Bayesian Classification for Rice Paddy Interpretation

In this paper, we present a case study of interpreting paddy distributions of three counties on Northern Taiwan during two crop seasons on year 2000 using multitemporal imageries together with cadastre GIS by Bayesian posteriori probability classifier. In order to integrating Bayesian conditional probability, priori probabilities of paddy's attributes were estimated from photogrammettric interpretation results provided by the Food Bureau, and the spectrum reflectance from different growth stages was used. Due to the spatial heterogenous of paddy's distribution, classifier parameters were established individually on each map-quadrangle. Temporal change of NDVI from different growth stages pass through rice's life cycle has been measured and we find two-stage images make significant improvement on classification results. Results of the study help us to evaluate the accuracy of the classifier. Imagery classification results were compared with aerial photo's interpreting results for assessing accuracy. Overall accuracy of first crop of Tao-yuan, Hsin-chu, and Miao-li were 89.93% 92.83% 95.33% respectively. Bayesien classifier has advantages including easy-to-adjusted and easy-to-computed rules and comparative stable results when limited SPOT satellite imageries available. Bayesin method also provides results with probability that help the operator to assess the places having least confidence. These advantages allow us to suggest Bayesian method be used in paddy-area investigation in Taiwan.