AI-Based early warning and landslide Risk Monitoring System in NER
Ministry of Development of North Eastern Region (MDoNER) · Disaster Management · Software
The engineering is achievable and the sponsor is genuine, but this theme is crowded and there is no supplied dataset, so decide early what your ground truth is and be ready to defend it, because that is where this PS is won or lost.
What it actually is
Landslides cut off hill villages in the Northeast every monsoon, and right now nobody knows a slope is failing until someone phones it in. The ask is a system that watches rainfall, soil moisture, terrain and satellite imagery and warns district officials before a slope goes. It also has to reach villages that have almost no mobile signal.
What to build
A landslide susceptibility and early-warning platform with an ingestion layer pulling IMD rainfall APIs, satellite imagery, DEM-derived slope/aspect/curvature and a historical landslide inventory, an ML model that outputs a per-grid-cell risk score refreshed against live rainfall, a GIS heatmap dashboard overlaying vulnerable roads, villages and infrastructure with the four panels the description names — risk severity, road connectivity status, weather-linked forecast, emergency response priority — plus a field mobile app where citizens and officials upload geo-tagged crack/slope-movement photos into the same map, an automated SMS and push alert dispatcher, and multilingual, offline-capable operation with sync on reconnect.
Smallest thing that wins the room
Replay the rainfall record of a real past NER landslide day and show the risk map going red over that exact slope hours before the recorded failure time.
How crowded this one gets
A guess, projected from the 2025 statements — the last year where both the submission counts and the winners were published.
Quieter than 48% of the 226 · #119 of 226 by expected field
A normal-sized field. Your idea has to be good, not miraculous.
Why: central ministry statements sat below the average.
This is a guess, not a fact
Nobody has published 2026’s numbers yet. This is an analysed estimate from last year’s pattern, so please do not take it as the truth — check the live counter on the SIH portal before you decide anything. The range covers the middle half of likely outcomes, so one statement in two lands outside it. Entry closes at 500 ideas per statement, so no range goes past that — a statement that reaches the cap fills and shuts rather than drawing an unlimited crowd. The model reads only three things a team can see before choosing — software or hardware, the theme, and what kind of body posted it — and those explain about a quarter of the variation in last year’s field sizes (R² 0.25 on held-out statements). Trust the band more than the number, and the ordering more than either. It cannot see how good your idea is, which is the part that actually decides it.
The scores
The number is the shorthand. The line under it is the reason.
Acceptance potential
3/5Solid and genuinely wanted, but landslide and flood early-warning is one of the most repeatedly attempted themes in SIH history, and with no linked dataset your accuracy claim rests on labels you assembled yourself.
Feasibility
3/5DEM, slope and satellite layers are free from Bhuvan and SRTM/Copernicus and GSI's landslide inventory gives you labels, but the description's soil-moisture sensor network does not exist for you to read from and NER landslide records are sparse and coarsely geolocated, so your training labels will be thin where it matters most.
Innovation scope
2/5Points a through f fix the data sources, the alert recipients, the GIS layer, the citizen upload channel and all four dashboard panels, so what is left to you is model choice and nothing about the product shape.
Clarity
4/5The deliverables are enumerated tightly down to the four dashboard panels and the integration targets, but the description never says what counts as a correct prediction — no lead time, no spatial resolution, no accuracy floor — which is exactly the thing a judge will press you on.
Effort
HeavyMulti-source ingestion, a trained risk model, a GIS dashboard, a separate field-reporting mobile app, an SMS gateway, multilingual content and offline sync are six distinct workstreams, and the ML is the smallest of them.
Demo-ability
MediumYou cannot produce a landslide, so the whole demo rests on a historical replay being convincing; it works, but only if you have picked a well-documented event and can show the model was not simply trained on that day.
In its favour
- Green flag: Landslide susceptibility mapping is a mature published field, so you can anchor your feature set in accepted literature rather than inventing predictors and defending them from scratch
- Green flag: The citizen geo-tagged upload requirement gives you a second, cheap demo surface that works even when the model is not convincing
- Green flag: MDoNER as a sponsor means the evaluation panel will know NER terrain specifically, so naming real corridors like the Imphal–Jiribam route buys credibility that a generic map does not
Against it
- Red flag: The description lists soil moisture sensors as an input and there is no such deployed network you can read — teams either quietly drop it or fake it, and a judge who notices will treat the whole input stack as suspect
- Red flag: Landslide early warning is heavily attempted every SIH cycle; you will be one of many, and the marginal team is separated by validation rigour, not by dashboard polish
- Red flag: Offline plus multilingual plus SMS is a full day of unglamorous plumbing that teams always schedule last and always run out of time for
- Red flag: Without a stated lead time you have no target — predicting a landslide six hours out and six days out are different problems, and picking the wrong one makes your evaluation meaningless
What you will be writing
- XGBoost / Random Forest susceptibility model
- Google Earth Engine or Sentinel-2 imagery
- SRTM/Cartosat DEM with GDAL terrain derivatives
- PostGIS + Leaflet/Mapbox GL web-GIS
- IMD / OpenWeather rainfall API
- Flutter offline-first app with SQLite sync
- Disaster management
- Geospatial analytics
- Public safety infrastructure
Prior art to read before you start
landslide susceptibility mapping · rainfall-triggered early warning · geo-tagged citizen hazard reporting
Analysed by Claude Opus. Every score above is a judgment call with its reasoning attached — kindly cross-check this against the official statement on the SIH portal before your team commits to it.