RECLAIM: End-to-End Reservoir Sedimentation ML Pipeline

RECLAIM is an end-to-end Machine Learning deployment pipeline designed to automate the estimation of reservoir capacity loss. The system orchestrates raw GIS data ingestion, automated feature engineering, and high-fidelity inference to serve actionable hydrological insights.

---

🏗️ Pipeline Architecture

The RECLAIM pipeline is divided into four distinct phases:

phase 1: Data Ingestion & Inflow

• GIS Ingestion: Support for multi-source uploads including GeoJSON catchment boundaries and reservoir polygons.
• Time-Series Inflow: Integrated processing of meteorological CSVs (Precipitation, Temperature, Wind) and reservoir operational data (AEC curves, Inflow/Outflow).
• Sample Data Pipeline: Pre-validated datasets for rapid benchmarking and automated form populating.

Phase 2: Automated Feature Engineering

Located in the reclaim/ core, this stage transforms raw spatial and temporal data:

• Spatial Alignment: Automated CRS re-projection to EPSG:4326 for global consistency.
• Feature Generation: Extraction of catchment-specific attributes and environmental variables from global NetCDF datasets (Land cover, Terrain, Soil types).
• Session-Based Storage: Isolated data preprocessing in unique job directories to ensure multi-user concurrency.

Phase 3: Model Inference & Synthesis

• Predictive Engine: Execution of the RECLAIM sedimentation model using calculated features and user parameters.
• Trend Synthesis: Interpolation of sedimentation rates across selected observation periods (1900–Current).
• Data Validation: Schema enforcement for meteorological and operational files to ensure inference stability.

Phase 4: Serving & Visualization (The Interface)

• Interactive Serving: A Django-driven REST/Web interface for model triggering.
• GIS Visualization: Real-time Leaflet.js rendering of processed catchment and reservoir geometries.
• Insight Delivery: Professional Plotly visualization of time-series forecasts and metadata synthesis.
• Export Layer: Automated CSV generation of predicted time-series for downstream MLOps consumption.

---

🚀 Deployment Guide

Environment Initialization

git clone https://github.com/SanchitMinocha/RECLAIM-Web-Application.git
cd RECLAIM-Web-Application
python3 -m venv .reclaim_webapp_env
source .reclaim_webapp_env/bin/activate
pip install -r requirements.txt

Serving the Pipeline

1. Apply Migrations: python manage.py migrate
2. Launch Server: python manage.py runserver
3. Admin Monitoring: Access /admin to monitor model metadata and trigger system-wide cleanup.

---

🛠️ System Configuration

Storage Management

The pipeline includes an Automated Cleanup Routine to manage heavy GIS files:

• Toggle: Controls available via FileCleanupSettings in the Admin Dashboard.
• Mechanism: Auto-purges job-specific directories in uploaded_files/ post-inference.

Heavy Data Optimization

Large global datasets (NetCDF/Heavy GeoJSON) are handled outside the main Git tree to keep the deployment package optimized. Refer to .gitignore for data exclusion rules.

---

📊 Pipeline Components

Component	Responsibility	Technology
Ingestion Engine	Geo-spatial & CSV file handling	geopandas, django-fs
Feature Generator	NetCDF data extraction & engineering	reclaim.gf, numpy
Inference Layer	Core estimation logic	reclaim.Reclaim
Service UI	Glassmorphic dashboard & serving	vanilla-js, leaflet

Lead Developer: Joshua Zhao
Project Lead & Mentor: Sanchit Minocha