no
matter what. Her successor could have her wardrobe. They had
bookings through Iowa, Illinois, Missouri, and Kansas. She might
even get to New York. (Incredibly enough, she did actually play
Juliet through the Mid-west, to audiences of the bewildered yokelry.)
She was sorry to leave Cap in the lurch like this. And she would
close, and begged to remain his loving Wife (this inked out but still
decipherable)—begged to remain, his truly, Elly Chipley. Just below
this signature the added one of Lenore La Verne, done in
tremendous sable downstrokes and shaded curlecues, especially
about the L’s.
It was a crushing blow for Schultzy, who loved her. Stricken, he
thought only of her happiness. “She can’t get along without me,” he
groaned. Then, in a stunned way, “Juliet!” There was nothing of
bitterness or rancour in his tone; only a dumb despairing wonder.
“Juliet! And she couldn’t play Little Eva without making her out a
slut.” He pondered this a moment. “She’s got it into her head she’s
Bernhardt, or something. . . . Well, she’ll come back.”
“Do you mean to say you’d take her back!” Parthy demanded.
“Why, sure,” Schultzy replied, simply. “She never packed a trunk
in her life, or anything. I done all those things for her. Some ways
she’s a child. I guess that’s how she kept me so tight. She needed
me all the time. . . . Well, she’ll come back.”
Captain Andy sent to Chicago for an ingénue lead. It was then,
pending her arrival, that Magnolia stepped into the breach—the step
being made, incidentally, over what was practically Parthy’s dead
body. For at Magnolia’s calm announcement that she knew every line
of the part and all the business, her mother stormed, had hysterics,
and finally took to her bed (until nearly time for the rise of the
curtain). The bill that night was The Parson’s Bride. Show-boat
companies to this day still tell the story of what happened during
that performance on the Cotton Blossom.
They had two rehearsals, one in the morning, another that lasted
throughout the afternoon. Of the company, Magnolia was the
calmest. Captain Andy seemed to swing, by invisible pulleys, from
the orchestra pit below to Parthy’s chamber above. One moment he
would be sprawled in the kerosene footlights, his eyes deep in
wrinkles of delight, his little brown paws scratching the mutton-chop
whiskers in a frenzy of excitement.
“That’s right. That’s the stuff! Elly never give it half the——’Scuse
me, Schultzy—I didn’t go for to hurt your feelings, but by golly,
Nollie! I wouldn’t of believed you had it in you, not if your own
mother told——” Then, self-reminded, he would cast a fearful glance
over his shoulder, that shoulder would droop, he would extricate
himself from the welter of footlights and music racks and prompt
books in which he squatted, and scamper up the aisle. The dim
outline of a female head in curl papers certainly could not have been
seen peering over the top of the balcony rail as he fancied, for when
he had clattered up the balcony stairs and had gently turned the
knob of the bedroom door, there lay the curl-papered head on the
pillow of the big bed, and from it issued hollow groans, and
plastered over one cheek of it was a large moist white cloth soaked
in some pungent and nostril-pricking stuff. The eyes were closed.
The whole figure was shaken by shivers. Mortal agony, you would
have said (had you not known Parthy), had this stricken and
monumental creature in its horrid clutches.
In a whisper—“Parthy!”
A groan, hollow, heartrending, mortuary.
He entered, shut the door softly, tiptoed over to the bed, laid a
comforting brown paw on the shivering shoulder. The shoulder
became convulsive, the shivers swelled to heaves. “Now, now,
Parthy! What you taking on so for? God A’mighty, person’d think
she’d done something to shame you instead of make you mighty
proud. If you’d see her! Why, say, she’s a born actress.”
 The groans now became a wail. The eyes unclosed. The figure
raised itself to a sitting posture. The sopping rag rolled limply off.
Parthy rocked herself to and fro. “My own daughter! An actress!
That I should have lived to see this day! . . . Rather have . . . in her
grave . . . why I ever allowed her to set foot on this filthy scow . . .”
“Now, Parthy, you’re just working yourself up. Matter of fact, that
time Mis’ Means turned her ankle and we thought she couldn’t step
on it, you was all for going on in her part, and I bet if Sophy Means
hadn’t tied up her foot and gone on like a soldier she is, we’d of had
you acting that night. You was rarin’ to. I watched you.”
“Me! Acting on the stage! Not that I couldn’t play better than any
Sophy Means, and that’s no compliment. A poor stick if I couldn’t.”
But her defence lacked conviction. Andy had surprised a secret
ambition in this iron-armoured bosom.
“Now, come on! Cheer up! Ought to be proud your own daughter
stepping in and saving us money like this. We’d of closed. Had to.
God knows when that new baggage’ll get here, if she gets here at
all. What do you think of that Chipley! Way I’ve treated that girl, if
she’d been my own daughter—well! . . . How’d you like a nice little
sip of whisky, Parthy? Then you come on down give Nollie a hand
with her costumes. Chipley’s stuff comes up on her like ballet skirts.
—Now, now, now! I didn’t say she——Oh, my God!”
Parthy had gone off again into hysterics. “My own daughter! My
little girl!”
The time for severe measures had come. Andy had not dealt with
actresses for years without learning something of the weapons with
which to fight hysteria.
“All right. I’ll give you something to screech for. The boys
paraded this noon with a banner six feet long and red letters a foot
high announcing the Appearance Extraordinaire of Magnolia the
Mysterious Comedy Tragedienne in The Parson’s Bride. I made a
special spiel on the corner. We got the biggest advance sale we had
this season. Yessir! Doc’s downstairs raking it in with both hands and
you had the least bit of gumption in you, instead of laying here
whining and carrying on, you’d——”
“What’s the advance?” spake up Parthy, the box-office expert.
