Phenology-based crop classification algorithm and its implications on agricultural water use assessments in California's Central Valley

Abstract

The overarching goal of this study was to map specific crop types in the Central Valley, California and estimate the effect of classification uncertainty on the calculation of crop evapotranspiration (ETc). A phenology-based classification (PBC) approach was developed to identify crop types based on phenological and spectral metrics derived from the time series of Landsat TM/ETM+ imagery. Phenological metrics, calculated by fitting asymmetric double sigmoid functions to temporal profiles of enhanced vegetation index (EVI), were capable of separating crop types with distinct crop calendars. An innovative method was used to compute spectral metrics to represent crops' spectral characteristics at certain phenological stages instead of any specific imaging date. Crop mapping using these metrics showed a stable performance without influences of low-quality data and inter-annual differences in imaging dates. The requirement for ground reference data by the PBC approach was low because classification algorithms were mostly built according to the knowledge on crop calendars and agricultural practices. Techniques including image segmentation, data fusion with MODIS imagery, and decision tree were incorporated to make the approach effective and efficient. Though moderate accuracy (~65.0 percent) was achieved, ETc calculated by the Food and Agriculture Organization (FAO) 56 method showed that the estimate of water use was not likely to be significantly affected by the classification error in PBC. All these advantages imply the strength of the PBC approach in the regular crop mapping of the Central Valley. © 2012 American Society for Photogrammetry and Remote Sensing.