Regional Biomass Estimation And Application Based On Remote Sensing

With the improvement of living standard and the rising awareness of environment protection,ecological values of natural resource including forest ecosystem are becoming increasingly prominent.Biomass is one basic parameter and indicator of forest productivity and carbon storage.Accurate quantification of spatiotemporal biomass information makes great contributions to global carbon balance and sustainable development.In the past decades,the development of computer science and remote sensing technology has facilitated more efficient and scientific monitoring and management of forest resources.Although the remote sensing-based biomass estimation provides macroscopical and efficient results,there still exists great uncertainty.Combing the widely used Landsat imagery and limited forest field investigations with proper features and approaches to realize the mapping and analysis of biomass will be meaningful for scientific forest management and sustainable development.In this study,above ground biomass(AGB)was estimated through a comparison of five different regression methods(Partial Least Square method,K-Nearest Neighbor,Support Vector Regression,Random Forest and Stochastic Gradient Boosting)by combining Landsat data,point and planar field information and topographic data in the middle part of Hangzhou,China to investigate the importance of modeling approach,features and sample size.Afterwards,the optical dataset and modeling methods were combined to quantify the local spatiotemporal AGB.Furthermore,the characteristics of AGB distribution and change were analyzed to get a better understanding of local forest.The main contents and conclusions of this research are as follows:(1)Different predicted variables derived from remote sensing imagery,topography data and field measurement were compared to find the optimal feature assemblage.Based on the Landsat image,multi-spectral bands showed negative relations with biomass,and the shortwave band had the strongest correlation.Among the vegetation indices,which containing short wave infrared hand had positive correlations with aboveground biomass,while most of those without short wave infrared band had negative correlations with biomass.Although spearman analysis showed that the NDVI didn't have a significant relation with biomass,wetness,NDVIc and the mean texture band of shortwave bands had high correlations.Tree height and diamter at breast height had an evident positive relation with biomass.(2)During the modeling process,the size of sampling plots,feature combinations and machine learning methods were investigated using four accurate assessment measurements to choose the best combination.Furthermore,the importance of all the factors was compared using the ANOVA analysis.Cross validation was used for accuracy assessment to compare the performance of different regression methods.In terms of RMSE,R2 and variance ratio,Random Forest and Stochastic Gradient boosting achieved the best performance,followed by Support Vector Regression,K-Nearest Neighbor and Partial Least Square method in a descending order.With regard to bias,Partial Least Square and Stochastic Gradient boosting obtained the most stable predictions.From the perspective of sample size,accuracy increase range of PLS and KNN was less than that of SVM,RF and SGB.Among the latter three methods,SGB had the highest improvement.For different features,the addition of texture features and field investigation data improved the overall prediction accuracy,however,the increase were not significant.Compared with other two feature selection methods.Boruta feature selection reduced the time of regression process and produced similar accuracy with all candidate features.Using all the features will promote machine learning methods to make use of all the potential feature information,but Boruta is more pratical.According to the ANOVA result,the most important factor determing the estimation accuracy was regression approach.Therefore,the modeling approached played the most significant role in biomass estimation based on Landsat imagery.(3)On the basis of AGB map produced by synthesizing the most appropriate regression approach,features,samples and regression parameters,AGB spatial distribution characteristic was investigated through stratified topography conditions and different forest types,landscape indices and geostatistics method(including semi-variance analysis and Moran'I index).In general,AGB increased with the increase of elevation and slope,which may due to the fact that forests located in higher elevation and slope are far away from human interference and keep more primitive growth,which indicated that establishing natural conservation area will have positive effect for local forest biomass accumulation.For differen forest types,there is a significant difference in AGB distribution between arbor forest,bamboo and shrubs.The mean values of AGB decrease from broadleaf forest,Chinese fir forest,pine forest and economic forest.AGB statistics showed that natural forest had higher biomass than man-made forest,but the development of man-made bamboo industry improved the biomass of artificial forest.The landscape index analysis of reclassified AGB showed that the middle level AGB(60-90ton/ha)in the study area had a largest coverage with high fragmentized and complex distribution.Semi-variance analysis and Moran'I index showed that AGB presented moderate spatial autocorrelation,means that local biomass distribution was mainly dependent on natural environment,but the influence of human beings shouldn't be ignored.Hotspot analysis showed that along rivers and areas with higher vegetation coverage higher AGB clusters were located,while shrub forest was distributed with lower AGB clusters.(4)The relationship between remote sensing bands and field data in 20]0 was applied to other periods and multitemporal spatial distributions of AGB were generated.On this basis,the spatiotemporal change of estimated AGB was linked to natural factors and anthropogenic interference to investigate the characteristics of AGB distribution.The mean valuw of AGB was gradually increased from 1984 to 2013,especially after the year 2000.The change rate of AGB was calculated in seven periods to compare the influence of different elevation grades and forest management modes.As a result,the greatest AGB increase took place in the regions with lower elevations(below 100m),which demonstrated the great influence of human activities.Moreover,national and provincial ecological forests had higher AGB values and change rate than normal forest,illustrating that the current forest delimitation of ecological forest was reasonable and had a positive effect on biomass.(5)The equivalent value per unit area of forest ecosystem service value(ESV)was improved by taking the industrial structure change into consideration by means of equivalent value estimation method.In addition,the social development coefficient and spatiotemporal AGB map were combined to evaluate the whole forest ecosystem service value from 1984 to 2013.As a result,the total forest ESV showed rapid increase tendency,the proportion of supply service was decreasing and aesthetic value in the cultural service was increasing.The quantitative hierarchy of forest ecosystem service value showed that the most high ESV was located in the north bank of Fuchun River,where was most planted with coniferous forest.The relationship between ESV increase with population,GDP and AGB indicated that ESV increased faster than population density and AGB before 2010,but slower than economic development and AGB after 2010,which indicated that ESV was sensitive to population density.Therefore,during the process of urban construction and development,the bearing capacity of forest ecosystem for population should be taken into consideration.