Available at: https://digitalcommons.calpoly.edu/theses/1668
Date of Award
MS in Engineering - Water Engineering
Bioresource and Agricultural Engineering
Ground truthing actual crop types in an area can be expensive and time-consuming. The California Department of Water Resources attempts to ground truth land use in each county in California every five years. However, this is limited by budgetary constraints and often results in infrequent (more than every ten years) surveying of many counties. An accurate accounting of crops growing in a region is important for a variety of purposes including farm production estimates, groundwater and surface water modeling, evapotranspiration estimation, water planning, research applications, etc. Agricultural land use is continually changing due to development and environmental factors.
Currently, USDA NASS provides georeferenced land use maps of regions throughout the U.S. While these are beneficial, the accuracy is not very high for California due to the wide variety of crops grown throughout the state. California has an increasingly complex agricultural system which includes multi-crops changing on an annual and even semiannual basis, long growing seasons, and complex and flexible irrigation schedules.
Remotely sensed data from available satellites are used to more accurately classify crop types within the Madera and Merced Counties of California’s Central Valley. An initial classification approach utilizing a simplified decision tree for a data subset of the area considered is presented. In order to accommodate the larger dataset at hand, a computer based approach is applied using the Nearest Neighbor classification algorithm in the computer program eCognition. Iterative analyses were performed to consider a range of scenarios with varying spectral inputs. The results show the methods presented can be beneficial in discriminating 24 of the major crop types from multi-temporal spectral data.