Track alignment
Primary track: Track 6 — Resilience Informatics. Kraków Sponge connects drought monitoring with practical landscape-retention screening for the Rudawa, Prądnik (Białucha), and Dłubnia catchments. It brings together IMGW gauges and warnings, historical flow records since 1991, ERA5 climate indicators, mapped drainage, and terrain-based survey candidates.
It also contributes to Data-to-Insight through transparent geospatial calculations, and Citizen Science UX through bilingual, location-based observations that users control and can export.
The One Health connection is shared freshwater resilience: upstream water availability matters to aquatic habitats and communities, including the city's drinking-water catchments. The prototype supports investigation and coordination; it does not claim demonstrated ecological or public-health outcomes.
Inspiration
Kraków's water story begins beyond the city boundary. The Rudawa and Dłubnia feed water-treatment plants, while the Prądnik connects protected landscapes with the city. Looking only at urban streets misses the fields, forests, drainage ditches and river corridors upstream.
The project's dated drought analysis compares rainfall, climatic water balance and river flows. These indicators tell different parts of the story: near-normal rainfall totals can coexist with very low flows. They motivate a better question than simply whether it rained: where should we investigate the landscape's capacity to retain water?
Kraków Sponge brings these scattered datasets into one explorable map and shows the assumptions behind each proposed survey priority.
What it does
Problem: residents and local decision-makers need to connect drought evidence with specific places to investigate. Monitoring and mapping are incomplete, and an attractive map can easily overstate what terrain data prove.
Solution: a working English/Polish web application with four connected workflows:
- Understand the drought. Compare gauge readings and warnings with historical flow records, rainfall, reference evapotranspiration and soil moisture. Source dates and freshness indicators distinguish current readings from saved snapshots.
- Find survey priorities. Explore drainage ditches ranked by an explained heuristic using land cover, slope, length and building distance. Inspect six ponding survey candidates refined with official 1 m ground terrain. Changing fill height redraws connected pond footprints and calculated added capacity; open drainage routes limit the result.
- Compare and share scenarios. Compare candidate settings, copy a scenario link, and download its summary. Explore river corridors screened against mapped buildings and land cover, and compare current channels with calculated bend concepts.
- Add observations from the ground. Report a ditch, stream, spring or culvert as flowing, standing, dry or blocked, with a date, note and optional photo. Reports remain in the browser until the user exports GeoJSON or chooses to open and submit a prefilled GitHub issue. Example reports are labelled and excluded from real exports.
The map includes a guided story tour, methods and source explanations, mobile layouts, and Polish and English interfaces.
Target users: residents, walkers and anglers; local environmental groups; municipal and catchment staff; and farmers or foresters considering places for professional field assessment.
Intended impact: shorten the path from fragmented data to a documented shortlist for field visits. Shareable scenarios make assumptions reviewable, while citizen observations can help identify monitoring gaps. No retention intervention or measured environmental benefit is claimed as completed.
How we built it
A Python geospatial pipeline processes BDOT10k topography, Copernicus elevation, GUGiK 1 m terrain and hydrological/climate data into compact map layers and statistics. It uses GeoPandas, Rasterio, Pysheds, Shapely and related scientific libraries. Reconstructed catchments are compared with the official MPHP reference.
The frontend uses JavaScript, Vite, MapLibre GL and Chart.js. Processed data ship with a static GitHub Pages deployment, so judges can use the prototype without accounts, API keys or a separate application server. GitHub Actions builds the site and refreshes drought snapshots; the interface exposes data dates and stale-data states.
The terrain workflow computes connected standing-water footprints, checks mapped building clearance, accounts for drainage outlets, and separates existing depression capacity from added capacity. River-corridor screening includes building proximity as well as land-cover proportions. Citizen observations use local browser storage, GeoJSON export and an explicit GitHub handoff.
The repository includes setup instructions, methodology, source attribution, and browser and geometry checks. The submitted demo records the deployed application. Its narration is AI-generated with Gemini TTS via OpenRouter; the footage and calculations are from the actual application.
Challenges we ran into
- Different resolutions and missing data. A coarse regional elevation model is useful for catchment context but cannot resolve small ditch structures. We keep regional screening separate from selected 1 m terrain calculations and show the remaining uncertainty.
- False confidence from map colours. Corridor averages can hide a building close to the river. Building proximity therefore overrides an apparently open corridor, and green means fewer mapped constraints rather than available land.
- Storage is not a water forecast. Raising a fill can expose an outlet and stop increasing the pond's capacity. The app shows this behaviour and distinguishes geometric standing-water capacity from rainfall, groundwater recharge and flood predictions.
- Operational data can fail or lag. Snapshot dates and stale-state messages are essential when external services are unavailable.
- Useful citizen science without premature infrastructure. Browser-local drafts and explicit export make the workflow usable while preserving the distinction between a private observation and a shared report.
Accomplishments that we're proud of
- A deployed prototype connects drought evidence, candidate screening, scenario comparison, sharing and observations in one bilingual interface.
- Selected survey candidates respond to fill-height changes using 1 m terrain, with visible drainage limits and calculation assumptions.
- Regional catchments are compared with an official reference; the documented area differences are approximately 0.2–7.1%, with remaining urban-routing limitations.
- The demo shows actual application behaviour in 4:26, within the required 3–5 minutes.
- The public repository includes reproducible setup, methods, data attribution and an MIT code licence.
What we learned
The most useful output is often a better field question. A terrain model can show where water would stand under stated assumptions, but it cannot establish landowner consent, hidden drains, reliable inflow, ecological suitability or engineering safety.
We also learned that reporting uncertainty is part of the interface: dates, spill routes, missing monitoring and the difference between saved and submitted observations need to be visible where decisions are made.
What's next
Validate candidate locations with field surveys and land managers; add drainage, soil and groundwater evidence; and compare the screening results with professional hydraulic assessment. Develop a moderated shared observation layer with repeat-visit records, while retaining explicit consent and provenance. Extend high-resolution terrain work only after validating the pilot workflow.
Potential partners include local municipalities, catchment authorities, conservation organisations and water-service teams. These are intended collaborators, not claimed partnerships.
Limits and responsible use
All retention and bend outputs are screening concepts for investigation. Geometric capacity assumes water reaches the site; it is not a prediction of runoff, recharge, flood reduction or drinking-water yield. Ownership, permissions, infrastructure, field conditions and ecological/hydraulic suitability need independent assessment before intervention. The project does not establish that land is free or that a ditch should be blocked.
Built with
javascript, python, vite, maplibre-gl, chart.js, geopandas, rasterio, pysheds, shapely, scikit-image, playwright, github-pages, github-actions, geojson
Data and licences
Code: MIT. Source datasets retain their own licences and attribution; see the repository's data-sources section. Sources include GUGiK/BDOT10k and ground terrain, Copernicus elevation and ERA5, IMGW-PIB, MPHP/PGW Wody Polskie, Open-Meteo, and OpenStreetMap/OpenFreeMap basemaps. No proprietary OneAquaHealth sandbox data are included or required.
Log in or sign up for Devpost to join the conversation.