Transferability of hydrological model parameter spaces in the estimation of runoff in ungauged catchments

Abstract

In this study, transferability options of the Hydrologiska Byråns Vattenbalansavdelning (HBV) hydrological model parameter (MP) spaces were investigated to estimate ungauged catchment runoff. Three approaches were applied in the study: MP space transfer from single, neighbouring and all potential donor catchments. The model performance was evaluated by a jackknife procedure, where one catchment at a time was treated as if ungauged, and behavioural MP sets from candidate donor catchments were used to estimate the “ungauged” runoff. The results showed that ungauged catchment runoff estimation could not be guaranteed by transferring MP sets from a single physiographically nearest donor catchment. Integrating MP sets typically from one to six donor catchments supplemented the lack of effective MP sets and improved the model performance at the ungauged catchments. In addition, the analysis results revealed that the model performance converged to an average performance when the MP sets of all potential donor catchments were integrated.     

Development of a one-parameter variable source area runoff model for ungauged basins

This research develops a one-parameter model of saturated source area dynamics and the spatial distribution of soil moisture. The single required parameter is the maximum soil moisture deficit within the catchment. The concept behind the development of the model comes from the fact that the complexity of topographically-driven runoff generation can be reduced through the use of geomorphological scaling relations. The scaling formulation allows the prediction of the dynamics of saturated source areas as a function of basin-wide soil moisture state. This model offers a number of potential advantages. Firstly, the model parameter is independent of topographic index distribution and its associated scale effects. Secondly, it may be possible to measure this single parameter using field measurements or perhaps remote sensing, which gives the model significant potential for application in ungauged basins. Finally, the fact that this parameter is a physical characteristic of the basin, estimation of this parameter avoids regionalization and parameter transferability problems. The model is tested using rainfall–runoff data from the 10.4 ha experimental catchment known as Tarrawara in Australia, the 37 km² Town Creek catchment in U.S.A., and the 620 km² Balaphi and the 850 km² Likhu sub-catchments of the Koshi river in Nepal. In sub-catchments of Koshi river, the simulation results compare favorably against the calibrated TOPMODEL both in terms of direct runoff and the spatial distribution of soil moisture state. In the Tarrawara and Town Brook catchments, simulation results compare favorably against observed storm runoff using all observed data, without calibration.     

Modelling daily streamflow at ungauged catchments: what information is necessary?

Rainfall–runoff modelling at ungauged catchments often involves the transfer of calibrated model parameters from ‘donor’ gauged catchments. However, in any rainfall–runoff model, some parameters tend to be more sensitive to the objective function, whereas others are insensitive over their entire feasible range. In this paper, we analyse the effect of selectively transferring sensitive versus insensitive parameters on streamflow predictability at ungauged catchments. We develop a simple daily time‐step rainfall–runoff model [exponential bucket hydrologic model (EXP‐HYDRO)] and calibrate it at 756 catchments within the continental USA. Nash–Sutcliffe efficiency of (NS) is used as the objective function. The model simulates satisfactorily at 323 catchments (NS > 0.6), most of which are located in the eastern part of the USA, along the Rocky Mountain Range, and near the western Pacific coast. Of the six calibration parameters, only three parameters are found to be sensitive to NS. Two of these parameters control the hydrograph recession behaviour of a catchment, and the third parameter controls the snowmelt rate. We find that when only sensitive parameters are transferred, model performance at ungauged catchments is almost at par with that of transferring all six parameters. Conversely, the transfer of only insensitive parameters results in a significant deterioration in model performance. Results suggest that streamflow predictability at ungauged catchments using rainfall–runoff models is largely dependent on the transfer of a small subset of parameters. We recommend that, in any modelling framework, such parameters should be identified and further characterized to better understand the information controlling streamflow predictability at ungauged catchments. Copyright © 2012 John Wiley & Sons, Ltd.     

