Model and Predict Species Proportions

This repository contains an R Markdown tutorial for modeling and predicting tree species proportions using remote sensing data. The methodology combines airborne laser scanning (ALS) and Sentinel-2 imagery, toguether with Random Forest modeling, to predict species proportions at the stand level.

Repository Contents

• species_proportions_analysis.Rmd – R Markdown file containing the full analysis pipeline.
• README.md – This file, providing an overview of the repository.

Data description

The datasets used in this analysis should include:

• Sample plot dataset - Training data containing field-measured species proportions and remote sensing-derived predictors.
• Validation plot dataset - Independent dataset used to assess model performance.

Methodology Overview

The analysis follows a multi-step approach:

1. Load Data – Sample plot and validation datasets are loaded.
2. Prepare Data – Remote sensing metrics are selected and cleaned for modeling.
3. Train Random Forest Model – A multi-response Random Forest model is trained to predict species proportions (Spruce, Pine, and Deciduous).
4. Predict on Validation Data – The trained model is applied to the validation dataset.
5. Calibrate Predictions (Optional) – A correction model is applied to adjust biases in the predictions.
6. Normalize Predictions – Predictions are normalized to ensure they sum to one.
7. Aggregate at Stand Level – Plot-level predictions are aggregated to stands.
8. Evaluate Performance – Error metrics such as RMSE and relative RMSE are calculated.
9. Visualization – Observed vs. predicted species proportions are plotted.

Required Dependencies

Make sure to install the following R packages before running the analysis:

install.packages(c("dplyr", "tidyr", "randomForestSRC", "data.table", "ggplot2"))