Utilising grassland management and climate data for more accurate prediction of herbage mass using the rising plate meter

Abstract

Efficient grass-based livestock production depends on precise allocation of pasture to the herd in the form of herbage mass (HM). Accurate measurement of HM results in increased utilisation of grass in the herd’s diet and consequently reductions in whole-farm feed inputs, emissions and costs. The rising plate meter (RPM) is an established method of estimating HM, but there is scope to improve its accuracy. Real-time meteorological data and pasture management information have never been analysed in combination with the RPM. This study aimed to utilise such data to improve the accuracy of HM prediction using multiple linear regression (MLR) and machine learning through the random forest (RF) algorithm. Seventeen variables were assessed and models were evaluated in terms of relative prediction error (RPE). Decreases of 6–12% RPE were observed for the MLR models compared with conventional models. Further decreases of 11–17% were recorded for RF models. An MLR model comprising of management data that were readily available to farmers was deemed optimum for on-farm use and included coefficients for: compressed sward height (mm), nitrogen fertiliser rate (kg ha−1) and grazing rotation number (RMSE = 324 kg DM ha−1). The addition of meteorological variables resulted in a further 0.9% decrease in RPE (RMSE = 312 kg DM ha−1), but was not practical considering the expense of on-farm meteorological sensors. The RF model with meteorological variables (RMSE = 262 kg DM ha−1) had 1.5% lower RPE compared with the RF model without (RMSE = 243 kg DM ha−1).