Spatial variation analysis of soil organic matter in karst area

As an important parameter for soil quality assessment, soil organic matter (SOM) can provide nutrients for crops, strengthen soil fertility, and improve soil physicochemical properties. Besides, SOM also plays an important role in restricting the activity of heavy metal elements and pesticide residues, and in regulating CO₂ in soil. Therefore, the study on spatial distribution of SOM, its influence factors and prediction by GIS and geostatistical analysis is important in agricultural practice, environmental management and the measurement of soil carbon storage.The content of SOM is easily affected by the factors such as geological background, climate change, agricultural activities and land use change. Hence the complexity and non-stationarity of the spatial variability of SOM will make it difficult for the quantitative prediction. The high heterogeneity of SOM and insufficient sampling data in the karst area also leads to low accuracy of spatial prediction. Guangxi is one of the areas with the most widely distributed karst areas in China. Typical karst landforms are developed in the northern part of Mashan county, with a range of land use and parent materials. Different patterns of land use include paddy field, dry land, shrub land, forest land, orchard and grassland, accounting for 10.55%, 29.39%, 29.49%, 12.86%, 0.70% and 0.66% respectively. Paddy fields, consisting of paddy soil, are located in karst valleys; dry lands, consisting of red soil, brown lime soil, and alluvial soil, are distributed at the bottom of slopes and karst depressions on both sides of the valley; shrubland and forest land, mainly consisting of brown lime soil and red soil, spread over mountains. A total of 441 SOM data, with 8 high-density topsoil (0-20 cm) samples per km² on average, were obtained in the geochemical survey of land quality in the northern karst area of Mashan county, Guangxi in 2017. The data facilitates the spatial prediction of highly heterogeneous SOM in karst areas.In this paper, qualitative variables such as land uses and soil types were converted into quantitative variables through the assignment method of dummy variables, and terrain factors were used as auxiliary variables to meet the requirement of geostatistical regression analysis. With the purpose of exploring the applicability of geostatistical SOM mapping in karst areas, this study respectively established five method models—ordinary Kriging (OK), regression Kriging (RK), Geographically Weighted Regression Kriging (GWRK), Ordinary Kriging of Median Centralization (MC_OK) and Mean Modified Ordinary Kriging (MM_OK)—to compare the prediction results, and select the most suitable prediction model for the karst area.. Results show that the SOM content in the study area ranges from 0.81% to 5.03%, with an average of 2.36%, and the coefficient of variation is 37.3%. The spatial distribution is moderate spatial variation. Results of geostatistical analysis by GS+9.0 show that the fitted theoretical variogram models are all exponential ones. Both nugget value and abutment value are less than 25%, which indicates that there is a certain spatial autocorrelation, and the SOM content is mainly affected by structural factors; hence correlation interpolation analysis can be carried out.In addition, the spatial variation of SOM in the karst area is jointly affected by land use, soil types and topographical factors. The areas with high content of SOM are located respectively in the karst areas and paddy fields where lime soil is distributed in the northwest, west and east of the study area, and the areas with low content are located in the alluvial land along the Hongshui river in the north of the study area. The SOM content shows significantly positive correlation with paddy fields because the reducing environment of paddy fields is conducive to the preservation and accumulation of SOM. But SOM content shows significantly negative correlation with dry land due to the oxidative environment of this land type where frequent disturbance and rapid degradation of SOM will occur.Finally, RK, GWRK, MM_OK and MC_OK can be used for SOM prediction and mapping in karst areas because of their good interpretability. The results of internal verification show that the root-mean-square of GWRK is the smallest, and the average standard error is close to root-mean-square, indicating the better fitness of GWRK model. The results of external verification demonstrate that the mean absolute error and root mean squares error of GWRK are the smallest, suggesting the highest precision and the best stability of GWRK model. Meanwhile, MC_OK shows its best accuracy. Combined with auxiliary variable factors such as land use, soil types, and topography, GWRK model can effectively eliminate the influence of spatial variation factors and overcome the spatial non-stationarity of SOM content in karst areas, thereby improving the stability and precision of the SOM prediction model, while MC_OK model can improve the prediction accuracy.