Un método sencillo para predecir las crecidas del Ebro/Simple method to predict Ebro River level rises

Un equipo de investigadores de la Universidad de Castilla-La Mancha (UCLM) ha desarrollado un nuevo modelo matemático que permite predecir de forma sencilla las crecidas del río Ebro en Zaragoza a partir del flujo de agua que se registra en Castejón (Navarra). En una predicción a 20 horas, el sistema acierta el 97,5% de los casos.
El estudio, que se publicará en abril en la revista Environmental Modelling & Software, aborda el problema de cómo modelizar y predecir los flujos y niveles del Ebro en la capital aragonesa, a partir del análisis de un tramo de 110 km que comienza en la localidad navarra de Castejón.
Los investigadores han desarrollado el sistema predictivo en dos etapas. Primero han utilizado una función matemática“de transferencia” para dimensionar el modelo e identificar de forma aproximada el retardo con el que el flujo del agua en Castejón afecta al nivel de las aguas en Zaragoza.
“Pero la dinámica del río no se comporta de forma lineal”, apunta Pedregal. El profesor señala, por ejemplo, que el agua circula más rápido cuando el caudal es medio o bajo, que las tormentas no afectan por igual a un lecho seco (absorbe más el agua) que a uno húmedo, o que los deshielos rápidos arrastran un aguanieve que se comporta de forma diferente al agua líquida.
Todos estos factores, además de la propia física de los fluidos, hacen que la dinámica de los ríos sea “no lineal”, por lo que los investigadores han aplicado sistemas matemáticos de este tipo y han propuesto un algoritmo específico que recoja la influencia de las variables.
La capacidad predictiva del modelo se ha contrastado con la información de la base de datos histórica de la Confederación Hidrográfica del Ebro, produciendo unos errores medios del 2,5% en un horizonte de predicción de 20 horas.

A team of researchers from the University of Castilla-La Mancha (UCLM) has developed a new mathematical method to easily predict rises in the level of the Ebro River in Zaragoza based on water flow recorded in Castejón (Navarre). The system has a 97.5% success rate for 20-hour predictions.
The study, to be published in April in the journal Environmental Modelling & Software, looks at the problem of how to model and predict the flow and levels of the Ebro in the regional capital of Aragon, based on analysis of a 110km section that starts in the town of Castejón, in Navarre.
The researchers developed the predictive system in two stages. First of all they used a mathematical “transfer” function to determine the scale of the model and get an approximate idea of the time delay before the water flowing through Castejón has an impact on water levels in Zaragoza.
“But the river’s dynamics don’t behave in a linear way,” points out Pedregal: the water flows more quickly when it is at an average or low level, that storms do not have the same effect on a dry river bed (which absorbs more water) as a wet one, and that rapid thawing of snow introduces slushy water into the river, which behaves in a different way than liquid water.
All these factors, along with the physics of fluid itself, means that the dynamics of rivers are “not linear”, resulting in the researchers having had to use mathematical systems to reflect these factors and produce a specific algorithm to incorporate the impact of all these variables.
The predictive ability of the model has been contrasted with information from the historical database of the Hydrographic Confederation of the Ebro River, resulting in an average error rate of 2.5% over a 20-hour prediction timescale.

Tomado de/Taken from Plataforma SINC/Science Daily

Resumen de la publicación/Abstract of the paper
D.J. Pedregal, R. Rivas, V. Feliu, L. Sánchez y A. Linares. A non-linear forecasting system for the Ebro River at Zaragoza, Spain. Environmental Modelling & Software, 2009; 24 (4): 502
Abstract
This paper addresses the problem of modelling and forecasting river flows and levels based on flood routing type models. Though this is generally considered as a non-linear problem, very often it is treated by linear models. A forecasting system is built for the level and flow measurements registered in the Ebro River at the station of Zaragoza (Spain), with the main purpose of preventing floods in an early stage of development. The model takes advantage of the wealth of data available at the Ebro Hydrographical Confederation and is non-linear in essence. The system is obtained by application of system identification tools, starting from a linear specification and relating the parameters of the model estimated to some transformation of the input in the system. Such transformation requires the application of a Kalman Filter in a particular set up and the full estimation algorithm involves an iterative procedure. The model is fully developed on a data set and is thoroughly validated on a different span of data