Programming code for: Unlocking value: mapping invasive alien tree biomass with satellite data and machine learning to support ecological restoration

Programming code for: Unlocking value: Mapping invasive alien tree biomass with satellite data and machine learning for a restoration economy 

This repository hosts the code for the processing of raw biomass and satellite-derived datasets to develop an invasive alien tree biomass model.

Summary
In this study we aim to test the performance of multi-source remote sensing-derived datasets for the estimation of biomass of invasive alien trees in South Africa.

LiDAR and field-measured invasive alien tree metrics were used to develop a regional biomass model for an alien infested region of the Aguhlas Plain, in the Cape Florisitc Region of South Africa. This study leveraged satellite-derived datasets (Sentinel-1, Sentinel-2, SRTM, GEDI-L2A, and Global Canopy Height) and machine learning to develop a regional-scale model for estimating invasive alien tree biomass at a 25m resolution. Performance of the models were assessed using the coeffiecent of determination (R-squared), Spearman correlation coeffiecent (rs), PBias, and RMSE. Predictor variable importance were also tested for. Once the models were developed, they were validated with an independant alien infested ground truth site. Additionally, the most effective model was applied to ecologically distinct catchments to test model transferability and to demonstrate how the model could be used in a practical management scenario.

• Note: 'Invasive alien trees' refer to taxa such as Pinus, Eucalyptus, Neltuma, and Acacia spp that are non-native to South Africa that can displace native vegetation and cause environmental and economic harm.

Repository Components
The components of this repository include four primary phases each including a set of method workflow notes and code scripts
The four phases include the following:

(1) Model Development
- Collate a target variable dataset (representing invasive alien tree biomass)
- Collate a predictor variable dataset (representing satellite-derived features)
- Develop an invasive alien tree biomass model with machine learning

(2) Model Validation
- Assess model performance with various metrics

(3) Ground-truthing
- Compare model estimates to harvested biomass from an independant ground-truthing site