Identification of spatially distributed precipitation-runoff response routines for hourly simulation in gauged and ungauged basins

Abstract

Due to sudden and considerable flow fluctuations during hydropeaking operation, forecasting of inflow to hydropower reservoirs at high temporal resolution (e.g. hourly) is required. Prediction of streamflow for both gauged and ungauged basins using Precipitation-Runoff (P-R) models is widely employed for operational purposes. However, there are various challenging factors and inherent uncertainties in the P-R modelling. Moreover, for the boreal Norwegian catchments there are research gaps for prediction of hourly streamflow related to identification of suitable parameterizations, model structures and regionalization methods for prediction in ungauged basins (PUB). Therefore, to address some of the research gaps, comprehensive calibration and postcalibration comparative evaluation of the performances of the different runoff response routines for both catchment and regional scale modelling are required. This is required to improve the runoff simulation based on observed input climate forcing (simulation mode) and hence for the improvement of hydrological forecast (forecast mode). The objective of the first paper (P1) in this study was the identification of six different cases of explicitly resolved or probabilistic parameterizations of the spatial heterogeneity of a single state subsurface storage capacity. We conducted semidistributed and distributed simulations based on the ‘fill-and-spill’ saturation excess, infiltration excess and subsurface drainage runoff mechanisms. Equivalent performances of simulation from the different cases indicate the unidentifiably of the parameterizations and hence a preference for a parsimonious simple distributed parameterization. Identification requires more representative input climate data than a mere calibration problem. In addition, calibration only to streamflow data cannot fully identify the parameterizations. In light of the findings from the first paper, we conducted a study on identification of parametrical parsimonious and more complex configurations of the widely used conceptual Hydrologiska Byråns Vattenbalansavdelning (HBV) runoff response routine in the second paper (P2). Despite equivalent streamflow simulations for the tested HBV variants, a parametrical parsimonious HBV routine (HBV-Parsim) provided better parameter identifiability and more reliable baseflow simulation. In the other variants, considerable interactions between the soil moisture accounting and the response routine parameters and compensation between the outflow from the upper reservoir and the baseflow from the lower reservoir affect the reliability of the simulation. Hence, evaluation of the reliability of internal simulations of baseflow and soil moisture by the widely used HBV routines against tailor-made analytical methods or observations is necessary. Inspired by the preference to parsimony in P1 and P2 and the compensation between the fluxes from multiple storage reservoirs of the conceptual model in P2, our objective in P3 was geared towards the evaluation of the performance of a distributed version of a ‘top-down’ parsimonious single storage routine (hereafter named Kirchmod). We applied the principle of catchments as simple dynamical systems following Kirchner (2009) for a macroscale (3090 km2) mountainous catchment of considerable runoff delay compared to the hourly simulation. In this case, we both set the response routine parameters by estimation from streamflow recession analysis and by calibration. We obtained simulated streamflow hydrographs and flow duration curves that are in good agreement with the observed and transferability of the optimal parameter sets to the interior subcatchments validated the model. However, the parameter calibration provides slightly better simulation of peak flows than estimation from the recession analysis. In addition, the various sources of uncertainty in parameter estimation needs thorough assessment. There is no marked influence of the runoff delay due to the correlation among the free parameters, which indicates problems of parameter nonidentifiability even for the parsimonious routine. Based on the findings from the catchment scale performances of the P-R response routines in P1, P2 and P3, we wished to address the issues of Prediction in Ungauged Basins (PUB) through multi-model identification of different regionalization methods on 26 catchments in a mid-Norway study region in P4. We found that the best performing regionalization methods for the catchments vary among the model structures and evaluation metrics. However, based on the regional performances, the regionalization methods based on the single-donor physical similarity and the multidonor regional calibration corresponding to maximum regional weighted average (MRWA) performance measures (PM) performed better than the nearest neighbor and regional median parameters. The lack of data on the subsurface physical attributes and high- density hourly hydro-climatic gauging networks for the region can affect the performances of the regionalization methods, which needs scrutiny in future endeavors. The fifth objective was the identification of distributed P-R response routines relevant to operational purposes based on a multi-basin (26 catchments) local and regional calibration in P5. The best performing model structure(s) vary among the catchments and the evaluation metrics and hence there is no unique model structure that performs best for all catchments in the region. However, the Kirchmod followed by the BGM perform better than the various configurations of the HBV routine for the majority of the catchments and in terms of the regional calibration (MRWA) for the PUB. Therefore, flexible models and a multi-basin modelling framework, which allow identification of models for hourly simulation among a pool of plausible options for several catchments in the region, is better than the common single catchment model for operational purposes. In P1 to P5, we observed the challenges in identifying a unique regional P-R response routine due to the uniqueness of catchments runoff response and various sources of uncertainties. The last objective of the thesis was the development of data based statistical model and comparative evaluations against the best performing P-R response routine for hourly prediction in an ungauged and regulated basin for ecological applications in P6. A simple regional regression model based on the relationship among streamflow percentiles and catchment drainage areas, and regional transfer of streamflow information to the nearest neighbor catchment performed better than the MRWA based transfer of model parameters using the Kirchmod. We found the simple regional regression model to be useful to predict a natural time series of streamflow in a regulated river to derive ecologically relevant streamflow metrics, for assessing hydrological alterations due to regulation and hydropeaking and environmental flows.