Machine Learning for the Classification and Analysis of Biomass Indices and their relationship with Climate Change, Atacama Desert

Abstract

In this work we use Machine Learning (Randon Forest) as a tool to classify biomass and calculate vegetation indices seeking to identify the characteristics of the vegetation cover at the head of the Atacama Desert. The aim is to establish the correlation between vegetation indices and precipitation, in order to know their reliability on the climatology in this region. The geospatial analysis based on Google Earth Engine (GEE) and the processing of Landsat 5 ETM and Landsat 8 OLI/TIRS images was important, for the period 1985 - 2022, which made it possible to characterize climate change. The NDVI, SAVI, GVI and RVI have been tested and validated in arid systems. The NDVI responds positively to precipitation in the wet season and weakly in the winter rainy season. It is confirmed that the high NDVI corresponds to summer, after a prolonged drought. Towards the years 2020 and 2022, an increase in vegetation cover is recorded in places with higher temperatures, evidencing climate change and reflected in biomass indices.