Spatial-temporal Pattern Change And Driving Force Analysis Of Vegetation Coverage In East Africa

East Africa,which is rich in natural resources,is an important gateway to the African continent and a key area of concern for the serious degradation of vegetation in Africa.It mainly includes nine complete national administrative regions,such as Ethiopia,Burundi and Eritrea.The terrain in the study area is obviously undulating,the degree of vegetation cover change is high,coupled with the intensification of global climate change,the sudden increase of population pressure and the rapid increase of food demand,the problem of vegetation degradation and soil erosion in the region poses a great threat to the sustainable development of ecology and has become the focus of attention of relevant countries in the region.In this study,nine countries in East Africa were selected as the study area,and factors such as precipitation,surface temperature,evapotranspiration,surface radiation,and population density from 2002 to 2021 were selected as the influencing factors of vegetation change.Based on Sen-MK trend test,coefficient of variation method,spectral clustering method,BP neural network algorithm and other methods,the spatial and temporal variation of vegetation change and its influencing factors in East Africa were analyzed,and the adaptability of each machine learning algorithm in cluster analysis and simulation prediction of vegetation change in East Africa was discussed.According to the optimal combination algorithm,the vegetation change in East Africa in the past 20 years was clustered.The results can provide scientific basis for vegetation restoration,comprehensive prevention and control of soil erosion and ecological environment protection in East Africa.The main conclusions are as follows:(1)The temporal and spatial differentiation of vegetation coverage and its influencing factors in East Africa from 2002 to 2021 was clarified.Vegetation coverage in East Africa is low,and the annual average NDVI showed a fluctuating upward trend.Except for the fire mask index,the vegetation impact factors showed an increasing trend with time.The NDVI during the year showed a bimodal curve distribution,with double peaks in May and December,respectively.There were four factors with the same distribution trend,namely evapotranspiration(ET),gross primary productivity(GPP),leaf area index(LAI)and precipitation(PRE).Spatially,the distribution of NDVI interannual mean is dominated by the Ethiopian Plateau and the East African Plateau,and decreases to the east and north respectively.The high-value areas of aerosol optical depth(AOD),land surface temperature(LST),photosynthetically active radiation(PAR)and other factors are mainly concentrated in the coastal areas of Eritrea and northeast Djibouti,where the NDVI multi-year mean is low.The distribution of high-value areas of ET,GPP,and LAI is similar to that of NDVI distribution areas,all of which are at higher altitudes such as the Ethiopian Plateau and the East African Plateau.In the past 20 years,NDVI showed a relatively strong increasing trend in more than half of the East African region,mainly concentrated in the dense distribution of arbor and shrub and the eastern coastal area of Tanzania.Most of the vegetation reduction occurred in the economically developed regions of East Africa.The stability of NDVI in East Africa is relatively high,and the area with coefficient of variation < 0.15 accounts for 77.25%.(2)The main driving factors of vegetation change in East Africa and the applicability of each clustering algorithm in driving force analysis of vegetation change were explored.The main driving forces of vegetation growth in East Africa are ET,GPP,LAI and PRE.AOD,LST and PAR inhibit vegetation growth in East Africa.Among them,the main driving forces of low vegetation coverage in the eastern Ethiopian Plateau and central Somalia are high temperature,high radiation and low precipitation.The driving forces of vegetation change in western Ethiopia,southwestern Uganda and eastern coastal areas of Tanzania are mainly high altitude,high precipitation and high evapotranspiration.The highest change rate of vegetation change driving factors is in the central region of Kenya.The combination of spectral clustering method and variance weighted distance calculation method is the most suitable for the clustering analysis of vegetation change in East Africa.The spectral clustering method has the best clustering contour coefficient when the clustering number is 5 and 6.The K-means clustering method has the lowest clustering quality and is not suitable for this study.In different distance selection algorithms,the variance weighted distance method has the largest average contour coefficient under the multi-clustering method,but the clustering quality under other conditions shows a significant downward trend with the increase of clustering number.In the spectral clustering method,the clustering contour coefficients corresponding to the six distance calculation methods have the highest mean value and the strongest stability.With the increase of the number of clusters,the factors causing vegetation change in East Africa gradually become prominent.(3)The future trends of vegetation and its influencing factors in East Africa in the next10 years and the optimal simulation prediction algorithm are clarified.In the next ten years,NDVI,ET,LST and PAR in East Africa will continue to rise slowly,and the rising rate will be higher than the corresponding growth rate of each factor from 2002 to 2021.The growth rate of LAI was lower than that of the previous 20 years.The future development trend of the firemask is good.The GPP and PRE fluctuate greatly in the next decade,but the overall trend is on the rise.There are obvious abnormal values of each factor in the second dry season of2024 and the rainy season of 2030,and the fluctuation values are mostly distributed in the rainy season.The NDVI value in the dry season will decrease significantly,and the vegetation coverage in the rainy season of East Africa will gradually deteriorate since the end of December 2029.In the future trend prediction simulation of East African vegetation and its influencing factors,the MLP / BP algorithm has high stability and applicability.In the prediction simulation of each factor,the minimum difference of the mean absolute error of the training sets,the experiment sets and the verification sets is 0.0202,only 4.70 %.Decision Tree algorithm,Random Forest algorithm,Generalized Regression Neural Networks algorithm followed by goodness of fit and accuracy;the Gate Recurrent Unit algorithm and Bi-directional Long Short-Term Memory algorithm perform well in the prediction simulation of each factor.The Kernel Method algorithm and the Long-Short Term Memory algorithm have the lowest accuracy.The overall performance of the Convolutional Neural Networks algorithm is relatively general,but it performs worse in specific factors.