Claude Michel

Improvement of a parsimonious model for streamflow simulation

Abstract Hydrologists have been struggling over the past decades to improve rainfall-runoff models. As a consequence, models proposed 20–30 years ago still keep evolving as progress is made in the understanding of catchment hydrological behaviour. Here we present the GR4J model, a daily lumped rainfall-runoff model which is the result of a continuous improvement process over the last 15 years. The article provides the mathematical formulation of a new four-parameter version of the model. Model performance is assessed on a large sample of catchments: compared to other rainfall-runoff models, the GR4J performance is among the best ones. It also gives better results than the previous three-parameter model version, especially in the simulation of low flows. The tests indicate that a four-parameter structure corresponds to the maximum level of complexity that could be afforded in the model. Adding more free parameters did not bring significant improvements. The gain in model robustness with this new version should enhance the confidence in the practical use of this simple model for water engineering and resource management. The discussion underlines the potential limits introduced in the modelling process when one relies on a priori concepts in building a model structure and it stresses the value of large catchment samples to assess models.

Does a large number of parameters enhance model performance? Comparative assessment of common catchment model structures on 429 catchments

Abstract Hydrological models must be reliable and robust as these qualities influence all applications based on model output. Previous studies on conceptual rainfall–runoff models have shown that one of the root causes of their output uncertainty is model over-parameterisation. The problem of poorly defined parameters has attracted much attention but has not yet been satisfactorily solved. We believe that the most fruitful way forward is to improve the structures where these parameters act. The main objective of this paper is to examine the role of complexity in hydrological models by studying the relation between the number of optimised parameters and model performance. An extensive comparative performance assessment of the structures of 19 daily lumped models was carried out on 429 catchments, mostly in France but also in the United States, Australia, the Ivory Coast and Brazil. Bulk treatment of the data showed that the complex models outperform the simple ones in calibration mode but not in verification mode. We argue that the main reason why complex models lack stability is that the structure, i.e. the way components are organised, is not suited to extracting information available in hydrological time-series. An inadequate complexity typically results in model over-parameterisation and parameter uncertainty. Although complexity has been used as a response to the challenge of predicting the hydrological effects of environmental changes, this study suggests that such models may have been developed with excessive confidence and that they could face difficulties of parameter estimation and structure validation when confronted with hydro-meteorological time-series. This comparative study indicates that some parsimonious models can yield promising results and should be further developed, although they are not able to tackle all types of problems, which would be the case if their complexity were ideally adapted.

Spatial proximity, physical similarity, regression and ungaged catchments: A comparison of regionalization approaches based on 913 French catchments

[1] Given the contradictory results from recent studies, this paper compares classical regionalization schemes of catchment model parameters over the wide range of hydroclimates found in France. To ensure the generality of the conclusions, we used two lumped rainfall-runoff models applied to daily data over a large set of 913 French catchments. Three types of approaches were considered: regionalization using regression, regionalization based on spatial proximity and regionalization based on physical similarity. This comparison shows that in France, where a dense network of gauging stations is available, spatial proximity provides the best regionalization solution. The regression approach is the least satisfactory, with results very close to those obtained using one median parameter set for the whole country. The physical similarity approach is intermediary. However, the results obtained with these three methods lag far behind those obtained by full model calibration. Our results also show that some improvement could be made by combining spatial proximity and physical similarity, and that there is still considerable room for progress in the field of ungaged catchment modeling.

GR3J: a daily watershed model with three free parameters

This paper describes a new empirical watershed model, the prime feature of which is its parsimony. It involves only three free parameters, a characteristic unparalleled by continuous process models able to work on a wide array of catchments. In spite of its crude simplicity, it achieved, on average, worthwhile results on a set of 140 French catchments and overwhelmingly outperformed a linear model involving 16 parameters. It performed roughly as well as a conceptual model with five free parameters, derived from the well-known TOPMODEL.