Machine learning based identification of dominant controls on runoff dynamics

Hydrological models used for flood prediction in ungauged catchments are commonly fitted to regionally transferred data. The key issue of this procedure is to identify hydrologically similar catchments. Therefore, the dominant controls for the process of interest have to be known. In this study, we applied a new machine learning based approach to identify the catchment characteristics that can be used to identify the active processes controlling runoff dynamics. A random forest (RF) regressor has been trained to estimate the drainage velocity parameters of a geomorphologic instantaneous unit hydrograph (GIUH) in ungauged catchments, based on regionally available data. We analyzed the learning procedure of the algorithm and identified preferred donor catchments for each ungauged catchment. Based on the obtained machine learning results from catchment grouping, a classification scheme for drainage network characteristics has been derived. This classification scheme has been applied in a flood forecasting case study. The results demonstrate that the RF could be trained properly with the selected donor catchments to successfully estimate the required GIUH parameters. Moreover, our results showed that drainage network characteristics can be used to identify the influence of geomorphological dispersion on the dynamics of catchment response.     

Modified NLPM-ANN model and its application

The non-linear perturbation model based on artificial neural network (NLPM-ANN) takes advantage of the consideration of seasonal information by the linear perturbation model (LPM) and the notable non-linear simulation capability of artificial neural network (ANN). However, this model does not take account of antecedent catchment wetness that may effect the simulation and forecasting accuracy. A modified NLPM-ANN model is proposed and developed to take the consideration of antecedent catchment wetness. The output perturbing terms of the response function in the simple linear model (SLM) in an auxiliary component are taken as inputs of ANN to represent catchment wetness. The simulated total runoff is obtained by integrating the outputs of ANN with that of the seasonal model. The rainfall–runoff data of eight catchments were selected and used to compare the modified NLPM-ANN with the NLPM-ANN models. Results show that the modified NLPM-ANN is significantly superior to the NLPM-ANN, and the model component efficiency index values are 16.82% and 16.74% over the NLPM-ANN during calibration and verification periods, respectively.     

Vegetation characteristics and the prediction of runoff: Analysis of an experiment in the New Forest,Hampshire

Vegetation characteristics have not been sufficiently utilized in catchment runoff models. An analysis of storm hydrograph data from nested subareas of the Highland Water catchment, New Forest, U.K., indicates that depth of runoff and peak discharge from areas under heathland cover is substantially greater than from areas under woodland cover at several spatial scales. The significance of heath vegetation composition in the identification of runoff contributing areas is illustrated by an analysis of vegetation composition, water table depth, baseflow discharge and storm runoff from areas predominantly covered by heathland. Methods are proposed to employ the hydrological characteristics of heathland to refine and develop the Flood Studies Approach to discharge estimation in ungauged heathland catchments. Such an approach is greatly facilitated by the use of remotely-sensed data.     

Storm‐event rainfall–runoff modelling approach for ungauged sites in Taiwan

This work develops a top‐down modelling approach for storm‐event rainfall–runoff model calibration at unmeasured sites in Taiwan. Twenty‐six storm events occurring in seven sub‐catchments in the Kao‐Ping River provided the analytical data set. Regional formulas for three important features of a streamflow hydrograph, i.e. time to peak, peak flow, and total runoff volume, were developed via the characteristics of storm event and catchment using multivariate regression analysis. Validation of the regional formulas demonstrates that they reasonably predict the three features of a streamflow hydrograph at ungauged sites. All of the sub‐catchments in the study area were then adopted as ungauged areas, and the three streamflow hydrograph features were calculated by the regional formulas and substituted into the fuzzy multi‐objective function for rainfall–runoff model calibration. Calibration results show that the proposed approach can effectively simulate the streamflow hydrographs at the ungauged sites. The simulated hydrographs more closely resemble observed hydrographs than hydrographs synthesized using the Soil Conservation Service (SCS) dimensionless unit hydrograph method, a conventional method for hydrograph estimation at ungauged sites in Taiwan. Copyright © 2008 John Wiley & Sons, Ltd.     

