Hydrological Digital Twin for Flood Prediction Through Community-driven Early Warning Systems for All: A Case Study on the Guaíba River Basin, Rio Grande do Sul, Brazil

Código

I-EBHE0185

Autores

Marília de Oliveira Felten, Caline Cecília Oliveira Leite, Mateo Hernández Sánchez, Luis Miguel Castillo Rápalo, Paul Munoz, Ruth Sofia La Fuente Pillco, Ann Van Griensven, Danielle de Almeida Bressiani, Eduardo Mário Mendiondo

Tema

WG 1.12: Development & application of river basin simulators

Resumo

Climate change is increasing the frequency and intensity of hydrological extremes, such as heavy precipitation events. Urban areas are particularly vulnerable to flooding due to these events, impacting social, economic, and environmental aspects. In this context, Early Warning Systems for All (EWS4All) are crucial for mitigating flood risks by providing timely evacuation and preparedness alerts to the civilian population. However, most EWS rely solely on rainfall amount or intensity as input parameters, whether measured in real-time or numerically forecasted, to establish alert thresholds. While these systems can be effective, they often fail in two key areas: 1) predicting flood magnitude and extent due to inadequate representation of initial hydrological conditions with updated information, sometimes resulting in false alarms; and 2) updating information collected from local communities potentially affected by floods, promoting inclusiveness, diversity, and participatory schemes of polycentric governance. To enhance the reliability and effectiveness of a new generation of EWS4All, integrating hydrological and hydraulic models with multiple rainfall datasets can address current gaps. These models can create a digital representation of watershed spatial hydrological conditions (such as soil water content and river water depths), including anthropogenic alterations like dams, and assimilate online hydrological data to generate a Hydrological Digital Twin (HDT). Recent studies demonstrate that some models can predict hazards using hydrological monitoring networks, online data, or satellite data. One such model is the Hydrodynamic Pollutant 2D Model (HydroPol2D), an open-source distributed hydrodynamic and quality model capable of generating an HDT based on near real-time data and comparing results with current monitoring gauge stations on a dashboard. This study aimed to compare the accuracy of HydroPol2D in providing near-real-time monitoring using different databases: the ANA (National Water and Basic Sanitation Agency) telemetric monitoring network and satellite information from PERSIANN PDIR Now. Conducted in the Guaíba River Basin, Rio Grande do Sul, Brazil, which experienced unprecedented flooding from April to June 2024, the study followed these steps: (i) executing the model for 2018?2023, comparing observed and simulated data to demonstrate significant representation of Guaíba River basin behavior; (ii) simulating HDT for the remaining months of 2024 up to May using telemetric and forecast satellite data from PERSIANN PDIR Now; (iii) comparing model-generated flows with recorded data using the Nash-Sutcliffe Efficiency Index (NSE), Kling-Gupta Efficiency (KGE) index, and percentage of bias to assess accuracy; and (iv) comparing generated flood maps with actual flooded locations. Preliminary results indicate the HydroPol2D model's capability to monitor current hydrological status during extreme events. However, PERSIANN PDIR Now's information may be limited as its artificial intelligence algorithms were trained with past events, causing near-real-time data to underestimate rainfall compared to unprecedented episodes observed in April?May 2024.