The sources of uncertainty is everywhere and every steps of the modelling. You have errors in the measurements of meteo forcing (mainly rainfall) due to many reasons, the errors of interpolation from measurement points to the point of interest (usually to the modelling units). The errors of interpolation comes form two sides: lack of enough rainfall sample to capture the real rainfall variability (too small station) and the interpolation model errors (both deterministic and geo-statistics). The errors of ETP is very huge, because most of the time we don't have measurements to cross check our estimations. I personally also believe that in some models, the topographic information and the use of basin topography and its network also introduce errors. These are just the input uncertainties. The model uncertainty also there, i.e the model structure and model parameters. So one should take care of those issues in the modelling step. There is also a measurement errors in the discharge data to which we train our model during calibration processes. All those factors need to be take care, if we are really want to cover all the uncertainty.
If you really are interested to cover all those errors, the recent pioneering work in this direction is the use of information theory by Gupta and his co-workers.
I had identified few hydrologic model uncertainties in my MS thesis, which includes physical characteristics of the watershed, input data, modeler decision, and the model capability. I hope this will help you to start.
There are uncertainties in measurement/estimation of every component of a water balance to varying degrees, depending on the component. For instance, inadequate spatiotemporal resolution of rainfall data, river discharge, groundwater flow and human water abstraction/return flows. But perhaps the largest uncertainty in estimation lies within evapotranspiration, especially in forests. Then, there are interannual variation, which leads to further uncertainty in the absence of a long time series. The use of satellite-imagery based land cover maps used for watershed level water balances and hydro modeling also depends upon the accuracy of classification.Tuning hydrological models is a fine art !
Lots of good info above - I would only stress that it would depend on what model you're using, and where you're applying it (e.g. urbanised or rural catchment?). However, if you're new to whatever model you're using, search around for literature (a manual or whatever else you can find) on the most sensitive parameters of your selected model - or better still, do some sensitivity analysis on the parameters yourself. It can be very informative!
The sources of uncertainty is everywhere and every steps of the modelling. You have errors in the measurements of meteo forcing (mainly rainfall) due to many reasons, the errors of interpolation from measurement points to the point of interest (usually to the modelling units). The errors of interpolation comes form two sides: lack of enough rainfall sample to capture the real rainfall variability (too small station) and the interpolation model errors (both deterministic and geo-statistics). The errors of ETP is very huge, because most of the time we don't have measurements to cross check our estimations. I personally also believe that in some models, the topographic information and the use of basin topography and its network also introduce errors. These are just the input uncertainties. The model uncertainty also there, i.e the model structure and model parameters. So one should take care of those issues in the modelling step. There is also a measurement errors in the discharge data to which we train our model during calibration processes. All those factors need to be take care, if we are really want to cover all the uncertainty.
If you really are interested to cover all those errors, the recent pioneering work in this direction is the use of information theory by Gupta and his co-workers.
Over simplification of the modeled system, numerical error such as truncation error, roundoff error and artificial oscillations that might grow up with time inducing misleading results. Uncertainity could be due to incorrect conceptualization and the values of parameters used. Moreover, numerical alogarithm used for solving the equations could be very sensitive to some conditions. There are many publications in this regards.
Viewing model components and the errors in them within a dynamic systems framework helps to better understand and organize the different uncertainty sources in hydrologic modeling. Uncertainty in hydrologic modeling may arise from several sources: model structure, parameters, initial conditions, and observational data used to drive and evaluate the model.
Hydrological models are composed of different components: system boundary, inputs, initial states, parameters, structure, states, and outputs. One of the important method for system identification, parameter estimation, state estimation, and combined state and parameter estimation are data assimilation method.
The critical aspects of addressing hydrologic uncertainty may include understanding, quantifying, and reducing uncertainty, to arrive at a general context for hydrologic data assimilation.
