摘要：Time series of coarse resolution imagery offer the advantage of free global coverage but have to deal with mixed pixels. The study uses neural nets as modelling tool for sub-pixel crop acreage estimation. Nets are trained with reference crop acreage information derived from 30 m Landsat images and CORINE LC map for interpreting changes in the shapes of coarse resolution AVHRR NDVI profiles. Using official AGRIT statistics for Tuscany (Italy) as reference information, the network portability across years was evaluated. Using 3 images acquired before 2002 nets were trained. Subsequent application to 2002–2009 data explained roughly half of the inter-annual variance.
