How does inclusion of weather forecasting impact in-season crop model predictions?
Accurately forecasting crop yield in advance of harvest could greatly benefit decision makers. However, few evaluations have been conducted to determine the effectiveness of including weather forecasts, as opposed to using historical/climatology data, into crop models. We tested a combination of short-term weather forecasts from the Weather Research and Forecasting Model (WRF) to predict in season weather variables, such as, maximum and minimum temperature, precipitation, and radiation at four different forecast lengths (14 days, 7 days, 3 days, and 0 days). This forecasted weather data along with the current and historic (previous 35 years) data were combined to drive Agricultural Production Systems sIMulator (APSIM) in-season forecasts of corn [Zea mays L] and soybean [Glycine max] crop yield and phenology in Iowa, USA. The overall goal of this research was to determine how the inclusion of weather forecasting impacts in-season crop model predictions. To achieve this goal we had two objectives 1) to determine the dependence of the accuracy of APSIM yield and phenology predictions on weather forecast length, and 2) the impact of weather forecasts accuracy on APSIM prediction accuracy. APSIM simulations of biomass accumulation and phenology were evaluated against bi-weekly field measurements across 16 field trials (two years, 2015 and 2016; two sites, central and northwest Iowa, USA; two crops, corn and soybean; and two planting dates; early May vs early June). We hypothesized that 1) the accuracy and variability of crop yield predictions will be inversely proportional to the weather forecast length and 2) the inclusion of an explicit weather forecast will reduce crop yield prediction uncertainty and produce a reliable estimate with more lead time relative to using historical variation alone. The accuracy of in-season yield forecasts of corn and soybean varied by treatment, but overall the accuracy was inversely proportional to forecast length (P < 0.05). Our analysis indicated that the most accurate forecast length varied greatly among the 16 treatments, but that the 0 day and 3 day forecasts were, on average, the most accurate. That the 0 day forecast was most accurate meant that a weather forecast from WRF was not better than a weather forecast based on historical weather, however in these cases the difference between the accuracy of the 0 day forecast and the other forecast lengths was not enough to rule out using short-term weather forecasts. Our analysis indicated that there was not sufficient evidence to suggest forecasts of up to 14 days do not on average cause the APSIM predictions to be too inaccurate to use. This means that 14 day length forecasts could be used for management decisions that require lead time, but a combination of all of the forecast lengths should be used to make final decisions.