Use of artificial neural networks for estimating crop acreage from MODIS data in large-scale

Population growth, urban expansion, and land degradation may place plant natural resources for food and agriculture at risk. Crop acreage monitoring is basic information necessary for wise management of these resources. Recent developments in remote sensing technologies have created promising opportunities for improving agricultural statistics systems. The Moderate Resolution Imaging Spectroradiometer (MODIS) is one detector board on Terra's (EOS-AM1), which was launched on December 18, 1999 by NASA. It offers a unique combination of spectral, temporal, and spatial resolution compared to previous global sensors, making it a good candidate for large-scale crop acreage estimating. However, because of subpixel heterogeneity, the application of traditional hard classification approaches to MODIS data may result in significant errors in crop area estimation, especially in China. This paper describes the application of an ANN (artificial neural network) classifier to differentiate different crops. A classic multilayer feedforward neural network with backpropagation algorithm was used throughout the experiment. In the experiment the model responds to subpixel class composition in MODIS data in Henan province of China. The approach with MODIS data estimates subpixel fractions of crop area based on the temporal signature of reflectance throughout the growing season. In the experiment a zone that can get LANDSAT/TM data was chosen to be the train dataset in the ANN. The paper assumes that the crop area estimating from LANDSAT/TM data is correct; in the training set the crop area based on MODIS data can be obtained from the classification results of LANDSAT/TM data. After the complication of training ANN, we can estimate the entire crop area based on the MODIS data in Henan province. Compared to national statistic data, the relative error of winter wheat's planting acreage is 4.2% in Henan province of China in 2002.