Deep Learning for Sea Ice Roughness/Thickness Prediction

Please find the paper for further details.

Authors: Sangwon Lim and Omar Kawach

Getting Started

Install the Package in a Python Virtual Environment

To avoid conflicts, the first step is to isolate this project by creating a Python virtual environment called venv. The virtual environment will have it's own python interpreter, dependencies, and scripts. Commands should only be entered in a terminal that has venv active.

Apple Silicon (M1)

sh M1_install.sh
source ~/miniforge3/bin/activate

Linux

python -m venv venv
source venv/bin/activate
pip install .
pip install -r requirements.txt

Windows

python -m venv venv
venv/Scripts/activate
pip install .
pip install -r requirements.txt

Processes

create_dataset

Creates training and test datasets based on the configurations provided.

# To run all configurations
$ create_dataset --config ./Configs/Dataset_configs/*
# To run specific configuration
$ create_dataset --config ./Configs/Dataset_configs/thickness.yml

CAE

3D-CAE processing based on configurations.

# To run all configurations
$ CAE --config ./Configs/CAE_configs/*
# To run specific configuration
$ CAE --config ./Configs/CAE_configs/CAE.yml

CAE.sh

Content of CAE.sh:

#!/bin/bash

CAE --config ./Configs/CAE_configs/*

for parent_dir in thk rms; do 
    for file in CAE SOBEL_CAE canny_CAE; do
        echo "************************* ./data/"$parent_dir"_tr_data/"$file".csv *************************" 
        normalize_csv --input ./data/"$parent_dir"_tr_data/"$file".csv --std-data ./data/"$parent_dir"_tr_data/"$file".csv
        normalize_csv --input ./data/"$parent_dir"_te_data/"$file".csv --std-data ./data/"$parent_dir"_tr_data/"$file".csv    
    done
done

To run the script:

$ sh ./bash_scripts/CAE.sh

NN_regression / NN_classification

Training processes for Deep Neural Network regression and classification.

$ NN_classification --ml-config ./Configs/ML_configs/NN_type_class/CNN.yml --result-dir ./results/NN_type_class/CNN
$ NN_regression --ml-config ./Configs/ML_configs/NN_thk_regr/FYI_MYI.yml --result-dir ./results/NN_thk_regr/FYI_MYI

train.sh

Content of train.sh:

#!/bin/bash

for file in CNN CNN_spk CAE CAE_spk backscatter; do
    echo "************************* NN_type_class/$file *************************" 
    NN_classification --ml-config ./Configs/ML_configs/NN_type_class/"$file".yml --result-dir ./results/NN_type_class/"$file"
    test_classification --ml-config ./Configs/ML_configs/NN_type_class/"$file".yml --result-dir ./results/NN_type_class/"$file"
done
boxplot --result-dir ./results/NN_type_class/ --model-type classification --methods CNN CNN_spk CAE CAE_spk backscatter

for file in FYI_MYI FYI_MYI_likelihood; do
    echo "************************* NN_thk_regr/$file *************************" 
    NN_regression --ml-config ./Configs/ML_configs/NN_thk_regr/"$file".yml --result-dir ./results/NN_thk_regr/"$file"
    test_regression --ml-config ./Configs/ML_configs/NN_thk_regr/"$file".yml --result-dir ./results/NN_thk_regr/"$file"
done
boxplot --result-dir ./results/NN_thk_regr --model-type regression --methods FYI_MYI FYI_MYI_likelihood

for file in FYI_MYI FYI_MYI_likelihood; do
    echo "************************* NN_rms_regr/$file *************************" 
    NN_regression --ml-config ./Configs/ML_configs/NN_rms_regr/"$file".yml --result-dir ./results/NN_rms_regr/"$file"
    test_regression --ml-config ./Configs/ML_configs/NN_rms_regr/"$file".yml --result-dir ./results/NN_rms_regr/"$file"
done
boxplot --result-dir ./results/NN_rms_regr --model-type regression --methods FYI_MYI FYI_MYI_likelihood

To run the script:

$ sh ./bash_scripts/train.sh

ResNet.sh

Content of ResNet.sh:

#!/bin/bash

echo "************************* ResNet_rms_0 *************************" 
NN_regression --ml-config ./Configs/ML_configs/NN_rms_regr/ResNet/recursion_0.yml --result-dir ./results/NN_rms_regr/ResNet/recursion_0
test_regression --ml-config ./Configs/ML_configs/NN_rms_regr/ResNet/recursion_0.yml --result-dir ./results/NN_rms_regr/ResNet/recursion_0
predict --ml-config ./Configs/ML_configs/NN_rms_regr/ResNet/recursion_0.yml --result-dir ./results/NN_rms_regr/ResNet/recursion_0 --checkpoint 1 --classes rms_0 --output-dir ./results/NN_rms_regr/ResNet
predict --ml-config ./Configs/ML_configs/NN_rms_regr/ResNet/pred_0.yml --result-dir ./results/NN_rms_regr/ResNet/recursion_0 --checkpoint 1 --classes rms_0 --output-dir ./results/NN_rms_regr/ResNet

