Hydrological Model Can Help Predict Floods

floods
Researchers evaluated the application of a computer model for hydrological disaster prediction and early warning in the Doce river basin, which contains more than 200 Brazilian municipalities (photo: Wikimedia Commons)

The Doce, one of Brazil’s most important rivers, is some 850 km long and traverses Minas Gerais State and Espírito Santo State, with several major tributaries. Its catchment area, which covers 83,400 sq. km., contains more than 200 towns and cities. For the past year the river has suffered the consequences of the worst environmental disaster in Brazilian history. Over 60 million cu. m. of toxic mining waste cascaded into the Doce when the Bento Rodrigues iron ore tailings dam collapsed in Mariana, Minas Gerais, causing enormous damage to the local population and to ecosystems throughout the river’s basin.

Until the dam failed, the basin’s main problem was disasters caused by torrential rain: these were both constant and severe, with natural and human losses. Major floods occurred in Governador Valadares (Minas Gerais) in 1979, 1985, 1997, 2005 and 2008, Colatina (Espírito Santo) in 1997 and 2013, and Ponte Nova (Minas Gerais) in 2008. The river frequently overflows with less intense flooding. A flood early warning system for the Doce river basin was set up in 1997 by National Geological Service (CPRM).

“The Doce river basin flood early warning system generates water-level estimates based on a simple statistical linear propagation model, which uses measurements at upstream gauging stations to estimate streamflow for seven of the basin’s most important cities up to 24 hours in advance,” said Javier Tomasella, a researcher at Brazil’s Natural Disaster Surveillance & Early Warning Center (CEMADEN).

However, 24 hours notice is often too little, too late. “One-day flood warnings can be useless for many towns in the region, such as those in headwater basins where streams burst their banks quickly. These are typically associated with natural disasters that cost most in terms of human lives lost,” Tomasella said at FAPESP Week Montevideo, held on November 17-18 in the Uruguayan capital. The symposium was organized by FAPESP in collaboration with the Montevideo Group Association of Universities (AUGM) and Uruguay’s University of the Republic (UDELAR).

Reliable forecasting is just as important as early warning. Both systems (early warning and forecasting) should ideally work together so as to predict when water levels will rise in the rivers of a region enough time in advance to warn and help evacuate local communities.

A combined early warning and forecasting system requires a multidisciplinary team of collaborators in several fields, including meteorologists, hydrologists and disaster managers.

Tomasella’s presentation to the symposium included an assessment of a streamflow prediction system jointly developed by CEMADEN and Brazil’s National Space Research Institute (INPE), with funding from the National Scientific & Technological Development Council (CNPq). Predictions are based on INPE’s Distributed Hydrological Model (MHD), with weather forecasts derived from the regional Eta model run by the Center for Weather Forecasting & Climate Studies (CPTEC).

“According to our performance statistics, INPE’s MHD model produces promising results for up to 5 days ahead. The analysis shows that performance depends on basin scale and that the results are extremely dependent on initialization of the hydrological model, making joint operation with a real-time monitoring system essential,” Tomasella said.

A hydrological model is a simplified representation of water flow and behavior in a river system for the purposes of water management and scenario testing, in which mathematical equations are used to simulate the various processes that make up the hydrological cycle, such as streamflow, rainfall, evaporation, runoff etc.

Ensemble forecasting

Hydrological processes are complex and often not totally known, so that mathematical representation has limitations. Models differ with regard to structure. According to the researchers, the choice of a model should be based on the characteristics of the area studied and the purpose of modeling, while the choice of simulation details will depend on the basic information available.

Tomasella explained that INPE’s Distributed Hydrological Model (MHD) is designed to interact with atmospheric models in studies of global environmental changes. In the MHD model the basin represented is divided into a grid of regular cells to facilitate the transfer of information between models. Cell size may vary according to the region modeled and the density of the available information.

INPE’s MHD model comprises resolution modules such as “soil water balance”, “free water surface evaporation, saturated areas, vegetation interception and transpiration”, and “surface and subsurface flow and baseflow in each cell”.

The model represents the river’s daily and hourly hydrological cycle, Tomasella explained, and can be used for ensemble forecasting of hourly streamflows via different parts of CPTEC’s atmospheric Eta model. The Eta model is used in Brazil and many other countries for weather forecasting, climate prediction and climate variability studies, furnishing high-resolution ensemble forecasts that can be fed into the hydrological model.

“The atmosphere is a complex non-linear dynamic system, and forecasting its state for any future time entails uncertainty that’s inherent in the models and the limitations to our knowledge,” Tomasella told Agência FAPESP. “As a result, the methodology for simulating or estimating probable future states of the atmosphere is based on what’s known as ensemble forecasting.

“The atmosphere is a chaotic system, so minor uncertainties in initial conditions may entail huge variations in a forecast. In order to minimize these errors, numerical forecasting models are run a number of times with small perturbations to initial conditions. The collection of different runs from the series of initialization tweaks is called an ensemble, and each individual run is an ensemble member. Ensemble forecasts are fed into the hydrological model, which produces probabilistic predictions of river discharge.”

According to researchers affiliated with CEMADEN and INPE, the results of studies performed to date using the MHD model show that the model can reliably be used as a tool for hydrological disaster prediction and early warning. CEMADEN currently monitors 26 municipalities in the Doce river basin.