Comparison of Alternative Crop Phenology Detection Algorithms using MODIS NDVI Time Series Data in US Corn Belt Region

Predicting crop phenology is important for understanding of crop development and growth processes and improving the accuracy of crop model. Remote sensing offers a feasible tool for monitoring spatio-temporal patterns of crop phenology in region and continental scales. Various methods have developed to determine the timing of crop phenological stages using spectral vegetation indices (i.e. NDVI and EVI) derived from satellite data. In our study, it was compared four alternative detection methods to identify crop phenological stages (i.e. the emergence and harvesting date) using high quality NDVI time series data derived from MODIS. In threshold method assumes the emergence and harvesting date when NDVI values exceed and decreases down to a given threshold, respectively. Two kind of threshold values were applied for NDVI and it increment for eight days. The other two methods use a logistic fitting model and inflection points on fitted curve, respectively. It was compared the four methods for corn and soybean, respectively. For validation, three kinds of datasets were utilized: AmeriFlux biological data of planting and harvest dates, and emergence date estimated from growing degree days (AGDDs) at flux tower sites, and state-level USDA Crop Progress Report (CPR). All methods showed substantial uncertainty but the threshold method showed relatively better agreement against with both site- and state-level data for soybean phenology. For better NDVI-based regional estimation of crop phenology, factors of uncertainty were examined and discussed in this study.