High-Resolution Mapping Of Soil Organic Carbon And Its Spatial Multiscale Controlling Factors In Tibet

Soil organic carbon(SOC)is the largest carbon pool in the terrestrial ecosystems which contains more carbon than that in the atmosphere and biology.Soil is the substrate of material exchange between the atmospheric and biological community and is considered as a potential sink of greenhouse gases or emission source.The Qinghai-Tibet Plateau,familiarly known as '[the roof of the world',with an average altitude of 4500 m,is the largest ecosystem at this height on earth.Much of it is underlain by permafrost in which stored C is inactive.Of general concern is that with global warming some or all of the permafrost will melt,making the carbon available to microbial attack and the release of CO2 into the atmosphere.On the other hand,global change may make the growth of vegetation and make the SOC increase as a consequence.Therefore,assessing the spatial distribution of SOC and the controlling environmental factor is important as it is crucial for carbon dynamic monitoring and prediction in the future.In this study,we collected 1148 legacy soil samples from the Second National Soil Survey of China in the 1980s to predict the 0-30 cm soil organic carbon spatial distribution and its stock across Tibet.A data mining method was used to extract the relationship between soil organic carbon and environmental variables in Tibet based on the soil formation theory.The 2-dimensional empirical mode decomposition was used to reveal the components of SOC at different scale.Then he localized and scale-specific correlation between SOC and environmental factors was extracted to reveal the controlling factors of SOC at different location and scale.The major results and conclusions are as follows:(1)As the topography of the Plateau is far from flat;rather it is complex,a data mining method Cubist,which works as a form of regression tree,was introduced to predict the SOC across Tibet.Various covariables were used to predict SOC including terrain factors,soil orders,geology,climate and vegetation factors.We predicted the 90 m resolution SOC map and its uncertainty by bootstrapping 50 times at each node.The uncertainty was significantly related to the sampling density.The higher density of the samples are,the more reliable the map was predicted.The predicted SOC map showed a remarkable trend of higher SOC in southeast Tibet and lower in northwest.The Lin's Concordance Correlation Coefficient(LCCC)and RMSE were 0.66 and 0.19%respectively for the independent validation.By validated with the elevational sampled soil in Sygera Mountain,our fine resolution map was more accurate than the previous data Harmonized World Soil Database(HWSD)and SoilGrids.Our estimates of the carbon for the 1980s across Tibet can provide a baseline against which to judge change caused by global warming and the change of land use and land cover.(2)The 2-dimensional empirical mode decomposition was used to characterize the 2-dimensional scale-specific variations in soil organic carbon across Tibet.Three instinct mode function were derived from the original SOC data as well as the residual.In general,the residues contained most of the spatial variability while the IMFs had local oscillations.The range of spatial variability of IMFs were 7 km,95 km and 331 km,respectively.NDVI and elevation were significantly related with IMF1,which indicated that these factors impact SOC at smaller scale.Temperature,precipitation,radiation and ET were highly related to SOC at medium and large scale.Soil type,geology and climate variables accounted more variation of SOC at medium and large scale while elevation accounted more at small scale.In alpine condition,the relationships between environmental factors and SOC differ from those in other regions.The relationship was found varying with physiographic sections,permafrost type and landform,from which we can conclude that the restriction factors is different at different location in Tibet.(3)SOC stocks are usually estimated by multiplying the SOC density times the area of an individual cell on a grid map.However,land surface is tilted in mountain area rather than flat.The difference between the actual surface area and grid area can lead to an underestimated results.We introduced the adjustment coefficient calculated from slope to modify the surface area.The carbon stock was estimated 7.47 Pg with the adjustment coefficient,which was 9.6%higher without it.We found the SOC density of forest and grassland was similar with that in high latitude regions,higher than that in the similar latitude region compared with Tibet due to the geographic and climate.Comparing with the HWSD and SoilGrids data,this estimate differs substantially from the two previous coarser estimates based on global modeling which far exceed our 95%upper confidence limit.