Estimation Of Above-ground Biomass Of Arbors In The Typical Karst Rocky Desertification Comprehensive Management Area

For nearly half a century,Chinese government has invested a lot of manpower,material resources and financial resources to carry out comprehensive control of karst rocky desertification.So far,the area of rocky desertification land has been effectively controlled,and the control of rocky desertification has shifted from restraining the expansion of rocky desertification area to a new stage in which the resilience of rocky desertification ecosystems is improved.Therefore,assessment of the resilience of karst ecosystems and ecological restoration mechanisms have become key issues in the current comprehensive management of rocky desertification.Based on the spatial layout scale and time benefit scale of the comprehensive treatment project,the vegetation biomass is used as a substitute factor for the resilience of the ecosystem,and the above-ground biomass of arbor forests in the typical karst rocky desertification comprehensive management area is estimated to achieve rocky desertification.Quantitative assessment of ecosystem resilience can provide a scientific basis for comprehensive management of rocky desertification in the new period.Firstly,based on optical and radar data,this paper uses object-oriented,supervised classification,and random forest algorithms to extract the spatial distribution of arbor forests,compare them with the actual spatial distribution of arbor forests,and select the classification method with the best classification effect.Secondly,23 variables including 5 vegetation indices,8 texture features and their ratio differences were extracted,and the best variable was selected by factor analysis.Finally,the above-ground biomass inversion model of arbor forests was constructed based on the best variable,and the accuracy of the model was verified with 108verification data.The above-ground biomass of arbor forests was inverted based on the optimal inversion model,and then the above-ground biomass of arbor forests in 2010 and 2018 were analyzed quantity distribution.The result shows:(1)Use methods of object-oriented(K-nearest(KNN),support vector machine(SVM),and principal component analysis(PCA)),methods of supervised classification(minimum distance,Mahalanobis distance,maximum likelihood,neural network,and support vector machine),and random forest algorithm(RF)classify the land types in the study area,and then extract the spatial distribution of arbor forests.Based on the comparison of 9 classification accuracy and area,the classification accuracy shows that the SVM(overall classification accuracy(OA)is72.46%,Kappa coefficient(Kappa)is 0.67),the supervised classification support vector machine method(OA is 91.30%,Kappa is 0.65),and RF(OA is 77.17%,Kappa is 0.72)are better.The spatial distribution of arbor forests shows that the SVM,the supervised classification support vector machine method,and the RF extract arbor forests and actual comparisons.All show that arbor forests are concentrated in the study area of the central,the north-central and the southeast.According to statistics,the areas extracted by the three methods are 318.795 km~2,69.803 km~2and 235.292 km~2,respectively.Compared with the current forest resources of 267.921 km~2,the difference in the area of the arbor forest extracted by the RF is the smallest,which is only 32.629km~2.Therefore,the random forest algorithm is the best method and presents a relatively stable state,with an OA of 77.17%and a Kappa of 0.72.(2)The five vegetation indices of DVI,EVI,RVI,NDVI,OSAVI are extracted based on optical remote sensing images.The correlation results show that NDVI and OSAVI have a large redundancy with the rest of the vegetation indices,showing that the vegetation index that are significantly correlated exceed the total number.Half,there are 3 vegetation indices that are significantly related to NDVI and OSAVI,and the confidence level of 2 factors is 0.01.The eight texture features of Mean,Variance,Homogeneity,Contrast,Dissimilarity,Entropy,Second Moment,Correlation,and the ratios,the differences are extracted based on ALOS PALSAR data.According to the results of factor analysis,the 10 variables are divided into three categories.The first category is full choose Mean,Variance,Homogeneity,Correlation(HV:Mean,Contrast,Second Moment,Correlation)for modeling.This category of variables has less correlation with other texture variables and has low information redundancy.The second category is followed by Contrast,Dissimilarity,Entropy,Second Moment(HV:Variance,Homogeneity,Dissimilarity,Entropy),this category of variables is basically significantly related to other texture variables,if selected at the same time,it means that the amount of information redundancy increases.Third category is followed by HH/HV,HH-HV,this category of variables mainly integrates the information of the two polarization data.Which variables is selected as the final model construction factor depends on the time when it participates in the model construction that who has the higher goodness of fit.(3)Based on the estimated model goodness of fit of different combinations and 108verification data to verify the accuracy of the model,the model ZBWL＿4 built with mixed factors has the best effect,its goodness of fit is 0.934,relative error(σ)is 4.04 kg/m~2,and root mean square error(RMSE)is 4.05 kg/m~2.From 2010 to 2018,the comprehensive control project of rocky desertification has achieved obvious results,the area of arbor forests increased by25.489 km~2,and the five townships(towns)of Si’en town,Shuiyuan town,Da’an township and Xia’nan township have achieved remarkable results.The area of arbor forests in the townships(towns)are increasing,increasing by 10.531 km~2,9.287 km~2,2.197 km~2and 3.504 km~2,respectively,and decreasing by 0.029 km~2in Dacai township.Compared with 2010,the above-ground biomass of arbor forests in the study area decreased by 1051644.045 t,and the above-ground biomass of arbor forests in the five townships(towns)were all declining.Xia’nan township,Shuiyuan town,Si’en town,Dacai township,and Da’an township were respectively decrease by 152421.739 t,425994.953 t,10900.935 t,64045.505 t,and 398274.913 t.Therefore,the subsequent comprehensive control of rocky desertification should not only pay attention to the area of vegetation restoration,but also pay attention to the above-ground biomass of vegetation to improve the resilience of the regional ecosystem.