LANDSCAPE-BASED CROP MANAGEMENT FOR FOOD, FEED, AND BIOENERGY
Location: Cropping Systems and Water Quality Research
Title: Watershed-scale land-use mapping with satellite imagery
Submitted to: Proceedings of the American Society of Agricultural and Biological Engineers International (ASABE)
Publication Type: Proceedings
Publication Acceptance Date: June 25, 2012
Publication Date: July 29, 2012
Citation: Nguyen, A.T., Thompson, A.L., Sudduth, K.A. 2012. Watershed-scale land-use mapping with satellite imagery. In: American Society of Agricultural and Biological Engineers International Meeting Technical Papers. July 29 – August 1, 2012, Dallas, Texas. Paper No. 121338314.
Interpretive Summary: Soils of the Salt River Basin in Northeast Missouri are predominantly claypan soils that are poorly drained and have high runoff potential. The claypan soils are especially vulnerable to soil erosion, which has degraded soil and water quality throughout the basin, and to surface transport of herbicides. Because of this, the Salt River Basin was selected for the study of the effects of best management practices on water quality. Much of this research relies on computer modeling, which in turn requires information on spatial distribution of different crops. In this project, we used NASA’s Landsat 5 and MODIS remote sensing data to create crop type maps for the Salt River Basin in 2005 and 2006. Because few cloud-free Landsat images were available in those years, standard image analysis approaches were not successful and an innovative concept which combined the two satellite data sources with weather information was developed. Using this new approach, the accuracy of crop type discrimination was over 90 percent for both years, sufficient for water quality modeling efforts. This research will benefit environmental scientists wishing to investigate watershed-scale processes, as it provides them with a new method to obtain a portion of the input data needed for their environmental analysis models.
Satellite remote sensing data has many advantages compared with other data sources, such as field methods and aerial photography, for land cover classification. In particular,it is useful in evaluating temporal and spatial effects. In addition, remote sensing can offer a cost-effective means of providing more frequent crop/land cover data for short and long term planning. This research studied the application of Landsat 5 Thematic Mapper (TM) data to build crop type maps for the Salt River Basin in northeastern Missouri over a two-year period (2005 and 2006) for use in evaluating land-use management. MODIS Normalized Difference Vegetation Indices (NDVI) were used to generate crop phenological behaviors. This method analyzed both the unsupervised class from Landsat images and NDVI values derived from Landsat based on the MODIS NDVI threshold selection. Decision tree analysis, a combination of pixel and object based methods, was applied. In addition, statistical analysis on NDVI was used to build a regression model to predict Landsat NDVI from MODIS NDVI as an aid to define rules in decision tree analysis. Ground truth data in the Goodwater Creek sub-basin and Greenley, MO reference area were used to assess the accuracy of the procedure. Results from this method showed good agreement with limited cloud-free Landsat images during the growing season. The method gave an overall accuracy for the Salt River Basin of 94.1 percent in 2006 in classifying forest, pasture and crops, including soybean, corn, grain sorghum, winter wheat and double crop, when there was a lack of cloud-free images during the growing season. The overall accuracy results in 2005 for the two reference sites, the Goodwater Creek and the Greenley reference area, were 93.0 percent and 92.5 percent, respectively. These results indicate that a decision tree can be a useful method for crop separation if a single cloud-free Landsat 5 scene can be taken during the optimum crop discrimination period for a given area with the analysis of weather constraints and crop NDVI phenological behaviors if the decision tree is available from a previous year’s training.