for i in {1..50}; do
    echo "************************* ResNet_thk_$i *************************" 
    NN_regression --ml-config ./Configs/ML_configs/NN_thk_regr/ResNet/recursion.yml --result-dir ./results/NN_thk_regr/ResNet/recursion_"$i"
    test_regression --ml-config ./Configs/ML_configs/NN_thk_regr/ResNet/recursion.yml --result-dir ./results/NN_thk_regr/ResNet/recursion_"$i"
    predict --ml-config ./Configs/ML_configs/NN_thk_regr/ResNet/recursion.yml --result-dir ./results/NN_thk_regr/ResNet/recursion_"$i" --checkpoint 1 --classes thk_"$i" --output-dir ./results/NN_thk_regr/ResNet
    predict --ml-config ./Configs/ML_configs/NN_thk_regr/ResNet/pred.yml --result-dir ./results/NN_thk_regr/ResNet/recursion_"$i" --checkpoint 1 --classes thk_"$i" --output-dir ./results/NN_thk_regr/ResNet

    echo "************************* ResNet_rms_$i *************************" 
    NN_regression --ml-config ./Configs/ML_configs/NN_rms_regr/ResNet/recursion.yml --result-dir ./results/NN_rms_regr/ResNet/recursion_"$i"
    test_regression --ml-config ./Configs/ML_configs/NN_rms_regr/ResNet/recursion.yml --result-dir ./results/NN_rms_regr/ResNet/recursion_"$i"
    predict --ml-config ./Configs/ML_configs/NN_rms_regr/ResNet/recursion.yml --result-dir ./results/NN_rms_regr/ResNet/recursion_"$i" --checkpoint 1 --classes rms_"$i" --output-dir ./results/NN_rms_regr/ResNet
    predict --ml-config ./Configs/ML_configs/NN_rms_regr/ResNet/pred.yml --result-dir ./results/NN_rms_regr/ResNet/recursion_"$i" --checkpoint 1 --classes rms_"$i" --output-dir ./results/NN_rms_regr/ResNet
done

To run the script:

$ sh ./bash_scripts/ResNet.sh

test_regression / test_classification

Testing processes for regression and classification models.

$ test_classification --ml-config ./Configs/ML_configs/NN_type_class/CNN.yml --result-dir ./results/NN_type_class/CNN
$ test_regression --ml-config ./Configs/ML_configs/NN_thk_regr/FYI_MYI.yml --result-dir ./results/NN_thk_regr/FYI_MYI

Sea ice type classification result:

Sea ice surface roughness regression result:

Sea ice thickness regression result:

boxplot

Box plots to summarize test results.

boxplot --result-dir ./results/NN_rms_regr/FYI --model-type regression --methods backscatter corr_std GLCM CAE SOBEL_CAE canny_CAE
boxplot --result-dir ./results/NN_type_class/ --model-type classification --methods backscatter corr_std GLCM CAE SOBEL_CAE canny_CAE

boxplot.sh

Content of boxplot.sh:

#!/bin/bash

for parent_dir in NN_rms_regr NN_thk_regr; do
    for child_dir in FYI FYI_MYI MYI; do
        boxplot --result-dir ./results/"$parent_dir"/"$child_dir" --model-type regression --methods backscatter corr_std GLCM CAE SOBEL_CAE
    done
done

boxplot --result-dir ./results/NN_type_class --model-type classification --methods backscatter corr_std GLCM CAE SOBEL_CAE

To run the script:

$ sh ./bash_scripts/boxplot.sh

streamlining.sh

Content of streamlining.sh:

#!/bin/bash

pixel_select --x 9060 --y 9243
create_dataset --config ./Configs/streamlining_configs/dataset/*
CAE --config ./Configs/streamlining_configs/CAE/CAE_spk.yml
mkdir ./tmp/type
mkdir ./tmp/rms
mkdir ./tmp/thk
predict --result-dir ./results/NN_type_class/CNN_CAE --output-dir ./tmp/type --ml-config Configs/streamlining_configs/ML/type.yml --checkpoint 1 --classes FYI DFYI MYI

echo "************************* RNN_rms_0 *************************" 
predict --ml-config ./Configs/streamlining_configs/ML/rms/pred_0.yml --result-dir ./results/NN_rms_regr/RNN/recursion_0 --checkpoint 1 --classes rms_0 --output-dir ./tmp/rms

for i in {1..50}; do
    echo "************************* RNN_thk_$i *************************" 
    predict --ml-config ./Configs/streamlining_configs/ML/thk/pred.yml --result-dir ./results/NN_thk_regr/RNN/recursion_"$i" --checkpoint 1 --classes thk_"$i" --output-dir ./tmp/thk

    echo "************************* RNN_rms_$i *************************" 
    predict --ml-config ./Configs/streamlining_configs/ML/rms/pred.yml --result-dir ./results/NN_rms_regr/RNN/recursion_"$i" --checkpoint 1 --classes rms_"$i" --output-dir ./tmp/rms
done

output_image --input ./tmp/type/updated_tr_dataset.csv --label FYI DFYI MYI --center 9243 9060
output_image --input ./tmp/rms/updated_tr_dataset.csv --label rms_23 --center 9243 9060
output_image --input ./tmp/thk/updated_tr_dataset.csv --label thk_41 --center 9243 9060

To run the script:

$ sh ./bash_scripts/streamlining.sh

Streamlining process:

First year ice (FYI) probability map:

Deformed first year ice (DFYI) probability map:

Multi-year ice (MYI) probability map:

Sea ice type map:

Sea ice surface roughness map:

Sea ice thickness map: