Our interactive dashboard helps in dissemination of information regarding near-real time prediction of water availability in four dams of the Upper Cauvery river basin that supply water to Bengaluru City. We have adopted a physically described process-based approach that lends information to a machine learning model to come about with a 30, 60 and 90 day prediction of storage capacity and inflows. Prediction of water storage and inflow to a reservoir helps in deciding outflows from the dam to fulfil different water demands. This will be useful for the public and various stakeholders in making scientifically informed decisions.
- Python scripts used to extract data from satellite and reanalysis data
- Outputs of the Hydrological Model
- Outputs of the AI/ML Models for storage and inflow
- Dashboard frontend
- SWAT (Soil and Water Assesment Tool), a semi-distributed model has been used to understand the physical processes at play upstream of the four dams (Hemavathi, Harangi, Kabini and KRS)
- The model folder and files can be accessed here
- For more details on hydrological modeling, refer to the technical documentation
- Three different model architectures were explored : Fully Connected Dense Neural Networks, Recurrent Neural Networks and WaveNets.
- After much experimentation, we decided on Recurrent Neural Networks - specifically the Long-Short Term Memory architecture.
- Monte-Carlo Ensemble Model has been adopted to convert our single deterministic predictions into probability distribution functions to find an uncertainty estimate.
- For more details on our AI/ML model, refer to the technical documentation.
- The dashboard has two views - overview and dataview.
- Time-series data related to historical and future predictions are shown.
- Model comparison of predictions vs. observed has been done for the year Jan 2020 to Dec 2020
The RESERVOIRWATCH Dashboard can be accessed at www.aqrity.com/reservoirwatch