Deep Learning for the Earth Sciences: A Comprehensive Approach to Remote Sensing, Climate Science and Geosciences

✍ Scribed by Gustau Camps-Valls (editor), Devis Tuia (editor), Xiao Xiang Zhu (editor), Markus Reichstein (editor)

• Publisher: Wiley
• Year: 2021
• Tongue: English
• Leaves: 435
• Edition: 1
• Category: Library

✦ Synopsis

DEEP LEARNING FOR THE EARTH SCIENCES

Explore this insightful treatment of deep learning in the field of earth sciences, from four leading voices

Deep learning is a fundamental technique in modern Artificial Intelligence and is being applied to disciplines across the scientific spectrum; earth science is no exception. Yet, the link between deep learning and Earth sciences has only recently entered academic curricula and thus has not yet proliferated. Deep Learning for the Earth Sciences delivers a unique perspective and treatment of the concepts, skills, and practices necessary to quickly become familiar with the application of deep learning techniques to the Earth sciences. The book prepares readers to be ready to use the technologies and principles described in their own research.

The distinguished editors have also included resources that explain and provide new ideas and recommendations for new research especially useful to those involved in advanced research education or those seeking PhD thesis orientations. Readers will also benefit from the inclusion of:

• An introduction to deep learning for classification purposes, including advances in image segmentation and encoding priors, anomaly detection and target detection, and domain adaptation
• An exploration of learning representations and unsupervised deep learning, including deep learning image fusion, image retrieval, and matching and co-registration
• Practical discussions of regression, fitting, parameter retrieval, forecasting and interpolation
• An examination of physics-aware deep learning models, including emulation of complex codes and model parametrizations

Perfect for PhD students and researchers in the fields of geosciences, image processing, remote sensing, electrical engineering and computer science, and machine learning, Deep Learning for the Earth Sciences will also earn a place in the libraries of machine learning and pattern recognition researchers, engineers, and scientists.

