Type of hydrological models

The models that have been typically used are the lumped models. These models describe the watershed as a single entity with a single rainfall input (mean rainfall). The discharge at the watershed outlet is described based on a global dynamic of the system. There are numerous lumped hydrologic models. These models are usually based on the concept of the unit hydrogram, UH. This concept is valid within a framework that assumes that the watershed is a linear causative and time invariant system, where only part of the rainfall, the rainfall excess or "Funció de Producció" or "Funció de pèrdues", produces runoff (Figure 1).

 

esquema_classic_peq

Figure 1: Typical diagram of the components of a lumped model with a unitarian hydrogram.


This type of models are not very physically based. They do not take into account that part of the surface runoff may infiltrate when it travels through a permeable river bed and they, therefore, underestimate the subsurface component of river flow. In addition, the identification of the UH is not simple. The UH may be derived from the solution to the inverse problem using hydrologic information available in the watershed (rainfall and flow data). However, there are not always enough data or data of enough quality. In addition, the information on the spatial distribution of the rainfall is limited. To try to avoid this problem, the watershed has been conceptualized as a simpler system so that a single unit hydrogram may be derived based on few parameters. These parameters may be obtained through statistical techniques. These type of models are the lumped conceptual models. A further step has been to relate the parameters of the UH to geomorphologic characteristics of the watershed. This has resulted in the Geomorphologic Unit Hydrogram which is derived solely from the physical characteristics of the watershed, that is, is not based on hydrologic data.

A new set of models were developed in the 60s, the lumped models physically based. These models aim to represent the processes that occur in the watershed with a more solid physical base. These models consist of a series of conceptual elements interconnected, each element represents a subsystem within the hydrologic cycle (evaporation, surface runoff when the profile is saturated, rill surface runoff). However, eventhough the number of parameters is great, the processes description is still quite simplified, so much so, that there is not much difference between these models and the lumped conceptual models.

A new approach to modeling the hydrologic processes was developed in the 70s. The model structure adapts to the inputs and outputs of the system. For this reason, these models are called black box models. The first type of these models are the autoregressive models, based on time series analysis. A newer type of model, based on the autoregressive models, but with the possibility of incorporating information, based on newer data are called artificial neural network models, eventhough there are cases where this type of models have given good results, more work needs to be done to investigate its possibilities.

The critera that are the base of a model structure are still unclear, this makes difficult the use of a model in watersheds where there are no data or the task of parameter regionalization. However, nowadays technology allows us to have an exhaustive knowledge of certain variables with good spatial resolution and to operationally work with volumes of information in a short time.

This makes us think that a model that will incorporate the spatial distribution of the variables, will give better results than a lumped one. The distributed models incorpore spatial variability of the variables more accurately and they reproduce more faithfully the processes that occur in a watershed than lumped models do. The physically based distributed models describe with great detail the processes of the watershed. These models may describe the hydrologic processes with a fine resolution (100-500m). The equations of the processes are solved in each unit (cell) and the output of the processes of neighbour cells are combined. This leads to very complex models that require a great amount of information, and at least, up to now, the calibration of a tremendous amount of parameters if not all the variables may be estimated from field data.

Nowadays, we may follow a line in between both major modeling lines. The so called models of distributed parameters. It is a characteristic of these type of models to divide the watershed in smaller units. These units are more homogeneus than the whole watershed. The model is applied to them. The watershed response is a composite of the responses of the units. A first classification of this type of models may be done based on the type of units (Figure 2). These units may be subwatersheds or artificial units based on a preestablished criterion. The fact that a lumped model is applied to each unit is the reason why this type of model is called distributed model (but) conceptually lumped. Distributed hydrologic modeling is a developing field around the world and there are many types, phylosophies, and lines of work. As an example, one type of distributed hydrologic model is that which tries to integrate the units in order to represent the spatiallity of the variables in a simple way. TopModel is such a model.

model_peq

Figure 2: Schematic of a watershed discretization and of the flow network in subwatersheds (left); and in artificial units (right). Example based on the Besòs River watershed.


The previous text was extracted from Carles Corral's dissertation, titled: "Desenvolupament d'un model hidrològic per incorporar la informació del radar meteorològic. Aplicació operacional a la conca del riu Besòs" (Development of a hydrologic model that incorporates weather radar data. Application to the Besòs River watershed). Civil Engineering Dept. , Universitat Politècnica de Catalunya.