Impact of limited streamflow data on the efficiency and the parameters of rainfall—runoff models

Abstract Streamflow data are essential for the calibration of continuous rainfall—runoff (RR) models. The quantity and quality of streamflow data can significantly influence parameter calibration and thus model robustness. Most existing sensitivity analysis studies on the role of streamflow data have used continuous periods to calibrate model parameters, with a minimum of one year, though ideally much longer periods are generally advised. However, in practical model applications, streamflow data series available for model calibration may be rather short or non-continuous. This study aims at assessing the sensitivity of continuous RR models to the quantity of information used during model calibration when it is randomly sampled in the observed hydrograph, i.e. using non-continuous calibration periods. This sampling provides less auto-correlated streamflow information for model calibration than continuous records. Two daily RR models with four and six free parameters were tested on a sample of 12 basins in the USA to obtain more general conclusions. The results showed that, in general, 350 calibration days sampled out of a longer data set including dry and wet conditions are sufficient to obtain robust estimates of model parameters. The more parsimonious model requires fewer calibration data to obtain stable and robust parameter values. Stable parameter values prove more difficult to reach in the driest catchments.

A two-parameter monthly water balance model for French watersheds

Abstract A very simple water balance model has been developed for water resources assessment and management. Only two parameters are to be calibrated or estimated from physical characteristics for its use on a given watershed. The parsimony of the model should guarantee its robustness and make the research for parameter explanation easier. In spite of its obvious crudeness, this model compares well with other well-known monthly water balance models.

A distribution-free test to detect gradual changes in watershed behavior

[1] Thispaperpresentsadistribution-free statistical testaimedatdetectinggradualchanges inthehydrological behaviorofwatersheds.Theproposedtestusesarainfall-runoffmodelto identify watershed behavior over successive time periods and a resampling approach to quantify the significance of trends. The method can be applied with any model deemed suitable for the studied watershed. To assess test efficiency, we used three different case studies: An afforested agricultural watershed, a burnt-over forested watershed, and a watershed covered by old-growth forest. All three watersheds had a long period of rainfall andrunoffrecords(60,35,and40years,respectively), onwhichstationaritycouldbetested. The test was shown to adequately detect gradual changes, and it can therefore be useful to identify hydrological trends, wherever rainfall and streamflow time series are available. INDEX TERMS: 1803 Hydrology: Anthropogenic effects; 1860 Hydrology: Runoff and streamflow; 1821 Hydrology: Floods; 1812 Hydrology: Drought; KEYWORDS: rainfall-runoff modeling, trend

Flood forecasting with a watershed model : a new method of parameter updating

Abstract Flood forecasting is of prime importance when it comes to reducing the possible number of lives lost to storm-induced floods. Because rainfall-runoff models are far from being perfect, hydrologists need to continuously update outputs from the rainfall-runoff model they use, in order to adapt to the actual emergency situation. This paper introduces a new updating procedure that can be combined with conceptual rainfall-runoff models for flood forecasting purposes. Conceptual models are highly nonlinear and cannot easily accommodate theoretically optimal methods such as Kalman filtering. Most methods developed so far mainly update the states of the system, i.e. the contents of the reservoirs involved in the rainfall-runoff model. The new parameter updating method proves to be superior to a standard error correction method on four watersheds whose floods can cause damage to the greater Paris area. Moreover, further developments of the approach are possible, especially along the idea of combining parameter updating with assimilation of additional data such as soil moisture data from field measurements and/or from remote sensing.

Flood routing in a wide channel with a quadratic lag-and-route method

Abstract In hydrological practice, flood routing is rarely achieved using the complete Saint-Venant equations. Instead, hydrological methods have been developed to provide users with simple methods that are robust yet efficient. The lag-and-route method has been widely used but its parameters have never been related to the physical parameters of channels. Moreover, the linear reservoir in this method is not well suited to reproduce effectively actual channel routing. A quadratic reservoir coupled with a lag component results in a far more accurate method while remaining numerically workable. This latter method depends on only two parameters which have been successfully related to physical channel features. This simple and efficient method enables one to gain insight into the main features of flood propagation in river channels.