Improving AGBD Models: Combatting Overfitting with a Data-Centric Approach in Machine Learning

In the previous tutorial, we identified overfitting issues in our random forest model. In this tutorial, we will tackle this challenge by improving both the quality and quantity of our training dataset. We will demonstrate how to use the quality_mask, error_mask, and slope_mask functions to filter out unreliable Global Ecosystem Dynamics Investigation (GEDI) Level 4A (L4A) aboveground biomass density (AGBD) measurements. By excluding data points with high uncertainty and measurements taken on steep slopes, we aim to enhance the accuracy of GEDI L4A AGBD estimates. Additionally, we will perform a scale sensitivity analysis to determine the optimal scale for the most accurate model results. Course, script, and blog post links: https://aigeolabs.com/courses/course-8/module-1/1-1-2/ https://github.com/ck1972/Python-Geospatial_Model1/blob/main/2b_Modeling_AGBD_GEDI_S2_SpectralIndices_RF_Model_GEE_MafungautsiForestReserve_Scale_Sensitivity_Analysis_Tutorial.ipynb https://aigeolabs.com/boosting-aboveground-biomass-density-agbd-model-accuracy-leveraging-eo-data-and-machine-learning-with-data-centric-strategies/