Differential Split-sample Test Research Articles

Comparative analysis of lumped and semi-distributed hydrological models in humid Mediterranean environments

ABSTRACT This study conducts a comparative analysis of four hydrological models—GR4J, HBV-light, HEC-HMS, and WEAP—to evaluate their effectiveness in simulating surface water responses in humid Mediterranean climates. The Longaví basin, located in the Maule River basin of central Chile, was selected for its characteristic climate. The models used lumped and semi-distributed approaches to simulate the basin’s hydrological cycle. The study area was divided into three zones, considering climate, soil properties, and altitude. Results showed that GR4J, HEC-HMS, and WEAP performed well during the winter months (June to October), while HBV-light was more effective in summer (November to May). WEAP model, with its extensive parameterization, effectively captured streamflow variability during the rainy season. The Differential Split Sample Test indicated that GR4J and WEAP offered the best fit for analysing climatic variability. These models are highly recommended for climate change impact studies, given their strong performance in capturing seasonal hydrologic responses across similar regions.

Assessing rainfall-runoff models for climate change: simple and differential split-sample tests for conceptual and artificial intelligence models

ABSTRACT In this study, the performance of two popular conceptual models, a classic artificial intelligence (AI) model, and a deep learning model is compared using simple and differential split-sample tests (SSST and DSST). Finally, the two conceptual models and the classic AI model are combined employing a model ensemble technique. The SSST and DSST schemes confirmed the ability of both types of models (conceptual and AI) to capture the hydrological response to rainfall events, although the conceptual models showed a better performance in the DSST scheme. The ensemble technique enhanced the modelling performance up to 22% and 35% in the calibration and verification steps. This study highlights the importance of applying DSST using existing data to build further confidence in model projections, and suggests that the use of model ensemble techniques combining conceptual and AI models can enhance the performance of rainfall-runoff modelling in the context of climate change.

Dynamic quantitative assessment of multiple uncertainty sources in future hydropower generation prediction of cascade reservoirs with hydrological variations

Hydropower is the dominant renewable energy source, which can significantly reduce carbon dioxide emissions while providing electric energy. It is a crucial way to achieve the dual-carbon goal of carbon peak and carbon neutrality. However, climate change inevitably threatens hydropower, either directly or indirectly. Global climate models (GCMs) and hydrological models (HMs) under different representative concentration pathways (RCPs) have been combined to investigate future hydropower generation responses to climate change. The contribution of multiple uncertainty sources to multi-scale hydropower generation uncertainties with hydrological variations remains unclear. Therefore, to address the above challenge, a three-step framework was proposed in this paper: (1) predicting future inflows to reservoirs using three HMs (HBV, SWAT and XAJ) with different structures driven by six GCMs (BNU-ESM, GFDL-ESM2G, GFDL-ESM2M, CanESM2, MIROC-ESM-CHEM and CSIRO-MK3-6-0) under three RCPs (i.e., RCP2.6, RCP4.5 and RCP8.5); (2) diagnosing future hydrological conditions using the Comprehensive Differential Split-sample Test; (3) predicting future hydropower generation of cascade reservoirs by developing a cascade hydropower generation model and dynamically quantifying the contribution of multiple uncertainty sources to multi-scale hydropower generation uncertainties with hydrological conditions based on the variance decomposition method through subsampling. The results indicate that GCMs, HMs, climate emission scenarios and their interactions significantly impacted the uncertainty of hydropower generation with hydrological variations. In the flood season, the relative contributions of GCMs and HMs were the largest, accounting for about 40 %. The interaction effects of multiple uncertainty sources gradually increased before, after and during the non-flood season, with its relative contribution accounting for about 36 %. These findings can facilitate the active response to the new challenges of strategic hydropower energy development under climate change, reducing the losses caused by climate change.

Hydrological models weighting for hydrological projections: The impacts on future peak flows

