Developing Methods For Vegetation Change Mapping In The Purple Mountains From High Spatial Resolution Cross-sensor Remotely Sensed Images

Accurate understanding of vegetation changes taking place in the Purple Mountains,including the location,area,and change direction(vegetation gain or loss)can provide reliable data support for scenic area management and resources allocation.The development of high spatial resolution remote sensing images makes change detection more subtle.The major objective of this study was to map and quantify vegetation changes in the Purple Mountains in Nanjing City,using multi-temporal cross-sensor high spatial resolution remote sensing images to further identify the main drivers of the vegetation changes and to provide a valuable reference for sustainable management.Quickbird images acquired on July 2004,IKONOS images acquired on June 2009,WorldView-2 images acquired on July 2015 and GF-2 images acquired on April 2018 were used as the data source.These images were first fused to adequately combine the high spatial information from panchromatic band and the spectral information from multispectral bands.Based on the fused images,four pixel-based direct change detection methods including the normalized difference vegetation index(NDVI)difference method,the multi-index integrated change analysis(MIICA),the principal component analysis(PCA)and the spectral gradient difference(SGD)analysis were performed to detect vegetation changes with change thresholds determined by objective receiver operating characteristic(ROC)curve method.The stratified random sampling was used for accuracy verification,thus the best pixel-based change detection method could be determined for the subsequent object-oriented vegetation change detection to reduce salt-and-pepper noise in detection results.And in object-oriented change detection,image segmentation was performed with optimal segmentation scales determined by the precision(P)and the recall rate(R)metrics.The segmentation images were then used as the input data,in partnership with the best pixel-based direct change detection method,an object-oriented change detection analysis was performed.The accuracies of the results derived from the best pixel-based detection method and the object-oriented change detection method were compared.Accordingly,the optimal change detection method was finalized to extract multi-temporal vegetation change information in Purple Mountains.Finally,the perpendicular impervious index(PII)was used to extract impervious surfaces.Combined with vegetation change detection results,the vegetation changes were mapped and the quantitative parameters such as the change area and proportion were calculated to analyze vegetation change status and the main driving factors.The results showed that:(1)MIICA outperformed in the vegetation change detection among four pixel-based direct change detection methods,and its overall accuracy and Kappa coefficient were 0.884 and 0.804,respectively,thus it was determined for the subsequent object-oriented vegetation change detection.Further,the object-oriented MIICA detection with an overall accuracy of 0.902 and a Kappa coefficient of 0.836 was superior to the pixel-based MIICA,which largely reduced the impact of salt-and-pepper noise on the detection results to some extent.(2)In the analysis of vegetation change situations in Purple Mountains,from 2004 to 2009,the proportion of vegetation gain was low(0.92%)and the proportion of vegetation loss was high(2.16%).The main cause of vegetation change was the demand for tourism development.From 2009 to 2015,due to the reduction of the available land for development and utilization in the periphery of Purple Mountains combined with the enhancement of environmental protection and greening measures,the vegetation gain rate increased at 1.36%,while the proportion of vegetation loss decreased by 0.67%.Despite this,there were still obvious addition and expansion of constructions and roads.And in 2015-2018,the proportion of vegetation gain and loss in Purple Mountains were 0.92% and 1.21%,respectively.The vegetation changes circumstances during this period were relatively stable,which was less influenced by the change of impervious surfaces.Vegetation change detection results indicated that,although the degree of vegetation loss showed a downward trend,the conversion from vegetation to impervious surface still needs to be noted by managers.Relevant departments should strengthen the environmental awareness education actions,reinforce multi-agency administrative law enforcement inspections regarding natural resource utilization,standardize the approval of addition and expansion of the constructions,and coordinate the relationship between economic benefits and ecological benefits in the forest management process.