Time-space Process Fusion Method For Land Cover Simulation In Mining Area

Under the direct and indirect influence of multiple driving factors,land cover status in mining area,in terms of time series,show its diversity in volatility pattern and the stability in status,whereas,in terms of space,a high uncertainty.The lack of process information and the existence of simulation rules such as linear assumption and overfitting in current simulation model limit the application of the model in land cover simulation in mining area.Therefore,this thesis assumed that land cover change is composed of the inertial evolution and the spatial self-organization process.Based on above hypothesis,a novel land cover simulation model composed of inertial evolution and spatial self-organization process was developed to adapt to the complex land cover change simulation.The main work of the study is as follows:(1)The PSTV selective spatio-temporal fusion algorithm was developed to generate NDVI time series with 30 m spatial resolution at 16-day intervals between 2000 and 2016,which was regarded as basic data for the inertial evolution simulation of land cover.The results showed that the NDVI time series constructed by proposed fusion algorithm can effectively retain the abrupt information in land cover change and high phenological stability in mining area.(2)On the basis of NDVI time series,the inertial evolution simulation model of land cover was proposed by combining the BFAST-Monitor algorithm and SVM algorithm.The results showed that the overall accuracy and Kappa coefficients of the model were 62.70% and 49.94%,respectively.The model is suitable for the simulation of land cover types which have relatively stable space and time status.(3)In order to solve the problem of low simulation accuracy of traditional RF-CA algorithm,caused by the assumption of static conversion rules and the lack of area control factor,the FU-RF-CA model was developed by introducing feature update module and area control module into traditional RF-CA algorithm to simulate land cover of the spatial self-organizing process.The results indicated that the FU-RF-CA model can effectively improve the simulation accuracy.The OA and Kappa coefficients of this model were 75.36% and 66.49%,respectively.The FU-RF-CA have demonstrated a great ability than Logistics-CA and MLP-CA in land cover simulation in mining area,and it improved the OA and Kappa respectively by 3.93% and 4.73%,3.8% and 4.51%,when compared with Logistics-CA and MLP-CA.(4)On the basis of integrated kalman filter algorithm(En KF),an assimilation simulation model was proposed by integrating the sequence inertial evolution of land cover and spatial self-organizing process.The results showed that the OA and Kappa coefficients of proposed model were 79.84% and 71.61%,respectively,with improvements of 17.14% and 21.67%,4.48% and 5.12%,when compared with the single use of sequence inertial evolution model and the single use of FU-FR-CA model.