✦ Table of Contents

Cover
Title Page
Copyright
Contents
Foreword
Acknowledgments
List of Contributors
List of Acronyms
Chapter 1 Introduction
1.1 A Taxonomy of Deep Learning Approaches
1.2 Deep Learning in Remote Sensing
1.3 Deep Learning in Geosciences and Climate
1.4 Book Structure and Roadmap
Part I Deep Learning to Extract Information from Remote Sensing Images
Chapter 2 Learning Unsupervised Feature Representations of Remote Sensing Data with Sparse Convolutional Networks
2.1 Introduction
2.2 Sparse Unsupervised Convolutional Networks
2.2.1 Sparsity as the Guiding Criterion
2.2.2 The EPLS Algorithm
2.2.3 Remarks
2.3 Applications
2.3.1 Hyperspectral Image Classification
2.3.2 Multisensor Image Fusion
2.4 Conclusions
Chapter 3 Generative Adversarial Networks in the Geosciences
3.1 Introduction
3.2 Generative Adversarial Networks
3.2.1 Unsupervised GANs
3.2.2 Conditional GANs
3.2.3 Cycle‐consistent GANs
3.3 GANs in Remote Sensing and Geosciences
3.3.1 GANs in Earth Observation
3.3.2 Conditional GANs in Earth Observation
3.3.3 CycleGANs in Earth Observation
3.4 Applications of GANs in Earth Observation
3.4.1 Domain Adaptation Across Satellites
3.4.2 Learning to Emulate Earth Systems from Observations
3.5 Conclusions and Perspectives
Chapter 4 Deep Self‐taught Learning in Remote Sensing
4.1 Introduction
4.2 Sparse Representation
4.2.1 Dictionary Learning
4.2.2 Self‐taught Learning
4.3 Deep Self‐taught Learning
4.3.1 Application Example
4.3.2 Relation to Deep Neural Networks
4.4 Conclusion
Chapter 5 Deep Learning‐based Semantic Segmentation in Remote Sensing
5.1 Introduction
5.2 Literature Review
5.3 Basics on Deep Semantic Segmentation: Computer Vision Models
5.3.1 Architectures for Image Data
5.3.2 Architectures for Point‐clouds
5.4 Selected Examples
5.4.1 Encoding Invariances to Train Smaller Models: The example of Rotation
5.4.2 Processing 3D Point Clouds as a Bundle of Images: SnapNet
5.4.3 Lake Ice Detection from Earth and from Space
5.5 Concluding Remarks
Chapter 6 Object Detection in Remote Sensing
6.1 Introduction
6.1.1 Problem Description
6.1.2 Problem Settings of Object Detection
6.1.3 Object Representation in Remote Sensing
6.1.4 Evaluation Metrics
6.1.4.1 Precision‐Recall Curve
6.1.4.2 Average Precision and Mean Average Precision
6.1.5 Applications
6.2 Preliminaries on Object Detection with Deep Models
6.2.1 Two‐stage Algorithms
6.2.1.1 R‐CNNs
6.2.1.2 R‐FCN
6.2.2 One‐stage Algorithms
6.2.2.1 YOLO
6.2.2.2 SSD
6.3 Object Detection in Optical RS Images
6.3.1 Related Works
6.3.1.1 Scale Variance
6.3.1.2 Orientation Variance
6.3.1.3 Oriented Object Detection
6.3.1.4 Detecting in Large‐size Images
6.3.2 Datasets and Benchmark
6.3.2.1 DOTA
6.3.2.2 VisDrone
6.3.2.3 DIOR
6.3.2.4 xView
6.3.3 Two Representative Object Detectors in Optical RS Images
6.3.3.1 Mask OBB
6.3.3.2 RoI Transformer
6.4 Object Detection in SAR Images
6.4.1 Challenges of Detection in SAR Images
6.4.2 Related Works
6.4.3 Datasets and Benchmarks
6.5 Conclusion
Chapter 7 Deep Domain Adaptation in Earth Observation
7.1 Introduction
7.2 Families of Methodologies
7.3 Selected Examples
7.3.1 Adapting the Inner Representation
7.3.2 Adapting the Inputs Distribution
7.3.3 Using (few, well‐chosen) Labels from the Target Domain
7.4 Concluding Remarks
Chapter 8 Recurrent Neural Networks and the Temporal Component
8.1 Recurrent Neural Networks
8.1.1 Training RNNs
8.1.1.1 Exploding and Vanishing Gradients
8.1.1.2 Circumventing Exploding and Vanishing Gradients
8.2 Gated Variants of RNNs
8.2.1 Long Short‐term Memory Networks
8.2.1.1 The Cell State ct and the Hidden State ht
8.2.1.2 The Forget Gate ft
8.2.1.3 The Modulation Gate vt and the Input Gate it
8.2.1.4 The Output Gate ot
8.2.1.5 Training LSTM Networks
