Search

Abstract: To date, industry and crop forecasters have had a good idea of the potential crop yield for a specific season, but early-season information on crop area for a shire or region has been mostly unavailable. The question of “how early and with what accuracy?” area estimates can be determined using multi-temporal Moderate Resolution Imaging Spectroradiometer (MODIS) enhanced vegetation index (EVI) imagery was investigated in this paper. The study was conducted for two shires in Queensland, Australia for the 2003 and 2004 seasons, and focused on deriving total winter crop area estimates (including wheat, barley and chickpea). A simple metric (), which measures the green-up rate of the crop canopy, was derived. Using the unsupervised k-means classification algorithm, the accumulated difference of two consecutive images (one month apart) for three EVI threshold cut-offs (, where , 500 and 750) at monthly intervals from April to October was calculated. July showed the highest pixel accuracy with percent correctly classified for all thresholds of 94% and 98% for 2003 and 2004, respectively. The differences in accuracy between the three cut-offs were minimal and the T500 threshold was selected as the preferred cut-off to avoid measuring too small or too large fluctuations in the differential EVI values. When compared to the aggregated shire data (surveyed) on crop area across shires and seasons, average percent differences for the for July and August ranged from −19% to 9%. To capture most of the variability in green-up within a region, the average of July and August was used for the early-season prediction of total winter crop area estimates. This resulted in high accuracy (R2=0.96; RMSE = 3157 ha) for predicting the total winter crop from 2000 to 2004 across both shires. This result indicated that this simple multi-temporal remote sensing approach could be used with confidence in early-season crop area prediction at least one to two months ahead of anthesis. [Copyright &y& Elsevier]