Agricultural monitoring is a major concern to economies largely based on agriculture like Uruguay. In order to improve crop yield forecasts, identification of crop types must be performed early in the planting season. However, this task is generally difficult because of the spatial heterogeneity of the landscape, the different crop cycles, the spectral similarity with grassland, and the inter-annual variability due to climatic events and fallow periods. In collaboration with INIA, this study investigates remote sensing methods for dynamic mapping of cropland areas and for producing a map of winter and summer crops at 250m using MODIS time series. The originality of the approach consists of: (i) exploiting all the multi-spectral information using an adaptive compositing method for a better discrimination of cropland types and to better capture their spatio-temporal variability; (ii) a spatio-temporal analysis of various land use types prior to the classification for a better knowledge of crops behaviours and the selection of the most discriminating seasons in the classification; and (iii) combining NDVI profiles, multi-spectral composites with reference dataset, high resolution images and expert knowledge. The accuracy of the product is assessed based on a reference dataset of crop fields. The results confirm the relevance of MODIS time series in term of spatial and temporal resolutions for mapping cropland areas and characterizing the inter-annual variability. Thanks to a good reference dataset and an analysis of crops spectro-temporal behaviour, it was possible to identify cropland areas from other land use types and discriminate summer crops from winter crops.
- Contributions to Conferences
Second International Conference on Agro-Geoinformatics p. 292-295