Urban-PREDICT

Working Package 2

Prediction and EWS Across Spatial Scales

Introduction

Urban areas are increasingly vulnerable to multi-hazard risks, ranging from extreme air pollution, extreme heat, heavy precipitation, and flooding to cascading weather events. Given the spatial variability of these hazards, the resolution of their prediction may significantly affect the accuracy and usability of early warning systems for diverse urban populations. Accurate and reliable hazard predictions are critical for informing decision-makers and communities to implement timely and effective responses, particularly in heterogeneous urban environments.

For instance, in the PRAISE-HK system, ultra-high spatiotemporal resolution of air pollutant concentrations at street-level (~10 meters) is required to assess exposure and mobility and to warn vulnerable groups such as asthma patients. This is particularly relevant at traffic hotspots, where large gradients in air quality may result in inequalities in population exposure. Such spatial variability depends not only on local emissions and atmospheric conditions but also on the interactions between wind and complex urban morphology. The urban background contribution may also be a relevant factor.

Understanding these gradients and their main drivers for different hazards and types of cities is key to identifying model and data requirements for their accurate representation. However, there is still a lack of standardized benchmarks and good practices for multi-hazard modelling at different spatial scales. Benchmarks could consist of coarse resolution (or coarse-grained) multi-hazard model predictions, supplemented with observations from selected case studies, to evaluate the performance of high-resolution multi-hazard model predictions.

In addition, with the global push for impact-based forecasts and warnings to make warnings more meaningful and relevant, the inclusion of exposure data has been identified as a required data source for setting impact-based warning thresholds. However, setting thresholds based on exposure may introduce an urban-rural bias, where warnings are issued in areas of higher population density, excluding low-density areas. This introduces a need to explore and test the role and influence of exposure data in designing impact-based warnings.

Activities

  1. Conduct a focused review examining the spatial variability of key hazards—air quality, urban heat stress, cold stress and freezing infrastructure, urban heavy precipitation, urban wind, flash flooding, and compound/cascading hazards. Include case study examples of past urban hazards. Develop a preliminary analysis table of optimal prediction scales for each hazard type.
  2. Coordinate with WP3 to recommend multiple case studies from different geographical regions, representing a diverse range of hazards and contexts. Selected cases will include those with available observational data and engaged communities. Participants will run hazard prediction models at varying resolutions.
  3. Assess the performance of different models and configurations for providing urban multi-hazard warnings, including differences in physics settings, downscaling, input datasets, and assimilation techniques.
  4. Identify appropriate scales for effective hazard prediction and EWS. Develop open-access benchmarks and evidence-based recommendations for ultra-high-resolution urban hazard models.
  5. Use a risk-modelling approach with exposure data (assets and vulnerability functions) to test how these influence threshold-setting for impact-based warnings. Address issues such as urban-rural bias and evaluate performance in selected case studies.