Distributed watershed models are increasingly being used to support decisions about alternative management strategies in the areas of land use change, climate change, water allocation, and pollution ...control. For this reason it is important that these models pass through a careful calibration and uncertainty analysis. To fulfil this demand, in recent years, scientists have come up with various uncertainty analysis techniques for watershed models. To determine the differences and similarities of these techniques we compared five uncertainty analysis procedures: Generalized Likelihood Uncertainty Estimation (GLUE), Parameter Solution (ParaSol), Sequential Uncertainty FItting algorithm (SUFI-2), and a Bayesian framework implemented using Markov chain Monte Carlo (MCMC) and Importance Sampling (IS) techniques. As these techniques are different in their philosophies and leave the user some freedom in formulating the generalized likelihood measure, objective function, or likelihood function, a literal comparison between these techniques is not possible. As there is a small spectrum of different applications in hydrology for the first three techniques, we made this choice according to their typical use in hydrology. For Bayesian inference, we used a recently developed likelihood function that does not obviously violate the statistical assumptions, namely a continuous-time autoregressive error model. We implemented all these techniques for the soil and water assessment tool (SWAT) and applied them to the Chaohe Basin in China. We compared the results with respect to the posterior parameter distributions, performances of their best estimates, prediction uncertainty, conceptual bases, computational efficiency, and difficulty of implementation. The comparison results for these categories are listed and the advantages and disadvantages are analyzed. From the point of view of the authors, if computationally feasible, Bayesian-based approaches are most recommendable because of their solid conceptual basis, but construction and test of the likelihood function requires critical attention.
Using the parameters associated with the best-fit simulation (i.e., the simulation with the highest objective function value) to represent a calibrated hydrological model is inadequate. The reason is ...that the calibrated models best objective function value is usually not significantly different from the next best value or the values after that. This non-uniqueness of the objective function values causes a problem because the best solution's parameters are often significantly different from the next best set of parameters. Therefore, only using the best simulation parameters as the calibrated model's sole parameters to interpret the watershed processes or perform further modeling analyses could produce misleading results. Furthermore, the lack of pristine watersheds makes the task of watershed-scale calibration increasingly challenging. Subjective thresholds of acceptable performance criteria suggested by some researchers, based on comparing the measured and the best solution signals, are often not achievable. Hence, to obtain a satisfactory fit, researchers and practitioners are often forced to compromise the science behind their work. This article discusses the fallacy in using the best-fit solution in hydrologic modeling. A two-factor statistic to assess the goodness of calibration/validation is discussed, considering model output uncertainty.
•We modeled hydrology of the entire European continent with SWAT.•We included river discharge and nitrate loads as well as crop yield in the model.•We provide a protocol for calibration of ...large-scale models with uncertainty analysis.•We modeled blue and green water resources of Europe at subbasin level.•We improved SWAT-CUP to include parallel processing and visualization.
A combination of driving forces are increasing pressure on local, national, and regional water supplies needed for irrigation, energy production, industrial uses, domestic purposes, and the environment. In many parts of Europe groundwater quantity, and in particular quality, have come under sever degradation and water levels have decreased resulting in negative environmental impacts. Rapid improvements in the economy of the eastern European block of countries and uncertainties with regard to freshwater availability create challenges for water managers. At the same time, climate change adds a new level of uncertainty with regard to freshwater supplies. In this research we build and calibrate an integrated hydrological model of Europe using the Soil and Water Assessment Tool (SWAT) program. Different components of water resources are simulated and crop yield and water quality are considered at the Hydrological Response Unit (HRU) level. The water resources are quantified at subbasin level with monthly time intervals. Leaching of nitrate into groundwater is also simulated at a finer spatial level (HRU). The use of large-scale, high-resolution water resources models enables consistent and comprehensive examination of integrated system behavior through physically-based, data-driven simulation. In this article we discuss issues with data availability, calibration of large-scale distributed models, and outline procedures for model calibration and uncertainty analysis. The calibrated model and results provide information support to the European Water Framework Directive and lay the basis for further assessment of the impact of climate change on water availability and quality. The approach and methods developed are general and can be applied to any large region around the world.
Inversely obtained hydrologic parameters are always uncertain (nonunique) because of errors associated with the measurements and the invoked conceptual model, among other factors. Quantification of ...this uncertainty in multidimensional parameter space is often difficult because of complexities in the structure of the objective function. In this study we describe parameter uncertainties using uniform distributions and fit these distributions iteratively within larger absolute intervals such that two criteria are met: (i) bracketing most of the measured data (>90%) within the 95% prediction uncertainty (95PPU) and (ii) obtaining a small ratio (<1) of the average difference between the upper and lower 95PPU and the standard deviation of the measured data. We define a model as calibrated if, upon reaching these two criteria, a significant R2 exists between the observed and simulated results. A program, SUFI-2, was developed and tested for the calibration of two bottom ash landfills. SUFI-2 performs a combined optimization and uncertainty analysis using a global search procedure and can deal with a large number of parameters through Latin hypercube sampling. We explain the above concepts using an example in which two municipal solid waste incinerator bottom ash monofills were successfully calibrated and tested for flow, and one monofill also for transport. Because of high levels of heavy metals in the leachate, monitoring and modeling of such landfills is critical from environmental points of view.
