Urban Forest Based On Multi-Source And Multi-Temporal Remote Sensing

Urban forests are an important component of urban ecosystems.They can effectively maintain the carbon-oxygen balance within cities and promote harmonious coexistence between humans and nature.Therefore,fully understanding the coverage and changes of urban forests is crucial to maintaining and improving the urban ecological environment,and ensuring the sustainable development of urban forests.As a recent research hotspot in the field of remote sensing,multi-source and multi-temporal remote sensing data integration provides the possibility for high-precision urban forest extraction.However,there are many problems in the current urban forest information extraction,such as interference and redundancy of multiple feature information,fragmentation of extraction results,and ignoring geographic spatial factors,which pose great challenges to the application of multi-source and multi-temporal remote sensing data in urban forest information extraction.This article selected Nanjing city,which have relatively high forest vegetation coverage in the Yangtze River Delta region,as typical study areas,and explores high-precision urban forest extraction and change detection methods based on multisource and multi-temporal Sentinel-1 and Sentinel-2 remote sensing data.The main research content and conclusions are as follows:(1)To solve the problem of information redundancy and interference caused by mutual influence of multi-source and multi-temporal remote sensing features,an improved artificial bee colony feature selection algorithm(ABC-LIBSVM)was proposed.Based on the LIBSVM support vector machine library,the fitness in the ABC algorithm was constrained to solve the problem of over-fitting in traditional swarm intelligence algorithms.Then,a random forest classifier was used to extract urban forests from the selected optimal feature set.The proposed method not only improved the feature selection rate but also achieved high accuracy in land cover extraction and classification.The forest extraction accuracy was as high as 98.06%,the overall classification accuracy of land cover in the study area was 86.20%,and the Kappa coefficient was 0.8116.This experiment verified the importance of screening multi-source remote sensing features and proved the effectiveness of the proposed feature selection method in urban forest extraction and classification.(2)Pixel-based analysis and classification methods are prone to "salt and pepper" phenomena,and the deeper information in the optimal feature sequence needs to be further explored.Based on the ABC-LIBSVM feature selection method,an object-oriented optimal feature weighted attribute sequence similarity analysis and classification method was proposed.This method can flexibly find the best matching point between feature sequences and create a weighted attribute sequence while considering relevant feature attributes such as spatial distribution and texture of urban forests.The experimental results showed that compared with the ABC-LIBSVM method,this method can obtain better urban forest and other land cover classification results,with a forest extraction accuracy of up to 90.02% and an overall classification accuracy of 83.53%.In addition,in areas with high forest cover density,some phenological parameters such as the accumulation of biota during the growing season were of great help to high-precision forest extraction.(3)Due to the extreme reliance on classification accuracy and the focus only on temporal changes of pixels or objects in traditional post-classification change detection methods,it is easy to overlook changes in geographic space.Therefore,this study adopted the classification method from a previous study to obtain the urban forest extraction results for two years,2018 and 2022,in Nanjing City.Then,based on street vector data,a self-organizing map neural network model(SOM)was used for secondary clustering of the classified results for the two years to obtain a land structure type map at the street scale.Finally,the traditional change detection results were combined for comprehensive analysis.The experimental results showed that the forest vegetation area and street land structure type in the central urban area of Nanjing remained stable between2018 and 2022,while the land structure type of the edge streets changed from farmland/forest to farmland/building.In addition,influenced by policies,the increase in forest area in Nanjing mainly concentrated around some scenic areas.This research provided technical support for urban forest resource monitoring,and provided scientific basis for maintaining the forest ecological environment in the Yangtze River Delta region,stabilizing urban carbon cycling,and protecting biodiversity.