8.2.2 Other Gated Variants
8.3 Representative Capabilities of Recurrent Networks
8.3.1 Recurrent Neural Network Topologies
8.3.2 Experiments
8.4 Application in Earth Sciences
8.5 Conclusion
Chapter 9 Deep Learning for Image Matching and Co‐registration
9.1 Introduction
9.2 Literature Review
9.2.1 Classical Approaches
9.2.2 Deep Learning Techniques for Image Matching
9.2.3 Deep Learning Techniques for Image Registration
9.3 Image Registration with Deep Learning
9.3.1 2D Linear and Deformable Transformer
9.3.2 Network Architectures
9.3.3 Optimization Strategy
9.3.4 Dataset and Implementation Details
9.3.5 Experimental Results
9.4 Conclusion and Future Research
9.4.1 Challenges and Opportunities
9.4.1.1 Dataset with Annotations
9.4.1.2 Dimensionality of Data
9.4.1.3 Multitemporal Datasets
9.4.1.4 Robustness to Changed Areas
Chapter 10 Multisource Remote Sensing Image Fusion
10.1 Introduction
10.2 Pansharpening
10.2.1 Survey of Pansharpening Methods Employing Deep Learning
10.2.2 Experimental Results
10.2.2.1 Experimental Design
10.2.2.2 Visual and Quantitative Comparison in Pansharpening
10.3 Multiband Image Fusion
10.3.1 Supervised Deep Learning‐based Approaches
10.3.2 Unsupervised Deep Learning‐based Approaches
10.3.3 Experimental Results
10.3.3.1 Comparison Methods and Evaluation Measures
10.3.3.2 Dataset and Experimental Setting
10.3.3.3 Quantitative Comparison and Visual Results
10.4 Conclusion and Outlook
Chapter 11 Deep Learning for Image Search and Retrieval in Large Remote Sensing Archives
11.1 Introduction
11.2 Deep Learning for RS CBIR
11.3 Scalable RS CBIR Based on Deep Hashing
11.4 Discussion and Conclusion
Acknowledgement
Part II Making a Difference in the Geosciences With Deep Learning
Chapter 12 Deep Learning for Detecting Extreme Weather Patterns
12.1 Scientific Motivation
12.2 Tropical Cyclone and Atmospheric River Classification
12.2.1 Methods
12.2.2 Network Architecture
12.2.3 Results
12.3 Detection of Fronts
12.3.1 Analytical Approach
12.3.2 Dataset
12.3.3 Results
12.3.4 Limitations
12.4 Semi‐supervised Classification and Localization of Extreme Events
12.4.1 Applications of Semi‐supervised Learning in Climate Modeling
12.4.1.1 Supervised Architecture
12.4.1.2 Semi‐supervised Architecture
12.4.2 Results
12.4.2.1 Frame‐wise Reconstruction
12.4.2.2 Results and Discussion
12.5 Detecting Atmospheric Rivers and Tropical Cyclones Through Segmentation Methods
12.5.1 Modeling Approach
12.5.1.1 Segmentation Architecture
12.5.1.2 Climate Dataset and Labels
12.5.2 Architecture Innovations: Weighted Loss and Modified Network
12.5.3 Results
12.6 Challenges and Implications for the Future
12.7 Conclusions
Chapter 13 Spatio‐temporal Autoencoders in Weather and Climate Research
13.1 Introduction
13.2 Autoencoders
13.2.1 A Brief History of Autoencoders
13.2.2 Archetypes of Autoencoders
13.2.3 Variational Autoencoders (VAE)
13.2.4 Comparison Between Autoencoders and Classical Methods
13.3 Applications
13.3.1 Use of the Latent Space
13.3.1.1 Reduction of Dimensionality for the Understanding of the System Dynamics and its Interactions
13.3.1.2 Dimensionality Reduction for Feature Extraction and Prediction
13.3.2 Use of the Decoder
13.3.2.1 As a Random Sample Generator
13.3.2.2 Anomaly Detection
13.3.2.3 Use of a Denoising Autoencoder (DAE) Decoder
13.4 Conclusions and Outlook
Chapter 14 Deep Learning to Improve Weather Predictions
14.1 Numerical Weather Prediction
14.2 How Will Machine Learning Enhance Weather Predictions?
14.3 Machine Learning Across the Workflow of Weather Prediction
14.4 Challenges for the Application of ML in Weather Forecasts
14.5 The Way Forward
Chapter 15 Deep Learning and the Weather Forecasting Problem: Precipitation Nowcasting
