Estimation And Mapping Of Forest Aboveground Biomass Based On K-NN Model And Remote Sensing

Forests are an important part of terrestrial ecosystems.However,due to climate change,their growth changes affect the energy exchange,carbon cycle and water cycle of ecosystems.Therefore,the remote sensing estimation of above-ground biomass of regional forests has gradually become a research hotspot in the field of forestry remote sensing.Traditional parametric statistical methods cannot effectively describe the nonlinear relationship between forest aboveground biomass and remote sensing features,so nonparametric methods such as k-nearest neighbor(k-NN),support vector machine(SVM)and neural network(NN)have emerged for the study of forest biomass remote sensing estimation.In Shangri-La City,four typical forest types(Picea asperata and Abies fabri,Pinus densata,Pinus yunnanensis,and Quercus semicarpifolia)were used as research objects,a k-NN biomass estimation model was established using Landsat 8OLI images,sample plot data and forest manager survey data.The k-NN regression model was compiled and optimized by genetic algorithm based on MATLAB platform.And the uncertainty of sample size in the aboveground biomass estimation model based on k-NN method is analyzed and discussed.The main research contents and results are as follows:(1)For the traditional k-nearest neighbor,there are insufficient problems that give the weight of the feature variables equally when searching the nearest neighbor population units,and lack a weight vector for the feature variables.On MATLAB Platform,we use genetic algorithm to optimize the k-nearest neighbor model in the early stage and the k-NN three parameters(k,t and distance metric)were repeatedly tested and optimized by encoding,then the technological process of Forest Aboveground Biomass remote sensing estimation based on optimized k-NN model was established.The main algorithms include k-NN regression algorithm and genetic algorithm.Among them,the k-NN algorithm uses leave-one method to cross-validate and that Euclidean distance to calculate the similarity of space feature.And the main operations of genetic algorithm optimization model include binary conversion,selection,crossover,mutation and fitness calculation etc..(2)The Landsat 8 OLI image was pre-processed through radiation calibration,FLASSH atmospheric correction and 3×3 window mean filtering,and a total of 306spatial feature variables,including spectral,texture and topographic factors,were extracted from the image based on pixels to be used as alternative parameters for Forest Aboveground Biomass Estimation model.Then,Pearson correlation between remote sensing characteristic variables of different tree species and aboveground biomass was analyzed,and the strong correlation variables were selected as modeling factors.Correlation analysis results show that the correlation between aboveground biomass and spectral features is weak,and the correlation between texture variables is strong.(3)In order to explore the uncertainty of sample size in quantitative remote sensing.we analyzed the effect of different sample size on the accuracy of the model based on Geostatistics semi-variogram theory and k-NN model.The results show that the accuracy of the model tends to be stable with the increase of sample size,and the reasonable range of sample size is between 60 and 70.On the whole,it shows the following rules:When the sample size is less than 60,the accuracy of the model is unstable and fluctuates greatly,while when the sample size is more than 60,the fluctuation of model accuracy decreases and tends to be stable.Finally,the best sample size of Pinus densata and Picea asperata and Abies fabri in this paper are calculated to be 72 and 65 respectively.(4)The forest aboveground biomass in Shangri-La City was estimated and retrieved based on the optimized k-NN model.The estimation accuracy of optimized k-NN model was improved and the percentage of root mean square error RMSE%of Pinus yunnanensis,Picea asperata and Abies fabri,Quercus semicarpifolia and Pinus densata were 52.81%,41.64%,48.57%and 44.66%respectively.The estimation of forest above-ground biomass based on pixel scale is realized,in which the biomass of Pinus densata is 1.21×10~7 Mg,Picea asperata and Abies fabri is 2.91×10~7Mg,Quercus semicarpifolia is 1.30×10~7Mg,Pinus yunnanensis is 0.68×10~7 Mg.Finally,a map of the spatial distribution of forest biomass inversion in Shangri-La was generated.