Reliable hydrological projections are of great importance for climate change impact studies, especially knowing that these analyses can allow identifying regional adaptation and mitigation strategies for the future. However, the literature has highlighted that hydrological models used under climate conditions that are contrasting to those used during their calibrations can lower their performance and reliability, an issue that lowers the confidence in hydrological projections and adds uncertainty to their analyses. More studies are needed to explore this issue and evaluate potential strategies that might improve hydrological models’ reliability in climate change impact studies. Thus, the present study evaluates the robustness of hydrological models under contrasting climatic conditions and investigates the use of weighting techniques to improve their combined performance and reliability. The robustness of five lumped hydrological models is analysed using a Differential Split-Sample Testing (DSST) that evaluates their performance under cold, warm, humid and dry historical contrasting conditions over 77 basins covering different hydroclimatic conditions (two domains, one located in Quebec, Canada, and one in Mexico). Additionally, four basins were selected from the study area to evaluate the robustness of a more complex semi-distributed and more physically-based hydrological model and compare its simulations against the simpler lumped hydrological models. Based on the resulting performance of each hydrological model, five different weighting methods were applied to evaluate the potential improvements in the multi-model ensemble performance and quantify their effects on hydrological projections, particularly on future peak flows. For each basin, these streamflow projections were produced using two regional climate simulations (one per studied domain) issued from the Canadian Regional Climate Model version 5 (CRCM5) under the Representative Concentration Pathway (RCP) 8.5 for the 1976–2005, 2041–2070 and 2070–2099 periods. The results showed that weighting hydrological models, even with the most simplistic methods, showed better performances over historical contrasting conditions than the best-performing lumped hydrological model. Between the different weighting methods, the Granger-Ramanathan type A showed the overall best performance among the different basins and climate conditions, particularly in peak streamflows. Over the CRCM5-driven peak flow projections, the weighting methods Granger-Ramanathan types A and B produced the largest impacts on the projected floods magnitudes and climate change signals. On the other hand, the additional tests using the semi-distributed and more physically-based hydrological model revealed that this model showed more robust simulations than the weighted lumped hydrological models on low flows over the four selected basins. Additionally, more robust high-flow simulations were observed over a small snow-dominated basin, suggesting a potential added value in adding more complex hydrological models to simulate conditions under a changed climate.

Hydrology and hydrological extremes under climate change scenarios in the Bosque watershed, North-Central Texas, USA

This study evaluates hydrology and hydrological extremes under future climate change scenarios. The climate change scenarios were developed from multiple Global Circulation Models (GCMs), Representative Concentration Pathway (RCP) scenarios, and statistical downscaling techniques. To ensure hydrological model robustness, the Soil Water Assessment Tool (SWAT) was calibrated and validated using the Differential Split Sample Test (DSST) approach. The model was also calibrated and validated at the multi-gauges of the watershed. Future climate change scenarios revealed a reduction in precipitation (in the order of −9.1% to 4.9%) and a consistent increase in maximum temperature (0.34°C to 4.10°C) and minimum temperature (−0.15 °C to 3.7°C) in different climate model simulations. The climate change scenarios triggered a reduction of surface runoff and streamflow and a moderate increase in evapotranspiration. Future climate change scenarios projected a decrease in high flow (Q5) and low flow (Q95). A higher reduction of Q5 and annual minimum flow is also simulated in future climate scenarios, whereas an increase in annual maximum flow is simulated in climate change scenarios developed from the RCP8.5 emission scenario. The study suggests optimal water management structures which can reduce the effect of change in high and low flows.

Multi-model based soil moisture simulation approach under contrasting weather conditions

We developed and tested a multi-model based soil moisture simulation approach for improving parameterization scheme and its transferability under contrasting weather conditions. Three popular hydrologic models in combination with three optimization schemes were adapted to consider uncertainties in physical and optimization model structures for estimating near-surface soil moisture dynamics. In order to improve parameterization and its transferability under contrasting weather conditions, a multiple set of weights based on different land surface wetness conditions (wet, normal, and dry) were used using a Multiple Weighting Algorithm (MWA). A Differential Split Sample Testing (DSST) scheme was used to test the transferability of parameterizations in the optimization and physical model domains. Data from three experimental sites (Little Washita, LW13 in Oklahoma, Everglades in Florida and Bondville/Olney sites in Illinois) were used to verify our proposed approach. Our findings indicated that the multi-model outputs were highly influenced both by the physical and optimization model structures. Overall, our approach performed well in the test of transferability under different weather conditions. However, we confirmed that the overfitted parameters (to the in-situ measurements during the calibration period) due to the optimization model structures can undermine the transferability of multi-model approach. Also, our findings indicate that the estimated soil hydraulic properties during the dry years can be limited in representing the soil wetness conditions for the wet years.

