Application Of Sobol’s To The Prediction Model For Soil Solute Transfer Into Surface Runoff

The solute transfer from soil to surface runoff has been recognized as a major source of non-point source pollution,and it is of great significance to predict this process,for which the model parameters are very important to the model prediction accuracy.Sensitivity analysis is a common method to study the uncertain model parameter importance to the prediction data,which can provide important reference for the model parameters calibration in future.However,model parameters interact with each other,and the variance-based global sensitivity analysis method of Sobol’s can explore the parameter importance considering model parameters interactions.Therefore,the Sobol’s was applied to the prediction model of the soil solute Cl^-and paddy nitrogen loss into rainfall induced surface runoff based on the soil mixing layer theory,and the corresponding sensitivities of uncertain model parameters and parameters interaction to the prediction data of solute concentrations in surface runoff or underlying the soil mixing layer.In the prediction model of the soil solute Cl^-loss into rainfall induced surface runoff,the main uncertain model parameters are the surface runoff related and soil infiltration related incomplete mixing parametersαandγand soil mixing layer depth parameter d_mix.The solute in the d_mixwith solute concentration C_w（t）is the solute source in surface runoff.And the Sobol’s method was applied to the prediction model of the soil solute Cl^-loss into rainfall induced surface runoff for three study cases under different initial conditions,to determine the importance of the model parametersα,γand d_mixto the predicted solute concentrationαC_w（t）in surface runoff and the parameters interactions in the three cases.The study results showed that the importance of parametersα,γ,and d_mixto preditedαC_w（t）varied with time.In study cases with initially saturated soil or with initially unsaturated soil,the first order index ofαand d_mixvaried oppositely,and the interactions betweenαand d_mixbecame stronger with time.In cases1and 2,parameterγwas the most sensitive and important for the predictedαC_w（t）at later time,and it had strong interactions withαand d_mix.Andγwas the most sensitive parameter for the predictedαC_w（t）in Case 1 in the whole time.In case 3,under the condition of restricted subsurface drainage without downward infiltrated water for the initially saturated soil,γwas not sensitive to the predictedαC_w（t）,with no interactions ofγwithαor d_mix.Furthermore,in the prediction model of the paddy nitrogen loss into rainfall induced surface runoff,the main uncertain model parameters are the urea hydrolysis rate constantμ’_w,1,the nitrification rate constants of solid and liquidμ’_s,2,2 andμ’_w,2,nitrogen denitrification rate constantμ_w,3,the soil mixing layer depth d_mix,soil separation coefficient a,rainfall intensity p.It was found thatμ’_w,1,μ’_s,2,μ’_w,2,μ_w,3were not sensitive to the predicted nitrogen concentration in surface runoff or underlying the soil mixing layer with local sensitivity analysis,while parameters of p,a and d_mixwere sensitive.Therefore,Sobol’s method was applied to the prediction model to determine the sensitivity of p,a and d_mixand parameter interactions to the predicted nitrogen concentration in surface runoff or underlying the soil mixing layer.The study results showed that p was the most sensitive to the predicted nitrogen concentration in surface runoff or underlying the soil mixing layer in the early simulation time,while d_mixwas the most sensitive parameter in the later simulation time.And there was a strong interaction between p and d_mixduring the whole simulation time.However,the influence of parameter a on the predicted nitrogen concentration in surface runoff was greater than that underlying the soil mixing layer,and there was a certain interaction between a and p or d_mix.