Remote Sensing Estimation Methods Of Forest Canopy Closure

Forest canopy closure(FCC)is an important factor for assessing the quality of forest resources,grasping the characteristics of forests and their changing rules,and providing a basis for forest ecosystem management.Meanwhile,it is also a basic index to reflect forest effectiveness and forest spatial utilization degree,which can also provide a support to forest management decision-making.Traditional forest canopy closure estimation methods have some limitations,which are not only time-consuming and labor-cost,but also difficult to timely present the spatial-temporal information of large-scale FCC.Remote sensing technology has become an important alternative method for estimating FCC due to its large-area coverage,economic,dynamic,and repeatability advantages.In this study,Genhe City(hereafter,Genhe study area)of the Great Khingan,located at Inner Mongolia,and Gaofeng Forest Farm(hereafter,Gaofeng study area)in Guangxi were selected as the study areas.Taking the main forest area of Genhe as the key study area,and then based on the threshold method,object-oriented information extraction,support vector regression(SVR),multiple linear regressions(SMLR),k-nearest neighbor with fast iterative features selection(KNN-FIFS),the Gaofen-1 wide field view data(GF-1 WFV)combined with Sentinel-2A red edge indexes,Gaofen-2(GF-2)data,and airborne LiDAR point cloud data,with assistance of land cover types and ASTER GDEM products,the forest canopy information was extracted from individual tree scale to a stand scale in the Genhe study area,and the feasibility and applicability of above-mentioned methods were terrified by the measured FCC data.The results obtained by this study were as follows:(1)Based on airborne LiDAR CHM products,forest canopy density was estimated by both threshold extraction and object-oriented information extraction methods.And then FCC was obtained at stand scale according to the canopy density information.Validated by forest measurements,the estimation accuracy from threshold extraction method was R~2=0.85 and RMSE=0.06,which was higher than that from object-oriented information extraction method with R~2=0.77 and RMSE=0.08.In addition,based on GF-2 data and object-oriented information extraction method,individual tree canopy information was also extracted to estimate forest canopy density.The accuracy of inversion results was R~2=0.49 and RMSE=0.10,which was lower than those from the airborne LiDAR.(2)Combining two Sentinel-2A red-edge bands(RE)with the GF-1 WFV multispectral bands,the simulated GF-6 data was obtained.After that,the extracted relevant texture information(TI),vegetation index(VI)and red edge index(RI),and a k-nearest neighbor with fast iterative features selection(KNN-FIFS)method,was applied to estimate FCC at stand scale in the Genhe study area.Besides that,the impact of terrain was further explored by adding topographic factors(TF)into the KNN-FIFS inputting feature compositions.The verification with leave-one-out(LOO)method showed that:FCC estimates based on GF-1WFV was in reasonable agreement with measurements,with R~2=0.52,RMSE=0.08;and the GF-1 WFV+VI+TI's has R~2=0.56,RMSE=0.08;and the GF-1 WFV+RE+RI+TI's has been significantly improved with R~2=0.63 and RMSE=0.07.The highest accuracy was obtained by the GF-1 WFV+RE+RI+TI+TF composition with R~2=0.68 and RMSE=0.07,which was superior to the results from both stepwise multiple linear regressions(SMLR)(R~2=0.39,RMSE=0.10)and support vector machine(SVM)(R~2=0.49,RMSE=0.10)methods.It indicated that the KNN-FIFS method is more reliable for FCC estimation than both SMLR and SVM methods,and taking the topographic factors into account can effectively improve the accuracy of FCC estimation.(3)A comparison among the results of FCC from GF-1 WFV+Sentinel 2A data,GF-2 data and airborn CHM products were conducted at plot scale and regional scale.Although,the estimation accuracy of CHM data was highest,the acquisition of high-precision LiDAR data is relatively laborious and costly,which limits its large-scale application.Although the estimation accuracy of GF-2 data was the lowest,it does not depend on the measured data using as training samples,and it can be applied when the measured data was not available.Compared with the CHM's estimates,the GF-2's results were highly under-or over-estimated at the areas with frequent anthropic disturbance,such as along the roadside.The estimation results based on GF-1 WFV+Sentinel-2A data were satisfactory,whose the overall trend was consistent with that of CHM data,and the absolute errors of them were less than 0.2.Moreover,the LOO cross validation method applied in the KNN-FIFS takes full advantages of each sample measurement to the training and validation,which is applicable to the large-scale application of quantitative retrieval.(4)In order to further explore the applicability of GF satellite data and KNN-FIFS method in FCC inversion for the typical forests in the southern China,by use of GF-1 WFV+Sentinel2A data,FCC estimation in the Gaofeng study area were carried out.The accuracy of FCC estimation was R~2=0.49 and RMSE=0.07,which was much lower than that obtained in the Genhe study area.It can be explained by the complex structure of forest plantation stand and the time difference between the field survey and the remote sensing image acquisition.Previous studies have mostly focused on a single study area based on multiple data and single method,or on single data source and multiple methods.This study appliedmulti-model remote sensing data and several methods,to retrieve a multi-scale forest canopy information,in order to explore a operational method suitable for the typical forest areas in the both North and South China.In addition,combining the"red edge"information to simulate GF-6 satellite data for the testing and analysis cast a light for the coming-soon quantitative application of GF-6data in the quantitative inversion and precise monitoring forest parameters.