Using empirical mode decomposition to analyze time-series MODIS data for monitoring rice phenology

Monitoring crop phenology using temporal remotely-sensed data provides necessary information for evaluating crop productivity and crop management. The main objective of this research is to access the utilization of time-series MODIS 250-m data by empirical mode decomposition (EMD) for monitoring rice crop phenology in the Vietnamese Mekong Delta for the year 2007. There are three main procedures in data processing: 1) the time-series MODIS NDVI data for the year 2007 was constructed. Noise present in time-series data was filtered with the EMD method; 2) a crop phenology detection method was developed for determination of phenological dates (sowing, heading and harvesting dates) of rice cropping patterns. Based on smooth NDVI profile, we first identified the heading date using the local maximum algorithm. From the estimated heading date and analysis of agricultural practices in the study, we then established proper thresholds to determine sowing and harvesting dates. The rice phenological patterns were defined based on these estimated phenological dates; and 3) the results were eventually compared with the 2007 field investigation data. The initial findings indicate that the root mean square errors (RMSE) calculated for the estimated phenological dates and field survey data are 8.5 for the sowing date and 9.6 days for the harvesting date, respectively. The results confirm that EMD acts a well-fitted filter for noise reduction of the time-series MODIS NDVI data, which can be applied for determination of phenogical stages of rice crops in the region.