Optimization Of Soil Sampling Density In A Hilly Area

The spatial distribution of soil nutrients are the basic information for the study of soil resources.It is very important to master the variation of soil nutrients in the terrain on agricultural production and environmental simulation.The terrain is the main factor affecting the spatial differentiation of soil nutrients under the condition of single climatic conditions,cultivation method and homogeneous soil parent material.In the complex hilly area,the acquisition of soil samples takes a lot of cost,and it is particularly important to ensure the integrity of spatial information of soil nutrients.Effective and reasonable soil sampling pattern can not only fully reflect the spatial information of soil nutrients,but also can greatly reduce the cost.In this paper,the spatial variation of soil pH,organic matter,available nitrogen,available phosphorus and available potassium and the relationship with terrain factors were studied by spatial analysis theory in the typical hilly area of Yongxing Town,Jiangjin District,Chongqing City(2 km2),and the optimal sampling density and the best sampling unit are determined by simulated annealing method.Terrain factors were employed to predict the distribution of soil nutrients using the neural network method.The main results are:(1)There are interactions between soil nutrients.There was a significant positive correlation between soil pH and soil organic matter,soil available nitrogen and soil available phosphorus.There was a significant positive correlation between soil organic matter and available nitrogen and soil available potassium,and there was a significant positive correlation between soil available phosphorus and soil available potassium.There is spatial autocorrelation for soil nutrients.Soil available nitrogen and soil organic matter contents have strong spatial autocorrelation,soil available phosphorus,soil available potassium and soil pH value have moderate autocorrelation.There are correlations between soil nutrients with topographic factors.The results showed that there was a significant negative correlation between soil pH with TWI,which was positively correlated with HORIZC and Slope.The results showed that soil organic matter contents and soil available nitrogen contents were negatively correlated with Elevation,HORIZC,slope and RPI,which was significantly correlated with the TWI,SlpLen and SCA.Soil available potassium contents were positively correlated with the TWI and Slp Len,and the correlation with other topographic factors was weak.With the reducing of terrain roughness,contents of soil organic matter,soil available nitrogen and soil available potassium gradually increase.The correlation between soil available phosphorus contents and topographic factors is weak.(2)The spatial pattern of original 200 soil samples for each soil nutrient in the training set was optimized by simulated annealing algorithm combined with the neural network model,and the corresponding optimal spatial pattern combinations was given.At the same time,the corresponding prediction error(mean square error)was given.From the error results,original 200 samples of soil pH,soil organic matter,soil available nitrogen,soil available phosphorus and soil available potassium could be instead by 5,6,7,6,5 optimized samples,respectively,and the error is not higher than the original mean square error.In addition,the optimal number of soil pH,soil organic matter,soil available nitrogen,soil available phosphorus and soil available potassium were 68,118,87,86,60,respectively.(3)The BP neural network model of soil nutrients was constructed based on topographic auxiliary variables under the condition of original samples and the best sample pattern.The result showed that the forecast ability of the model with the optimal sampling pattern is improved and the prediction accuracy increased,the model complexity is reduced comparing with the model before optimizing.On the RMSE,Soil pH,soil organic matter,soil available phosphorus,soil available phosphorus and soil available potassium were decreased by 16.88%,14.85%,5.29%,104.26% and 59.94%,respectively.On the MAE,soil pH,soil organic matter,soil available nitrogen,soil available phosphorus and soil available potassium decreased by 39.49%,53.49%,0.26%,3.52% and 10.41%,respectively.(4)In order to improve the irrational situation of soil sampling in the traditional sampling scheme,sampling units were determined based on terrain position in the study area.According to the terrain characteristics,the terrain position is divided into upper slope,lower slope and valley,and sampling units of soil nutrients in different terrain positions were determined on the basis of optimized samples.The results show that sampling units of soil nutrients in sloping fields are approximate,and sampling units were large in valley.Sampling units are smaller in sloping fields than valley.Considering sampling units in different terrain positions and the actual situation of sampling,the average value of sampling units on the different terrain position could be taken as the sampling unit in the study area,the sampling unit of upper slope is 2.06 hm2,the sampling unit of lower slope is 1.81 hm2,and the sampling unit of valley is 4.91 hm2.(5)The spatial distribution of soil nutrients was digitally mapped by using the soil samples combined with BP neural network model,which was optimized by simulated annealing algorithm.The accuracy of the model was reliable,and the distribution of soil nutrients is consistent with the actual situation.It is of great theoretical and practical value for the implementation of soil nutrient management,digital agriculture and precision agriculture in hilly area.