To expand the forage biomass database and develop practical management tools, two innovative methodologies were developed combining rapid field sampling with satellite information.

Accurate estimation of forage biomass requires calibration data collected under a wide range of environmental and management conditions. Previous studies indicate that six to seven years of data are needed to build robust models with relative errors close to 22%, especially under variable environmental conditions (Kearney et al., 2022 – doi.org/10.1016/j.rse.2022.112907).
Using a subset of 755 field-collected biomass data points from the first year—corresponding to various forage resources and regions across Argentina—we evaluated the ability of machine learning models (Random Forest) to predict biomass using spectral indices derived from Sentinel-1 and Sentinel-2 satellite imagery. The general model yielded a prediction error of 1100 kg DM/ha (99%), while narrowing the focus to a subset of data from pastures in Balcarce reduced the error to 650 kg DM/ha (55%). These results suggest that reducing environmental heterogeneity greatly improves prediction accuracy, although further expansion of the calibration database is still needed.
To address this challenge, the project designed and is currently evaluating two methodologies that efficiently link field observations with satellite data:
1. Spatial Simplification: This method consists of an application developed in Google Earth Engine, where biomass values measured at selected field points are uploaded. These data are used to calibrate a correlation model with spectral indices derived from satellite images taken on nearby dates. The resulting model can be applied visually over user-defined areas within the interface to estimate biomass across entire fields (Figure 1). Validation conducted in natural grasslands at INTA’s Concepción del Uruguay Experimental Station showed a good level of agreement (R² = 0.6). In 21 evaluations over Festuca and Alfalfa pastures at INTA Balcarce, 17 achieved R² values above 0.6.
2. Temporal Simplification: This methodology combines field measurements of residual biomass after grazing with forage growth models driven by satellite data. This approach enables real-time estimation of biomass availability per paddock in rotational grazing systems, provided residual biomass is measured. Users can visualize the current availability and weekly evolution of forage stocks (Figure 2A). Evaluations in two systems in Balcarce showed good agreement between observed and predicted values (R² = 0.8, Figure 2B).
The implementation of these tools in livestock systems helps test their usefulness as decision-making support, promotes the adoption of satellite technology in the agricultural sector, and contributes to consolidating a robust monitoring network to support increasingly accurate and representative predictive models.