“Three hundred; and not anywheres near four o’clock.”
With one movement Parthy had flung aside the bedclothes and
stepped out of bed revealing, rather inexplicably, a complete lower
costume including shoes.
Andy was off, down the stairs, up the aisle, into the orchestra pit
just in time to hear Magnolia say, “Schultzy, please! Don’t throw me
the line like that, I know it. I didn’t stop because I was stuck.”
“What’d you stop for, then, and look like you’d seen spooks!”
“I stopped a-purpose. She sees her husband that she hates and
that she thought was dead for years come sneaking in, and she
wouldn’t start right in to talk. She’d just stand there, kind of frozen
and stiff, staring at him.”
“All right, if you know so much about directing, go ahead and di
——”
She ran to him, threw her arms about him, hugged him, all
contrition. “Oh, Schultzy, don’t be mad. I didn’t go to boss. I just
wanted to act it like I felt. And I’m awfully sorry about Elly and
everything. I’ll do as you say, only I just can’t help thinking, Schultzy
dear, that she’d stand there, staring kind of silly, almost.”
“You’re right. I guess my mind ain’t on my work. I ought to know
how right you are. I got that letter Elly left for me, I just stood there
gawping with my mouth open, and never said a word for I don’t
know how long——Oh, my God!”
“There, there, Schultzy.”
By a tremendous effort (the mechanics of which were not entirely
concealed) Schultzy, the man, gave way to Harold Westbrook, the
artist.
“You’re right, Magnolia. That’ll get ’em. You standing there like
that, stunned and pale.”
 “How’ll I get pale, Schultzy?”
“You’ll feel pale inside and the audience’ll think you are.” (The
whole art of acting unconsciously expressed by Schultzy.) “Then
Frank here has his sneery speech—so and so and so and so and so
and so—and thought you’d marry the parson, huh? And then you
open up with your big scene—so and so and so and so and so and
so——”
Outwardly calm, Magnolia took only a cup of coffee at dinner,
and Parthy, for once, did not press her to eat. That mournful matron,
though still occasionally shaken by a convulsive shudder, managed
her usual heartening repast and actually spent the time from four to
seven lengthening Elly’s frocks for Magnolia and taking them in to fit
the girl’s slight frame.
Schultzy made her up, and rather overdid it so that, as the
deserted wife and school teacher and, later, as the Parson’s
prospective bride, she looked a pass between a healthy Camille and
Cleopatra just before she applied the asp. In fact, in their effort to
bridge the gap left by Elly’s sudden flight, the entire company
overdid everything and thus brought about the cataclysmic moment
which is theatrical show-boat history.
Magnolia, so sure of her lines during rehearsal, forgot them a
score of times during the performance and, had it not been for
Schultzy, who threw them to her unerringly and swiftly, would have
made a dismal failure of this, her first stage appearance. They were
playing Vidallia, always a good show-boat town. The house was filled
from the balcony boxes to the last row downstairs near the door,
from which point very little could be seen and practically nothing
heard. Something of the undercurrent of excitement which pervaded
the Cotton Blossom troupe seemed to seep through the audience; or
perhaps even an audience so unsophisticated as this could not but
sense the unusual in this performance. Every one of the troupe—
Schultzy, Mis’ Means, Mr. Means, Frank, Ralph, the Soapers
(Character Team that had succeeded Julie and Steve)—all were
trembling for Magnolia. And because they were fearful for her they
threw themselves frantically into their parts. Magnolia, taking her
cue (literally as well as figuratively) from them, did likewise. As
ingénue lead, her part was that of a young school mistress earning
her livelihood in a little town. Deserted some years before by her
worthless husband, she learns now of his death. The town parson
has long been in love with her, and she with him. Now they can
marry. The wedding gown is finished. The guests are invited.
This is her last day as school teacher. She is alone in the empty
schoolroom. Farewell, dear pupils. Farewell, dear schoolroom,
blackboard, erasers, water-bucket, desk, etc. She picks up her key.
But what is this evil face in the doorway! Who is this drunken,
leering tramp, grisly in rags, repulsive—— My God! You! My
husband!
(Never was villain so black and diabolical as Frank. Never was
heroine so lovely and frail and trembling and helpless and white—as
per Schultzy’s directions. As for Schultzy himself, the heroic parson,
very heavily made up and pure yet brave withal, it was a poor stick
of a maiden who wouldn’t have contrived to get into some sort of
distressing circumstance just for the joy of being got out of it by this
godly yet godlike young cleric.)
Frank, then: “I reckon you thought I was dead. Well, I’m about
the livest corpse you ever saw.” A diabolical laugh. “Too damn bad
you won’t be able to wear that new wedding dress.”
Pleadings, agony, despair.
Now his true villainy comes out. A thousand dollars, then, and
quick, or you don’t walk down the aisle to the music of no wedding
march.
“I haven’t got it.”
“No! Where’s the money you been saving all these years?”
“I haven’t a thousand dollars. I swear it.”
“So!” Seizes her. Drags her across the room. Screams. His hand
stifles them.
Unfortunately, in their very desire to help Magnolia, they all
exaggerated their villainy, their heroism, their business. Being a trifle
uncertain of her lines, Magnolia, too, sought to cover her deficiencies
by stressing her emotional scenes. When terror was required her
face was distorted with it. Her screams of fright were real screams of
mortal fear. Her writhings would have wrung pity from a fiend. Frank
bared his teeth, chortled like a maniac. He wound his fingers in her
long black hair and rather justified her outcry. In contrast, Schultzy’s
nobility and purity stood out as crudely and unmistakably as white
against black. Nuances were not for show-boat audiences.
