Long-term Evolution Of Winter Habitats In Poyang Lake Derived From Satellite Imagery Using Machine Learning Methods

In order to obtain the long-term evolution of winter habitat and explore the driving force in Poyang Lake,we use Landsat satellite imagery from 1990 to 2021 to systematically assess habitat characteristic changes over the past decades.Four machine learning methods including random forest,gradient boosting tree,support vector machine and classification and regression trees and convolutional neural network are analyzed by comparing the overall accuracy and Kappa coefficients.NDVI(Normalized Difference Vegetation),MNDWI(Modified Normalized Difference Water Index),NDBI(Normalized Difference Building Index),and EVI(Enhanced Vegetation Index)are used to identify habitat types.We use the land type transfer matrix to describe the land change.The landscape index was analyzed in the Fragstats and Arc GIS.The results showed that:(1)The overall classification accuracy of the four machine learning algorithms,Random Forest,Support Vector Machine,Gradient Boosting Tree,and Classification and Regression Tree,is above 98 %,and the Kappa coefficients are all greater than 0.98.Among these four algorithms,Random Forest algorithm demonstrates the best classification performance.Based on the aforementioned research,it was discovered that the Convolutional Neural Network achieves higher classification accuracy than Random Forest when applied to classify the ecological and economic area of Poyang Lake.The accuracy on the training set remains above 99 % with minimal fluctuations.(2)The Random Forest algorithm was employed to classify the winter habitat in Poyang Lake using Landsat remote sensing images from 1990 to 2021.Based on the percentage map of winter habitat types and the area change map from 1990 to 2021,it was found that during periods of reduced water surface area in winter,mudflats were the predominant habitat type.(3)According to the thematic map of winter habitat types in Poyang Lake from1990 to 2021,it was observed that grasslands generally appear surrounding mudflats,while sandy areas are primarily distributed near the main river channel of Poyang Lake.Cultivated land surrounding Poyang Lake is mainly concentrated in the southeastern part of the study area,and there was a declining trend in the cultivated land area from1998 to 2004,which may be associated with the local policy of “returning farmland to lake”.(4)In order to reveal the impact of the Three Gorges Dam on the ecological environment of Poyang Lake,a comparative study was conducted on the habitat types of Poyang Lake before and after dam construction.Through the analysis of land use transition matrices,it was found that the mudflat area significantly increased before dam construction during 1990-2002,while more mudflats were converted into grasslands after the dam construction.Overall,the mudflat area decreased after the dam construction compared to before.(5)The changes in landscape pattern indices from 1990 to 2021 were analyzed.The landscape patch density showed a fluctuating trend from 1990 to 1997,indicating further dispersion of winter habitat types in Poyang Lake.However,from 1997 to 2021,both the landscape patch density and landscape disturbance index exhibited a declining trend,while the landscape aggregation index showed an increasing trend,indicating the recovery of habitat integrity since 1997-1998.(6)By studying the influence of water level variations on the classification results of Poyang Lake habitats in different years,it was found that the long-term changes in water surface area did not align with water level variations.This suggests that the water surface area changes in the classification results of this study are unlikely to be primarily influenced by water level fluctuations,indicating that water level is not the main factor affecting the variation of different habitat types in Poyang Lake.