Yield prediction of malting barley based on meteorological data

To estimate the import of malting barley from foreign markets a reliable prediction of grain yield is required by the malting industry. In the present study the feasibility to predict malting barley yield based on statistical yield and daily meteorological data as well as phenological data sets was tested for six counties in southwestern Germany. The data sets were available for the years 1987-2001. Additionally, NDVI data from NOAA satellite were obtained. The growing season was divided into the following phases: pre-sowing, emergence, tillering, stem elongation, grain filling and maturation. For each individual year and for each of these phases, parameters were calculated using daily maximum temperature, mean temperature, air humidity, potential evapotranspiration, global radiation and precipitation. The duration of the phases was also computed. Backward statistical calculations with all parameters for all phases are conducted for 1987-2000. Those parameters proven to provide yield estimates within a given confidence interval were used to construct an empirical model. The resulting model was used to predict the 2001 yield. The accuracy of the predictions varied between counties but the deviation from the reported yield never exceeded 15%. In further investigations with the premises of either an early stage or a "low effort" prediction, the meteorological parameters of the phases grain filling and maturation and, in the second case, those data causing high cost of obtaining as well as those being late available were not taken into account any more. Finally, NOAA-AVHRR NDVI maximum composite values were combined with meteorological parameters. Unfortunately, NOAA NDVI data were not available for all years. This required a reduction of the number of the selected meteorological parameters to enable proper statistical analysis. This approach yielded less accurate predictions as those based on meteorological data only. The deviation from the reported yield increased up to 22%. All parameters of all conducted calculations resulting in the empirical models are tested on their frequency of occurrence. This may allow a further reduction of the necessary parameters.