Fusion of Modis and Radarsat Data for Crop Type Classification — an Initial Study

Agricultural land use mapping and change detection are important for environmental assessment and crop yield estimation. An effective mapping and change detection of crop land use requires large image coverage with sufficient resolutions in spatial, spectral and temporal dimensions. But few earth observation sensors can solely collect image data that meet all these requirements. The MODIS (Moderate Resolution Imaging Spectroradiometer) satellite images provided by NASA are an optimal data source for collecting high resolution spectral and temporal information of the earth’s surface, because of its large coverage (2300 km ground swath), sufficient spectral bands (7 bands between visible and mid infrared at the spatial resolution of 250 to 500 m; and 28 bands between visible and thermal infrared at the spatial resolution of 1000 m), daily revisit rate, and low cost (free of charge). But its spatial resolutions (250 m, 500 m and 1000 m) are too coarse to delineate the crop field boundaries. On the other hand, the Radarsat images provided by the Canadian Space Agency are a good data source for obtaining high resolution spatial information (from 3 m to 100 m depending on the beam mode used) at very frequent repeat rate, because of its all-whether and day-night collection capability, low cost, and large coverage (from 50km×50km to 500km×500km depending on the beam mode). However, Radarsat images, like other radar images, are noisy and have limited spectral information. Even though field boundaries are recognizable in many beam modes, it is not effective to differentiate crop types just using a single radar image. To find a cost-effective solution for frequent, large coverage crop land use mapping, this paper presents an initial study on combination of low spatial resolution MODIS mulitspectral images and high spatial resolution Radarsat amplitude images for crop classification. The MODIS spectral bands 1 through 7 with a spatial resolution of 250 m and 500 m respectively and the Radarsat standard Mode with a spatial resolution of 12.5 m are used for the combination. A wavelet and IHS transform integrated technique was developed to effectively fuse (combine) the multispectral information from MODIS and the spatial information from Radarsat, to produce a image which contains spectral (colour) information of crop types and spatial (boundary) information of crop fields. Pixel-based and object-based classification techniques are then employed to classify the MODIS-Radarsat fused images for obtaining crop land use classes. The classification results demonstrate that pixel-based classification techniques fail to classify the crop types due to radar noise in the fused images, while object-based classification yields promising classification results.