► Several methods for correcting biased climate model output were analyzed. ► All reviewed correction methods are somewhat able to improve original climate model data. ► More sophisticated ...higher-skill approaches outperform simpler approaches. ► The choice of bias correction method is relevant for assessing hydrological change. ► The application of bias correction algorithms can considerably improve hydrological simulations.
Despite the increasing use of regional climate model (RCM) simulations in hydrological climate-change impact studies, their application is challenging due to the risk of considerable biases. To deal with these biases, several bias correction methods have been developed recently, ranging from simple scaling to rather sophisticated approaches. This paper provides a review of available bias correction methods and demonstrates how they can be used to correct for deviations in an ensemble of 11 different RCM-simulated temperature and precipitation series. The performance of all methods was assessed in several ways: At first, differently corrected RCM data was compared to observed climate data. The second evaluation was based on the combined influence of corrected RCM-simulated temperature and precipitation on hydrological simulations of monthly mean streamflow as well as spring and autumn flood peaks for five catchments in Sweden under current (1961–1990) climate conditions. Finally, the impact on hydrological simulations based on projected future (2021–2050) climate conditions was compared for the different bias correction methods. Improvement of uncorrected RCM climate variables was achieved with all bias correction approaches. While all methods were able to correct the mean values, there were clear differences in their ability to correct other statistical properties such as standard deviation or percentiles. Simulated streamflow characteristics were sensitive to the quality of driving input data: Simulations driven with bias-corrected RCM variables fitted observed values better than simulations forced with uncorrected RCM climate variables and had more narrow variability bounds.
Goodness-of-fit measures are important for an objective evaluation of runoff model performance. The Kling-Gupta efficiency (RKG), which has been introduced as an improvement of the widely used ...Nash-Sutcliffe efficiency, considers different types of model errors, namely the error in the mean, the variability, and the dynamics. The calculation of RKG is implicitly based on the assumptions of data linearity, data normality, and the absence of outliers. In this study, we propose a modification of RKG as an efficiency measure comprising non-parametric components, i.e. the Spearman rank correlation and the normalized flow-duration curve. The performances of model simulations for 100 catchments using the new measure were compared to those obtained using RKG based on a number of statistical metrics and hydrological signatures. The new measure resulted overall in better or comparable model performances, and thus it was concluded that efficiency measures with non-parametric components provide a suitable alternative to commonly used measures.
Hydrological modeling for climate-change impact assessment implies using meteorological variables simulated by global climate models (GCMs). Due to mismatching scales, coarse-resolution GCM output ...cannot be used directly for hydrological impact studies but rather needs to be downscaled. In this study, we investigated the variability of seasonal streamflow and flood-peak projections caused by the use of three statistical approaches to downscale precipitation from two GCMs for a meso-scale catchment in southeastern Sweden: (1) an analog method (AM), (2) a multi-objective fuzzy-rule-based classification (MOFRBC) and (3) the Statistical DownScaling Model (SDSM). The obtained higher-resolution precipitation values were then used to simulate daily streamflow for a control period (1961–1990) and for two future emission scenarios (2071–2100) with the precipitation-streamflow model HBV. The choice of downscaled precipitation time series had a major impact on the streamflow simulations, which was directly related to the ability of the downscaling approaches to reproduce observed precipitation. Although SDSM was considered to be most suitable for downscaling precipitation in the studied river basin, we highlighted the importance of an ensemble approach. The climate and streamflow change signals indicated that the current flow regime with a snowmelt-driven spring flood in April will likely change to a flow regime that is rather dominated by large winter streamflows. Spring flood events are expected to decrease considerably and occur earlier, whereas autumn flood peaks are projected to increase slightly. The simulations demonstrated that projections of future streamflow regimes are highly variable and can even partly point towards different directions.
It is expected that an increasing proportion of the precipitation will fall as rain in alpine catchments in the future. Consequently, snow storage is expected to decrease, which, together with ...changes in snowmelt rates and timing, might cause reductions in spring and summer low flows. The objectives of this study were (1) to simulate the effect of changing snow storage on low flows during the warm seasons and (2) to relate drought sensitivity to the simulated snow storage changes at different elevations. The Swiss Climate Change Scenarios 2011 data set was used to derive future changes in air temperature and precipitation. A typical bucket‐type catchment model, HBV‐light, was applied to 14 mountain catchments in Switzerland to simulate streamflow and snow in the reference period and three future periods. The largest relative decrease in annual maximum SWE was simulated for elevations below 2,200 m a.s.l. (60–75% for the period 2070–2099) and the snowmelt season shifted by up to 4 weeks earlier. The relative decrease in spring and summer minimum runoff that was caused by the relative decrease in maximum SWE (i.e., elasticity), reached 40–90% in most of catchments for the reference period and decreased for the future periods. This decreasing elasticity indicated that the effect of snow on summer low flows is reduced in the future. The fraction of snowmelt runoff in summer decreased by more than 50% at the highest elevations and almost disappeared at the lowest elevations. This might have large implications on water availability during the summer.
