A study of MODIS and AWiFS multisensor fusion for crop classification enhancement

Accurate, robust, timely and complete remote sensing-based crop classification results are critical to the mission of the National Agricultural Statistics Service (NASS), United States Department of Agriculture. However, due to cloud coverage and limited budget, in many cases, there are not enough quality AWiFS image data available for performing a reliable multitemporal crop classification. To solve this problem, extra image data from other sensors are sought for fusing with AWiFS images for temporal compensation while preserving the high spatial and spectral resolutions. This paper attempts to assess the crop classification accuracy enhancement with AWiFS and MODIS multisensor, multispectral and intertemporal fusion. Three different image fusion methods: principal component analysis (PCA), intensity-hue-saturation (IHS) and image band stacking (IBS) are applied to perform intertemporal image fusion between the 56m AWiFS and the 8-day composited reflectance MODIS data (Red and NIR bands only) with 250m resolution from NASA to incorporate more spectral dynamic information from MODIS images for better crop classification. To make the two-band MODIS data applicable to IHS fusion, this paper proposes a novel combined fusion process, in which the MODIS green band is replaced with the AWiFS green band to create a new multispectral image for IHS transformation. The fused image from AWiFS and MODIS images, together with the original AWiFS multispectral image, are then fed into the decision tree classifier for multitemporal crop classifications in accordance with different fusion methods and temporal combinations. The crop classification accuracies of various classification experiments are assessed with respect to different image fusion methods and different temporal combinations and compared with the reference single AWiFS classification results. The experimental results indicate that properly using the fusion of intertemporal MODIS and AWiFS data improves the crop classification accuracy in large crop area when enough fused temporal images are used.