Relative Importance of the Different Rainfall Statistics in the Calibration of Stochastic Rainfall Generation Models (original) (raw)
Related papers
Journal of Hydrometeorology, 2005
A major goal in quantitative precipitation estimation and forecasting is the ability to provide accurate initial conditions for the purposes of hydrologic modeling. The accuracy of a streamflow prediction system is dependent upon how well the initial hydrometeorological states are characterized. A methodology is developed to objectively and quantitatively evaluate the skill of several different precipitation algorithms at the scale of application-a watershed. Thousands of hydrologic simulations are performed in an ensemble fashion, enabling an exploration of the model parameter space. Probabilistic statistics are then utilized to compare the relative skill of hydrologic simulations produced from the different precipitation inputs to the observed streamflow. The primary focus of this study is to demonstrate a methodology to evaluate precipitation algorithms that can be used to supplement traditional radar-rain gauge analyses. This approach is appropriate for the evaluation of precipitation estimates or forecasts that are intended to serve as inputs to hydrologic models.
Sensitivity of monthly rainfall-runoff models to input errors and data length
Hydrological Sciences Journal, 1994
Two problems are addressed which arise when using monthly water balance models as an aid to making decisions in water resources engineering: what is the influence of data errors on model performance, and what is the data length required in order to obtain reliable models? Two previously defined types of models are used: in PE type models the input series are precipitation and potential évapotranspiration; in P type models the only input is precipitation. The main conclusions are:
Uncertainty in watershed response predictions induced by spatial variability of precipitation
Environmental Monitoring and Assessment, 2007
Negligence to consider the spatial variability of rainfall could result in serious errors in model outputs. The objective of this study was to examine the uncertainty of both runoff and pollutant transport predictions due to the input errors of rainfall. This study used synthetic data to represent the "true" rainfall pattern, instead of interpolated precipitation. It was conducted on a synthetic case area having a total area of 20 km 2 with ten subbasins. Each subbasin has one rainfall gauge with synthetic precipitation records. Six rainfall storms with varied spatial distribution were generated. The average rainfall was obtained from all of the ten gauges by the arithmetic average method. The input errors of rainfall were induced by the difference between the actual rainfall pattern and estimated average rainfall. The results show that spatial variability of rainfall can cause uncertainty in modeling outputs of hydrologic, which would be transport to pollutant export predictions, when uniformity of rainfall is assumed. Since rainfall is essential information for predicting watershed responses, it is important to consider the properties of rainfall, particularly spatial rainfall variability, in the application of hydrologic and water quality models.
Journal of Hydrologic Engineering, 2009
This study investigated the impact of temporal and spatial sampling of rainfall on runoff predictions using a physically based ͓System Hydrologique European ͑MIKE SHE͔͒ and conceptual ͓hydrologic modeling system ͑HMS͔͒ hydrologic models. The numerical models were applied to Goodwin Creek watershed in northern Mississippi. The drainage area of Goodwin Creek is approximately 21.4 km 2 . This study showed that MIKE SHE was more sensitive to both the spatial and the temporal samplings of rainfall than the HMS. The study also showed that errors introduced by coarse sampling scenarios can be significant. Overall, for this particular watershed size, increasing the rain gauge density from 1 to 2 resulted in the most significant improvement for both models. Similarly, a temporal sampling frequency beyond 1 h showed significant deterioration in the quality of the runoff prediction. The combined spatial-temporal sampling experiment showed that increasing the temporal sampling compensates, at least partially, for the loss of rainfall spatial information. It also showed that under poor temporal sampling frequency, the gain in model performance by increasing the spatial sampling density is negligible.
On the deterministic and stochastic use of hydrologic models
Environmental simulation models, such as precipitation-runoff watershed models, are increasingly used in a deterministic manner for environmental and water resources design, planning, and management. In operational hydrology, simulated responses are now routinely used to plan, design, and manage a very wide class of water resource systems. However, all such models are calibrated to existing data sets and retain some residual error. This residual, typically unknown in practice, is often ignored, implicitly trusting simulated responses as if they are deterministic quantities. In general, ignoring the residuals will result in simulated responses with distributional properties that do not mimic those of the observed responses. This discrepancy has major implications for the operational use of environmental simulation models as is shown here. Both a simple linear model and a distributed-parameter precipitation-runoff model are used to document the expected bias in the distributional properties of simulated responses when the residuals are ignored. The systematic reintroduction of residuals into simulated responses in a manner that produces stochastic output is shown to improve the distributional properties of the simulated responses. Every effort should be made to understand the distributional behavior of simulation residuals and to use environmental simulation models in a stochastic manner.
