EARSeL eProceedings Vol. 3, No. 3, 347-353, 2004

Abstract
Yield forecasts are of high interest to the malting and brewing industry in order to allow the most convenient organisation of the respective purchasing policy of raw materials. Within this investigation, malting barley yield forecasts (Hordeum vulgare L.), in Germany mostly grown as spring barley, are performed for typical growing regions in South-Western Germany. Multitemporal remote sensing data on the one hand and ancillary data such as meteorological, agrostatistical, topographical and pedological data on the other hand are used as input data for two versions of prediction models, which were based on an empirical-statistical modelling approach.

Since spring barley production is dependent on acreage and on the yield per area, classification is needed, which was performed by a supervised multitemporal classification algorithm, utilizing optical remote sensing data (LANDSAT TM/ETM+). The classification algorithm considers spectral data, topographical data (Digital Elevation Model) and expert knowledge input. The latter is important with regard to the particular phenological development of the observed crop, an expertise which was used to distinguish it from similar crops.

The basic version of the yield estimation model was conducted by means of linear correlation of remote sensing data (NOAA-AVHRR NDVI Maximum Value Composites), CORINE land cover data and agrostatistical data. In an extended version meteorological data (temperature, evapo­transpiration) and soil data were incorporated. Both basic and extended prediction systems led to feasible results depending on the selection of the time span for NDVI accumulation. For NDVI accumulation across the grain-filling period, the mean deviation of the reported yield from the simulated one was 7.0% and 6.4% for the basic and extended yield estimation model, respectively.

ISSN 1729-3782