Web Tool Aims to Pair Hydrometeorological Forecasting with Pandemic Water Monitoring in Brazil

Researchers in Brazil report that building a web-based platform for hydrometeorological forecasting in the MATOPIBA region—Maranhão, Tocantins, Piauí and Bahia—is technically feasible, according to a preliminary feasibility analysis published as a research article in October 2021. The study, led by José Roberto Dantas da Silva Júnior and colleagues including Davidson Martins Moreira, evaluates whether the WRF-Hydro modeling system can be integrated with artificial intelligence tools and a geographic information system interface so that researchers who are not computing specialists can run simulations, monitor water bodies and generate predictive analyses of flows with reduced computational cost.

The motivation sits at the intersection of two problems. The first is water management: the authors note that the latest United Nations World Water Development Report highlighted that only 10 percent of developing countries have water quality monitoring systems, and that in Brazil roughly 35 million people lack treated water while more than 100 million lack basic sanitation. The second is the COVID-19 pandemic: the authors cite studies indicating that municipalities with poor sanitation recorded the highest numbers of infections and deaths from SARS-CoV-2, and they argue that better tools for understanding the dynamics, displacement and reach of contaminants in water bodies could serve as an auxiliary instrument for managing both water resources and the pandemic.

The scientific core of the proposal is the WRF-Hydro system, the hydrological extension of the Weather Research and Forecasting model developed by the National Center for Atmospheric Research in partnership with NASA. The system couples an atmospheric model with a hydrological model, allowing detailed simulation of rainfall, surface runoff, streamflow and flood forecasting. The authors describe it as state of the art in hydrometeorological modeling, and they review its application worldwide: flood forecasting and reservoir studies in Greece and the United States, hydrometeorological analysis of the Tana River basin in Kenya, soil moisture simulation in hyper-arid environments of the United Arab Emirates, and flow estimation for the Brahmaputra River between India and Bangladesh, where simulated and observed data showed good correlation. In Brazil, previous work applied the system in Pernambuco state and in the MATOPIBA region itself.

That pedigree comes with a practical drawback, and it is the technical crux of the new paper. Running WRF-Hydro involves preprocessing, processing and postprocessing phases that demand large volumes of high-resolution input data—topography, land use and occupation, lake and river information—and a robust computing infrastructure. In Brazil, the authors argue, such infrastructure and measured validation data are not always available, which limits who can use the model and where. Their proposed answer is a web tool that wraps the model’s execution steps behind an intuitive interface, integrating WRF-Hydro with ESRI’s ArcGIS geoprocessing toolkit and exposing the workflow through REST architecture, the SOAP protocol and an application programming interface developed in languages including Java.NET and Python.

The methodological demonstration centers on a nested modeling configuration. The weather model was set up in three nested domains with horizontal resolutions of 9 kilometers, 3 kilometers and 1 kilometer. The domain of interest, D03, covers a grid of 448 by 454 cells at 1-kilometer resolution, with geoprocessing tools refining the water-scope grid to 250 meters. The simulation used the WRF-ARW core version 3.9.1, initialized at 12 UTC on March 10, 2019 and extending to 18 UTC on March 20, 2019—246 hours of simulation intended to generate preliminary data for geoprocessing in ArcGIS. Initial and boundary conditions came from NCEP’s GDAS-FNL global analysis at 0.25-degree resolution and 6-hour intervals, while land use and topography data were supplied by the United States Geological Survey.

The artificial intelligence component is presented as a planned accelerator rather than a completed result. The authors outline how convolutional neural networks and long short-term memory networks could be trained on measured data and on the history of WRF-Hydro outputs, so that once trained, forecasts of water body flows could be obtained faster and at lower computational cost than running the full physical model. They note that neural networks have already delivered good results for short- and medium-term simulations and pollutant concentration modeling, and that AI has been applied to the WRF model without its water module, but that combining AI with WRF-Hydro remains under development. The proposed architecture would host previously trained models in an AI module, validate them against WRF-Hydro results, and support faster forecasting of extreme events such as droughts and floods.

