Search

Lack of discharge data for model calibration is challenging for flood prediction in ungauged basins. Since establishment and maintenance of a permanent discharge station is resource demanding, a possible remedy could be to measure discharge only for a few events. We tested the hypothesis that a few flood-event hydrographs in a tropical basin would be sufficient to calibrate a bucket-type rainfall–runoff model, namely the HBV model, and proposed a new event-based calibration method to adequately predict floods. Parameter sets were chosen based on calibration of different scenarios of data availability, and their ability to predict floods was assessed. Compared to not having any discharge data, flood predictions improved already when one event was used for calibration. The results further suggest that two to four events for calibration may considerably improve flood predictions with regard to accuracy and uncertainty reduction, whereas adding more events beyond this resulted in small performance gains.

Discharge observations and reliable rainfall forecasts are essential for flood prediction but their availability and accuracy are often limited. However, even scarce data may still allow adequate flood forecasts to be made. Here, we explored how far using limited discharge calibration data and uncertain forcing data would affect the performance of a bucket-type hydrological model for simulating floods in a tropical basin. Three events above thresholds with a high and a low frequency of occurrence were used in calibration and 81 rainfall scenarios with different degrees of uncertainty were used as input to assess their effects on flood predictions. Relatively similar model performance was found when using calibrated parameters based on a few events above different thresholds. Flood predictions were sensitive to rainfall errors, but those related to volume had a larger impact. The results of this study indicate that a limited number of events can be useful for predicting floods given uncertain rainfall forecasts.

Central America is frequently affected by droughts that cause significant socio-economic and environmental problems. Drought characterisation, monitoring and forecasting are potentially useful to support water resource management. Drought indices are designed for these purposes, but their ability to characterise droughts depends on the characteristics of the regional climate and the quality of the available data. Local comprehensive and high-quality observational networks of meteorological and hydrological data are not available, which limits the choice of drought indices and makes it important to assess available datasets. This study evaluated which combinations of drought index and meteorological dataset were most suitable for characterising droughts in the region. We evaluated the standardised precipitation index (SPI), a modified version of the deciles index (DI), the standardised precipitation evapotranspiration index (SPEI) and the effective drought index (EDI). These were calculated using precipitation data from the Climate Hazards Group Infra-Red Precipitation with Station (CHIRPS), the CRN073 dataset, the Climate Research Unit (CRU), ECMWF Reanalysis (ERA-Interim) and a regional station dataset, and temperature from the CRU and ERA-Interim datasets. The gridded meteorological precipitation datasets were compared to assess how well they captured key features of the regional climate. The performance of all the drought indices calculated with all the meteorological datasets was then evaluated against a drought index calculated using river discharge data. Results showed that the selection of database was more important than the selection of drought index and that the best combinations were the EDI and DI calculated with CHIRPS and CRN073. Results also highlighted the importance of including indices like SPEI for drought assessment in Central America.

Lack of discharge data for model calibration is challenging for flood prediction in ungauged basins. Since establishment and maintenance of a permanent discharge station is resource demanding, a possible remedy could be to measure discharge only for a few events. We tested the hypothesis that a few flood-event hydrographs in a tropical basin would be sufficient to calibrate a bucket-type rainfall-runoff model, namely the HBV model, and proposed a new event-based calibration method to adequately predict floods. Parameter sets were chosen based on calibration of different scenarios of data availability, and their ability to predict floods was assessed. Compared to not having any discharge data, flood predictions improved already when one event was used for calibration. The results further suggest that two to four events for calibration may considerably improve flood predictions with regard to accuracy and uncertainty reduction, whereas adding more events beyond this resulted in small performance gains.

The performance of hydrological models is affected by uncertainty related to observed climatological and discharge data. Although the latter has been widely investigated, the effects on hydrological models from different starting times of the day have received little interest. In this study, observational data from one tropical basin were used to investigate the effects on a typical bucket-type hydrological model, the HBV, when the definitions of the climatological and discharge days are changed. An optimization procedure based on a genetic algorithm was used to assess the effects on model performance. Nash-Sutcliffe efficiencies varied considerably between day definitions, with the largest dependence on the climatological-day definition. The variation was likely caused by how storm water was assigned to one or two daily rainfall values depending on the definition of the climatological day. Hydrological models are unlikely to predict high flows accurately if rainfall intensities are reduced because of the day definition.

The performance of hydrologicalmodels is affected by uncertainty related to observed climatological and discharge data. Although the latter has been widely investigated, the effects on hydrological models from different starting times of the day have received little interest. In this study, observational data from one tropical basin were used to investigate the effects on a typical bucket-type hydrological model, the HBV, when the definitions of the climatological and discharge days are changed. An optimization procedure based on a genetic algorithm was used to assess the effects on model performance. Nash-Sutcliffe efficiencies varied considerably between day definitions, with the largest dependence on the climatological-day definition. The variation was likely caused by how storm water was assigned to one or two daily rainfall values depending on the definition of the climatological day. Hydrological models are unlikely to predict high flows accurately if rainfall intensities are reduced because of the day definition.

