Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Off to new Shores: A Dataset & Benchmark for (near-)coastal Flood Inundation Forecasting

Brandon Victor, Mathilde Letard, Peter Naylor, Karim Douch, Nicolas Longépé, Zhen He, Patrick Ebel

Abstract

Floods are among the most common and devastating natural hazards, imposing immense costs on our society and economy due to their disastrous consequences. Recent progress in weather prediction and spaceborne flood mapping demonstrated the feasibility of anticipating extreme events and reliably detecting their catastrophic effects afterwards. However, these efforts are rarely linked to one another and there is a critical lack of datasets and benchmarks to enable the direct forecasting of flood extent. To resolve this issue, we curate a novel dataset enabling a timely prediction of flood extent. Furthermore, we provide a representative evaluation of state-of-the-art methods, structured into two benchmark tracks for forecasting flood inundation maps i) in general and ii) focused on coastal regions. Altogether, our dataset and benchmark provide a comprehensive platform for evaluating flood forecasts, enabling future solutions for this critical challenge. Data, code & models are shared at https://github.com/Multihuntr/GFF under a CC0 license.

Off to new Shores: A Dataset & Benchmark for (near-)coastal Flood Inundation Forecasting

Abstract

Floods are among the most common and devastating natural hazards, imposing immense costs on our society and economy due to their disastrous consequences. Recent progress in weather prediction and spaceborne flood mapping demonstrated the feasibility of anticipating extreme events and reliably detecting their catastrophic effects afterwards. However, these efforts are rarely linked to one another and there is a critical lack of datasets and benchmarks to enable the direct forecasting of flood extent. To resolve this issue, we curate a novel dataset enabling a timely prediction of flood extent. Furthermore, we provide a representative evaluation of state-of-the-art methods, structured into two benchmark tracks for forecasting flood inundation maps i) in general and ii) focused on coastal regions. Altogether, our dataset and benchmark provide a comprehensive platform for evaluating flood forecasts, enabling future solutions for this critical challenge. Data, code & models are shared at https://github.com/Multihuntr/GFF under a CC0 license.
Paper Structure (17 sections, 5 figures, 3 tables)

This paper contains 17 sections, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. River streamflow & runoff forecasting
  4. Rapid mapping
  5. Data
  6. Identifying candidate Regions Of Interest
  7. Capturing the diverse drivers of flood & flood extent forecasting
  8. Generating floodmaps anywhere
  9. Label post-processing
  10. Selecting representative ROI via stratification
  11. Data splits
  12. Benchmarks
  13. Baselines
  14. Track 1: Global flood map prediction
  15. Track 2: Coastal versus near-coastal & inland flood map prediction
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Exemplary data. Columns: Three ERA5 and ERA5-Land time series samples, DEM, Height Above Nearest Drainage (HAND). Pre-flood Sentinel-1 (S1) overlaid with target-time flood map. Columns 1-3 provide context data at a coarse scale, red dots indicates the coverage of columns 4-6 at fine scale. Rows: Three examples, showcasing floods near a river, settlement and coastline. The events are due to heavy rain, tropical storms and a storm surge, illustrating the diversity of GFF.
  • Figure 2: Conceptual similarities and differences between the two tasks of a) in-event rapid flood mapping and b) pre-event forecasting potential flood maps. While the former focuses on detecting change of water coverage in observations at prior dates versus now, the latter is about predicting such change at a given lead time $d$ where no observation is yet available. Beyond internalizing the physical signatures of moisture and water, this requires learning the dynamics of associated flood drivers.
  • Figure 3: Map of dataset. Points are centers of curated ROI and their associated flood events. Red shadings indicate the distribution of global flood hazard frequency dilley2005natural. Many of the endangered regions are close to the sea, especially in the (sub-)tropics. Unlike most prior work addressing well-monitored or upstream regions, these areas are particularly well represented in our dataset.
  • Figure 4: Empirical distributions of floods in our dataset regarding their frequency across different a) climate zones and b) continents. In both regards, the histogram of cumulative flood (orange) and no flood observations (green) qualitatively mirrors the expected occurrence of all global flood events reported by the DFO (blue), with minor discrepancies due to the (near-)coastal focus.
  • Figure 5: Baseline model platform design, accommodating for i) a context network (top) which processes a spatio-temporal sequence of coarse resolution context data and whose output feature embeddings are then processed by ii) a local network (bottom), concatenated with local high resolution data. The two network backbones (in dark grey) are placeholders for the different baselines benchmarked herein. The final output is a flood segmentation forecast at a given lead time.