Analysis Of Influencing Factors And Prediction Of Soil Organic Carbon At Agricultural Landscape In Hilly Area

As the biggest carbon stock of terrestrial ecosystems, soil organic carbon(SOC) plays a crucial role in global warming and agricultural production, and as an important indicator of soil quality, soil organic carbon can maintain good physical of soil structure, soil stability, and soil biodiversity. Both of natural and human factors can bring soil organic carbon great influence, which varies in different scales and regional. Influencing factors of soil organic carbon of soil organic carbon varies in different geographical environment and topography, the same as to the spatial distribution.In this paper, the agricultural soil of mountain and hilly area was selected as the object of study, the study area is located in Jingjin County, to be more precisely. Under the supports of a serious soft tools, such as, Geographic Information Systems, computer technology, classical statistical analysis and machine learning methods were used to discuss the impacts of soil type, land use type and other topographic factors on soil organic carbon density comprehensively, as well as we discussed impacts of different Influencing factors on the different soil types.At last, four different machine learning methods, Classification and Regression tree(CART), Support vector machine(SVM), Gradient Boosting regression tree(GBRT) and Random Forest(RF), were used to predicted soil organic carbon density(SOCD), prediction precisions were measured. The best predictive model was used to descript importance of each factor on the soil organic carbon density,and generate soil organic carbon density map. The main conclusions are as follows:(1) The average, 0-20 cm, agricultural SOCD ranges from 0.45 to 6.18 kg/m2, with the arithmetic mean of 2.64 kg/m2,and the coefficient of variation shown SOCD has a relatively moderate C.V.(38.49%). The predictive agricultural SOCD ranges from 1.03 to 3.92 kg/m2, with the arithmetic mean of 2.74 kg/m2 is below the whole country’s corresponding figure(3.0kg/m2).(2) There are four types of soil in Jiangjin, paddy soil, alluvial soil, purple soil, yellow soil. Different type of soil has different arithmetic mean of SOCD. Paddy soil has the highest value of SOCD, followed by alluvial soil, and purple soil has lowest, The average content was 2.68 kg/m2, 2.53 kg/m2,2.21 kg/m2, respectively.(3) Major land-use type in Jiangjin are paddy field, dry land, Orchard, Others. The results showed Paddy soil has the highest value of SOCD, followed by Orchard, and Others has lowest, The average content was 3.17 kg/m2 kg/m2, 2.53 kg/m2, 2.13 kg/m2, respectively.(4) There are six slope position in the study area: ridge, upper slope, middle slope, flat slope, lower slope and valley. The distribution of soil samples are concentrated in middle slope and flat slope position, rarely in valley. Slope position has a moderate importance to SOCD, Effect of slope position on paddy soil organic carbon was drastically, conversely, alluvial soil organic carbon slightly.(5) The results of the correlation analysis between SOCD and topographical factors matched the trend obtained by Random Forest approximately, however, local trends of impact varied at different intervals. According to the results, there were highly significant((p<0.01) correlations of SOCD with all the selected topographical factors. SOCD was positively correlated to elevation, SAGA wetness index, as well as Multiresolution index of valley bottom flatness(MRVBF) and those factors had drastically effect on SOCD. Vertical Distance to Channel Network(Vdist) and aspect had no effect on SOCD.(6) Build Random Forest Regression model to determine the importance of different variables on SOCD. The most important factors of SOCD are soil type, land use type, elevation. The most important factors of paddy soil organic carbon density are SAGA wetness index, Valley Depth, length-slope factor. The most important factors of purple soil organic carbon density are land use type, Multiresolution index of valley bottom flatness. SAGA wetness index.The most important factors of yellow soil organic carbon density are land use type, Multiresolution index of valley bottom flatness、SAGA wetness index. The most important factors of alluvial soil organic carbon density are SAGA wetness index, Valley Depth, Multiresolution index of valley bottom flatness. Soil type had greatest effect on SOCD, and relative importance of each factor varied in different soil type. As far as Purple soil and yellow soil concerned, land use type had greatest effect on SOCD, and Paddy soil and alluvial soil concerned, SAGA wetness index had greatest effect on SOCD. In addition, land use types had little effect on alluvial soil organic carbon density.(7) In our study, we compared accuracy of predictive models of soil organic carbon density, built by Classification and Regression tree(CART), Support vector machine(SVM), Gradient Boosting regression tree(GBRT) and Random Forest(RF). The results shown that RF was the best method to predict SOC density of whole region,Paddy soil and purple soil. CART was the best method to predict SOC density of yellow soil and alluvial soil.(8) These results indicated that the RF model was suitable for predicting SOC density in our study area, and the prediction map of SOC density shown that the variation trend of SOC density distribution was conformed to variation trends of topographic factors. The high levels of SOCD was mainly distributed in the northwest part, and low levels of SOCD was mainly distributed in southeast part and along the middle rive.