Study On Finer Mapping Of Wetlands Based On High Temporal And High Spatial Resolution Data

As one of the three most important natural ecosystems on earth,wetlands play a key role in the global carbon cycle.However,as a result of excessive human activities,a large number of wetland ecosystems have been reported to degrade and shrink rapidly over the past few decades.There is an urgent need to monitor the status of wetland ecosystems,which also calls for precise delineation of wetlands.Being in a transitional location between terrestrial and hydrological ecosystems,wetlands share the characteristics of transitional complexity,spatial heterogeneity,and hydrological dynamics.Consequently,wetland classification based on single-date or coarse-resolution satellite imagery is no longer effective.In order to map wetland at a finer scale,we proposed a viable scheme employing satellite data with high temporal and spatial resolution.The selected study areas included the Poyang Lake which is the biggest fresh water lake in China and the Yancheng coastal wetland.The main contents and conclusions of the paper are as follows:1.Fine-scale classification of wetlands based on high temporal resolution imagery.Although NDVI time series(NTS)shows great potential in wetland classification,the effectiveness of various NTS with different spatial and temporal resolution has not been evaluated for seasonal wetland classification.To address this issue,we conducted comparisons of those NTS,including the moderate-resolution imaging spectroradiometer(MODIS)NTS with 500 m resolution,NTS fused with MODIS and Landsat data(MOD_LC8-NTS),and HJ-1 NDVI compositions(HJ-1-NTS)with finer resolution,for wetland classification of Poyang Lake.Results showed that the HJ-1-NTS outperformed the fused NTS and the MODIS-NTS with an overall accuracy of 88.12%.Compared with the MODIS-NTS,the fused NTS has improved the classification of seasonal wetlands by 5.66%(product accuracy)/ 9.1%(user's accuracy).In addition,the fused NTS improved the identification of floating plants,submerged plants,line features(e.g.rivers),and small patches(e.g.ponds)as well as capturing the fast changed environmental gradients.However,the classification of highly heterogeneous areas of the fused NTS was unsatisfying especially in narrow zones and small patches.The HJ-1-NTS had remarkable superiority on classifying highly heterogeneous wetlands over the fused NTS,but the uneven time interval of composited images led to missing some key period features for identifying specific vegetation types.2.Fine-scale classification of wetlands based on high spatial resolution imagery.Object-based image analysis is an effective method for high spatial resolution image classification.Segmentation parameter,input features,and classifiers are key factors influencing the classification accuracy.In this study,we used the ESP Tool and the ROC-LV curve to automatically determine the optimal segmentation parameter(SP).Results showed the optimal SP of GF-1 imagery with 2m resolution,Sentinel-2A imagery with 10 m resolution,Landsat 15 m,and Landsat 30 m data was 145,110,85,and 72,respectively.In order to select the best fitted features,the Recursive Feature Elimination method was used to rank the features according to their importance and the least important features were eliminated.Then,a correlation analysis was conducted to remove highly correlated features.For the GF-1 imagery,Sentinel-2A imagery,Landsat 15 m,and Landsat 30 m data,there were 13,20,18,and 18 optimal features selected respectively.Results also showed that the spectral information was the most important feature for wetland classification,especially the red edge and SWIR bands.A comparison between the two decision tree-based algorithm: the Random Forest and the C5.0 showed that the Random Forest was more suitable for wetland classification.3.The influence of different spatial resolution images on the analysis of wetland landscape characteristics.The landscape index is commonly used for quantitative analysis of ecosystems.However,these analyses often overlook the differences in spatial resolution of satellite images,which may cause uncertainty and incomparability of the results.As the spatial resolution decreasing from 2m to 30 m,the number of patches decreased,the mean patch size increased,the area weighted shape index decreased,and the SHDI as well as the SHEI increased,indicating that the landscape tended to be more fragmented,the shape of patches tended to be simpler,and the distribution of different types tended to be more uniform.The dominant species determined by the GF-1 imagery was the Spartina alterniflora,while it was the Phragmites australis determined by Sentinel-2A and Landsat imagery.Therefore,it is necessary to indicate the spatial resolution of the data source and select appropriate data while employing high spatial resolution data for landscape analysis.