So then, screams, protestations, snarls, ha-ha’s, pleadings,
agony, cruelty, anguish.
Something—intuition—or perhaps a sound from the left upper
box made Frank, the villain, glance up. There, leaning over the box
rail, his face a mask of hatred, his eyes glinting, sat a huge hairy
backwoodsman. And in his hand glittered the barrel of a businesslike
gun. He was taking careful aim. Drama had come late into the life of
this literal mind. He had, in the course of a quick-shooting rough-
and-tumble career, often seen the brutal male mishandling beauty in
distress. His code was simple. One second more and he would act
on it.
Frank’s hand released his struggling victim. Gentleness and love
overspread his features, dispelling their villainy. To Magnolia’s staring
and open-mouthed amazement he made a gesture of abnegation.
“Well, Marge, I ain’t got nothin’ more to say if you and the parson
want to get married.” After which astounding utterance he slunk
rapidly off, leaving the field to what was perhaps the most abject
huddle of heroism that every graced a show-boat stage.
The curtain came down. The audience, intuitively glancing
toward the upper box, ducked, screamed, or swore. The band struck
up. The backwoodsman, a little bewildered but still truculent,
subsided somewhat. A trifle mystified, but labouring under the
impression that this was, perhaps, the ordinary routine of the
theatre, the audience heard Schultzy, in front of the curtain,
explaining that the villain was taken suddenly ill; that the concert
would now be given free of charge; that each and every man,
woman, and child was invited to retain his seat. The backwoodsman,
rather sheepish now, took a huge bite of Honest Scrap and looked
about him belligerently. Out came Mr. Means to do his comic
Chinaman. Order reigned on one side of the footlights at least,
though behind the heaving Venetian lagoon was a company saved
from collapse only by a quite human uncertainty as to whether tears
or laughter would best express their state of mind.
The new ingénue lead, scheduled to meet the Cotton Blossom at
Natchez, failed to appear. Magnolia, following her trial by firearms,
had played the absent Elly’s parts for a week. There seemed to be
no good reason why she should not continue to do so at least until
Captain Andy could engage an ingénue who would join the troupe at
New Orleans.
A year passed. Magnolia was a fixture in the company. Now, as
she, in company with Parthy or Mis’ Means or Mrs. Soaper, appeared
on the front street of this or that little river town, she was stared at
and commented on. Round-eyed little girls, swinging on the front
gate, gazed at her much as she had gazed, not so many years
before, at Elly and Julie as they had sauntered down the shady path
of her own street in Thebes.
She loved the life. She worked hard. She cherished the
admiration and applause. She took her work seriously. Certainly she
did not consider herself an apostle of art. She had no illusions about
herself as an actress. But she did thrive on the warm electric current
that flowed from those river audiences made up of miners, farmers,
Negroes, housewives, harvesters, backwoodsmen, villagers, over the
footlights, to her. A naïve people, they accepted their theatre without
question, like children. That which they saw they believed. They
hissed the villain, applauded the heroine, wept over the plight of the
wronged. The plays were as naïve as the audience. In them,
onrushing engines were cheated of their victims; mill wheels were
stopped in the nick of time; heroes, bound hand and foot and left to
be crushed under iron wheels, were rescued by the switchman’s
ubiquitous daughter. Sheriffs popped up unexpectedly in hidden
caves. The sound of horses’ hoofs could always be heard when
virtue was about to be ravished. They were the minstrels of the
rivers, these players, telling in terms of blood, love, and adventure
the crude saga of a new country.
Frank, the Heavy, promptly fell in love with Magnolia. Parthy,
quick to mark the sheep’s eyes he cast in the direction of the
ingénue lead, watched him with a tigress glare, and though he lived
on the Cotton Blossom, as did Magnolia; saw her all day, daily;
probably was seldom more than a hundred feet removed from her,
he never spoke to her alone and certainly never was able to touch
her except in the very public glare of the footlights with some
hundreds of pairs of eyes turned on the two by the Cotton Blossom
audiences. He lounged disconsolately after her, a large and
somewhat splay-footed fellow whose head was too small for his
shoulders, giving him the look of an inverted exclamation point.
His unrequited and unexpressed passion for Magnolia would have
bothered that young lady and her parents very little were it not for
the fact that his emotions began to influence his art. In his scenes
on the stage with her he became more and more uncertain of his
lines. Not only that, his attitude and tone as villain of the piece took
on a tender note most mystifying to the audience, accustomed to
seeing villainy black, with no half tones. When he should have been
hurling Magnolia into the mill stream or tying her brutally to the
track, or lashing her with a horsewhip or snarling at her like a wolf,
he became a cooing dove. His blows were caresses. His baleful glare
became a simper of adoration.
“Do you intend to speak to that sheep, or shall I?” demanded
Parthy of her husband.
“I’ll do it,” Andy assured her, hurriedly. “Leave him be till we get
to New Orleans. Then, if anything busts, why, I can always get some
kind of a fill-in there.”
They had been playing the Louisiana parishes—little Catholic
settlements between New Orleans and Baton Rouge, their
inhabitants a mixture of French and Creole. Frank had wandered
disconsolately through the miniature cathedral which each little
parish boasted and, returning, had spoken darkly of abandoning the
stage for the Church.
New Orleans meant mail for the Cotton Blossom troupe. With
that mail came trouble. Schultzy, white but determined, approached
Captain Andy, letter in hand.
