Digital Soil Mapping Of PH Based On MGWR

Soil pH is an important basic property of soil,which is closely related to the level of soil fertility,the activity of microorganisms and fauna and the formation of humus.Digital soil mapping is an important means to obtain soil pH,and its accuracy is of great significance for rational land planning and management.Soil is affected by climate,biology,topography,parent material,time,soil itself and spatial location.The effect of environmental factors on soil pH is spatially non-stationary,so the traditional linear regression model is difficult to accurately describe the complex relationship between them.Based on the data of 599 sampling sites in Anhui,Henan,Jiangsu and Shandong,this paper comprehensively considered the effects of soil,topography,climate and biology on soil pH,and selected 9 related environmental factors.The spatial distribution of soil pH in the study area was modeled and compared by using four models: multiple linear regression(MLR),geographically weighted regression(GWR),mixed geographically weighted regression(Mixed GWR)and multi-scale geographically weighted regression(MGWR).The prediction mapping of soil pH with multiple models was realized.Combined with quantile regression and geographic detectors,the effect of environmental factors on soil pH was revealed.The main results are as follows:(1)The average value of soil pH in the study area is 7.10,which ranges from 4.20 to 9.18,which belongs to the middle level of variation,including 6.37 in Anhui,7.89 in Henan,6.90 in Jiangsu and 7.06 in Shandong.The results of spatial autocorrelation showed that under the spatial distance of 56.3 km,100 km,200 km,300 km,400 km,500 km and 600 km,the global Moran’s I values of soil pH in the study area were 0.70,0.65,0.58,0.52 0.47,0.43 and 0.39,respectively,indicating that there was a significant global spatial positive correlation in soil pH samples,and the global Moran’ SI value decreased gradually with the increase of spatial distance threshold.That is,the global spatial correlation is gradually weakened.Significant local spatial clustering was found in 274 of the 599 sample points under the spatial distance of 56.3km,which showed that there was significant local autocorrelation in soil pH samples.(2)The order of model accuracy from high to low is PC-MGWR > MGWR > PCMixed GWR > Mixed GWR > PC-GWR > GWR > MLR > PC-MLR.Among them,the effect of PC-MGWR model is the best,the AICc value on the modeling set is1050.12,and the sum of squares of the residual is 181.80,which can explain 64% of the soil pH variation on the modeling set.The CCC,MAE and RMSE on the verification set are 0.78,0.49 and 0.63 respectively,and 62% of the soil pH variation on the verification set can be explained.The autocorrelation level of model residual between MGWR and PC-MGWR was the lowest,in which the global Moran’s I value of MGWR model was 0.07,only 22 samples had significant local spatial clustering,and the global Moran’s I value of PC-MGWR model was 0.07(P < 0.01).The spatial distribution of soil pH in the study area decreased gradually from north to south,with the highest in northern Henan and the lowest in southern and western Anhui,in which the pH value in northern Henan and Huaibei plain in Anhui was higher than that in the same latitude,and the pH in the mountainous and hilly region of central and southern Shandong was lower than that in the same latitude.(3)The results of geographic detector analysis showed that the explanatory ability of soil pH from high to low was MAP > MAT > NDVI > MRRTF > MRVBF > TWI >Elevation,and the q values were 0.382,0.208,0.119,0.105,0.105 and 0.044.Among them,MAP is one of the most important factors affecting soil pH,and it has a significant negative correlation with soil pH,that is,with the increase of MAP,soil pH decreases gradually,and its ability to explain soil pH variation reaches 38%.The most interactive combination is climate factor MAT and MAP,which account for 51% of soil pH variation,followed by MAP and MRVBF,which account for 49% of soil pH variation.(4)MRVBF had a significant positive effect on soil pH on MLR model.The results of quantile regression showed that MRVBF had a significant positive effect on soil pH at low quantile level(θ = 0.1 ～ 0.5),while soil pH had no significant effect on soil pH at high quantile level(θ = 0.6 ～ 0.9).MAP had a significant negative effect on soil pH,and the effect of MAP on soil pH decreased gradually with the increase of quantile level,that is,the soil pH at the higher quantile level was less sensitive to the change of MAP.The results of MGWR regression showed that there was strong spatial heterogeneity in the effect of environmental factors on soil pH.For example,the effect of average annual rainfall on soil pH is stronger in northern Jiangsu and most areas of Shandong,and weaker in western Henan and southern Anhui;the effect of MRRTF on soil pH is stronger in Shandong Peninsula,eastern Anhui and Jiangsu,and weaker in northwest Henan.