The pandemic linkage is framed cautiously as an additional, potential use. The authors review evidence that coronaviruses can persist in water and wastewater—survival of up to 10 days in water at 23 degrees Celsius and up to 100 days at 4 degrees Celsius was reported for coronaviruses before the SARS-CoV-2 outbreak, and studies of SARS-CoV-2 itself suggest survival ranging from roughly 1.57 days to 14 days or more, depending strongly on temperature, pH, organic matter and disinfectant presence. They also note that approximately 54 percent of Brazil’s sewage is collected and only 69 percent of that collected sewage is treated, meaning the virus can be released into water bodies, and that wastewater monitoring has been identified as a useful indicator of contamination levels—although one study they cite found laboratory tests alone insufficient to predict rises or falls in infection levels, and another reported that transmission through contaminated water and feces remains uncertain and needs further study.

Against that backdrop, the proposed platform would associate pandemic indicators with characteristics of water bodies modeled by WRF-Hydro and optimized by AI, offering an auxiliary rapid-response complement to wastewater-based epidemiology. The authors are explicit that direct monitoring of water bodies requires sampling campaigns, laboratory analyses and trained personnel, all of which can be costly; a modeling tool would not replace those measurements but could help target them and interpret their spatial dynamics. One cited study even found significant correlations between hydrological variables—lake area, river length, precipitation and water resource volume—and confirmed SARS-CoV-2 case numbers, suggesting, in the authors’ reading, that such correlations merit attention even as the transmission route itself remains under investigation.

What the study actually demonstrates is narrower than the ambition. The team completed regional geoprocessing with ArcGIS, obtained partial detailed regional information, and produced flow, precipitation and other variables that they say show WRF-Hydro is an efficient hydrometeorological simulation model. They also present a digital elevation model of the Rio Grande with simulated river flows, including an animated visualization of a micro-basin, and describe a dashboard integrating the HPC-hosted coupled model, automated Python simulation scripts and the GIS web layer. The paper states plainly that the research is in development and represents preliminary results with future perspectives, with the group still working on automated and efficient data calibration to improve simulation accuracy.

The conclusion the authors draw is one of feasibility, not finished capability: consolidating the described processes, they write, would allow implementation of a solution usable largely by scholars who are not computing experts, with promising prospects for reducing computational cost and for correlating water management with contamination indicators during pandemics such as COVID-19. For the MATOPIBA region—home to large shares of Brazil’s freshwater in the Northwest Atlantic, Parnaíba, Tocantins-Araguaia and São Francisco basins, and an agricultural frontier of strong socioeconomic interest to the Brazilian government—the value of such a tool lies in making sophisticated hydrometeorological modeling accessible for drought and flood forecasting and water resource planning. Whether the AI-accelerated, pandemic-aware version of the platform delivers on that promise will depend on the validation and calibration steps the team has yet to complete.

Subject of Research: Feasibility of a web-based hydrometeorological forecasting tool combining WRF-Hydro and artificial intelligence for water resource management in Brazil's MATOPIBA region

Article Title: Feasibility analysis on the construction of a web solution for hydrometeorological forecasting considering water body management and indicators for the SARS-COV-2 pandemic

Article References: Dantas da Silva Júnior, J. R., Pedruzzi, R., de Souza, F. M., Ferraz, P. S., Silva, D. G., Vieira, C. S., de Moraes, M. R., Nascimento, E. G. S., & Moreira, D. M. (2021). Feasibility analysis on the construction of a web solution for hydrometeorological forecasting considering water body management and indicators for the SARS-COV-2 pandemic. AI Perspectives, 3(1), Article 4. https://doi.org/10.1186/s42467-021-00011-0

Image Credits: AI Generated

DOI: 10.1186/s42467-021-00011-0

Keywords: WRF-Hydro, hydrometeorological modeling, MATOPIBA, Brazil, water resources, artificial intelligence, COVID-19, SARS-CoV-2, wastewater, ArcGIS, flood forecasting, web platform