Parameter uncertainty of the AWBM model when applied to an ungauged catchment

In this study, a quantitative assessment of uncertainty was made in connection with the calibration of Australian Water Balance Model (AWBM) for both gauged and ungauged catchment cases. For the gauged catchment, five different rainfall data sets, 23 different calibration data lengths and eight different optimization techniques were adopted. For the ungauged catchment case, the optimum parameter sets obtained from the nearest gauged catchment were transposed to the ungauged catchments, and two regional prediction equations were used to estimate runoff. Uncertainties were ascertained by comparing the observed and modelled runoffs by the AWBM on the basis of different combinations of methods, model parameters and input data. The main finding from this study was that the uncertainties in the AWBM modelling outputs could vary from ?1.3% to 70% owing to different input rainfall data, ?5.7% to 11% owing to different calibration data lengths and ?6% to 0.2% owing to different optimization techniques adopted in the calibration of the AWBM. The performance of the AWBM model was found to be dominated mainly by the selection of appropriate rainfall data followed by the selection of an appropriate calibration data length and optimization algorithm. Use of relatively short data length (e.g. 3 to 6 years) in the calibration was found to generate relatively poor results. Effects of different optimization techniques on the calibration were found to be minimal. The uncertainties reported here in relation to the calibration and runoff estimation by the AWBM model are relevant to the selected study catchments, which are likely to differ for other catchments. The methodology presented in this paper can be applied to other catchments in Australia and other countries using AWBM and similar rainfall–runoff models. Copyright © 2014 John Wiley & Sons, Ltd.     

Streamflow prediction under extreme data scarcity: a step toward hydrologic process understanding within severely data-limited regions

Streamflow prediction in ungauged basins is necessary to support water resources management decisions. Herein we refine and evaluate the Streamflow Prediction under Extreme Data-scarcity (SPED) model, a framework designed for streamflow prediction within regions of sparse hydrometeorological observation. With the SPED framework, inclusion of soft data directs optimization to balance runoff efficiency with the selection of hydrologically representative parameters. Here SPED is tested in catchments around the world, including four well-gauged catchments, by mimicking data-scarcity and comparing against data-intensive approaches. By differentiating equifinal models, SPED succeeds where traditional approaches are likely to fail: partially dissimilar reference/target catchments. For instance, in a pair of reference/target catchments with different base flow regimes, SPED outperforms a model calibrated only to maximize efficiency (NSE of 0.54 versus 0.08). SPED performs consistently (NSE range: 0.54–0.74) across the diverse climatological and physiographic settings tested and proves comparable to state-of-the-science methods that use robust data networks.     

新安江模型在资料匮乏的长江中下游山区中小流域洪水预报应用

水文资料匮乏流域的洪水预报(PUBs)是水文科学与工程中一个尚未解决的重大挑战.中国湿润山区中小流域大多是水文资料匮乏的流域,在此地区进行洪水预报的重要手段之一就是水文模型参数的估计.对基于参数物理意义的估算方法(以下简称物理估算法)及两种区域化方法进行了研究,将其用于新安江模型参数的估算及移植.皖南山区的29个中小流...     

How informative are stream level observations in different geographic regions?

Simple runoff models with a low number of model parameters are generally able to simulate catchment runoff reasonably well, but they rely on model calibration, which makes their use in ungauged basins challenging. In a previous study it has been shown that a limited number of streamflow measurements can be quite informative for constraining runoff models. In practice, however, instead of performing such repeated flow measurements, it might be easier to install a stream level logger. Here, a dataset of 600+ gauged basins in the USA was used to study how well models perform when only stream level data, rather than streamflow data, are available. A runoff model (the HBV model) was calibrated assuming that only stream level observations were available, and the simulations were evaluated on the full observed streamflow record. The results indicate that stream level data alone can already provide surprisingly good model simulation results in humid catchments, whereas in arid catchments some form of quantitative information (e.g. a streamflow observation or a regional average value) is needed to obtain good results. These results are encouraging for hydrological observations in data scarce regions as level observations are much easier to obtain than streamflow measurements. Based on runoff modelling, it might even be possible to derive streamflow time series from the level data obtained from loggers, satellites or community‐based approaches. Copyright © 2016 John Wiley & Sons, Ltd.     

