Gross Primary Productivity estimation through remote sensing and machine learning techniques in the high Andean Region of Ecuador

Abstract

Accurate estimations of Gross Primary Productivity (GPP) are a valuable tool for simulating the carbon cycle and therefore play a significant role in addressing the challenges posed by climate change. However, the estimation of GPP is a large challenge, owing to the lack of direct. In this study, GPP was estimated using machine learning models (ML), such as Random Forest (RF) and Support Vector Regression (SVR), in páramo. The strength and complex nonlinear relationships that dominate the GPP are fundamentally important to conduct an uncertainty analysis for future climate projections. This study evaluated the relationship between biometeorological variables, remote sensing data and GPP estimation. The methodology used to estimate GPP confirmed that ML-based models performed better than traditional models. The performance of ML models varied significantly among seasons, with R ranging from 0.24-0.86. The RF model performed better in capturing the temporal changes and magnitude of GPP in less humid season, displaying the highest R (0.86), lowest RMSE (0.37 g C/m2), and lowest PBIAS (-3 %). In addition, the importance of the variables showed that solar radiation was the most significant predictor of GPP. This suggests that production in the Páramo biome is not limited by water, but by the amount of incident solar radiation. The study provided an approach to derive daily GPP fluxes over a 2-year study period. This study has attempted to assess the impact of various variables on GPP estimates. It can be used to further the development of vegetation prediction models