Research On Local Accuracy Estimation For Land Cover Change Information

Accuracy assessment or validation is becoming increasingly important with the accumulation of land cover change information.Accuracy assessment has entered the development stage featuring a shifting from global into local and from static land cover information into dynamic land cover change information.The research reported here is very timely.Even though the general methods for the global and the local accuracy evaluation of static land cover information is relatively well established,there are still numerous difficulties to overcome in accuracy validation of land cover change information.In this paper,two localized accuracy estimation methods of land cover change information are proposed based on summarizing the basic principle and methods of accuracy estimation of land cover change information,and their feasibility and effectiveness have been validated with real data.The result is very valuable for studies and applications in national geographical conditions monitoring and land resources inventory,etc.The main research work is summarized as the following:(1)The method to estimate the accuracy of land cover change classification based on logistic regression model.In this method,the landscape indexes with statistical significance of ground truth reference data are selected as explanatory variables to construct logistic regression model to estimate local accuracy,otherwise known as per-pixel accuracy estimation.This result would provide support for accuracy evaluation and spatial analysis.In this paper,two experimental schemes are designed in accordance with two ways to choose explanatory variables:the direct scheme uses the landscape indexes of land cover change classification map as explanatory variables of logistic regression model,and indirect scheme selects the landscape indexes of bi-temporal land cover classification maps as explanatory variables of logistic regression model.The experiment results indicate that there is no significant difference between two schemes,but the computation efficiency of direct scheme is higher,which is recommend to use.(2)The accuracy estimation method of land cover change based on maximum posterior probability.This method makes full use of training sample data and selects maximum posteriori probability as an indirect index to estimate the classification accuracy(i.e.,the classification of change category)through correction method based on regression analysis.In the process,single data image is performed tasseled cap transformation and kernel density estimation,and the expectation maximization algorithm is used to determine the prior probability of change category so as to perform Bayesian classification on bi-temporal images.Maximum posteriori probability of change category is chosen as the indirect index for the probability of correct classification.Some reference sample data are used to calculate the linear correction coefficients,and then the accuracy estimation results after calibration are obtained.