SBPproperties

© Copyright 2022 Tiago Morais

This repository includes the algorithm (Random Forest) that was developed to estimate pasture properties of Portuguese Sown biodiverse pastures rich in legumes (SBP) derived from Sentinel-2 data. The considered properties are:

• Standing biomass (kg/ha)
• Legumes fraction (-)
• Nitrogen content (g/kg)
• Phosphorus content (g/kg)

Used scripts are in folder "train_models" and the best trained models are in folder "models"

Two cross-valiadtion approaches were used, whcih lead to set of models:

• Random cross-validation (RN-CV): random approach with random selection of the folds
• Leave-location-and-year-out cross-validation (LLYO-CV): each fold is the set of all observations from each farm and year

Usage

• Requires scikit-learn library
• Requires joblib library
• Download folder "models" folder
• Prepare input data (see raw data file used in "Raw data folder"
• Use the individual models in the "models" folder using the input data file

Example of application of the models

Context of raw data file

Reference

• Morais et al. (2022). Characterization of Portuguese sown rainfed grasslands using remote sensing and machine learning. Precision Agriculture X(XX), XXX-XXX; (https://doi.org/10.1007/XX)