| With the acceleration of urbanization and economic globalization,the improvement of people's living standards is accompanied by many social and environmental problems.As the basic data reflecting the urban spatial pattern,development change and an important index of urban planning,environmental monitoring and protection,the impervious surface data has been widely concerned by scholars at home and abroad.The traditional research on impervious surface mainly focuses on classification mapping,ignoring the common problem of mixed pixels in the middle resolution remote sensing images,and the extracted data is still significantly different from the actual distribution of impervious surface.In order to describe the actual distribution and change of impervious surface more accurately,the precision inversion of sub-pixel impervious surface percentage(ISP)has become a research hotspot.At the present stage,the Landsat satellite images of the United States was mostly used in the percentage inversion of impervious surface,and few researchers use Chinese satellite images for research,lacking the full utilization and application potential evaluation of Chinese satellites.Therefore,this research would carry out the sub-pixel impervious surface percentage inversion based on the Chinese high resolution satellite(GF-1 and GF-2)images and the environment and disaster mitigation satellite(HJ-1A/B)images,evaluate the application potential of Chinese satellite images in the precision inversion of impervious surface,and build a single time phase and time series impervious surface precision inversion model based on Chinese satellite images.In this paper,the accuracy and effectiveness of common machine learning inversion algorithms were compared,the advantages and disadvantages of each algorithm were summarized,and the optimal single-phase impervious surface precision inversion method was confirmed.Then,based on the optimal single time inverse algorithm,a time series impervious precision inversion model was built.Finally,the optimal single-phase impervious surface percentage model was used to retrieval the impervious surface accurately in large areas,and the adaptability of the model under different terrain and environmental conditions was tested.Through comparison,analysis and experiment,the following conclusions were obtained:(1)The three machine learning algorithms used in this study had no significant difference in accuracy:RMSE of SVM model was 17.46%,RMSE of ANN model was 17.43%,and RMSE of RF model was 15.78%.In contrast,RF algorithm was slightly more accurate.Based on the analysis of the inversion results,the following conclusions were drawn:?RF algorithm could more accurately reflect the spatial distribution of the percentage of impervious surface,while SVM algorithm and ANN algorithm were significantly underestimated.?RF algorithm was not as effective as SVM algorithm and ANN algorithm in the recognition of bare soil and bare rock.?RF algorithm had the best recognition effect on road network,followed by ANN algorithm,while SVM algorithm had the worst recognition effect on road.In general,RF algorithm had the best inversion accuracy and effect,and the process of model parameter setting and optimization was relatively simple.(2)Using GF-1/2,HJ-1A/B images and random forest algorithm,the precision inversion model of time series impervious surface based on multi-temporal remote sensing images was built and was confirmed that it had a high accuracy.Using this model to estimate the percentage of impervious surface in Gwadar city from 2009 to 2017,RMSE of the model was 12.82%to 16.03%,which was close to the accuracy of domestic and foreign studies on impervious surface percentage.This research had proved that the multi-source Chinese satellite images could be effectively used for the inversion of impervious surface percentage.HJ-1A/B satellite had great application potential in the dynamic monitoring of impervious surface percentage due to its advantages of high time resolution.(3)The single-phase impervious surface percentage inversion model based on random forest algorithm was used to obtain the impervious surface percentage data of the whole region of the China-Pakistan Economic Corridor(CPEC),which proved that the impervious surface percentage inversion model built in this study had better inversion effect under different terrain and environmental conditions.At the same time,the study made full use of the advantages of the environment and disaster mitigation satellite in terms of its large width and high temporal resolution,and selected 6 scenes of the same day to realize the proportion inversion of impervious surface in key cities in Pakistan,which proved the application potential of Chinese satellites in impervious surface inversion once again. |
PDF Full Text Request