Forest Remote Sensing Classification And Spatiotemporal Variation Analysis Of Biomass In Tianma Nature Reserve

The changes in forest resources have important impacts on global carbon cycles,climate change,biodiversity and ecological environment.The current development of various remote sensing big data platforms and advanced artificial intelligence provides unprecedented potential and opportunities for dynamic monitoring of forest resources.Quantifying and monitoring the spatiotemporal changes in forest vegetation distribution and biomass can help strengthen forest management and is of great significance for analyzing and evaluating forest productivity,ecological functions and responding to climate change.This thesis chooses the typical national nature reserve forest located in the northern subtropical zone as the research object.Combined with the GEE(Google Earth Engine)platform,using Landsat data from two phases of summer and winter,the forest type structure of the reserve is identified and extracted through a random forest classifier.Using three machine learning algorithms(XGBoost,SVM,MLP)to construct four scenarios of coniferous forest,broad-leaved forest,mixed coniferous and broad-leaved forest,and all forests(integration of three forest types)for remote sensing quantitative inversion models of AGB.On this basis,combined with the Random forest classification results,the spatial distribution and trend changes of AGB in the study area from 1998 to2022 are extrapolated.According to the AGB change trend,the response relationship with climate factors is analyzed.The results show that:(1)In terms of forest type extraction,the classification accuracy of multi-temporal remote sensing images is better than that of single-temporal classification results.In terms of overall classification accuracy:summer-winter dual-phase>winter phase>summer phase.Using only summer Landsat image data for classification,the overall accuracy(OA)is 88.94%,and the Kappa coefficient is 0.8486;while using only winter single-phase data,the overall accuracy is 92.83%,and the Kappa coefficient is 0.9014.Based on summer-winter dual-phase classification,the overall accuracy is 96.88%,and the Kappa coefficient is 0.9573.The multi-temporal method can provide a high-precision forest classification product,which can help applications based on forest type mapping.After remote sensing interpretation,the forest coverage rate in the reserve has slowly increased during the study period,with broad-leaved forests being the main forest type,followed by mixed coniferous and broad-leaved forests,and the distribution range of coniferous forests being the smallest.(2)Among the three machine learning algorithms in the AGB estimation model optimization,the MLP algorithm performs best in the AGB estimation of broad-leaved forests,while the XGBoost model performs best in the modeling of coniferous forests,mixed coniferous and broad-leaved forests,and all forests.The modeling accuracy of differentiating forest types is better than that of not differentiating forest types.(3)Combined with the forest type classification results,according to the optimal model of each forest type,the spatial distribution maps of AGB of three forest types:broad-leaved forest,coniferous forest,and mixed coniferous and broad-leaved forest in 1998,2002,2006,2010,2014,2018 and 2022 were inferred and inverted.The results show that the AGB per unit area of the entire forest in the reserve is maintained between 133-147t/hm~2 during the study period,with some fluctuations.The total AGB of the reserve forest is between 3.38×10~6 t and 3.80×10~6 t,of which the AGB of broad-leaved forests accounts for about 40%of the total amount,the AGB of coniferous forests accounts for about 30%of the total amount,and the AGB of mixed coniferous and broad-leaved forests accounts for about 30%of the total amount.According to the analysis results of the spatiotemporal change trend of forest AGB in the reserve for 24 years,there is a certain downward trend in forest AGB in two major areas(Mazongling area and Tianzhai area)in the reserve,and there is an obvious increase trend in AGB in high mountain areas far from human activity areas.(4)There is a certain response relationship between forest AGB and average temperature and average precipitation,but during extreme climate periods(i.e.1998 and 2022),forest AGB is not effectively coupled with the two.However,the response relationship between forest AGB and the extreme precipitation index CDD(Consecutive Dry Days)is better than that of average temperature and average precipitation.