Evaluating the reliability of time series land cover maps by exploiting the hidden Markov model

Time series land cover maps are important materials for the work related to land use and land cover change. Satellite remote sensing images prove advantageous in fast mapping with low cost. In most time series land cover products yielded by the satellite remote sensing images, a number of illogical transitions exist between different time phases. The time series land cover products cannot exactly reflect the real land cover types and land cover changes for each pixel. The accuracy evaluation based on the limited ground truth cannot well guide the users because the reliability of different pixels of the land cover products is unknown. A generic model for the reliability evaluation of time series land cover products should be developed based on a strong theoretical frame. In order to better guide the use of the land cover products, this paper proposed an approach to evaluate the reliability of time series land cover products by exploiting the joint probability of hidden Markov model (HMM), in which the classification performance and the spatio-temporal relationships were taken into account. We applied the proposed evaluation method on the time series land cover maps of Poyang Lake Eco-economic Region in China. The reliability of the land cover products was presented by the grading of the joint probability of HMM. The results effectively reflected the classification performance, the spatio-temporal relationships and even the quality of the data source.