Research On Vegetation Diversity Of Forest Arbor Species Based On Airborne Hyperspectral Remote Sensing Data

The richness of forest biodiversity is an important indicator of forest ecosystem.The protection of forest biodiversity not only helps protect the ecosystem on which human beings live,but also benefits the harmonious development of human and nature.The use of remote sensing technology for the estimation of forest biological diversity has become an important method at home and abroad.The technology not only provides timely and quantitative information collection for all weather,real-time,dynamic and non-destructive,but also effectively avoids human-induced interference,this provides more accurate information support for the forestry sector to carry out quantified evaluation of forest biodiversity,mapping and protection decision-making planning.With the development of new remote sensing sensors,airborne hyperspectral image data acquired by aircraft equipped with sensors has great potential for remote sensing estimation of forest biological diversity.This paper took the Gutian Mountain Nature Reserve in Quzhou,Zhejiang Province as the research base,and took forest arbor species as the research object,based on the ground actual measured plots,the diversity index of forest arbor species were calculated,including the species richness index,Shannon-Wiener index,Simpson index,and Plieou index.Combining airborne hyperspectral image data,relevant remote sensing feature factors were extracted from hyperspectral remote sensing images,including the first,second,and third principal components after the principal component analysis,texture features after principal component analysis,and the narrowband vegetation index.Owing to there were many remote sensing factors extracted and there were correlations between the features,this paper selected the characteristic factors that make a great contribution to the regression modeling by the random forest iterative feature selection method.In this paper,the Multivariate Linear Regression(MLR),Random Forest(RF),and Support Vector Regression(SVR)were used to compare the ability to estimate the diversity of forest arbor species with remote sensing.The ten-fold cross validation method was used to verify the model's estimation accuracy.The following conclusions:(1)From the results of principal component analysis of hyperspectral images,this method effectively avoid the redundancy between the band information and makes the image information be enhanced.At the same time,from the important rankings of remote sensing feature variables participating in the diversity index,there is a strong correlation between the first principal component and the four diversity indices after the principal component analysis,indicating that the first principal component contains rich species information.(2)Based on the random forest iterative feature selection method,it is possible to effectively filter out the redundant variables from the redundant regression models.In the modeling process of the regression model,if all feature factors are directly added to the regression model,it will not only slow down the calculation speed of the model,but also lead to a good fitting effect of the model on the training sample data.The phenomenon of poor application of the verification sample data will affect the model's prediction results and estimation accuracy.Owing to there are many feature factors extracted in this paper,it is necessary to filter the feature variables effectively.The combination of feature variables selected by the random forest iterative feature method not only effectively avoid the arbitrariness of factor selection,but also ensure that the feature factors involved in modeling had a great correlation with the model.(3)Using the three mathematical models to construct a regression model between the diversity of forest arbor species and remote sensing characteristic factors,in terms of estimation accuracy,the species richness index(RF model:R~2=0.65,RMSE=2.58;SVR model:R~2=0.56,RMSE=2.70;MLR model:R~2=0.44,RMSE=3.42);with Shannon-Wiener index(RF model:R~2=0.40,RMSE=0.53;SVR model:R~2=0.54,RMSE=0.59;MLR model:R~2=0.43 RMSE=0.65);for the Simpson index(RF model:R~2=0.61,RMSE=0.052;SVR model:R~2=0.55,RMSE=0.059;MLR model:R~2=0.41,RMSE=0.072);for the Pielou index(RF model:R~2=0.63,RMSE=0.054;SVR model:R~2=0.57,RMSE=0.063;MLR model:R~2=0.33,RMSE=0.081).Compared with MLR and SVR models,the accuracy of the RF model is improved,and the accuracy of the RF model estimation is significantly better than that of the MLRM and SVR methods.Compared to the MLR and SVR models,the accuracy of the RF model in estimating the four diversity indices increased by 21%and9%,17%and 6%,20%and 6%,30%and 6%,respectively.In particular,the estimation accuracy of the species richness index is significantly higher than that of the other two models,and the forecasting effect is best.Therefore,the RF model has the strongest and most stable estimation ability,the SVR model prediction accuracy is less effective.(4)From the effect of inversion,the three models have the phenomenon of underestimation of high diversity index and overestimation of low diversity index.Among them,RF model has the best inversion effect and strong generalization ability,the problem high value underestimate and low value overestimation is the lightest,the SVR model is the second;the degree of this phenomenon is more serious in the MLR model,and the generalization ability of the model is not strong.Therefore,the RF model can be used to invert the species diversity of forest arbor vegetation in the study area.