Study Of Temporal Variability Of Model Parameters In Monthly Water Balance Modeling

Stationarity assumption of watershed conditions has been challenged under changing climate and human activities.As a simplified representation of the physical characteristic in hydrologic processes,model parameters were treated as time-invariant,which is no longer appropriate when watershed conditions change.Identification of the temporal variation of model parameters and estimation of the time-variant parameters based on watershed dynamics can contribute to an improvement of model performance,and further provide the theoretical basis and technical support for hydrological mechanism understanding and water resources planning and management.In this thesis,the technique of data assimilation was proposed to identify the temporal variation of model parameters by assimilating the runoff observations.On the basis of watershed properties,various relationships between model parameters and vegetation and climate indicators were established to improve the accuracy of runoff simulation.Furthermore,transferability of hydrological model parameters at mean annual and monthly time scale was investigated based on water balance models derived through the maximum entropy production principle.The main contents and results were summarized in the following:(1)The technique of data assimilation was proposed to identify the temporal variation of parameters for a two-parameter monthly water balance model(TWBM)by assimilating the runoff observations.Firstly,the capability of the ensemble Kalman filter(EnKF)in hydrological states updating was assessed by an application to the reservoir inflow estimation.Then the identification of hydrological model parameters variation using EnKF was started with a synthetic experiment,where different types of parameter variations and various levels of observation uncertainty were designed to examine the performance of the EnKF.The results show that the temporal variations of the parameters can be successfully captured even under a high level of observation uncertainties,which would have an influence on the performance of the EnKF.The proposed method was applied to the Wudinghe and Tongtianhe basins in China.It was found that the temporal variations of the model parameters can be interpreted from the perspective of the catchment characteristic changes.(2)Based on the watershed properties,different scenarios of hydrological model parameters were built as simple linear functions of vegetation and climate factors(i.e.,NDVI,P12 and PET12).Parameters from different scenarios were estimated using SCE-UA algorithm.All the performances were compared on the basis of case studies in Ganjiang and Tongtianhe basins.In summary,hydrological model with time-variant parameters,which were assumed to be functions of vegetation and climate indicators,can provide more accurate and credible simulations.The proposed method provided an effective tool for improving model's representativeness to the hydrological processes,which can enhance the capability of models to produce credible predictions in nonstationary hydrological modeling.(3)Model parameter transferability across time scales was investigated based on the mean annual one-parameter Budyko model and monthly water balance model,which share a similar derivation of the principle of maximum entropy production(MEP)and the generalized proportionality relationship.Two model parameters of the monthly water balance model,i.e.,k and m,were assumed to be time-variant only from month to month,namely the inter-annual variations of the parameters were not considered.Based on the hydrometeorological data set from 121 MOPEX catchments across the contiguous United States,the mean annual parameter ? was estimated from the model equation directly,while the monthly time-variant parameters k and m were calibrated using the SCE-UA algorithm,and the relationship between parameters k,m and NDVI and the frequency of rainfall events a were subsequently analyzed.The results showed that parameter k positively correlated with NDVI and a,while opposite results for parameter m.The multiple linear regression was used to fit the relationship between the s and the means and the coefficient of variations of parameters k and m,NDVI and the frequency of rainfall events a.The results showed that the estimated ? based on the mean values of k,m and a provides the best matches,and the correlation coefficients were 0.78 to 0.90.Based to the above empirical equation and the correlations between the time-variant parameters and NDVI,the monthly k and m were transferred from the mean annual parameter ?.The results showed that time-variant parameters provide improvements in monthly model efficiency;and model parameter transferability between mean annual and monthly time scales is feasible,it is useful for monthly runoff simulation in ungaged basins.