Stochastic bias-correction of daily rainfall scenarios for hydrological applications
Natural Hazards and Earth System Science, 2011
The accuracy of rainfall predictions provided by climate models is crucial for the assessment of climate change impacts on hydrological processes. In fact, the presence of bias in downscaled precipitation may produce large bias in the assessment of soil moisture dynamics, river flows and groundwater recharge. In this study, a comparison between statistical properties of rainfall observations and model control simulations from a Regional Climate Model (RCM) was performed through a robust and meaningful representation of the precipitation process. The output of the adopted RCM was analysed and re-scaled exploiting the structure of a stochastic model of the point rainfall process. In particular, the stochastic model is able to adequately reproduce the rainfall intermittency at the synoptic scale, which is one of the crucial aspects for the Mediterranean environments. Possible alteration in the local rainfall regime was investigated by means of the historical daily time-series from a dense rain-gauge network, which were also used for the analysis of the RCM bias in terms of dry and wet periods and storm intensity. The result is a stochastic scheme for bias-correction at the RCM-cell scale, which produces a realistic representation of the daily rainfall intermittency and precipitation depths, though a residual bias in the storm intensity of longer storm events persists.
Journal of Hydrology, 1996
The relationship between the spatial variability of rainfall and catchment response is investigated by conducting experiments with a stochastic rainfall field model and a physically based distributed modelling system, the Systeme Hydrologique Europeen (SHE), both of which are calibrated for a small upland catchment. The development and calibration of the rainfall field model is described in Part 1 of this paper, and the experiments with simulated rainfall fields and the SHE catchment model are described in Part 2. The rainfall field model is based on the use of the Turning Bands Method (TBM) incorporating a fractionally differenced line process to generate Gaussian random fields with a specified space-time correlation structure which can be isotropic or anisotropic. A transformation is then applied to the Gaussian field to reproduce the non-stationary temporal structure and skewed marginal distribution of observed rainfall. The transformed field is then propagated in space with the required velocity. The model is calibrated using hourly data for ten storms observed at three sites in the upper Wye catchment (area 10.55 km*) at Plynlimon, Wales. As rainfall over the catchment exhibits significant variation with altitude (and other factors), an altitude correction factor is applied to the simulated rainfall fields. Comparisons of the means, variances, skewnesses, cross-and autocorrelation functions of observed and simulated storms at the sampling points show good agreement, and realistic spatial patterns are observed in the simulated fields. A procedure is developed and applied for generating conditional simulations whereby the historical storm rainfall at the sampling points can be reproduced exactly in simulated fields, thus allowing several realizations of the unobserved spatial variability of rainfall at all other points in the catchment to be generated for any historical storm.
Error analysis for the evaluation of model performance: rainfall–runoff event summary variables
Hydrological Processes, 2007
This paper provides a procedure for the evaluation of model performance for rainfall-runoff event summary variables, such as total discharge or peak runoff. The procedure is based on the analysis of model errors, defined as the differences between observed values and values predicted by a simulation model. Model errors can (i) indicate whether and where the model can be improved, (ii) be used to measure the performance of a model, and (iii) be used to compare model simulations. In this paper, both statistical and graphical methods are used to characterize model errors. We explore model recalibration by relating model errors to the model predictions, and to external, independent variables. The R-5 catchment data sets that we used in this study include summary variables for 72 rainfall-runoff events. The simulations used in this study were previously conducted with the quasi-physically based rainfall-runoff model QPBRRM for 11 different characterizations of the R-5 catchment, each with increasing information or a refined spatial discretization of the overland flow planes. This paper is about proposing model diagnostics and not about procedures for using diagnostics for model modification.
Hydrology Research, 2013
The effect of the time step of calibration data on the performance of a hydrological model is examined through a numerical experiment where HYMOD, a rainfall–runoff model, is calibrated with data of varying temporal resolution. A simple scaling relationship between the parameters of the model and modelling time step is derived which enables information from daily hydrological records to be used in modelling at time steps much shorter than daily. Model parameters were found to respond differently depending upon the degree of aggregation of calibration data. A loss in performance, especially in terms of the Nash–Sutcliffe measure, is evident when behavioural simulators derived with one modelling time step are used for simulation at another time step. The loss in performance is greater when parameters derived from a longer time step were used for simulating flow with a shorter time step. The application of a simple scaling relationship derived from a multi-time step model calibration s...
Water Science and Technology, 2018
The technical lifetime of urban water infrastructure has a duration where climate change has to be considered when alterations to the system are planned. Also, models for urban water management are reaching a very high complexity level with, for example, decentralized stormwater control measures being included. These systems have to be evaluated under as close-to-real conditions as possible. Long term statistics (LTS) modelling with observational data is the most close-to-real solution for present climate conditions, but for future climate conditions artificial rainfall time series from weather generators (WGs) have to be used. In this study, we ran LTS simulations with four different WG products for both present and future conditions on two different catchments. For the present conditions, all WG products result in realistic catchment responses when it comes to the number of full flowing pipes and the number and volume of combined sewer overflows (CSOs). For future conditions, the ...