Estimación de la edad del cultivo de palma de aceite (elaeis guineensis) en función de índices de vegetación y humedad mediante imágenes satelitales y aprendizaje automático

Palm oil (Elaeis guineensis) is one of the most widely consumed vegetable oils in the world and is a key ingredient in profitable global value chains. In Costa Rica, it is one of the three crops with the largest occupied agricultural area. The objective of this study is to estimate the age of the palm oil crop based on vegetation and humidity indices generated with satellite images using machine learning techniques, for this information will be taken from 20 years of a palm oil crop located in the Central Pacific of Costa Rica and its LANDSAT satellite images. Two models were generated, the Random Forest (RF) model presented better results when using a ntree of 10000 in training, obtaining the following coefficients; NSE=0,9161, RMSE=1,6485 years, MAE=1,0178 years and r2=0,9476; and the XGBoost model using a max.depth of 4, nrounds of 50 and nthread of 10 in the training for which it was obtained: NSE=0,9530, RMSE=1,23 years, MAE=0,89 years and r2=0,9577. For the data obtained in the validation, the RF coefficients were: NSE=0,5395, RMSE=3,76 years, MAE= 2,46 years and r2=0,5533; for the XGBoost model NSE=0,6204, RMSE=3,4114 years, MAE= 2,23 years and r2=0,6210. The most important variable in both models was the NDMI 75th percentile. The most suitable model with the best performance is XGBoost, however it is not recommended when the age of the crop exceeds 25 years.