Grazed grass is the cheapest feed source available to Irish farmers. Daily grassland management is essential on Irish beef and dairy farms to best utilize this valuable resource. A predictive grass growth model would greatly aid farmers in their decision making. Grass growth models (GGMs) accounting for local parameters such as soil type, defoliation management and N fertilizer application have been developed with retrospective weather observations as inputs. However, forecasts have not yet been included. Daily forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) predicting up to ten days in advance were verified at 25 Irish weather stations between 2007 and 2013 to test their viability for inclusion in GGMs. Various forecast bias-correction methods were assessed to improve forecast quality. Weather variables relevant to grass growth were examined: maximum, minimum and mean 2 m air temperature, rainfall and soil temperatures at six depths. Predictions from a GGM using 1) observed weather and 2) bias-corrected forecasts from 2008 to 2013 were compared to each other and to observed on-farm grass growth data. This analysis was conducted using forecasts for up to ten days in advance to evaluate the length of the predictive ability of the model.
ECMWF forecasts of soil and air temperature variables generally performed well at all locations up to ten days. For example, the Root Mean Squared Error (RMSE) of forecasts with observations for 10cm soil temperatures rose from 1.62°C one day in advance to 2.46°C ten days in advance. However, ECMWF forecasts often struggled to predict large rainfall events. The Mean Absolute Error (MAE) for forecasts predicting five days in advance with observations between 40 and 50mm was 31.3mm, compared to 2.2mm for observations between 0 and 10mm. A bias correction approach using a regression model generally gave the best improvements in forecast accuracy for all temperature variables. However, none of the bias corrections resulted in more accurate forecasts for large rainfall events.
Predictions from the GGM were similar when weather forecasts and observations were included. However, grass growth predictions were poor when high rainfall events occurred. For example, N leached was modelled inaccurately in these cases. Overall, the work suggests that utilising weather forecasts, along with other relevant variables, in predicting future grass growth could be a valuable tool to farm management leading to better efficiency of grassland usage, increased productivity and reduced costs.