Key Points
In snow‐dominated areas, summer low flows will significantly decrease in the future
This decrease will be caused mostly by the decrease in snow storage and the shift of snowmelt season due to the increase in air temperature
Low flows at higher elevations are more sensitive to the decrease in snow storage than low flows at lower elevations
This article reviews recent applications of regional climate model (RCM) output for hydrological impact studies. Traditionally, simulations of global climate models (GCMs) have been the basis of ...impact studies in hydrology. Progress in regional climate modeling has recently made the use of RCM data more attractive, although the application of RCM simulations is challenging due to often considerable biases. The main modeling strategies used in recent studies can be classified into (i) very simple constructed modeling chains with a single RCM (S‐RCM approach) and (ii) highly complex and computing‐power intensive model systems based on RCM ensembles (E‐RCM approach). In the literature many examples for S‐RCM can be found, while comprehensive E‐RCM studies with consideration of several sources of uncertainties such as different greenhouse gas emission scenarios, GCMs, RCMs and hydrological models are less common. Based on a case study using control‐run simulations of fourteen different RCMs for five Swedish catchments, the biases of and the variability between different RCMs are demonstrated. We provide a short overview of possible bias‐correction methods and show that inter‐RCM variability also has substantial consequences for hydrological impact studies in addition to other sources of uncertainties in the modeling chain. We propose that due to model bias and inter‐model variability, the S‐RCM approach is not advised and ensembles of RCM simulations (E‐RCM) should be used. The application of bias‐correction methods is recommended, although one should also be aware that the need for bias corrections adds significantly to uncertainties in modeling climate change impacts.
•A few runoff measurements support model calibration in almost ungauged catchments.•Different strategies for the timing of these runoff measurements were evaluated.•Strategies focusing on high flow ...magnitudes were informative for hydrograph simulations.•Mean and low flow runoff data had most value for flow-duration curve simulations.
Applications of runoff models usually rely on long and continuous runoff time series for model calibration. However, many catchments around the world are ungauged and estimating runoff for these catchments is challenging. One approach is to perform a few runoff measurements in a previously fully ungauged catchment and to constrain a runoff model by these measurements. In this study we investigated the value of such individual runoff measurements when taken at strategic points in time for applying a bucket-type runoff model (HBV) in ungauged catchments. Based on the assumption that a limited number of runoff measurements can be taken, we sought the optimal sampling strategy (i.e. when to measure the streamflow) to obtain the most informative data for constraining the runoff model. We used twenty gauged catchments across the eastern US, made the assumption that these catchments were ungauged, and applied different runoff sampling strategies. All tested strategies consisted of twelve runoff measurements within one year and ranged from simply using monthly flow maxima to a more complex selection of observation times. In each case the twelve runoff measurements were used to select 100 best parameter sets using a Monte Carlo calibration approach. Runoff simulations using these ‘informed’ parameter sets were then evaluated for an independent validation period in terms of the Nash-Sutcliffe efficiency of the hydrograph and the mean absolute relative error of the flow-duration curve. Model performance measures were normalized by relating them to an upper and a lower benchmark representing a well-informed and an uninformed model calibration. The hydrographs were best simulated with strategies including high runoff magnitudes as opposed to the flow-duration curves that were generally better estimated with strategies that captured low and mean flows. The choice of a sampling strategy covering the full range of runoff magnitudes enabled hydrograph and flow-duration curve simulations close to a well-informed model calibration. The differences among such strategies covering the full range of runoff magnitudes were small indicating that the exact choice of a strategy might be less crucial. Our study corroborates the information value of a small number of strategically selected runoff measurements for simulating runoff with a bucket-type runoff model in almost ungauged catchments.
A drought index accounting for snow Staudinger, Maria; Stahl, Kerstin; Seibert, Jan
Water resources research,
October 2014, Letnik:
50, Številka:
10
Journal Article
Recenzirano
Odprti dostop
The Standardized Precipitation Index (SPI) is the most widely used index to characterize droughts that are related to precipitation deficiencies. However, the SPI does not always deliver the relevant ...information for hydrological drought management particularly in snow‐influenced catchments. If precipitation is temporarily stored as snow, then there is a significant difference between meteorological and hydrological drought because the delayed release of melt water to the stream. We introduce an extension to the SPI, the Standardized Snow Melt and Rain Index (SMRI), that accounts for rain and snow melt deficits, which effectively influence streamflow. The SMRI can be derived without snow data, using temperature and precipitation to model snow. The value of the new index is illustrated for seven Swiss catchments with different degrees of snow influence. In particular for catchments with a larger component of snowmelt in runoff generation, the SMRI was found to be a worthwhile complementary index to the SPI to characterize streamflow droughts.