15.1 Introduction
15.2 Formulation
15.3 Learning Strategies
15.4 Models
15.4.1 FNN‐based Models
15.4.2 RNN‐based Models
15.4.3 Encoder‐forecaster Structure
15.4.4 Convolutional LSTM
15.4.5 ConvLSTM with Star‐shaped Bridge
15.4.6 Predictive RNN
15.4.7 Memory in Memory Network
15.4.8 Trajectory GRU
15.5 Benchmark
15.5.1 HKO‐7 Dataset
15.5.2 Evaluation Methodology
15.5.3 Evaluated Algorithms
15.5.4 Evaluation Results
15.6 Discussion
Appendix
Acknowledgement
Chapter 16 Deep Learning for High‐dimensional Parameter Retrieval
16.1 Introduction
16.2 Deep Learning Parameter Retrieval Literature
16.2.1 Land
16.2.2 Ocean
16.2.3 Cryosphere
16.2.4 Global Weather Models
16.3 The Challenge of High‐dimensional Problems
16.3.1 Computational Load of CNNs
16.3.2 Mean Square Error or Cross‐entropy Optimization?
16.4 Applications and Examples
16.4.1 Utilizing High‐dimensional Spatio‐spectral Information with CNNs
16.4.2 The Effect of Loss Functions in Retrieval of Sea Ice Concentrations
16.5 Conclusion
Chapter 17 A Review of Deep Learning for Cryospheric Studies
17.1 Introduction
17.2 Deep‐learning‐based Remote Sensing Studies of the Cryosphere
17.2.1 Glaciers
17.2.2 Ice Sheet
17.2.3 Snow
17.2.4 Permafrost
17.2.5 Sea Ice
17.2.6 River Ice
17.3 Deep‐learning‐based Modeling of the Cryosphere
17.4 Summary and Prospect
Appendix: List of Data and Codes
Chapter 18 Emulating Ecological Memory with Recurrent Neural Networks
18.1 Ecological Memory Effects: Concepts and Relevance
18.2 Data‐driven Approaches for Ecological Memory Effects
18.2.1 A Brief Overview of Memory Effects
18.2.2 Data‐driven Methods for Memory Effects
18.3 Case Study: Emulating a Physical Model Using Recurrent Neural Networks
18.3.1 Physical Model Simulation Data
18.3.2 Experimental Design
18.3.3 RNN Setup and Training
18.4 Results and Discussion
18.4.1 The Predictive Capability Across Scales
18.4.2 Prediction of Seasonal Dynamics
18.5 Conclusions
Part III Linking Physics and Deep Learning Models
Chapter 19 Applications of Deep Learning in Hydrology
19.1 Introduction
19.2 Deep Learning Applications in Hydrology
19.2.1 Dynamical System Modeling
19.2.1.1 Large‐scale Hydrologic Modeling with Big Data
19.2.1.2 Data‐limited LSTM Applications
19.2.2 Physics‐constrained Hydrologic Machine Learning
19.2.3 Information Retrieval for Hydrology
19.2.4 Physically‐informed Machine Learning for Subsurface Flow and Reactive Transport Modeling
19.2.5 Additional Observations
19.3 Current Limitations and Outlook
Acknowledgments
Chapter 20 Deep Learning of Unresolved Turbulent Ocean Processes in Climate Models
20.1 Introduction
20.2 The Parameterization Problem
20.3 Deep Learning Parameterizations of Subgrid Ocean Processes
20.3.1 Why DL for Subgrid Parameterizations?
20.3.2 Recent Advances in DL for Subgrid Parameterizations
20.4 Physics‐aware Deep Learning
20.5 Further Challenges ahead for Deep Learning Parameterizations
Chapter 21 Deep Learning for the Parametrization of Subgrid Processes in Climate Models
21.1 Introduction
21.2 Deep Neural Networks for Moist Convection (Deep Clouds) Parametrization
21.3 Physical Constraints and Generalization
21.4 Future Challenges
Chapter 22 Using Deep Learning to Correct Theoretically‐derived Models
22.1 Experiments with the Lorenz '96 System
22.1.1 The Lorenz'96 Equations and Coarse‐scale Models
22.1.1.1 Theoretically‐derived Coarse‐scale Model
22.1.1.2 Models with ANNs
22.1.2 Results
22.1.2.1 Single‐timestep Tendency Prediction Errors
22.1.2.2 Forecast and Climate Prediction Skill
22.1.3 Testing Seamless Prediction
22.2 Discussion and Outlook
22.2.1 Towards Earth System Modeling
22.2.2 Application to Climate Change Studies
22.3 Conclusion
Chapter 23 Outlook
Bibliography
Index
EULA