The effect of weighting hydrological projections based on the robustness of hydrological models under a changing climate

Study regionThis study is developed in three catchments located in Denmark, France and Spain, covering different climate and physical conditions in Europe. Study focusThe simulation skill of hydrological models under contrasting climate conditions is evaluated using a Differential Split Sample Test (DSST). In each catchment, three different hydrological models are given a weight based on their simulation skill according to their robustness considering the DSST results for traditional and purpose-specific metrics. Four weighting approaches are used, each including a different set of evaluation metrics. The weights are applied to obtain reliable future projections of annual mean river discharge and purpose-specific metrics. New hydrological insightsProjections are found to be sensitive to model weightings in cases where the models show significantly different skills in the DSST. However, when the skills of the models are similar, there is no significant change when applying different weighting schemes. Nevertheless, the methodology proposed here increases the reliability of the purpose-for-fit hydrological projections in a climate change context.

Technical note: PMR – a proxy metric to assess hydrological model robustness in a changing climate

Abstract. The ability of hydrological models to perform in climatic conditions different from those encountered in calibration is crucial to ensure a reliable assessment of the impact of climate change on river regimes and water availability. However, most evaluation studies based on the differential split-sample test (DSST) endorsed the consensus that rainfall–runoff models lack climatic robustness. Models applied under climatologically different conditions typically exhibit substantial errors in streamflow volumes. In this technical note, we propose a new performance metric to evaluate model robustness without applying the DSST, and it can be performed with a single hydrological model calibration. The proxy for model robustness (PMR) is based on the systematic computation of model error on sliding sub-periods of the whole streamflow time series. We demonstrate that the PMR metric shows patterns similar to those obtained with the DSST for a conceptual model on a set of 377 French catchments. An analysis of the sensitivity to the length of the sub-periods shows that this length influences the values of the PMR and its equivalency with DSST biases. We recommend a range of a few years for the choice of sub-period lengths, although this should be context dependent. Our work makes it possible to evaluate the temporal transferability of any hydrological model, including uncalibrated models, at a very low computational cost.

Combining split-sample testing and hidden Markov modelling to assess the robustness of hydrological models

Abstract. The impacts of climate and land-use changes make the stationary assumption in hydrology obsolete. Moreover, there is still considerable uncertainty regarding the future evolution of the Earth’s climate and the extent of the alteration of flow regimes. Climate change impact assessment in the water sector typically involves a modelling chain in which a hydrological model is needed to generate hydrologic projections from climate forcings. Considering the inherent uncertainty of the future climate, it is crucial to assess the performance of the hydrologic model over a wide range of climates and their corresponding hydrologic conditions. In this paper, numerous, contrasted, climate sequences identified by a hidden Markov model (HMM) are used in a differential split-sample testing framework to assess the robustness of a hydrologic model. The differential split-sample test based on a HMM classification is implemented on the time series of monthly river discharges in the upper Senegal River basin in West Africa, a region characterized by the presence of low-frequency climate signals. A comparison with the results obtained using classical rupture tests shows that the diversity of hydrologic sequences identified using the HMM can help with assessing the robustness of the hydrologic model.

Modeling land use change impacts on a tropical river basin in Brazil

Water scarcity has affected the Atlantic Forest and other biomes in Brazil. To address this situation, the Brazilian Forest Code was created to control deforestation and promote sustainable development. However, how effective is this code for water production? After the cacao crises in 1980s, the Cachoeira River Basin has suffered from changes in water disponibility. Within this framework, the objective of this paper was to evaluate the Soil and Water Assessment Tool for modeling the flow behavior of the Cachoeira River Basin between 1985 and 2005 using manual calibration process under two alternative scenarios in order to obtain information that supports the guidelines regarding the future water resources management plans in the basin. Split sample test and differential split sample test were applied for calibration/validation process, and flow duration curve performance was used as a complementary procedure to attest the model. The scenario assessment results revealed that both scenarios proposed by the study did not affect significantly the hydrological services in the basin, suggesting that the land use changes that happened after the cacao crisis in the late 1980s did not affect substantially the water services in the Cachoeira River Basin.

