Hidden Markov Models for corn progress percents estimation in multivariate time series

Crop development information is critical to U.S. agricultural economy and decision making. In this paper, a general framework of Hidden Markov Models (HMMs) based corn progress percents esitmation method has been presented. Multivariate time series involving mean NDVI, fractal dimension, and Accumulated Growing Degree Days (AGDDs) are embedded into the modified HMM. Features of mean NDVI and fractal dimension are derived from MODIS (Moderate Resolution Imaging Spectroradiometer) NDVI (Normalized Difference Vegetation Index) time series, and AGDDs is from Automated Weather Data Network (AWDN). In our model, stage transition probabilities and observation probabilities are determined directly from supported data. Stage transition probabilities are corrected by AGDDs. Probability density function associated with three continuous features of each stage is modeled by multivariate Gaussian. It is worth mentioning that not only progress stages at a specific time slice is detected, but the proportion of corresponding progress stage can be estimated, simultaneously. Experimental studies have been conducted on state of Iowa, over a decade period (2002 through 2011) with assessment and validation by NASS's (National Agricultural Statistics Service) CPRs (Crop Progress Reports). The results demonstrate the feasibility of proposed solutions on corn progress percents estimation in the state-level.