“I got to go, Cap. She needs me.”
“Go!” squeaked Andy. His squeak was equivalent to a bellow in a
man of ordinary stature. “Go where? What d’you mean, she?” But he
knew.
Out popped Parthy, scenting trouble.
Schultzy held out a letter written on cheap paper, lined, and
smelling faintly of antiseptic. “She’s in the hospital at Little Rock.
Says she’s had an operation. He’s left her, the skunk. She ain’t got a
cent.”
“I’ll take my oath on that,” Parthy put in, pungently.
“You can’t go and leave me flat now, Schultzy.”
“I got to go, I tell you. Frank can play leads till you get
somebody, or till I get back. Old Means can play utility at a pinch,
and Doc can do general business.”
“Frank,” announced Parthy, with terrible distinctness, “will play no
leads in this company, and so I tell you, Hawks.”
“Who says he’s going to! A fine-looking lead he’d make, with that
pin-head of his, and those elephant’s hoofs. . . . Now looka here,
Schultzy. You been a trouper long enough to know you can’t leave a
show in the ditch like this. No real show-boat actor’d do it, and you
know it.”
“Sure I know it. I wouldn’t do it for myself, no matter what. But
it’s her. I wrote her a letter, time she left. I got her bookings. I said if
the time comes you need me, leave me know, and I’ll come. And she
needs me, and she left me know, and I’m coming.”
“How about us!” demanded Parthy. “Leaving us in the lurch like
that, first Elly and now you after all these years. A fine pair, the two
of you.”
 “Now, Parthy!”
“Oh, I’ve no patience with you, Hawks. Always letting people get
the best of you.”
“But I told you,” Schultzy began again, almost tearfully, “it’s for
her, not me. She’s sick. You can pick up somebody here in New
Orleans. I bet there’s a dozen better actors than me laying around
the docks this minute. I got to talking to a fellow while ago, down on
the wharf. The place was all jammed up with freight, and I was
waiting to get by so’s I could come aboard. I said I was an actor on
the Cotton Blossom, and he said he’d acted and that was a life he’d
like.”
“Yes,” snapped Parthy. “I suppose he would. What does he think
this is! A bumboat! Plenty of wharf rats in New Orleans’d like nothing
better——”
Schultzy pointed to where a slim figure leaned indolently against
a huge packing case—one of hundreds of idlers dotting the great
New Orleans plank landing.
Andy adjusted the pair of ancient binoculars through which he
recently had been scanning the wharf and the city beyond the levee.
He surveyed the graceful lounging figure.
“I’d go ashore and talk to him, I was you,” advised Schultzy.
Andy put down the glasses and stared at Schultzy in amazement.
“Him! Why, I couldn’t go up and talk to him about acting on no show
boat. He’s a gentleman.”
“Here,” said Parthy, abruptly, her curiosity piqued. She in turn
trained the glasses on the object of the discussion. Her survey was
brief but ample. “He may be a gentleman. But nobody feels a
gentleman with a crack in his shoe, and he’s got one. I can’t say I
like the looks of him, specially. But with Schultzy playing us this dirty
trick—well, that’s what it amounts to, and there’s no sense trying to
prettify it—we can’t be choosers. I’d just step down talk to him if I
was you, Hawks.”
 IX

T
his, then, turned out to be Magnolia’s first glimpse of
Gaylord Ravenal—an idle elegant figure in garments whose
modish cut and fine material served, at a distance, to conceal
their shabbiness. Leaning moodily against a tall packing case
dumped on the wharf by some freighter, he gazed about him and
tapped indolently the tip of his shining (and cracked) boot with an
exquisite little ivory-topped malacca cane. There was about him an
air of distinction, an atmosphere of richness. On closer proximity you
saw that the broadcloth was shiny, the fine linen of the shirt-front
and cuffs the least bit frayed, the slim boots undeniably split, the hat
(a delicate gray and set a little on one side) soiled as a pale gray hat
must never be. From the Cotton Blossom deck you saw him as the
son, perhaps, of some rich Louisiana planter, idling a moment at the
water’s edge. Waiting, doubtless, for one of the big river packets—
the floating palaces of the Mississippi—to bear him luxuriously away
up the river to his plantation landing.
The truth was that Gaylord Ravenal was what the river gamblers
called broke. Stony, he would have told you. No one had a better
right to use the term than he. Of his two possessions, save the sorry
clothes he had on, one was the little malacca cane. And though he
might part with cuff links, shirt studs and, if necessary, shirt itself, he
would always cling to that little malacca cane, emblem of good
fortune, his mascot. It had turned on him temporarily. Yet his was
the gambler’s superstitious nature. To-morrow the cane would bring
him luck.
Not only was Gaylord Ravenal broke; he had just politely notified
the Chief of Police of New Orleans that he was in town. The call was
not entirely one of social obligation. It had a certain statutory side as
well.
 In the first place, Chief of Police Vallon, in a sudden political
spasm of virtue, endeavouring to clear New Orleans of professional
gamblers, had given them all twenty-four hours’ shrift. In the second
place, this particular visitor would have come under the head of New
Orleans undesirables on his own private account, even though his
profession had been that of philanthropist. Gaylord Ravenal had one
year-old notch to his gun.
It had not been murder in cold blood or in rage, but a shot fired
in self-defence just the fraction of a second before the other man
could turn the trick. The evidence proved this, and Ravenal’s final
vindication followed. But New Orleans gathered her civic skirts about
her and pointed a finger of dismissal toward the door. Hereafter,
should he enter, his first visit must be to the Chief of Police; and
twenty-four hours—no more—must be the limit of his stay in the city
whose pompano and crayfish and Creoles and roses and Ramos gin
fizzes he loved.