Predicting dry‐season flows with a monthly rainfall–runoff model: Performance for gauged and ungauged catchments

Hydrologic models are useful to understand the effects of climate and land‐use changes on dry‐season flows. In practice, there is often a trade‐off between simplicity and accuracy, especially when resources for catchment management are scarce. Here, we evaluated the performance of a monthly rainfall–runoff model (dynamic water balance model, DWBM) for dry‐season flow prediction under climate and land‐use change. Using different methods with decreasing amounts of catchment information to set the four model parameters, we predicted dry‐season flow for 89 Australian catchments and verified model performance with an independent dataset of 641 catchments in the United States. For the Australian catchments, model performance without catchment information (other than climate forcing) was fair; it increased significantly as the information to infer the four model parameters increased. Regressions to infer model parameters from catchment characteristics did not hold for catchments in the United States, meaning that a new calibration effort was needed to increase model performance there. Recognizing the interest in relative change for practical applications, we also examined how DWBM could be used to simulate a change in dry‐season flow following land‐use change. We compared results with and without calibration data and showed that predictions of changes in dry‐season flow were robust with respect to uncertainty in model parameters. Our analyses confirm that climate is a strong driver of dry‐season flow and that parsimonious models such as DWBM have useful management applications: predicting seasonal flow under various climate forcings when calibration data are available and providing estimates of the relative effect of land use on seasonal flow for ungauged catchments.     

PAI-OFF: A new proposal for online flood forecasting in flash flood prone catchments

The Process Modelling and Artificial Intelligence for Online Flood Forecasting (PAI-OFF) methodology combines the reliability of physically based, hydrologic/hydraulic modelling with the operational advantages of artificial intelligence. These operational advantages are extremely low computation times and straightforward operation. The basic principle of the methodology is to portray process models by means of ANN. We propose to train ANN flood forecasting models with synthetic data that reflects the possible range of storm events. To this end, establishing PAI-OFF requires first setting up a physically based hydrologic model of the considered catchment and – optionally, if backwater effects have a significant impact on the flow regime – a hydrodynamic flood routing model of the river reach in question. Both models are subsequently used for simulating all meaningful and flood relevant storm scenarios which are obtained from a catchment specific meteorological data analysis. This provides a database of corresponding input/output vectors which is then completed by generally available hydrological and meteorological data for characterizing the catchment state prior to each storm event. This database subsequently serves for training both a polynomial neural network (PoNN) – portraying the rainfall–runoff process – and a multilayer neural network (MLFN), which mirrors the hydrodynamic flood wave propagation in the river. These two ANN models replace the hydrological and hydrodynamic model in the operational mode. After presenting the theory, we apply PAI-OFF – essentially consisting of the coupled “hydrologic” PoNN and “hydrodynamic” MLFN – to the Freiberger Mulde catchment in the Erzgebirge (Ore-mountains) in East Germany (3000 km²). Both the demonstrated computational efficiency and the prediction reliability underline the potential of the new PAI-OFF methodology for online flood forecasting.     

Effective impervious area for runoff in urban watersheds

Effective impervious area (EIA), or the portion of total impervious area (TIA) that is hydraulically connected to the storm sewer system, is an important parameter in determining actual urban runoff. EIA has implications in watershed hydrology, water quality, environment, and ecosystem services. The overall goal of this study is to evaluate the application of successive weighted least square (WLS) method to urban catchments with different sizes and various hydrologic conditions to determine EIA fraction. Other objectives are to develop insights on the data selection issues, EIA fraction, EIA/TIA ratio, and runoff source area patterns in urban catchments. The successive WLS method is applied to 50 urban catchments with different sizes from less than 1 ha to more than 2000 ha in Minnesota, Wisconsin, Texas, USA as well as Europe, Canada, and Australia. The average, median, and standard deviation of EIA fractions for the 42 catchments with residential land uses are found to be 0.222, 0.200, and 0.113, respectively. These values for the EIA/TIA ratio in the same 42 catchments are 0.50, 0.48, and 0.21, respectively. While the EIA/TIA results indicate the importance of EIA, 95% prediction interval of the mean EIA/TIA is found to be 0.07 to 0.93, which shows that using an average value for this ratio in each land use to determine EIA from TIA in ungauged urban watersheds can be misleading. The successive WLS method was robust and is recommended for determining EIA in gauged urban catchments. Copyright © 2016 John Wiley & Sons, Ltd.     

