Research On Forest Volume Estimation Based On Multi-source Remote Sensing Data

Combining remote sensing imagery with ground survey data to build a model for estimating the volume of forests is the main trend of current forest volume acquisition.With the maturity of remote sensing technology,remote and large-scale estimation of forest volume through remote sensing images Is a necessary means to replace manual field surveys.Among the many remote sensing data,the research practice of optical remote sensing data is the longest,the technology accumulation is the most,the data source is the most abundant,and the data performance is the most stable.However,due to the different imaging principles and the limitations of sensor technology conditions,the remote sensing data obtained by a single optical sensor cannot fully reflect the ground characteristics of the target,which has strong limitations.If multi-source remote sensing images of the same area and at the same time are intelligently synthesized It can generate new remote sensing data that is more accurate,more complete and richer in information than a single data source,and can obtain more reliable estimation results in the estimation of forest volume.In this paper,the Wangyedian Forest Farm in Inner Mongolia is used as the research area.The 76 standard plots surveyed in the research area in September 2017 were used as the research samples.The GF-2 and Landsat8 OLI remote sensing data of the same period were used.Methods are used to fuse,and the five types of regression models of random forest,support vector machine,K nearest neighbor,artificial neural network,and multiple linear regression are used to remotely sense the forest accumulation in Wangyedian Forest Farm using the fused remote sensing data and the ground samples investigated.Estimation and analysis of the potential of the two types of data in the estimation of forest stocks.By comparing the various types of fusion data and the results of the stocks of the first two types of data,the best fusion methods of the two types of data in terms of stock volume estimation and Fusion band,compare five regression models at the same time and analyze each model's ability to estimate forest volume,then developed an integrated regression model based on the estimation results of the five models,and based on the estimation results of the single model and the integrated model The volume distribution map of Wangyedian Forest Farm and the uncertainty distribution map of volume estimation were drawn.This study shows:(1)Blending the best band combination.Using GF-2 and Landsat8 as data sources,construct four machine learning regression models and a traditional linear empirical model to estimate the forest volume of Wangyedian Forest Farm.The best estimation result comes from the blue light band under the Brovey fusion method.The SVR regression model constructed by integration with Landsat8 has a coefficient of determination(R2)of 0.55,a relative root mean square error(RRMSE)of 26.4%,and an absolute deviation(MAE)of 57m3/hm2.(2)The support vector machine regression model has a better estimation of forest volume.Five types of accumulation volume estimation models were used in the study.Among them,four machine learning models are better than traditional linear models in estimating the accumulation volume.In particular,the support vector machine regression model performs best in all data.The root mean square error of its estimation results is between 26%-29%.(3)The GF-2 panchromatic image and Landsat image fusion did not play a good role in the estimation of the accumulation volume.In the NNDiffuse fusion method,the result is the best,the relative root mean square error is 32.1%,which is higher than the estimate before fusion The measurement results were 0.5 and 1.4 percentage points higher,respectively.It can be seen that the fusion of GF-2's panchromatic data and Landsat8 data played an opposite role in the estimation of accumulation.(4)Among the three data fusion methods,the BROVEY fusion method achieves the best estimation accuracy in the volume estimation.When the blue,green,and red visible light is fused with Landsat8,the GRAM-SCHMIDT fusion data is estimated to be the accumulation amount It is superior to NND fusion,and NND performs better than GRAM-SCHMIDT in the fusion of near-infrared and panchromatic bands.(5)Through regression analysis,the models added to the ensemble analysis are three regression models:support vector machine(SVR),K nearest neighbor(KNN),and random forest(RF).The integrated analysis of the weighted average of the coefficients of each model is not better than all models,but the integrated regression model can obtain more stable estimation results than all models.The estimation results and integration of the three models Uncertainty distribution map of the estimated forest volume in Wangyedian Forest Farm combined with the forest phase map of the forest farm shows that the estimation of the three models in the coniferous forest area in Wangyedian Forest Farm is relatively stable.For the mixed forests of coniferous and broad-leaved and some non-forest land,the estimation results of the three models are quite different,and the accumulation of this area cannot be estimated steadily in this study.