Carbon-Trading-Verification

This repository contains code used to the MSc Advanced Computing Thesis: Joint Study of Above Ground Biomass and Soil Organic Carbon for Total Carbon Estimation in Scotland.

Code Structure

Inside scotland_carbon/src contains our implementation:

• train.py: Trains the model, and output to file location specified by the user. Parameters include model, target variable, machine learning technique, isLog, output path.
• feature_importance.py: Generates the feature importance graph for specified models. Takes in model, model path, output path.
• carbon_maps.py: Provides two methods to generate carbon maps used in the report. The plot_graph function plots the prediction and error plots for the specified model. The plot_single_graph function plots for the carbon maps for total carbon estimation, total carbon ground truth and the total carbon error.
• grid_search.py: Contains the code for performing grid search on our models, we can tune hyperparameters for any model and ML technique. Parameters are model, target variable, ml technique and isLog
• utils.py contains utility functions and model definitions

File Structure

• archive contains previous work from Carla Estimating-AGB and Anish Estimating-SOC.
• scotland_carbon contains our implementation

Trained Models

The trained models and the results are available on AWS (https://bci-satellite-carbon.s3.eu-west-2.amazonaws.com/imperial-terrence/project_models.zip). Below is a brief description of the features used in different models

• VH_1 VV_1 corresponds to the Sentinel 1 data obtained
• The prefix BAND_ indicate data source from Sentinel 2
• The prefix L_ indicate data source from LandSat 8 satellite
• DEM_CS DEM_LSF DEM_TWI DEM_ELEV stands for digital elevation derivatives obtained from the digital elevation map, they corresponds to catchment slope, length slope factor , tropical wetness index and digital elevation respectively.
• NDVI EVI SATVI are vegetation indices calculated from Sentinel 2
• CATEGORY corresponds to data from the forest inventory data

Remote Sensing Data

All remote sensing data including Sentinel 1, Sentinel 2, LandSat 8, DEM, Inventory Data is available on AWS at:

• https://bci-satellite-carbon.s3.eu-west-2.amazonaws.com/imperial-terrence/satellite_data_part1.zip
• https://bci-satellite-carbon.s3.eu-west-2.amazonaws.com/imperial-terrence/satellite_data_part2.zip
• https://bci-satellite-carbon.s3.eu-west-2.amazonaws.com/imperial-terrence/satellite_data_part3-1.zip
• https://bci-satellite-carbon.s3.eu-west-2.amazonaws.com/imperial-terrence/satellite_data_part3-2.zip
• https://bci-satellite-carbon.s3.eu-west-2.amazonaws.com/imperial-terrence/satellite_data_part4.zip

Virtual Environment Location

https://bci-satellite-carbon.s3.eu-west-2.amazonaws.com/imperial-terrence/venv.zip

Running the code

Each file can be run separately by calling the corresponding functions.

• The csv file location is Carbon-Trading-Verification\scotland_carbon\data\S1AIW_S2AL2A_DEM_IDX_SOCS_SG_L_INVEN_AGB_300m_processed.csv.
• The Evaluation tif file for Model G is Carbon-Trading-Verification\scotland_carbon\data\MODEL_G_EVAL.tif
• The Evaluation tif file for Model H is Carbon-Trading-Verification\scotland_carbon\data\MODEL_H_EVAL.tif
• Note that for the isLog parameter, we use log for SOC estimation and no log for AGB estimation, this is based on prior experimentations that taking log improves model training and performance for SOC estimation but not the case for AGB estimation.