The evening before, he had stepped off the river packet Lady
Lee, now to be seen lying alongside the New Orleans landing
together with a hundred other craft. His twenty-four hours would
expire this evening.
Certainly he had not meant to find himself in New Orleans. He
had come aboard the Lady Lee at St. Louis, his finances low, his
hopes high, his erstwhile elegant garments in their present
precarious state. He had planned, following the game of stud poker
in which he immediately immersed himself, to come ashore at
Memphis or, at the latest, Natchez, with his finances raised to the
high level of his hopes. Unfortunately his was an honest and over-
eager game. His sole possession, beside the little slim malacca cane
(itself of small tangible value) was a singularly clear blue-white
diamond ring which he never wore. It was a relic of luckier days
before his broadcloth had become shiny, his linen frayed, his boots
split. He had clung to it, as he had to the cane, through almost
incredible hazards. His feeling about it was neither sentimental nor
superstitious. The tenuous streak of canniness in him told him that,
possessed of a clear white diamond, one can hold up one’s head and
one’s hopes, no matter what the state of coat, linen, boots, and hat.
It had never belonged, fiction-fashion, to his sainted (if any) mother,
nor was it an old Ravenal heirloom. It was a relic of winnings in
luckier days and represented, he knew, potential hundreds. In the
trip that lasted, unexpectedly, from St. Louis to New Orleans, he had
won and lost that ring six times. When the Lady Lee had nosed her
way into the Memphis landing, and again at Natchez, it had been out
of his possession. He had stayed on board, perforce. Half an hour
before coming into New Orleans he had had it again, and had kept
it. The game of stud poker had lasted days, and he rose from it the
richer by exactly nothing at all.
He had glanced out of the Lady Lee’s saloon window, his eyes
bloodshot from sleeplessness, his nerves jangling, his hands
twitching, his face drawn; but that face shaven, those hands
immaculate. Gaylord Ravenal, in luck or out, had the habits and
instincts of a gentleman.
“Good God!” he exclaimed now, “this looks like—it is New
Orleans!” It was N’Yawlins as he said it.
“What did you think it was?” growled one of the players, who had
temporarily owned the diamond several times during the journey
down river. “What did you think it was? Shanghai?”
“I wish it was,” said Gaylord Ravenal. Somewhat dazedly he
walked down the Lady Lee’s gangplank and retorted testily to a
beady-eyed giant-footed gentleman who immediately spoke to him
in a low and not unfriendly tone, “Give me time, can’t you! I haven’t
been twenty-four hours stepping from the gangplank to this wharf,
have I? Well, then!”
“No offence, Gay,” said the gentleman, his eyes still searching the
other passengers as they filed across the narrow gangplank. “Just
thought I’d remind you, case of trouble. You know how Vallon is.”
Vallon had said, briefly, later, “That’s all right, Gay. But by this
time to-morrow evening——” He had eyed Ravenal’s raiment with a
comprehending eye. “Cigar?” The weed he proffered was slim, pale,
and frayed as the man who stood before him. Gaylord Ravenal’s
jangling nerves ached for the solace of tobacco; but he viewed this
palpably second-hand gift with a glance of disdain that was a
triumph of the spirit over the flesh. Certainly no man handicapped
by his present sartorial and social deficiencies was justified in raising
a quizzical right eyebrow in the manner employed by Ravenal.
“What did you call it?” said he now.
Vallon looked at it. He was not a quick-witted gentleman. “Cigar.”
“Optimist.” And strolled out of the chiefs office, swinging the little
malacca cane.
So then, you now saw him leaning moodily against a wooden
case on the New Orleans plank wharf, distinguished, shabby, dapper,
handsome, broke, and twenty-four.
It was with some amusement that he had watched the crew of
the Mollie Able bring the flat unwieldy bulk of the Cotton Blossom
into the wharfside in the midst of the confusion of packets, barges,
steamboats, tugs, flats, tramp boats, shanty boats. He had spoken
briefly and casually to Schultzy while that bearer of evil tidings, letter
in hand, waited impatiently on the dock as the Cotton Blossom was
shifted to a landing position farther upstream. He had seen these
floating theatres of the Mississippi and the Ohio many times, but he
had never before engaged one of their actors in conversation.
“Juvenile lead!” he had exclaimed, unable to hide something of
incredulity in his voice. Schultzy, an anxious eye on the Mollie Able’s
tedious manœuvres, had just made clear to Ravenal his own position
in the Cotton Blossom troupe. Ravenal, surveying the furrowed
brow, the unshaven cheeks, the careless dress, the lack-lustre eye,
had involuntarily allowed to creep into his tone something of the
astonishment he felt.
Schultzy made a little deprecating gesture with his hands, his
shoulders. “I guess I don’t look like no juvenile lead, and that’s a
fact. But I’m all shot to pieces. Took a drink the size of this”—
indicating perhaps five fingers—“up yonder on Canal Street; straight
whisky. No drinking allowed on the show boat. Well, sir, never felt it
no more’n it had been water. I just got news my wife’s sick in the
hospital.”
Ravenal made a little perfunctory sound of sympathy. “In New
Orleans?”
“Little Rock, Arkansas. I’m going. It’s a dirty trick, but I’m going.”
“How do you mean, dirty trick?” Ravenal was mildly interested in
this confiding stranger.
“Leave the show flat like that. I don’t know what they’ll do. I——”
He saw that the Cotton Blossom was now snugly at ease in her new
position, and that her gangplank had again been lowered. He turned
away abruptly, without a good-bye, went perhaps ten paces, came
back five and called to Ravenal. “You ever acted?”
“Acted!”
“On the stage. Acted. Been an actor.”
Ravenal threw back his handsome head and laughed as he would
have thought, ten minutes ago, he never could laugh again. “Me! An
actor! N—” then, suddenly sober, thoughtful even—“Why, yes. Yes.”
And eyeing Schultzy through half-shut lids he tapped the tip of his
shiny shabby boot with the smart little malacca cane. Schultzy was
off again toward the Cotton Blossom.
If Ravenal was aware of the scrutiny to which he was subjected
through the binoculars, he gave no sign as he lounged elegantly on
the wharf watching the busy waterside scene with an air of
indulgent amusement that would have made the onlooker receive
with incredulity the information that the law was even then snapping
at his heels.
Captain Andy Hawks scampered off the Cotton Blossom and
approached this figure, employing none of the finesse that the
situation called for.
“I understand you’ve acted on the stage.”
Gaylord Ravenal elevated the right eyebrow and looked down his
aristocratic nose at the capering little captain. “I am Gaylord
Ravenal, of the Tennessee Ravenals. I failed to catch your name.”
 “Andy Hawks, captain and owner of the Cotton Blossom Floating
Palace Theatre.” He jerked a thumb over his shoulder at the show
boat.
“Ah, yes,” said Ravenal, with polite unenthusiasm. He allowed his
patrician glance to rest idly a moment on the Cotton Blossom, lying
squat and dumpy alongside the landing.
Captain Andy found himself suddenly regretting that he had not
had her painted and overhauled. He clutched his whiskers in
embarrassment, and, under stress of that same emotion, blurted the
wrong thing. “I guess Parthy was mistaken.” The Ravenal eyebrow
became interrogatory. Andy floundered on. “She said that no man
with a crack in the shoe——” he stopped, then, appalled.
Gaylord Ravenal looked down at the footgear under discussion.
He looked up at the grim and ponderous female figure on the
forward deck of the show boat. Parthy was wearing one of her most
uncompromising bonnets and a gown noticeably bunchy even in that
day of unsymmetrical feminine fashions. Black was not becoming to
Mrs. Hawks’ sallow colouring. Lumpy black was fatal. If anything
could have made this figure less attractive than it actually was,
Ravenal’s glance would seem to have done so. “That—ah—lady?”
“My wife,” said Andy. Then, mindful of the maxim of the sheep
and the lamb, he went the whole way. “We’ve lost our juvenile lead.
Fifteen a week and found. Chance to see the world. No
responsibility. Schultzy said you said . . . I said . . . Parthy said . . .”
Hopelessly entangled, he stopped.
“Am I to understand that I am being offered the position of—ah
—juvenile lead on the—” the devastating glance upward—“Cotton
Blossom Floating Palace——”
“That’s the size of it,” interrupted Andy, briskly. After all, even this
young man’s tone and manner could not quite dispel that crack in
the boot. Andy knew that no one wears a split shoe from choice.
“No responsibility,” he repeated. “A chance to see life.”
“I’ve seen it,” in the tone of one who did not care for what he has
beheld. His eyes were on a line with the Cotton Blossom’s deck. His
gaze suddenly became concentrated. A tall slim figure in white had
just appeared on the upper deck, forward—the bit of deck that
looked for all the world like a nautical veranda. It led off Magnolia’s
bedroom. The slim white figure was Magnolia. Preparatory to going
ashore she was taking a look at this romantic city which she always
had loved, and which she, in company with Andy or Doc, had
roamed a dozen times since her first early childhood trip on the
Creole Belle.
Her dress was bunchy, too, as the mode demanded. But where it
was not bunchy it was very tight. And its bunchiness thus only
served to emphasize the slimness of the snug areas. Her black hair
was drawn smoothly away from the temples and into a waterfall at
the back. Her long fine head and throat rose exquisitely above the
little pleated frill that finished the neckline of her gown. She carried
her absurd beribboned and beflowered high-crowned hat in her
hand. A graceful, pliant, slim young figure in white, surveying the
pandemonium that was the New Orleans levee. Columns of black
rose from a hundred steamer stacks. Freight barrels and boxes went
hurtling through the air, or were shoved or carried across the plank
wharf to the accompaniment of shouting and sweating and
swearing. Negroes everywhere. Band boxes, carpet bags, babies,
drays, carriages, wheelbarrows, carts. Beyond the levee rose the old
salt warehouses. Beyond these lay Canal Street. Magnolia was going
into town with her father and her mother. Andy had promised her
supper at Antoine’s and an evening at the old French theatre. She
knew scarcely ten words of French. Andy, if he had known it in his
childhood, had quite forgotten it now. Parthy looked upon it as the
language of sin and the yellow back paper novels. But all three
found enjoyment in the grace and colour and brilliance of the
performance and the audience—both of a sort to be found nowhere
else in the whole country. Andy’s enjoyment was tinged and
heightened by a vague nostalgia; Magnolia’s was that of one artist
for the work of another; Parthy’s was the enjoyment of suspicion.
She always hoped the play’s high scenes were going to be more
risqué than they actually were.
 From her vantage point Magnolia stood glancing alertly about
her, enjoying the babel that was the New Orleans plank wharves.
She now espied and recognized the familiarly capering little figure
below with its right hand scratching the mutton-chop whiskers this
side and that. She was impatient to be starting for their jaunt
ashore. She waved at him with the hand that held the hat. The
upraised arm served to enhance the delicate curve of the pliant
young figure in its sheath of white.