Studies for the prevention and mitigation of floods require information on discharge and extent of inundation, commonly unavailable or uncertain, especially during extreme events. This study was initiated by the devastating flood in Tegucigalpa, the capital of Honduras, when Hurricane Mitch struck the city. In this study we hypothesized that it is possible to estimate, in a trustworthy way considering large data uncertainties, this extreme 1998 flood discharge and the extent of the inundations that followed from a combination of models and post-event measured data. Postevent data collected in 2000 and 2001 were used to estimate discharge peaks, times of peak, and high-water marks. These data were used in combination with rain data from two gauges to drive and constrain a combination of well-known modelling tools: TOPMODEL, Muskingum-Cunge-Todini routing, and the LISFLOOD-FP hydraulic model. Simulations were performed within the generalized likelihood uncertainty estimation (GLUE) uncertainty-analysis framework. The model combination predicted peak discharge, times of peaks, and more than 90% of the observed highwater marks within the uncertainty bounds of the evaluation data. This allowed an inundation likelihood map to be produced. Observed high-water marks could not be reproduced at a few locations on the floodplain. Identifications of these locations are useful to improve model set-up, model structure, or post-event data-estimation methods. Rainfall data were of central importance in simulating the times of peak and results would be improved by a better spatial assessment of rainfall, e. g. from radar data or a denser rain-gauge net-work. Our study demonstrated that it was possible, considering the uncertainty in the post-event data, to reasonably reproduce the extreme Mitch flood in Tegucigalpa in spite of no hydrometric gauging during the event. The method proposed here can be part of a Bayesian framework in which more events can be added into the analysis as t

Most conceptual rainfall–runoff models use as input spatially averaged rainfall fields which are typically associated with significant errors that affect the model outcome. In this study, it was hypothesised that a simple spatially and temporally averaged event-dependent rainfall multiplier can account for errors in the rainfall input. The potentials and limitations of this lumped multiplier approach were explored by evaluating the effects of multipliers on the accuracy and precision of the predictive distributions. Parameter sets found to be behavioural across a range of different flood events were assumed to be a good representation of the catchment dynamics and were used to identify rainfall multipliers for each of the individual events. An effect of the parameter sets on identified multipliers was found; however, it was small compared to the differences between events. Accounting for event-dependent multipliers improved the reliability of the predictions. At the cost of a small decrease in precision, the distribution of identified multipliers for past events can be used to account for possible rainfall errors when predicting future events. By using behavioural parameter sets to identify rainfall multipliers, the method offers a simple and computationally efficient way to address rainfall errors in hydrological modelling.

Concentration times in small and medium-sized basins (similar to 10-1000 km(2)) are commonly less than 24 h. Flood-forecasting models are thus required to provide simulations at high temporal resolutions (1 h-6 h), although time-series of input and runoff data with sufficient lengths are often only available at the daily temporal resolution, especially in developing countries. This has led to study the relationships of estimated parameter values at the temporal resolutions where they are needed from the temporal resolutions where they are available. This study presents a methodology to treat empirically model parameter dependencies on the temporal resolution of data in two small basins using a bucket-type hydrological model, HBV-light, and the generalised likelihood uncertainty estimation approach for selecting its parameters. To avoid artefacts due to the numerical resolution or numerical method of the differential equations within the model, the model was consistently run using modelling time steps of one-hour regardless of the temporal resolution of the rainfall-runoff data. The distribution of the parameters calibrated at several temporal resolutions in the two basins did not show model parameter dependencies on the temporal resolution of data and the direct transferability of calibrated parameter sets (e.g., daily) for runoff simulations at other temporal resolutions for which they were not calibrated (e.g., 3 h or 6 h) resulted in a moderate (if any) decrease in model performance, in terms of Nash-Sutcliffe and volume-error efficiencies. The results of this study indicate that if sub-daily forcing data can be secured, flood forecasting in basins with sub-daily concentration times may be possible with model-parameter values calibrated from long time series of daily data. Further studies using more models and basins are required to test the generality of these results.

Studies for the prevention and mitigation of floods require information on discharge and extent of inundation, commonly unavailable or uncertain, especially during extreme events. This study was initiated by the devastating flood in Tegucigalpa, the capital of Honduras, when Hurricane Mitch struck the city. In this study we hypothesized that it is possible to estimate, in a trustworthy way considering large data uncertainties, this extreme 1998 flood discharge and the extent of the inundations that followed from a combination of models and post-event measured data. Postevent data collected in 2000 and 2001 were used to estimate discharge peaks, times of peak, and high-water marks. These data were used in combination with rain data from two gauges to drive and constrain a combination of well-known modelling tools: TOPMODEL, Muskingum-Cunge-Todini routing, and the LISFLOOD-FP hydraulic model. Simulations were performed within the generalized likelihood uncertainty estimation (GLUE) uncertainty-analysis framework. The model combination predicted peak discharge, times of peaks, and more than 90% of the observed highwater marks within the uncertainty bounds of the evaluation data. This allowed an inundation likelihood map to be produced. Observed high-water marks could not be reproduced at a few locations on the floodplain. Identifications of these locations are useful to improve model set-up, model structure, or post-event data-estimation methods. Rainfall data were of central importance in simulating the times of peak and results would be improved by a better spatial assessment of rainfall, e. g. from radar data or a denser rain-gauge net-work. Our study demonstrated that it was possible, considering the uncertainty in the post-event data, to reasonably reproduce the extreme Mitch flood in Tegucigalpa in spite of no hydrometric gauging during the event. The method proposed here can be part of a Bayesian framework in which more events can be added into the analysis as t

