Statistical Analysis On Temporal And Spatial Changes Of Vegetation In The Loess Plateau

Vegetation is an important part of land cover and the main factor affecting the ecological environment.The Loess Plateau is one of the ecologically sensitive areas in China,with significant problems of soil erosion and land desertification.Vegetation plays an irreplaceable role in maintaining the stability of ecological balance in the Loess Plateau.The systematic study on vegetation cover change and land cover classification in the Loess Plateau is very important for the construction of ecological environment.As the preferred index for the study of surface vegetation,Normalized Difference Vegetation Index(NDVI)is excellent in reflecting vegetation coverage.Based on the GIMMS NDVI 3g dataset from 1982 to 2015,this thesis studies the change characteristics of vegetation cover on the Loess Plateau from the temporal and spatial scale.On the spatial scale,a 408 dimensional dataset of 9150 pixels in the Loess Plateau is obtained by extracting NDVI values pixel by pixel.Using kmeans,MiniBatchKmeans and PAM algorithms to cluster the dataset respectively,comprehensively using elbow rule,Sihouette Coefficient and Calinski Harabasz Score,and referring to the land cover characteristics of the Loess Plateau,six types of land cover are finally determined:cultivated land,forest land,grassland,shrub land,wetland and bare land.It is found that the southeast of the Loess Plateau is dominated by forest land and cultivated land,the bare land is most widely distributed in the northwest,and the vegetation coverage is increasing from northwest to Southeast in space.Based on GlobeLand30 land cover product,the accuracy evaluation and comparison of these three clustering algorithms are completed by using Confusion Matrix,which not only confirms the rationality of using these three clustering algorithms for land cover classification in the Loess Plateau,but also shows that the clustering effect of PAM algorithm is the best.On the time scale,the clustering center obtained by PAM algorithm is used as the research data of time series analysis.Firstly,the time series data is decomposed and the period term is extracted to determine the change period of NDVI.Then the stationarity test is carried out,and the SARIMA model is established for different land cover sequences.Finally,the normality test and white noise test of the model are completed,and the NDVI value in 2016 is predicted,and the prediction effect is ideal.The results show that the NDVI values of all kinds of land cover in the Loess Plateau take one year as the cycle,and there are valley values from December to January of the next year.The NDVI of forest land reaches the peak in June,and that of other land cover reaches the peak in August;The vegetation activity increased significantly in spring and autumn,and was not significantly enhanced in summer and winter;From 1982 to 2016,the vegetation cover on the Loess Plateau was continuously improved.