Study On Impervious Surface Change Based On Time Series Optical Remote Sensing Images

As an important feature of urban areas,the Impervious Surface has become the most intuitive indicator of urbanization and is also closely related to the urban ecological environment.The rapid development of the city has accelerated the expansion of the impervious surface area,which has had an important impact on the regional ecological environment and also restricted the sustainable development of the region.Remote sensing technology has become an important means of studying the changes of impervious surface due to its advantages of low cost,large range,high spatial and temporal resolution.The acceleration of the urbanization process has made the impervious surface information cover a wide range and change rapidly.The traditional manual interpretation and index model extraction methods have a large workload and slow update speed,which is far from meeting the needs of modern urban management and decision-making.This paper focuses on the monitoring and analysis needs of the dynamic change of the impervious surface,relying on long-term optical satellite remote sensing data,and taking Tianjin as the experimental area to study the extraction method of the continuous change of the impervious surface and discuss the spatial and temporal distribution of its urban expansion.Excavating the driving factors of urban expansion,the results are of great significance for urban utilization planning(infrastructure construction,sustainable development of cities)and urban environmental assessment.The main work and research results completed in this paper are as follows:(1)Using cloud and cloud shadow detection algorithm and BRDF correction model,a continuous time series remote sensing data set is constructed.Mainly collected all Landsats 4-8 images with a cloud cover less than 80% available in Tianjin from 1985 to 2018,corrected the BRDF effect of the data using the RossThick-LiSparse-R BRDF model,and then applied cloud and cloud shadow detection algorithms FMask 3.3 performs cloud and cloud shadow detection on all Landsat data,and finally constructs a continuous time series remote sensing data set in the study area.Then,with the help of Google Earth's high spatial resolution historical images,simple random sampling and single-pixel time series curve comparison methods were used to quantitatively and qualitatively evaluate the results of change detection.The results show that the use of BRDF corrected LTS(Landsat Time Series)can significantly improve the change detection results,the overall accuracy is improved by 9.00%(84.5% VS 93.5%),and can reduce the leakage error(8.72%)and error Minute error(10.21%).In addition,the results show that the difference between the change detection results before and after BRDF correction is mainly distributed in the overlapping area between different Landsat scenes.(2)The linear density spatial feature extraction and the fusion classification of time,spectrum and spatial feature are realized,and then the effect and accuracy of different classification schemes are compared through verification experiments.The Continuous Change Detection and Classification(CCDC)algorithm was used to construct a time series model to monitor continuous land cover changes in the study area.The linear density spatial feature is extracted by using a combination of unsupervised classification model and morphological method.Subsequently,a random forest classifier is constructed by integrating time,spectrum and space characteristics,and continuous land cover classification is carried out.Finally,the time series data sets of impervious surface and land cover in the study area are extracted.Subsequently,using the high spatial resolution historical images of Google Earth and the visually interpreted sample data,the training verification experiment was repeated 20 times using the cross-validation method.The results show that compared with the original CCDC algorithm,the combined use of BRDF correction and linear density spatial features can improve the overall classification accuracy of 1.49%(88.91% VS.87.42%),and can achieve the relatively highest overall accuracy and classification effect;if only considering with BRDF correction or linear density,the overall accuracy is improved by 1.08% and 0.48%,respectively.Among the classification results after fusing the linear density,the misclassification error on the impervious surface is reduced by 1.70%,and the missed error is reduced by 1.49%.(3)The best results were used to establish the time-series data sets of land cover and impervious surface in Tianjin,and the spatial and temporal distribution of urban expansion was explored.At the same time,the driving factors for urban expansion were mined in combination with socio-economic data.First,the spatial and temporal distribution changes and land type transformation of the impervious surface in the study area are analyzed.The results show that: Tianjin has experienced rapid urban expansion in the past few decades.The urban expansion distribution and land type transformation mainly took 2005 as the time node.The main subjects before and after conversion are farmland in the central urban area and water body in the coastal area;in addition,the weighted average center and the standard deviation ellipse are used to analyze the urban expansion trend of Tianjin.The results show that the urban expansion as a whole has continuously moved about 3787 meters to the southeast coast,with obvious directionality,and the degree of urban expansion has gradually increased;finally,combined with socioeconomic data,correlation analysis was used to mine Tianjin urban expansion The main driving factor.The results show that the main driving factors include economy(GDP),population(resident population)and policy.The correlation coefficient between economy and impervious surface area is 0.948,and the correlation coefficient between population and it is 0.974.