Evaluating sources of uncertainty in Australian runoff projections

Generating estimates of the future impacts of climate change on human and natural systems is confounded by cascading uncertainties which propagate through the impact assessment. Here, a simple stochastic rainfall–runoff model representing 238 river basins on the Australian continent was used to assess the sensitivity of the risk of runoff changes to various sources of uncertainty. Uncertainties included global mean temperature change, greenhouse gas stabilisation targets, catchment sensitivities to climatic change, and the seasonality of runoff, rainfall, and evaporation. Model simulations provided estimates of the first-order risk of climate change to Australian catchments, with several regions having high likelihoods of experiencing significant reductions in future runoff. Climate uncertainty (at global and regional scales) was identified as the dominant driving force in hydrological risk assessments. Uncertainties in catchment sensitivities to climatic changes also influenced risk, provided they were sufficiently large, whereas structural assumptions of the model were generally negligible. Collectively, these results indicate that rigorous assessment of climate risk to water resources over relatively long time-scales is largely a function of adequately exploring the uncertainty space of future climate changes.     

Application of artificial neural networks in regional flood frequency analysis: a case study for Australia

Regional flood frequency analysis (RFFA) is widely used in practice to estimate flood quantiles in ungauged catchments. Most commonly adopted RFFA methods such as quantile regression technique (QRT) assume a log-linear relationship between the dependent and a set of predictor variables. As non-linear models and universal approximators, artificial neural networks (ANN) have been widely adopted in rainfall runoff modeling and hydrologic forecasting, but there have been relatively few studies involving the application of ANN to RFFA for estimating flood quantiles in ungauged catchments. This paper thus focuses on the development and testing of an ANN-based RFFA model using an extensive Australian database consisting of 452 gauged catchments. Based on an independent testing, it has been found that ANN-based RFFA model with only two predictor variables can provide flood quantile estimates that are more accurate than the traditional QRT. Seven different regions have been compared with the ANN-based RFFA model and it has been shown that when the data from all the eastern Australian states are combined together to form a single region, the ANN presents the best performing RFFA model. This indicates that a relatively larger dataset is better suited for successful training and testing of the ANN-based RFFA models.     

Effects of geological structures on rainfall-runoff responses in headwater catchments in a sedimentary rock mountain

Clarifying rainfall-runoff responses in mountainous areas is essential for disaster prediction as well as water resource management. Although runoff is considered to be significantly affected by topography, some previous studies have reported that geological structures also have significant effects on rainfall-runoff characteristics. Particularly in headwater catchments located in sedimentary rock mountains, dips and strikes may significantly affect rainwater discharge. In this study, the effects of geological structures on rainfall-runoff characteristics were investigated based on observed discharge hydrographs from 12 catchments, which lie radially from the summit of a sedimentary rock mountain. The results obtained were as follows: (1) Even though the topographic wetness index (TWI) distributions of the 12 catchments were similar, there were significant differences in their runoff characteristics; (2) Catchments with average flow direction oriented towards the strike direction (strike-oriented catchments) are characterized by large baseflows; (3) Catchments with average flow direction oriented towards the opposite dip direction (opposite dip-oriented catchments) are steep, and this results in quick storm runoff generation; (4) Catchments with average flow direction oriented toward the dip direction (dip-oriented catchments) are gentle, and this results in delayed storm runoff generation. It was presumed that in strike-oriented catchments, large quantities of groundwater flowing along the bedding planes owing to hydraulic anisotropy, exfiltrate and sustain the large amount of the observed baseflow, that is, in strike-oriented catchments, runoff is directly controlled by geological structures. Conversely, in opposite dip-oriented and dip-oriented catchments, runoff is indirectly controlled by geological structures, that is, geological structures affect slope gradients, which result in differences in storm runoff generation. Thus, this study clearly illustrates that geological structures significantly affect rainfall-runoff responses in headwater catchments located in sedimentary rock mountains.     