Most hydrologic models require daily weather data to run. While this information may be abundant in some parts of the world, in most parts such data is not available on daily basis. Distributed ...hydrologic models are particularly adversely affected by the lack of daily data or the existence of very inaccurate data as they impart large uncertainties to the model prediction. In this study we developed a daily weather generator algorithm (dGen) that uses the currently available 0.5° monthly weather statistics from the Climatic Research Unit (CRU). We tested dGen in two ways. First, we made a direct comparison of the measured and generated precipitation and maximum–minimum temperatures by looking at some long-term statistics in a few stations in West Africa. Second, we ran the model “Soil and Water Assessment Tool” (SWAT) with dGen-generated and measured daily weather data to simulate 25 years of annual and monthly river discharges at some gauging stations. The simulated river discharges were then compared with the measured ones. It was seen that using the dGen-simulated daily weather data resulted in a much better match with the measured discharge data than the measured daily weather data in combination with the SWAT internal weather generator WXGEN. WXGEN is used in SWAT to fill missing data using monthly statistics, which must be calculated from the existing daily data. For annual and monthly hydrological simulations, dGen-generated daily rainfall and temperature data appears to have a high degree of reliability.
In a national effort, since 1972, the Swiss Government started the “National Long-term Monitoring of Swiss Rivers” (NADUF) program aimed at evaluating the chemical and physical states of major rivers ...leaving Swiss political boundaries. The established monitoring network of 19 sampling stations included locations on all major rivers of Switzerland. This study complements the monitoring program and aims to model one of the program’s catchments – Thur River basin (area 1700
km
2), which is located in the north-east of Switzerland and is a direct tributary to the Rhine. The program SWAT (Soil and Water Assessment Tool) was used to simulate all related processes affecting water quantity, sediment, and nutrient loads in the catchment. The main objectives were to test the performance of SWAT and the feasibility of using this model as a simulator of flow and transport processes at a watershed scale. Model calibration and uncertainty analysis were performed with SUFI-2 (Sequential Uncertainty FItting Ver. 2), which was interfaced with SWAT using the generic iSWAT program. Two measures were used to assess the goodness of calibration: (1) the percentage of data bracketed by the 95% prediction uncertainty calculated at the 2.5 and 97.5 percentiles of the cumulative distribution of the simulated variables, and (2) the
d-factor, which is the ratio of the average distance between the above percentiles and the standard deviation of the corresponding measured variable. These statistics showed excellent results for discharge and nitrate and quite good results for sediment and total phosphorous. We concluded that: in watersheds similar to Thur – with good data quality and availability and relatively small model uncertainty – it is feasible to use SWAT as a flow and transport simulator. This is a precursor for watershed management studies.
Large-scale distributed watershed models are data-intensive, and preparing them consumes most of the research resources. We prepared high-resolution global databases of soil, landuse, actual ...evapotranspiration (AET), and historical and future weather databases that could serve as standard inputs in Soil and Water Assessment Tool (SWAT) models. The data include two global soil maps and their associated databases calculated with a large number of pedotransfer functions, two landuse maps and their correspondence with SWAT's database, historical and future daily temperature and precipitation data from five IPCC models with four scenarios; and finally, global monthly AET data. Weather data are 0.5° global grids text-formatted for direct use in SWAT models. The AET data is formatted for use in SWAT-CUP (SWAT Calibration Uncertainty Procedures) for calibration of SWAT models. The use of these global databases for SWAT models can speed up the model building by 75-80% and are extremely valuable in areas with limited or no physical data. Furthermore, they can facilitate the comparison of model results in different parts of the world.
Amid an increasing water scarcity in many parts of the world, virtual water trade as both a policy instrument and practical means to balance the local, national and global water budget has received ...much attention in recent years. Building upon the knowledge of virtual water accounting in the literature, this study assesses the efficiency of water use embodied in the international food trade from the perspectives of exporting and importing countries and at the global and country levels. The investigation reveals that the virtual water flows primarily from countries of high crop water productivity to countries of low crop water productivity, generating a global saving in water use. Meanwhile, the total virtual water trade is dominated by green virtual water, which constitutes a low opportunity cost of water use as opposed to blue virtual water. A sensitivity analysis, however, suggests high uncertainties in the virtual water accounting and the estimation of the scale of water saving. The study also raises awareness of the limited effect of water scarcity on the global virtual water trade and the negative implications of the global water saving for the water use efficiency and food security in importing countries and the environment in exporting countries. The analysis shows the complexity in evaluating the efficiency gains in the international virtual water trade. The findings of the study, nevertheless, call for a greater emphasis on rainfed agriculture to improve the global food security and environmental sustainability.
Accurate knowledge of freshwater availability is indispensable for water resources management at regional or national level. This information, however, has historically been very difficult to obtain ...because of lack of data, difficulties in the aggregation of spatial information, and problems in the quantification of distributed hydrological processes. The currently available estimates of freshwater availability by a few large international organizations such as FAO and UNESCO are often not sufficient as they only provide aggregated rough quantities of river discharge and groundwater recharge (blue water) at a national level and on a yearly basis. This paper aims to provide a procedure to improve the estimations of freshwater availability at subbasin level and monthly intervals. Applying the distributed hydrological model “Soil and Water Assessment Tool” (SWAT), the freshwater availability is quantified for a 4-million
km
2 area covering some 18 countries in West Africa. The procedure includes model calibration and validation based on measured river discharges, and quantification of the uncertainty in model outputs using “Sequential Uncertainty Fitting Algorithm” (SUFI-2) The aggregated results for 11 countries are compared with two other studies. It was seen that for most countries, the estimates from the other two studies fall within our calculated prediction uncertainty ranges. The uncertainties are, in general, within reasonable ranges but larger in subbasins containing features such as dams and wetlands, or subbasins with inadequate climate or landuse information. As the modelling procedure in this study proved quite successful, its application for quantification of freshwater availability at a global scale is already underway. There are, however, two limitations in the West African model: (1) not all the components of the water balance model such as soil moisture or deep aquifer recharge could be directly calibrated because of lack of data and (2) the full capabilities of the SWAT model could not be realized because of the lack of local water and agricultural management information.