Construction And Evaluation Of The Transfer Function Of Water Characteristic Curve Of Low Hill Red Soil In Western Zhejiang

Hilly red soil is distributed in central and western Zhejiang Province,most of Jiangxi Province,central and eastern Hunan Province,where is the important base of the tropical,subtropical economic forest and grain production in China.Due to the difference of spatial and temporal distribution of rainfall,severe seasonal drought in southern red soil restricts the sustainable and efficient development of agriculture.It is an important way to realize the sustainable use of red soil resources by increasing resources and reducing expenditures,and strengthening the scientific management of red earth's water resources.Studying the water properties of red soil and its migration rules is the key to improve the efficiency of water resources utilization.The soil water characteristic curve,which describes the relationship between soil potential and water content,is an essential parameter in the study of soil water.Due to the large volume,long period and high cost of determinating soil water characteristic curve in the field or lab,its large area field measurement is impractical.In this paper,the hilly red soil in western Zhejiang Province was taken as the research object.Through the determination of red soil water holding characteristics data,the water properties of red soil under different parental developmental levels,topography,erosive degree and tillage ways were analyzed.The red soil moisture characteristics curve was fitted.Based on the basic physical and chemical properties of soil and the parameters of the optimal fitting model,the indirect derivation method of soil water parameters,which called pedo-transfer function,was established by using multivariate linear regression and BP artificial neural network.The main research results are as follows:(1)Four common trend lines were used to fit the water characteristic curve of red soil,the determination coefficient R~2 from large to small was polynomial,linear,exponential and logarithmic.In the traditional trend line methods,the water characteristic curve of red soil was highest accurately fitted with polynomial form.(2)Two international common soil water characteristic curve models were used to fit the water characteristic curve of red soil,and the fitting precision were both higher than the traditional trend line fitting methods.In the two models,the accuracy of van Genuchten model was better than Mckee and Bumb model,which could describe the shape of the hilly red soil water characteristic curve accurately as well as whose parameters were of practical significance.(3)The red soil moisture characteristic curves with different parental developmental levels,topography,erosive degree and tillage ways would have some differences.The fundamental reason was that their soil structure,bulk density and organic matter content were different.The effect of different developmental degrees on the red soil water characteristic curve was mainly in the high suction section.Soil with high development degree was completely weathered,fine pores were well developed,and the water adsorption was strong.The water holding curve of the high suction section showed a gentle trend,and large changes in suction caused only small changes in water content.The difference of topographic position mainly affected the whole water content.As the content of sand grains increased and the soil structure became worse,the soil moisture content decreased.After the reclamation of red soil,the bulk density decreased,the content of sand grains and organic matter increased,so the water holding performance improved.The improvement degree was directly proportional to the reclamation time.Both the water and dry farming methods had improved the water capacity.After the artificial fertilization of the red soil,the grain content had increased,so the water holding capacity in the low suction section was strong.While the clay content of dryland was relatively high,the fine pores were well developed,and the water holding capacity was strong in the high suction section.(4)By using multivariate stepwise regression and BP neural network,four parameters(saturated water content?_s,residual water content?_r,parameter lg?related to intake air value,parameter lgn related to curve shape)of soil moisture curve model were set up to establish pedo-transfer function.The R~2 of the BP artificial neural network for modeling?_s was 0.96,which was higher than 0.76 of the multivariate stepwise regression,while their verification accuracy was equal.The BP's modeling accuracy of?_r and lg?was equivalent to multivariate linear regression,while the mean absolute error MAE and root mean square RMSE of validation were both less than multivariate stepwise regression,which meaned its verification precision was better.The BP's modeling and verification accuracy of lgn were both better than multivariate linear regression.Both methods could establish the pedotransfer functions of hilly red soil,while the BP artificial neural network method was slightly superior in precision.