Most conceptual rainfall–runoff models use as input spatially averaged rainfall fields which are typically associated with significant errors that affect the model outcome. In this study, it is hypothesized that a simple spatially and temporally averaged event–dependent rainfall multiplier can account for errors in the rainfall input. The potentials and limitations of this lumped multiplier approach are explored by evaluating the effects of multipliers on the accuracy and precision of the predictive distributions. Parameter sets found to be behavioural across a range of different flood events were assumed to be a good representation of the catchment dynamics and were used to identify rainfall multipliers for each of the individual events. An effect of the parameter sets on identified multipliers was found, however it was small compared to the differences between events. Accounting for event–dependent multipliers improved the reliability of the predictions. At the cost of a small decrease in precision, the distribution of identified multipliers for past events can be used to account for possible rainfall errors when predicting future events. By using behavioural parameter sets to identify rainfall multipliers, the method offers a simple and computationally efficient way to address rainfall errors in hydrological modelling.

There has been a significant increase in the skill and resolution of numerical weather prediction models (NWPs) in recent decades, extending the time scales of useful weather predictions. The land surface models (LSMs) of NWPs are often employed in hydrological applications, which raises the question of how hydrologically representative LSMs really are. In this paper, precipitation (P), evaporation (E), and runoff (R) from the European Centre for Medium-Range Weather Forecasts global models were evaluated against observational products. The forecasts differ substantially from observed data for key hydrological variables. In addition, imbalanced surface water budgets, mostly caused by data assimilation, were found on both global (P-E) and basin scales (P-E-R), with the latter being more important. Modeled surface fluxes should be used with care in hydrological applications, and further improvement in LSMs in terms of process descriptions, resolution, and estimation of uncertainties is needed to accurately describe the land surface water budgets.

Robust and reliable water-resource mapping in ungauged basins requires estimation of the uncertainties in the hydrologic model, the regionalisation method, and the observational data. In this study we investigated the use of regionalised flow-duration curves (FDCs) for constraining model predictive uncertainty, while accounting for all these uncertainty sources. A water balance model was applied to 36 basins in Central America using regionally and globally available precipitation, climate and discharge data that were screened for inconsistencies. A rating-curve analysis for 35 Honduran discharge stations was used to estimate discharge uncertainty for the region, and the consistency of the model forcing and evaluation data was analysed using two different screening methods. FDCs with uncertainty bounds were calculated for each basin, accounting for both discharge uncertainty and, in many cases, uncertainty stemming from the use of short time series, potentially not representative for the modelling period. These uncertain FDCs were then used to regionalise a FDC for each basin, treating it as ungauged in a cross-evaluation, and this regionalised FDC was used to constrain the uncertainty in the model predictions for the basin. There was a clear relationship between the performance of the local model calibration and the degree of data set consistency - with many basins with inconsistent data lacking behavioural simulations (i.e. simulations within predefined limits around the observed FDC) and the basins with the highest data set consistency also having the highest simulation reliability. For the basins where the regionalisation of the FDCs worked best, the uncertainty bounds for the regionalised simulations were only slightly wider than those for a local model calibration. The predicted uncertainty was greater for basins where the result of the FDC regionalisation was more uncertain, but the regionalised simulations still had a high reliability compared to the locally

Robust and reliable water-resource mapping in ungauged basins requires estimation of the uncertainties in the hydrologic model, the regionalisation method, and the observational data. In this study we investigated the use of regionalised flow-duration curves (FDCs) for constraining model predictive uncertainty, while accounting for all these uncertainty sources. A water balance model was applied to 36 basins in Central America using regionally and globally available precipitation, climate and discharge data that were screened for inconsistencies. A rating-curve analysis for 35 Honduran discharge stations was used to estimate discharge uncertainty for the region, and the consistency of the model forcing and evaluation data was analysed using two different screening methods. FDCs with uncertainty bounds were calculated for each basin, accounting for both discharge uncertainty and, in many cases, uncertainty stemming from the use of short time series, potentially not representative for the modelling period. These uncertain FDCs were then used to regionalise a FDC for each basin, treating it as ungauged in a cross-evaluation, and this regionalised FDC was used to constrain the uncertainty in the model predictions for the basin. There was a clear relationship between the performance of the local model calibration and the degree of data set consistency – with many basins with inconsistent data lacking behavioural simulations (i.e. simulations within predefined limits around the observed FDC) and the basins with the highest data set consistency also having the highest simulation reliability. For the basins where the regionalisation of the FDCs worked best, the uncertainty bounds for the regionalised simulations were only slightly wider than those for a local model calibration. The predicted uncertainty was greater for basins where the result of the FDC regionalisation was more uncertain, but the regionalised simulations still had a high reliability compared to the locally