Robustness of Process-Based versus Data-Driven Modeling in Changing Climatic Conditions

AbstractFuture climate projections require Earth system models to simulate conditions outside their calibration range. It is therefore crucial to understand the applicability of such models and their modules under transient conditions. This study assesses the robustness of different types of models in terms of rainfall–runoff modeling under changing conditions. In particular, two process-based models and one data-driven model are considered: 1) the physically based land surface model of the European Centre for Medium-Range Weather Forecasts, 2) the conceptual Simple Water Balance Model, and 3) the Long Short-Term Memory-Based Runoff model. Using streamflow data from 161 catchments across Europe, a differential split-sample test is performed, i.e., models are calibrated within a reference period (e.g., wet years) and then evaluated during a climatically contrasting period (e.g., drier years). Models show overall performance loss, which generally increases the more conditions deviate from the reference climate. Further analysis reveals that the models have difficulties in capturing temporal shifts in the hydroclimate of the catchments, e.g., between energy- and water-limited conditions. Overall, relatively high robustness is demonstrated by the physically based model. This suggests that improvements of physics-based parameterizations can be a promising avenue toward reliable climate change simulations. Further, our study illustrates that comparison across process-based and data-driven models is challenging due to their different nature. While we find rather low robustness of the data-driven model in our particular split-sample setup, this must not apply generally; by contrast, such model schemes have great potential as they can learn diverse conditions from observed spatial and temporal variability both at the same time to yield robust performance.

Sensitivity of discharge projections to potential evapotranspiration estimation in Northern Tunisia

Tunisia has a long history of coping with water scarcity, and the quantification of climate change impacts on runoff is important for future water management. A major requirement for such studies is an estimation of potential evapotranspiration (PET), which is challenging as many regions often lack the observational data needed for physically based PET equations. In this study, different PET estimation approaches were used to study the impact of PET estimation on discharge projections for catchments in Northern Tunisia. Discharge was simulated for five catchments using three rainfall-runoff models (RRMs): HBV, GR4 and IHACRES. A general differential split sample test (GDSST) was used for an RRM robustness evaluation based on subperiods with contrasting climatic conditions for the 1970–2000 period. Three cases with varying PET were considered: (1) daily calculated PET, (2) long-term daily mean PET with the same values for calibration and validation periods (calculated over the calibration period) and (3) long-term daily mean PET varying between calibration and validation periods (calculated over the calibration and validation period separately). Over the historical period, the comparison between cases 1 and 3 showed little impact of reduced PET information on the RRM performance and robustness. The comparison of cases 2 and 3 indicated a limited impact of varying PET between calibration and validation on the RRM results. The impact of varying levels of PET information on hydrological projections was also analysed over two future 30-year periods: mid-term period (2040–2070) and long-term period (2070–2100), with two representative concentration pathway scenarios (RCPs 4.5 and 8.5), by comparing cases 1 and 2. The projected discharge with constant PET (case 2) was generally lower than the projected discharge with variable PET (case 1) but the difference in volume change did not exceed 9% for both the time period and the RCP scenario considered. While PET slightly increased under the different climate change scenarios, actual evapotranspiration (AET) was found to decrease. These opposite trends of PET and AET can be attributed to the projected decrease in precipitation. Overall, our results demonstrate that discharge, in semi-arid regions like Northern Tunisia, is not sensitive to PET estimates since AET is mainly controlled by the availability of soil moisture. This finding is useful for performing studies of climate change impact on hydrological cycles in arid regions, as our study shows that simple PET estimation is a valid approach for such studies.

A general differential split-sample test to select sub-periods of discontinuous years gathering similar to different climate conditions

This article introduces a Matlab© code to implement the General Differential Split Sample Test (GDSST) (Dakhlaoui et al. [5]). As an illustration, the GDSST is applied to five catchments in northern Tunisia over 30-year reference period and compared to three benchmark Split Sample Test (SST) methods. The techniques are compared as regards to the number of validation exercises and to the differences in temperature (ΔT) and precipitation (ΔP) between the sampled sub-periods, whose length was set to 8-year. The GDSST allows a larger number of discontinuous periods to be sampled, and is computationally more effective than the basic bootstrap to identify the most climatically contrasting conditions. In addition, the GDSST offers a larger continuum of climatic conditions and a better spread of validation periods than the benchmark techniques, which is essential to test the parameter transferability of hydrological models. As supplementary material, a package file containing MATLAB© scripts to run the three benchmark SSTs and the proposed GDSST, as well as an application example on the five catchments, can be freely downloaded.•An enhanced split-sample test based on an oriented bootstrap to assess transferability of hydrological models.•The proposed split-sample test is computationally more effective than the basic bootstrap to identify the most climatically contrasting conditions.•MATLAB© code of the proposed GDSST and four benchmark SST, with application example.