Andy, catching sight of her, waved in return.
“Is that,” inquired Gaylord Ravenal, “a member of your
company?”
Andy’s face softened and glowed. “That? That’s my daughter
Magnolia.”
“Magnolia. Magnol—— Does she—is she a——”
“I should smile she is! She’s our ingénue lead, Magnolia is. Plays
opposite the juvenile lead. But if you’ve been a trouper you know
that, I guess.” A sudden suspicion darted through him. “Say, young
man—what’s your name?—oh, yes, Ravenal. Well, Ravenal, you a
quick study? That’s what I got to know, first off. Because we leave
New Orleans to-night to play the bayous. Bayou Teche to-morrow
night in Tempest and Sunshine. . . . You a quick study?”
“Lightning,” said Gaylord Ravenal.
Five minutes later, bowing over her hand, he did not know
whether to curse the crack in his shoe for shaming him before her,
or to bless it for having been the cause of his being where he was.
That he and Magnolia should become lovers was as inevitable as
the cosmic course. Certainly some force greater than human must
have been at work on it, for it overcame even Parthy’s opposition.
Everything conspired to bring the two together, including their being
kept forcibly apart. Himself a picturesque, mysterious, and romantic
figure, Gaylord Ravenal, immediately after joining the Cotton
Blossom troupe, became the centre of a series of dramatic episodes
any one of which would have made him glamorous in Magnolia’s
eyes, even though he had not already assumed for her the glory of a
Galahad.
She had never before met a man of Ravenal’s stamp. In this
dingy motley company he moved aloof, remote, yet irresistibly
attracting all of them—except Parthy. She, too, must have felt drawn
to this charming and magnetic man, but she fought the attraction
with all the strength of her powerful and vindictive nature. Sensing
that here lay his bitterest opposition, Ravenal deliberately set about
exercising his charm to win Parthy to friendliness. For the first time
in his life he received rebuff so bristling, so unmistakable, as to
cause him temporarily to doubt his own gifts.
Women had always adored Gaylord Ravenal. He was not a villain.
He was, in fact, rather gentle, and more than a little weak. His
method, coupled with strong personal attractiveness, was simple in
the extreme. He made love to all women and demanded nothing of
them. Swept off their feet, they waited, trembling deliciously, for the
final attack. At its failure to materialize they looked up, wondering,
to see his handsome face made more handsome by a certain wistful
sadness. At that their hearts melted within them. That which they
had meant to defend they now offered. For the rest, his was a
paradoxical nature. A courtliness of manner, contradicted by a bluff
boyishness. A certain shy boldness. He was not an especially
intelligent man. He had no need to be. His upturned glance at a
dining-room waitress bent over him was in no way different from
that which he directed straight at Parthy now; or at the daughter of
a prosperous Southern lawyer, or at that daughter’s vaguely uneasy
mama. It wasn’t deliberate evil in him or lack of fastidiousness. He
was helpless to do otherwise.
Certainly he had never meant to remain a member of this motley
troupe, drifting up and down the rivers. He had not, for that matter,
meant to fall in love with Magnolia, much less marry her. Propinquity
and opposition, either of which usually is sufficient to fan the flame,
together caused the final conflagration. For weeks after he came on
board, he literally never spoke to Magnolia alone. Parthy attended to
that. He saw her not only daily but almost hourly. He considered
himself lucky to be deft enough to say, “Lovely day, isn’t it, Miss
Magn——” before Mrs. Hawks swept her offspring out of earshot.
Parthy was wise enough to see that this handsome, graceful,
insidious young stranger would appear desirable and romantic in the
eyes of women a hundredfold more sophisticated than the childlike
and unawakened Magnolia. She took refuge in the knowledge that
this dangerous male was the most impermanent of additions to the
Cotton Blossom troupe. His connection with them would end on
Schultzy’s return.
Gaylord Ravenal was, in the meantime, a vastly amused and
prodigiously busy young man. To learn the juvenile leads in the plays
that made up the Cotton Blossom troupe’s repertoire was no light
matter. Not only must he memorize lines, business, and cues of the
regular bills—Uncle Tom’s Cabin, East Lynne, Tempest and Sunshine,
Lady Audley’s Secret, The Parson’s Bride, The Gambler, and others—
but he must be ready to go on in the concert after-piece, whatever it
might be—sometimes A Dollar for a Kiss, sometimes Red Hot Coffee.
The company rehearsed day and night; during the day they
rehearsed that night’s play; after the performance they rehearsed
next night’s bill. With some astonishment the Cotton Blossom troupe
realized, at the end of two weeks, that Gaylord Ravenal was acting
as director. It had come about naturally and inevitably. Ravenal had
a definite theatre sense—a feeling for tempo, rhythm, line, grouping,
inflection, characterization—any, or all, of these. The atmosphere
had freshness for him; he was interested; he wished to impress
Andy and Parthy and Magnolia; he considered the whole business a
gay adventure; and an amusing interlude. For a month they played
the bayous and plantations of Louisiana, leaving behind them a
whole countryside whose planters, villagers, Negroes had been
startled out of their Southern lethargy. These had known show boats
and show-boat performances all their lives. They had been visited by
this or that raffish, dingy, slap-dash, or decent and painstaking
troupe. The Cotton Blossom company had the reputation for being
the last-named variety, and always were patronized accordingly. The
plays seldom varied. The performance was, usually, less than
mediocre. They were, then, quite unprepared for the entertainment
given them by the two handsome, passionate, and dramatic young
people who now were cast as ingénue and juvenile lead of the
Cotton Blossom Floating Palace Theatre company. Here was Gaylord
Ravenal, fresh, young, personable, aristocratic, romantic of aspect.
