GIDPC.Green Infrastructure · Disaster Prevention
Solutions

Rural Climate Crisis

Rural Climate-Crisis AI · IoT Response

LSTM-based 6-hour runoff forecasting for farm catchments, LID effect verification, and IoT-sensor / Kakao-talk alerts — protecting irrigation infrastructure and farmland against flooding.

LSTMPyTorchIoT sensorsKMA AWSKakao alertsWindows Scheduler

What this solves

Rural micro-catchments are highly vulnerable to flash flooding. Physics-based models (SWMM, SCS-CN) are accurate but heavy on parameter calibration and compute — and they cannot be operated by field stakeholders (farmers, village leaders) directly.

An LSTM trained once predicts 6-hour runoff in 0.01 second. The result is effectively a "village AI forecaster" you build and own.

How we analyze

| Step | Method | Output | |---|---|---| | 1. Catchment definition | Area + land-cover analysis | Village-scale model | | 2. Input data synthesis | Korean storm + typhoon-grade time series | 3-yr hourly rainfall + soil moisture | | 3. Baseline / LID scenarios | SWMM · SCS-CN | runoff_baseline.csv, runoff_lid.csv | | 4. LSTM training | PyTorch, 2 layers, hidden 64 | lstm_baseline.pt, lstm_lid.pt | | 5. Validation | NSE / RMSE / peak error | Demo NSE 0.88 | | 6. Operational integration | Windows Scheduler + Kakao alerts | Automatic alerts when threshold exceeded |

Demo results (5 ha virtual village)

| Metric | Baseline | LID Applied | |---|---|---| | NSE | 0.641 | 0.880 | | Peak flow | 0.60 m³/s | 0.33 m³/s (−45%) | | Peak lag | 0 h | +5 h | | Annual runoff | 436,000 m³ | 390,000 m³ |

→ Five extra hours buys decision time for pump activation, drainage-gate operation, and resident evacuation.

Field deployment scenarios

  1. Real-time 6-hour forecast — rain observation → LSTM inference → Kakao alert
  2. Ex-ante LID evaluation — compare scenarios with vs without vegetated swales / ponds / permeable pavement
  3. County-scale expansion — integrate KMA AWS + Ministry of Environment hydrological monitoring data
  4. Training material — connects to the advisory-policy-education solution for rural-leader curricula

Typical deliverables

  • Trained LSTM model (.pt) for your village / micro-catchment
  • Operational alert scripts (02_train_lstm.py, 03_evaluate.py, etc.)
  • Quantitative NSE / peak-reduction report
  • Optional educational materials for farmers and rural leaders

Pilot-site partners welcome

We are actively recruiting pilot sites for this solution. If you have a rural micro-catchment (5–50 ha recommended) with rainfall and runoff observations, please contact us.

Solutions