Estimation of forage mass in a mixed pasture by machine learning, pasture management and satellite meteorological data

Abstract

Measuring forage mass (FM) in the pasture, prior to grazing, is critical to determining the daily allocation of forage in pastoral animal production systems. FM is estimated by cutting forage in known areas, using allometric equations, or with the use of remote sensors (RS); however, the accuracy and practicality of the different methods for estimating FM is variable. The objective was to obtain predictive models using environmental and pasture management variables to predict FM. Regression models were fitted to estimate FM based on variables of pasture management (PM) or measurements obtained by RS, such as reflectance, air temperature, and rainfall. A mixed pasture grazed by beef cattle was studied for three years. With 80 % of data, models were built by ordinary least squares (OLS) or by machine learning (ML) algorithms. The remaining 20 % of the data was used to validate the models using the coefficient of determination and average bias between estimated and observed values. The base model of study was the relationship between pasture height before grazing and FM, this model was fitted using OLS; the r2 was 0.43. When models that included PM variables were fitted, the r2 was 0.45 for OLS and 0.63 for ML. When fitting models with PM and RS variables, the r2 was 0.71 for OLS and 0.96 for ML. ML-fitted model ensembles reduced the bias of FM estimates of the examined pasture. Overall, ML models better represented the relationship between pasture height before grazing and FM than OLS models, when fitted with pasture management variables and RS information. ML models can be used as a tool for daily decision-making in pastoral production systems.