Key Points
SPI concept was expanded to account for temporary snow storage
The SMRI improves description of hydrological drought in snow‐dominated basins
SMRI is adjustable and allows for regional comparisons
A variety of landscape properties have been modelled successfully using topographic indices such as the topographic wetness index (TWI), defined as ln(
a/tan
β), where
a is the specific upslope area ...and
β is the surface slope. Previous studies have shown the influence of scale on TWI values when converting standard-resolution DEMs to coarser resolutions. In this study a high-resolution digital elevation model (DEM) with a 5
m grid size derived from LIDAR (light detection and ranging) data was used to investigate the scale-dependency of TWI values when converting from high-resolution elevation data to standard-resolution DEMs. First, a set of DEMs was generated from an initial DEM by thinning to resolutions of 10, 25, and 50
m grid sizes to study the effects of lower grid size and decreased information content. Next, to investigate the impact of different information content on DEMs with the same grid size, the three lower resolution DEMs were all interpolated to the original 5
m grid size. In addition to comparing index distribution functions, a second objective was to evaluate differences in spatial patterns. Thus the values of TWI and its components as computed for the seven different DEMs were compared in three different ways: (1) distribution functions and their statistics; (2) cell by cell comparison of four DEMs with the same resolution but different information content; and (3) comparison of blocks of cells within different resolution DEMs with different information content. Like previous TWI studies, the computed specific upstream area decreased on average for higher resolution DEMs while computed slope values followed a narrower distribution. TWI variation between neighbouring cells in 50
×
50
m areas decreased largely with increasing grid size. A cell by cell comparison of the TWI values of the four 5
m DEMs with different information content showed a clear decrease in correlation with the TWI based on the original DEM with decreasing information content. The results showed considerable differences between topographic indices computed for DEMs of different grid resolution. Interpolating the DEMs to a higher resolution (i.e. a smaller grid size) provided more similar TWI distributions, but the pixel by pixel comparison showed that different information contents caused clearly different TWI maps.
Projections of discharge are key for future water resources management. These projections are subject to uncertainties, which are difficult to handle in the decision process on adaptation strategies. ...Uncertainties arise from different sources such as the emission scenarios, the climate models and their postprocessing, the hydrological models, and the natural variability. Here we present a detailed and quantitative uncertainty assessment, based on recent climate scenarios for Switzerland (CH2011 data set) and covering catchments representative for midlatitude alpine areas. This study relies on a particularly wide range of discharge projections resulting from the factorial combination of 3 emission scenarios, 10–20 regional climate models, 2 postprocessing methods, and 3 hydrological models of different complexity. This enabled us to decompose the uncertainty in the ensemble of projections using analyses of variance (ANOVA). We applied the same modeling setup to six catchments to assess the influence of catchment characteristics on the projected streamflow, and focused on changes in the annual discharge cycle. The uncertainties captured by our setup originate mainly from the climate models and natural climate variability, but the choice of emission scenario plays a large role by the end of the 21st century. The contribution of the hydrological models to the projection uncertainty varied strongly with catchment elevation. The discharge changes were compared to the estimated natural decadal variability, which revealed that a climate change signal emerges even under the lowest emission scenario (RCP2.6) by the end of the century. Limiting emissions to RCP2.6 levels would nevertheless reduce the largest regime changes by the end of the century by approximately a factor of two, in comparison to impacts projected for the high emission scenario SRES A2. We finally show that robust regime changes emerge despite the projection uncertainty. These changes are significant and are consistent across a wide range of scenarios and catchments. We propose their identification as a way to aid decision making under uncertainty.
Key Points
Uncertainties in discharge projections assessed using a wide simulation range
Robust changes emerged in each catchment despite projection uncertainties
Emission policy enabled significant reductions of the projected impacts
Crowd‐based hydrological observations can supplement existing monitoring networks and allow data collection in regions where otherwise no data would be available. In the citizen science project ...CrowdWater, repeated water level observations using a virtual staff gauge approach result in time series of water level classes (WL‐classes). To investigate the quality of these observations, we compared the WL‐class data with “real” (i.e., measured) water levels from the same stream at a nearby gauging station. We did this for nine locations where citizen scientists reported multiple observations using a smartphone app and at 12 locations where signposts were set up to ask citizens to record observations on a paper form that could be left in a letterbox. The results indicate that the quality of the data collected with the app was better than for the forms. A possible explanation is that for each app location, a single person submitted the vast majority of the observations, whereas at the locations of the forms almost every observation was made by a different person. On average, there were more contributions between May and September than during the other months. Observations were submitted for a range of flow conditions, with a higher fraction of high flow observations for the locations were data were collected with the app. Overall, the results are encouraging for citizen science approaches in hydrology and demonstrate that the smartphone application and the virtual staff gauge are a promising approach for crowd‐based water level class observations.
The comic shows the two approaches to collect water level class estimates using the CrowdWater app or with paper forms throughout different seasons. The estimates were more accurate when the data were collected using the CrowdWater app. Most contributions to one app location were made by a single person, whereas at the locations of the forms almost every observation was made by a different person. Artwork by University of Zurich, Information Technology, MELS/SIVIC, Tara von Grebel.