Grazed grass is a cheap source of feed for Irish farm animals, and therefore to maximise profits it is important to have efficient use of grass on farms. A predictive grass growth model would help farmers with their grassland management. However to date, weather forecasts have not been included in Irish grass growth models; only retrospective observations have been used as weather inputs.
In this collaboration between Teagasc, Met Éireann and Maynooth University, ECMWF (European Centre for Medium-Range Weather Forecasts) deterministic weather forecasts from between 2007 and 2013 were evaluated to assess their potential usefulness in grass growth models for up to 10 days. The variables studied were daily rainfall, minimum, maximum and mean 2m air temperatures, as well as soil temperatures at depths of 5, 10, 20, 30, 50 and 100cm. The forecasts and observations were obtained at each of Met Éireann’s 25 synoptic stations. After studying the accuracy of the ECMWF forecasts, some basic bias correction techniques were applied to attempt to improve forecast precision.
It was found that air temperature forecasts performed well, usually giving lower RMSE values than low-skill forecasts such as persistence for up to a week. Although soil temperature forecasts gave low RMSE values, the day-1 forecasts at greater depths were often outperformed by persistence forecasts. This can be attributed to the homogeneity of soil temperatures at these depths. ECMWF rainfall forecasts generally predicted observations accurately up to seven days in advance, after which mean climatology forecasts were superior. There were substantial improvements in RMSE for both air and soil temperatures after the best bias correction methods were applied. However, none of the bias correction methods studied gave large improvements for rainfall forecasts.
Based on the results, it was concluded that ECMWF forecasts and the best bias corrected forecasts could be useful in a predictive grass growth model for up to a week. After this it may be the case that low skill forecasts such as mean climatology and persistence forecasts would be just as valuable.
This article was submitted by Jack McDonnell. Jack is a PhD fellow within the PrecisionGrazing
Stimulus Project at Maynooth University & TEAGASC supported by the DAFM Research Stimulus funding. Some of his provisional findings can be found on his poster.