Spatiotemporal Patterns Of Forest Disturbance And Recovery Using Integrated Land Trendr Algorithm And Machine Learning Method

Forests play an important role in mitigating climate change and maintaining global carbon balance.China’s forests are acting as a large carbon sink and play an important role in the global carbon budget.Subtropical China is the main forest distribution area of China,accounting for about44%of total forest area,and is the major base of China’s timberland and wood resources.During the past 50 years,the subtropical forests have been significantly disturbed by many natural and anthropogenic factors,including harvesting,land use change,fire,drought and insect&disease.In the meanwhile,the subtropical forests have also been affected by the implementations of many forestry policies and regulations,such as Collective Forest Right Reform,Natural Forest Protection Regulation,afforestation/reforestation policies,and grain-to-green policy.However,it has not yet been comprehensively and accurately assessed how these disturbance events and forestry policies/regulations have affected the forest structure and function in the subtropical China.The lack of long-term and high spatial and temporal resolution forest loss and gain information is a key reason.Based on the 30m Landsat TM/ETM/OLI images and Google Earth Engine(GEE)cloud platform,this study developed a two-step classification method by integrating the Land Trendr change detection algorithm with Random Forest classifier in machine learning method.This method was further applied to detect the spatial and temporal patterns of forest loss and gain in Zhejiang,Jiangxi and Guangxi Provinces in the subtropical China during 1986-2019.In addition,this study also analyzed the effects of multiple social and environmental factors on the spatiotemporal patterns of forest loss and gain.The main results were shown below:(1)Based on the national forest resource inventory plot data,local forest management inventory plot data,visual interpretation plots using Google Earth Pro high spatial resolution(<1m)images,and the global forest change(GFC)products,we conduct accuracy assessments to evaluate the classified forest loss and gain products at both spatial and temporal scales.The overall accuracy for forest loss and gain was 93.32%and 90.56%,respectively,and the Kappa coefficients were 0.89 and 0.86,respectively.The accuracy assessments indicated that our developed integrative classification method can efficiently and effectively detect the spatiotemporal patterns of forest loss and gains at 30m spatial resolution in the subtropical China.(2)During 1986-2019,the total forest loss area for Zhejiang,Jiangxi and Guangxi Provinces was7.76×10~4km~2,accounting for 40.68%and 23.70%of the total forest area in 1986 and 2019,respectively.The mean annual loss area and loss rate were 2282.55 km~2and 1.19%,respectively.The interannual variation patterns of forest loss area were similar between Zhejiang and Guangxi Provinces,displaying that forest loss area kept unobvious change tendency before 2003,a fast increasing rate during 2003-2017,stabilized around 2017,and a declining tendency after 2017.However,the forest loss area in Zhejiang Province showed relatively stable and no obvious tendency with time.The total forest loss area of the three provinces as a percentage of the forest area in 2019 is ranked as follows:Guangxi(36.08%)>Jiangxi(15.58%)>Zhejiang(10.14%).This suggested that the forest disturbance was the strongest in Guangxi Province and the smallest in Zhejiang Province.The lost forest area was scatteredly distributed across the entire three provinces,indicating that forest landscape fragmentation was enhancing during the past 50 years.(3)During 1986-2015,the total forest gain area for all three proviences was 2.03×10~5km~2,and the net gain area was 1.57×10~5km~2,with an annual increasing rate of 3.54%.The gained forest area was significantly greater than that of lost area,suggesting an increasing trend in forest area in the study region.Three provinces shared similar interannual variation patterns in forest gain area.The percentage of forest gain area in the three provinces in 2019 is:Guangxi(80.24%)>Jiangxi(50.33%)>Zhejiang(41.59%).(4)The spatiotemporal patterns in forest loss and gain area were affected by many socio-economic and environmental factors.The Collective Forest Reform policy,which was put forward in 2003,greatly promoted the forest loss area in Jiangxi and Guangxi Provinces,while it did not significantly affect Zhejiang Province.The successful implementations of Natural Forest Protection policy(2015)and public-welfare and ecological forest protection regulation have contributed to the declining trend of forest loss area after 2017.GDP in forestry section can reflect the utilization rate of forest resources and anthropogenic intervention intensity on forests.Our study found a significantly positive correlation(R~2=0.77;P<0.05)between forest loss area and forestry GDP.The continuously increasing forestry GDP/total GDP ratio further indicated that forest loss area in Jiangxi and Guangxi Provinces will still keep a high level in the future.The elevation(altitude)can affect the accessibility of human to forests and thus affect forest loss and gain area.In general,forest loss and gain area declined with elevation.However,the declining rates with elevation were different among three provinces.Jiangxi Province has the greatest declining rate,and the fraction of forest loss and gain area above 600m elevation was small.In contrast,Guangxi Province has the smallest declining rate,and considerable fraction of forest loss and gain area was distributed at higher elevation.Natural disasters,such as freezing disaster in 2008 and regional drought events in 1991,2000,2003,3007,and2019,have caused a sudden increase in forest loss area in the followby years.