Here was Magnolia, slim, girlish, ardent, electric, lovely. Their make-
believe adventures as they lived them on the stage became real;
their dangers and misfortunes set the natives to trembling; their
love-making was a fragrant and exquisite thing. News of this troupe
seeped through from plantation to plantation, from bayou to bayou,
from settlement to settlement, in some mysterious underground
way. The Cotton Blossom did a record-breaking business in a region
that had never been markedly profitable. Andy was jubilant, Parthy
apprehensive, Magnolia starry-eyed, tremulous, glowing. Her lips
seemed to take on a riper curve. Her skin was, somehow, softly
radiant as though lighted by an inner glow, as Julie’s amber
colouring, in the years gone by, had seemed to deepen into golden
brilliance. Her eyes were enormous, luminous. The gangling,
hobbledehoy, sallow girl of seventeen was a woman of eighteen,
lovely, and in love.
Back again in New Orleans there was a letter from Schultzy, a
pathetic scrawl; illiterate; loyal. Elly was out of the hospital, but
weak and helpless. He had a job, temporarily, whose nature he did
not indicate. (“Porter in a Little Rock saloon, I’ll be bound,” ventured
Parthy, shrewdly, “rubbing up the brass and the cuspidors.”) He had
met a man who ran a rag-front carnival company. He could use them
for one attraction called The Old Plantation; or, The South Before the
War. They were booked through the Middle West. In a few weeks, if
Elly was stronger . . .
He said nothing about money. He said nothing of their possible
return to the Cotton Blossom. That, Andy knew, was because of Elly.
Unknown to Parthy, he sent Schultzy two hundred dollars. Schultzy
never returned to the rivers. It was, after all, oddly enough, Elly
who, many many years later, completed the circle which brought her
again to the show boat.
 Together, Andy, Parthy, and Doc went into consultation. They
must keep Ravenal. But Ravenal obviously was not of the stuff of
show-boat actors. He had made it plain, when first he came aboard,
that he was the most impermanent of troupers; that his connection
with the Cotton Blossom would continue, at the latest, only until
Schultzy’s return. He meant to leave them, not at New Orleans, as
they had at first feared, but at Natchez, on the up trip.
“Don’t tell him Schultzy ain’t coming back,” Doc offered,
brilliantly.
“Have to know it some time,” was Andy’s obvious reply.
“Person’d think,” said Parthy, “he was the only juvenile lead left in
the world. Matter of fact, I can’t see where he’s such great shakes of
an actor. Rolls those eyes of his a good deal, and talks deep-voiced,
but he’s got hands white’s a woman’s and fusses with his nails. I’ll
wager if you ask around in New Orleans you’ll find something queer,
for all he talks so high about being a Ravenal of Tennessee and his
folks governors in the old days, and slabs about ’em in the church,
and what not. Shifty, that’s what he is. Mark my words.”
“Best juvenile lead ever played the rivers. And I never heard that
having clean finger nails hurt an actor any.”
“Oh, it isn’t just clean finger nails,” snapped Parthy. “It’s
everything.”
“Wouldn’t hold that against him, either,” roared Doc. The two
men then infuriated the humourless Mrs. Hawks by indulging in a
great deal of guffawing and knee-slapping.
“That’s right, Hawks. Laugh at your own wife. And you, too,
Doc.”
“You ain’t my wife,” retorted Doc, with the privilege of sixty-odd.
And roared again.
The gossamer thread that leashed Parthy’s temper dissolved
now. “I can’t bear the sight of him. Palavering and soft-soaping.
Thinks he can get round a woman my age. Well, I’m worth a dozen
of him when it comes to smart.” She leaned closer to Andy, her face
actually drawn with fear and a sort of jealousy. “He looks at
Magnolia, I tell you.”
“A fool if he didn’t.”
“Andy Hawks, you mean to tell me you’d sit there and see your
own daughter married to a worthless tramp of a wharf rat, or worse,
that hadn’t a shirt to his back when you picked him up!”
“Oh, God A’mighty, woman, can’t a man look at a girl without
having to marry her!”
“Having to marry her, Captain Hawks! Having——Well, what can
a body expect when her own husband talks like that, and before
strangers, too. Having——!”
Doc rubbed his leathery chin a trifle ruefully. “Stretching a point,
Mrs. Hawks, ma’am, calling me a stranger, ain’t you?”
“All right. Keep him with the show, you two. Who warned you
about that yellow-skinned Julie! And what happened! If sheriffs is
what you want, I’ll wager you could get them fast enough if you
spoke his name in certain parts of this country. Wait till we get back
to New Orleans. I intend to do some asking around, and so does
Frank.”
“What’s Frank got to do with it?”
But at this final exhibition of male obtuseness Parthy flounced
out of the conference.
On their return from the bayous the Cotton Blossom lay idle a
day at the New Orleans landing. Early on the morning of their arrival
Gaylord Ravenal went ashore. On his stepping off the gangplank he
spoke briefly to that same gimlet-eyed gentleman who was still
loitering on the wharf. To the observer, the greeting between them
seemed amiable enough.
“Back again, Gay!” he of the keen gaze had exclaimed. “Seems
like you can’t keep away from the scene of the——”
“Oh, go to hell,” said Ravenal.
He returned to the Cotton Blossom at three o’clock. At his
appearance the idler who had accosted him (and who was still
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebooknice.com