Yield estimation of corn with multispectral data and the potential of using imaging spectrometers

In the frame of the Special Yield Estimation, a regular procedure conducted for the European Union to more accuratelly estimate agricultural yield, a project was conducted for the State Minister for Rural Environment, Food and Forestry of Baden-Wurttemberg (Germany) to test remote sensing data with advanced yield formation models for accuracy and timeliness of yield estimation of corn. The methodology employed uses field based plant parameter estimation from atmospherically corrected multitemporal/multispectral LANDSAT-TM data. An agrometeorological plant-production-model is used for yield prediction. Based solely on 4 LANDSAT-derived estimates (between May and August) and daily meteorological data the grain yield of corn stands was determined for 1995. The modelled yield was compared with results independently gathered within the Special Yield Estimation for 23 test fields in the Upper Rhine Valley. The agreement between LANDSAT-based estimates (six weeks before harvest) and Special Yield Estimation (at harvest) shows a relative error of 2.3%. The comparison of the results for single fields shows, that six weeks before harvest the grain yield of single corn fields was estimated with a mean relative accuracy of 13% using satellite information. The presented methodolgy can be transferred to other crops and geographical regions. For future applications hyperspectral sensors show great potential to further enhance the results for yield prediction with remote sensing.