Statistical bias correction of regional climate model simulations for climate change projection in the Jemma sub-basin, upper Blue Nile Basin of Ethiopia

This study evaluates bias correction methods and develops future climate scenarios using the output of a better bias correction technique at the Jemma sub-basin. The performance of different bias correction techniques was evaluated using several statistical metrics. The bias correction methods performance under climate condition different from the current climate was also evaluated using the differential split sample testing (DSST) and reveals that the distribution mapping technique is valid under climate condition different from the current climate. All bias correction methods were effective in adjusting mean monthly and annual RCM simulations of rainfall and temperature to the observed rainfall and temperature values. However, distribution mapping method was better in capturing the 90th percentile of observed rainfall and temperature and wet day probability of observed rainfall than other methods. As a result, we use the future (2021–2100) simulation of RCMs which are bias corrected using distribution mapping technique. The output of bias-adjusted RCMs unfolds a decline of rainfall, a persistent increase of temperature and an increase of extremes of rainfall and temperature in the future climate under emission scenarios of Representative Concentration Pathways 4.5, 8.5 and 2.6 (RCP4.5, RCP8.5 and RCP2.6). Thus, climate adaptation strategies that can provide optimal benefits under different climate scenarios should be developed to reduce the impact of future climate change.

Conceptual Models and Calibration Performance—Investigating Catchment Bias

Many lumped rainfall-runoff models are available but no single model can account for the uniqueness and variability of all catchments. While there has been progress in developing frameworks for optimal model selection, the process currently selects a range of model structures a priori rather than starting from the hydrological data and processes. In addition, studies on differential split sample tests (DSSTs) have focused on objective function definitions and calibration approaches. In this study, seven hydrological signatures and 12 catchment characteristics from 108 catchments around Australia were extracted for two 7-year time periods: (1) wet and (2) dry. The data was modelled using the GR4J, HBV and SIMHYD models using three objective functions to explore the relationship between model performance, catchment features and identified parameters. The hypothesis is that the hydrological signatures and catchment characteristics reflect catchment behaviour, and that certain signatures and characteristics are associated with better calibration performance. The results show that a greater percentage of catchments achieved a better calibration performance in the wet period compared to the dry period and that better calibration performance is associated with catchments that have greater cumulative flow and a steeper flow duration curve. The findings are consistent across the three models and three objective functions, suggesting that there is a bias in the studied models to wetter catchments. This study echoes the need to develop a conceptual model that can accommodate a wide variety of catchments and climates and provides a foundation to optimise and improve model selection in catchments based on their unique characteristics.

A bootstrap-based differential split-sample test to assess the transferability of conceptual rainfall-runoff models under past and future climate variability

This study proposes a general differential split-sample test (GDSST) based on an oriented bootstrap to assess the transferability of conceptual rainfall-runoff models to climatically contrasting periods. Compared to existing benchmark techniques, the GDSST allows a larger number of climatically contrasted discontinuous periods to be sampled, and is computationally more effective than the basic bootstrap to identify the most contrasted periods. When applied to three hydrological models (GR4J, HBV and IHACRES) in five catchments in northern Tunisia, the GDSST provided clear limits of the transferability of the models under changing precipitation (P) and temperature (T) conditions towards drier and hotter conditions. According to the criteria and thresholds retained, approximate limits of model transferability are drawn. The models are roughly transferable for relative changes in precipitation ΔP < (0.08 ΔT − 0.18) with ΔP ∈ [−30%, +80%], and changes in temperature ΔT ∈ [−2 °C, +2 °C]. These transferability limits suggest selecting a past sub-period as close as possible to the future climate to identify calibration parameters, which can be used for hydrological projections. The limits of transferability were then compared to climate projections by eight high-resolution Regional Climate Model (RCM) simulations resulting from the EURO-CORDEX initiative. The RCMs’ precipitation and temperature simulations of the historical period 1970–2000 were first assessed to select the most realistic ones for future projections. A delta-change monthly correction was used to perturb the observed climate series according to climate simulations under two Radiative Concentration Pathway (RCP) scenarios (RCP4.5 and RCP8.5) for one medium-term horizon (2040–2070) and one long-term horizon (2070–2100). The effects of the selected past calibration period on the hydrological projections were then analysed. The RCP 8.5 climate projections fall outside the limits of transferability of all rainfall-runoff models tested. Models calibrated on the whole observed period were found to underestimate the impacts of climate change on runoff by 5%–20% in comparison with models calibrated on sub-periods with mean annual P and T closer to projected climate conditions.