Remotely sensed ET for streamflow modelling in catchments with contrasting flow characteristics: an attempt to improve efficiency

An efficient calibration with remotely sensed (RS) data is important for accurate predictions at ungauged catchments. This study investigates the advantages of streamflow-sensitive regionalization on calibration with RS evapotranspiration (ET). Regionalization experiments are performed at 28 catchments in Australia. The catchments are classified into three groups based on annual rainfall and runoff coefficients. Streamflow, RS ET, and a multi-objective RS ET-streamflow calibration are performed using the DiffeRential Evolution Adaptive Metropolis algorithm in each catchment. Simplified Australian Water Resource Assessment-Landscape model is calibrated for a selection of five parameters. Posterior probability distributions of parameters from three calibrations performed at donor catchments in each group are inspected to find the parameter for regionalization in the individual group. In group 1 of wetter catchments, regionalization of parameter FsoilEmax (soil evaporation scaling factor) helps to simplify the calibration without any deterioration in ET, soil moisture (SM) and streamflow predictions. Regionalization of parameter Beta (coefficient describing rate of hydraulic conductivity increase with water content) in group 2 assists to improve the streamflow predictions with no decrement in ET and SM predictions. However, regionalization is not able to provide satisfactory results in group 3. Group 3 includes low-yielding catchments, with average annual rainfall below 1000 mm/year and runoff coefficient less than 0.1, where traditional streamflow calibration also fails to produce accurate results. This study concludes that streamflow-sensitive regionalization is effective for improving the efficacy of RS ET calibration in wetter catchments.     

Controls on event runoff coefficients in the eastern Italian Alps

Analyses of event runoff coefficients provide essential insight on catchment response, particularly if a range of catchments and a range of events are compared by a single indicator. In this study we examine the effect of climate, geology, land use, flood types and initial soil moisture conditions on the distribution functions of the event runoff coefficients for a set of 14 mountainous catchments located in the eastern Italian Alps, ranging in size from 7.3 to 608.4 km². Runoff coefficients were computed from hourly precipitation, runoff data and estimates of snowmelt. A total of 535 events were analysed over the period 1989–2004. We classified each basin using a “permeability index” which was inferred from a geologic map and ranged from “low” to “high permeability”. A continuous soil moisture accounting model was applied to each catchment to classify ‘wet’ and ‘dry’ initial soil moisture conditions. The results indicate that the spatial distribution of runoff coefficients is highly correlated with mean annual precipitation, with the mean runoff coefficient increasing with mean annual precipitation. Geology, through the ‘permeability index’, is another important control on runoff coefficients for catchments with mean annual precipitation less than 1200 mm. Land use, as indexed by the SCS curve number, influences runoff coefficient distribution to a lesser degree. An analysis of the runoff coefficients by flood type indicates that runoff coefficients increase with event snowmelt. Results show that there exists an intermediate region of subsurface water storage capacity, as indexed by a flow–duration curve-based index, which maximises the impact of initial wetness conditions on the runoff coefficient. This means that the difference between runoff coefficients characterised by wet and dry initial conditions is negligible both for basins with very large storage capacity and for basins with small storage capacity. For basins with intermediate storage capacities, the impact of the initial wetness conditions may be relatively large.     

Water balance modeling over variable time scales based on the Budyko framework – Model development and testing

Partitioning of precipitation into evapotranspiration and runoff is controlled by climate and catchment characteristics. The degree of control exerted by these factors varies with the spatial and temporal scales of processes modeled. The Budyko framework or the “limits” concept was used to model water balance at four temporal scales (mean annual, annual, monthly and daily). The method represents a top-down approach to hydrologic modeling and is expected to achieve parsimony of model parameters. Daily precipitation, potential evapotranspiration, and streamflow from 265 catchments in Australia were used. On a mean annual basis, the index of dryness defined as the ratio of potential evapotranspiration to precipitation was confirmed to be a dominant factor in determining the water balance with one model parameter. Analysis of the data, however, suggested increased model complexity is necessary on finer time scale such as monthly. In response, the Budyko framework for mean annual water balance was extended to include additional factors and this resulted in a parsimonious lumped conceptual model on shorter-time scale. The model was calibrated and tested against measured streamflow at variable time scales and showed promising results. The strengths of the model are consistent water balance relationships across different time scales, and model parsimony and robustness. As result, the model has the potential to be used to predict streamflow for ungauged catchments.