Comparison of models estimating biomass in Cenchrus clandestinus (Hochst. ex Chiov.) Morrone pastures at high-altitude tropical farms

The use of remote sensing technologies has been increasingly recognized as a reliable tool for biomass estimation in tropical pastures. This study explores the application of predictive models that integrate agroclimatic variables, complementary field data, and satellite image processing to estimate biomass in rotational grazing systems of Kikuyu grass (Cenchrus clandestinus (Hochst. ex Chiov.) Morrone) in the Colombian high tropics. For this purpose, PlanetScope satellite imagery was used in combination with field-collected data, such as forage type and age, temperature, humidity, precipitation, and vegetation spectral index. Four regression models were compared: Generalized Linear Model with Poisson distribution, Multiple Linear Regression, Multilayer Perceptron Artificial Neural Network (MLP), and Random Forest Regression, in order to identify the most accurate statistical model for estimating forage biomass. Among the evaluated models, the MLP demonstrated the best performance (R² = 0.7). Variable importance analysis highlighted the Green Normalized Difference Vegetation index (NDVI), forage age, and the Green Chlorophyl Index (CIGreen) as the most relevant predictors. The results of this study suggest that remote sensing techniques, when complemented with forage sampling, offer an effective strategy for large-scale biomass monitoring and pasture management. The implementation of these models can enhance decision-making in livestock systems in the high tropics, improving grazing planning and forage resource management.