The comprehensive differential split-sample test: A stress-test for hydrological model robustness under climate variability

The choice of data periods for calibrating and evaluating conceptual hydrological models often seems ad-hoc, with no objective guidance on choosing calibration periods that produce the most reliable predictions. We therefore propose to systematically investigate the effects of calibration and validation data choices on parameter identification and predictive performance. We demonstrate our analysis on the Deggendorf/Kollbach catchment in Bavaria, Germany, for its long series of continuous hydrological and meteorological records. After classifying these data into three hydrological conditions (wet, dry and mixed) and combining them into periods of varied data length (2, 4, 8, 15, 20 and 25 years), we repeatedly calibrate a conceptual rainfall runoff hydrological model – Hydrologiska Byråns Vattenbalansavdelning (HBV) to these distinct data sets via Bayesian updating in a Monte Carlo setting. Then, we analyze predictive performance and posterior parameter statistics in various validation periods of distinct hydrological condition and time-series length. We call this the Comprehensive Differential Split-Sample Test (CDSST). Our results suggest that hydrological conditions in calibration tend to have a stronger impact than time-series length, and that calibrating on dry conditions might be a robust choice when aiming at predicting arbitrary future conditions (wet, dry or mixed). Furthermore, we found that posterior parameter estimates converged to a common optimum range with increasing data size under all investigated calibration scenarios, indicating that compensation of model structural errors by parameter fitting is independent of the chosen calibration condition. However, calibrating on time-series 8 years or longer led to overconfident predictions that failed to reliably envelope future data. While these findings are specific to our case study, we recommend using the CDSST to stress-test conceptual hydrological models to identify robust model parameters and/or deficiencies in the model structure. In general, we expect our proposed approach to be a valuable basis for model error diagnosis in any type of dynamic environmental system model, because it answers the following three questions: (1) what is the importance of physical processes not explicitly covered by the model? (2) How much overconfidence is present in the model? And (3), what are case-specific recommendations for appropriate calibration and validation setups?

Validation of a Hydrological Model Intended for Impact Study: Problem Statement and Solution Example for Selenga River Basin

The study is aimed to evaluate a hydrological simulation model intended for assessing climate change impact. A new test was suggested and applied to evaluate the performance of a physically based model of Selenga River runoff generation. In this test, to calibrate the model, an enhanced Nash–and-Sutcliffe efficiency (NSE) criterion was used, including trend-oriented reference (benchmark) models instead of the simple reference model used in the original NSE criterion. Next, modifications were made in the Differential Split Sample test (DSS-test) of V. Klemes (1986), focused on differences in the model performance criteria for climatically contrasting periods, and a new statistical measure was proposed to estimate the significance of these differences. After that, model performance was evaluated for four sites within the catchment, three indicators of interest (daily, monthly, and annual discharge series), and the model ability to reproduce the observed trends in annual and seasonal discharge values was assessed. The model proved robust enough to be applied to assessing climate change impact on the annual and monthly runoff in different parts of the Selenga River basin.

Hydrological model parameter (in)stability – “crash testing” the HBV model under contrasting flood seasonality conditions

ABSTRACTThis paper investigates the transferability of calibrated HBV model parameters under stable and contrasting conditions in terms of flood seasonality and flood generating processes (FGP) in five Norwegian catchments with mixed snowmelt/rainfall regimes. We apply a series of generalized (differential) split-sample tests using a 6-year moving window over (i) the entire runoff observation periods, and (ii) two subsets of runoff observations distinguished by the seasonal occurrence of annual maximum floods during either spring or autumn. The results indicate a general model performance loss due to the transfer of calibrated parameters to independent validation periods of −5 to −17%, on average. However, there is no indication that contrasting flood seasonality exacerbates performance losses, which contradicts the assumption that optimized parameter sets for snowmelt-dominated floods (during spring) perform particularly poorly on validation periods with rainfall-dominated floods (during autumn) and vice versa.

Groundwater recharge predictions in contrasted climate: The effect of model complexity and calibration period on recharge rates

We systematically evaluated the effect of model complexity and calibration strategy on estimated recharge using four varyingly complex models and a unique long-term recharge data set. A differential split sample test was carried out by using six calibration periods with climatically contrasting conditions in a constrained Monte Carlo approach. All models performed better during calibration than during validation due to differences in model structures and climatic conditions. The two more complex, physically-based models predicted the observed recharge with relatively small uncertainties, even when calibration and prediction periods had different climatic conditions. In contrast, the more simplistic soil-water balance model significantly underestimated the recharge rates. The fourth, semi-mechanistic model captured the observed recharge rates, but with a larger uncertainty range than the physically-based models. Our results may have relevant implications for a broad range of applications when recharge models are used as decision-making tools.