Page Banner

United States Department of Agriculture

Agricultural Research Service

Research Project: USING REMOTE SENSING & MODELING FOR EVALUATING HYDROLOGIC FLUXES, STATES, & CONSTITUENT TRANSPORT PROCESSES WITHIN AGRICULTURAL LANDSCAPES Title: Crop classification in the U.S. Corn Belt using MODIS imagery

Authors
item Doraiswamy, Paul
item Akhmedov, Bakhyt - SSAI
item Stern, Alan

Submitted to: International Geoscience and Remote Sensing Symposium Proceedings
Publication Type: Proceedings
Publication Acceptance Date: May 20, 2007
Publication Date: July 23, 2007
Citation: Doraiswamy, P.C., Akhmedov, B., Stern, A.J. 2007. Crop classification in the U.S. Corn Belt using MODIS imagery. In: Proceedings of the International Geoscience and Remote Sensing Symposium (IGARSS), July 23-27, 2007, Barcelona, Spain. 2007 CDROM.

Technical Abstract: Land cover classification is essential in studies of land cover change, climate, hydrology, carbon sequestration and yield prediction. Land cover classification uses pattern recognition technique that includes supervised / unsupervised approaches and decision tree technique. Land cover maps for regional scale are usually created using high resolution images such as Landsat TM, SPOT, and IRS. Because of infrequent data acquisition and cloud contamination, the operational using of high resolution imagery is difficult. The potential for using NASA Moderate Resolution Imaging Spectrometer (MODIS) sensor at 250 m resolution was investigated for USDA’s operational programs. The annual crop yield assessment program of the National Agricultural Statistics Agency of USDA requires classification of crops for acreage estimation and crop condition for the entire country. The MODIS sensor provided daily global coverage with the reflection data is provided on-line with some preliminary processing and also other landcover related products. Errors caused by atmospheric conditions, geo referencing and BRDF can sometimes cause large errors and therefore individual bands cannot be used in a standard manner for performing classification. Use of normalized difference vegetation indices (NDVI) for classification reduces errors caused by atmospheric conditions. This research was conducted over Iowa and Illinois to classify the corn and soybean crop. A non-crop mask was generated from earlier Landsat classified imagery to eliminate the non-crop area when using the MODIS imagery. Multi-temporal 8-day composite 250 m resolution surface reflectance product time series were used to generate the NDVI data which was used to differential between corn and soybean crop in the U.S. Corn Belt. The NDVI time series data for 2002 and 2005 from May 1 to November 1 was were used in this study since we had Landsat classification for these two years. The data was reprocessed to minimize cloud contamination and other factors that reduced the values. To eliminate these errors, the Savitzky-Golay filtering method was used to track the upper envelope of the NDVI seasonal profile. The filter was applied to each pixel using a moving window and approximated by a second order polynomial. The data was then applied to a phenology-based decision tree to separate the corn and soybean crop. The results of the MODIS-based classification were compared with the Landsat-based classification for the two year period. The over all classification accuracy for Iowa was 82-85% for the two years and an accuracy of about 79% for Illinois. In conclusion, this method has been used successively during the past 2 years to develop crop classification and products for crop condition and potential yield maps for Iowa and Illinois.

Last Modified: 11/26/2014
Footer Content Back to Top of Page