Above-ground Biomass Estimation In Hunan Province Based On Multi-Source Data And Artificial Intelligence Algorithms

Forests are an important component of terrestrial ecosystems and play a crucial role in absorbing greenhouse gases such as carbon dioxide,conserving soil and water,maintaining biodiversity,and mitigating global climate change.Fast and accurate inversion of spatial distribution maps of forest above-ground biomass(AGB)over large areas with high accuracy is essential for estimating the carbon balance of forest ecosystems,quantifying the effects of forest restoration measures,and predicting future changes in carbon sinks.Although remote sensing technology has been widely used in biomass estimation,it is still a challenge to improve the accuracy of biomass estimation.In this study,using Hunan Province as the study area,we used multiple sources of data such as literature data,remote sensing data and forest inventory data,and employed Meta-analysis,correlation analysis and stepwise regression analysis,combined with the currently popular artificial intelligence algorithm,to quantitatively invert the above-ground biomass,and then counted the basic conditions of forests in Hunan Province.Also,the main factors affecting the spatial distribution pattern of forest biomass in Hunan Province were analyzed from different scales.The main research results are as follows:(1)At the single wood scale,the fit of the anisotropic growth model constructed by tree species type was better.For coniferous species,the optimal model was AGB=0.0321·DBH2.7407 with R2=0.97 and MPB=-1.49.For broadleaved species,the optimal model was AGB=0.0298 ·(DBH2H)0.9952 with R2=0.95 and MPB=12.31.At the sample plot scale,the biomass prediction using the allometric growth model with subspecies was much better than that using the allometric growth model with all species.(2)When constructing biomass models,attention should be paid to the selection of data sources and algorithms.In terms of data source selection,models combining data from multiple sources are generally higher than those using only a single data source,and models combining active and passive remote sensing data sources perform better.In terms of model accuracy,the multiple linear regression(MLR)model was the least accurate and the gradient boosting machine(GBM)model was the most accurate.Combining the accuracy and operational speed of the models,this study finally decided to combine the active and passive remote sensing data and use the Extreme Gradient Boosting Machine(XGB)algorithm for AGB modelling.(3)Combined active and passive remote sensing data and wall-to-wall inversion of biomass using the Extreme Gradient Boosting Machine(XGB)model resulted in a refined forest above-ground biomass remote sensing map product for Hunan Province in 2020.The spatial resolution of this remote sensing product is 30 m and the overall accuracy is quite high(R2=0.79,RMSE=25.06,RMSE%=33.18).Among them,broadleaf forests had the highest accuracy(R2=0.82,RMSE=28.38,RMSE%=33.68),coniferous forests the second(R2=0.78,RMSE=22.77,RMSE%=32.07),and mixed forests the lowest(R2=0.73,RMSE=24.3,RMSE%=34.01).(4)High-precision remote sensing products using multi-source data and artificial intelligence algorithm inversion can quantify the regional forest resources.According to the high-precision remote sensing products of this study,the forest area of Hunan province is 13,277,200 ha,the forest cover is 62.69%,the average tree height is 16.13±3.28 m,the average above-ground biomass is 73.70±22.54 Mg/ha,the total above-ground biomass is 979 million tons,and the total forest carbon stock is 601 million tons.The forest resources in Hunan Province are mainly concentrated in western Hunan and southern Hunan.(5)Topography and growth condition are the main factors influencing forest biomass.Topography(elevation and slope)and vegetation growth condition(NDVI)are significantly and positively correlated with biomass,and there is heterogeneity in the correlation between biomass and each influencing factor in different regions.AGB is less influenced by temperature in western Hunan,while it is limited by moisture conditions in southern Hunan.(6)The government,research institutions and related personnel should carry out multiscale,all-round and scientific forest ecosystem management and management.At the singlewood scale,the protection of primary forests and large trees should be emphasized to minimize human interference.When planting trees at the site scale,broad-leaved species with strong carbon sequestration capacity are recommended,while the planting density of planted forests should also be reasonably regulated according to local conditions to prevent inefficient afforestation.At the regional scale,the macro control of forest management should be emphasized,and attention should be paid to the implementation of forest fire prevention and pest control measures to improve forest quality.These results show that the above-ground biomass of forests in Hunan Province has obvious regional characteristics,and the allometric growth models constructed in this paper can be widely used in forestry carbon sink research.The modelling idea of this study,based on multi-source data and artificial intelligence algorithms to construct biomass models,can be extended to national forest research.This study can provide theoretical support for forest management at national and regional scales,and contribute to the achievement of the "double carbon" target.