| Wetlands represent a precious natural resource that exerts an irreplaceable influence over the regulation of biological,ecological,and hydrological processes on both regional and global scales.However,the impact of both natural and human factors has led to the degradation or even disappearance of these invaluable wetlands to varying degrees.Hence,the significance of obtaining high-precision extraction of wetland information,scrutinizing the spatiotemporal evolution of wetlands,investigating the driving mechanisms of wetland evolution,and projecting future wetland distribution under different scenarios cannot be overstated for safeguarding and managing,rehabilitating and reconstructing wetland resources.Through the amalgamation of the Google Earth Engine and ensemble learning algorithms,this study accomplished rapid and high-precision extraction of wetland information using Landsat remote sensing data.Subsequently,wetland-related change indices and Geodetector techniques were utilized to quantitatively analyze the spatiotemporal evolution of wetlands and their drivers of change.Furthermore,the suitability of several wetland landscape simulation models was probed,and the distribution of wetlands under different future scenarios was predicted.The main conclusions are as follows:(1)The incorporation of Adaptive-Stacking with the Google Earth Engine yielded a notable increase in the accuracy of wetland information extraction.In comparison to Support Vector Machine(SVM)and Random Forest(RF),the Adaptive-Stacking algorithm proved superior(OA=89.74%,Kappa=0.87),exhibiting an increase in OA of 12.15%and 5.62%over SVM and RF,respectively,with Kappa showing an increase of 0.12 and 0.06 over SVM and RF,respectively.Classification accuracy varied with distinct combinations of image features.When solely using the spectral features of Landsat images,the OA and Kappa coefficient of wetland classification were 82.61%and 0.80,respectively.The addition of Landsat-derived index features increased the OA and Kappa coefficient to 85.39%and 0.83,respectively.The highest classification accuracy was achieved when the elevation and slope features were included,with an OA and Kappa coefficient of 89.74%and 0.87,respectively.(2)The study area exhibited a relatively sharp degree of wetland change,with wetland area displaying a trend of initial decrease and subsequent increase during the 1995-2020 period.The wetland area proportion decreased from 45.11%in 1995 to 44.69%in 2020,signifying a total decrease of 294.94 km2.Mudflats,reed swamps,and sedge swamps displayed the greatest degree of variability throughout the study period.Population density and GDP represented the most significant anthropogenic factors threatening wetland distribution,while temperature,precipitation,and sunshine hours were the primary natural factors impacting wetland change.The interaction between any two factors was more substantial than that of a single factor,with the precipitationsunshine hours interaction having the largest impact on the degree of wetland change(0.534),followed by the population density-precipitation(0.502)and population density-sunshine hours(0.520)interactions.(3)Compared with CA-Markov and FLUS,the PLUS model has better suitability for the simulation in the study area(Kappa=0.784,FoM=0.22),and Kappa is 0.191 and 0.069 higher than CA-Markov and FLUS,respectively,and FoM is 0.15 and 0.07 higher than CA-Markov and FLUS,respectively.The wetland areas under the three scenarios showed different trends,among which the wetland areas under the Natural Development Scenario(NDS)and the Wetland Protection Scenario(WPS)both showed an increasing trend,but the increase rate of the wetland areas under the NDS decreased year by year and approached a stable state by 2035,the wetland areas under the WPS still maintained a stable growth rate by 2035,and the wetland areas under the Economic Development Scenario(EDS)showed a significant decrease. |
PDF Full Text Request