In my experience the main sources of uncertainty in hydrological models are:
(1) conceptual model uncertainty (due to simplification in the conceptual model, or due to processes unknown and not included in the model);
(2) input data uncertainty (errors in rainfall data are a source of uncertainty as well as the spatial distribution of rainfall gauge and the method used to spatialize rainfall data);
(3) parameter uncertainty (there are many different sets of parameters which give similar discharge signals.
In addition errors could also exist in measurements used to calibrate the model.
It is very difficult to separate all these effects on the outputs, experience of modelers can make a bid difference in model calibration and can reduce significanthly the uncertainty.
"Broadly, the hydrologic forecast uncertainty is grouped into input forecast uncertainty and hydrologic model uncertainty; input forecast uncertainty implies to uncertainty of input forecasts (such as precipitation and temperature forecasts) that are supplied to a hydrologic model whereas uncertainty associated with model initial conditions, model structure and model parameters contribute to hydrologic model uncertainty."
The total hydrologic forecast uncertainty can be modeled in two ways, namely source-specifically (e.g., Demargne et al., 2014) or in a lumped fashion (Regonda et al., 2013 and references therein).
The techniques that account forecast uncertainty and generate ensemble forecasts are known as ensemble or probabilistic forecasting techniques. There are many ensemble forecasting techniques available (See Cloke and Pappenberger, 2009;
http://hepex.irstea.fr/about-hepex/).
Suggested readings:
Global Flood Aware System (GloFAS) (Alfieri, 2013);Hydrologic Ensemble Forecast Service (HEFS) (Demargne et al., 2014); Hydrologic Model Output Statistics (HMOS) approach (Regonda et al., 2014); European Flood Alert System (EFAS) (Thielen et al., 2009)
Alfieri et al, 2013: GloFAS –global ensemble streamflow forecasting and flood early warning, Hydrology and Earth System Sciences, 17, 1161-1175, doi:10.5194/hess-17-1161-2013.
Cloke, H.L. and Pappenberger, F., 2009. Ensemble Flood Forecasting: A Review. Journal of Hydrology, 375(1–4): 613–626.
Demargne et al., 2014: The science of NOAA's operational hydrologic ensemble forecast service, Bulletin of American Meteorological Society, http://dx.doi.org/10.1175/BAMS-D-12-00081.1.
Regonda et al., 2014 : Short-term ensemble streamflow forecasting using operationally-produced single-valued streamflow forecasts—A Hydrologic Model Output Statistics (HMOS) approach, Journal of Hydrology, 497, 80–96, doi:10.1016/j.jhydrol.2013.05.028
Thielen et al., 2009: The European Flood Alert System – Part 1: Concept and development, Hydrology and Earth System Sciences, 13, 125–140, doi:10.5194/hess-13-125-2009.
If garbage goes in, garbage goes out the model. Certainly, input data quality is of crucial importance as well as the capacity of the model to reproduce the processes it was designed for analyzing i.e. the mathematical structure if the model is empirical or processes based. The sensitivity of the model will give insight as to what variables one should parameterize in a precise manner that have the more impact in the outputs.
We started quite early to look at different sources of uncertainty in hydrological modelling. Have a look at https://www.researchgate.net/publication/242655301_Estimation_of_time-variant_hydrological_parameters concentrating on parameter uncertainty
A more recent paper on uncertainties related to model structure unceretainties can be found here https://www.researchgate.net/publication/243457593_LINEARITY_OF_HYDROLOGICAL_MODELS_AND_RELATED_UNCERTAINTY
Another paper deals with the uncertainty in rainfall prediction: https://www.researchgate.net/publication/222296857_Implications_of_radar_rainfall_estimates_uncertainty_on_distributed_hydrological_model_predictions
The largest problem still is the identification of different uncertainty sources in complex models
During the last decade hundreds of papers have been published about modelling uncertainties, mainly parameter estimation
Article Estimation of time-variant hydrological parameters
Conference Paper LINEARITY OF HYDROLOGICAL MODELS AND RELATED UNCERTAINTY
Article Implications of Radar Rainfall Estimates Uncertainty on Dist...