Sepehr Karimi, Ph.D.

Sepehr Karimi, Ph.D.

Director of AI and IT Solutions

Multi-Source Precipitation Data Fusion Visualization Dashboard

Overview

I led the development of the Multi-Source Precipitation Data Fusion Visualization Dashboard, a web-based geospatial platform for exploring fused precipitation estimates generated from the Precipitation Data Fusion Network (PDFN). The dashboard turns a large scientific precipitation dataset into an interactive, browser-accessible tool for researchers, modelers, and decision-support users working on hydrology, flood forecasting, drought monitoring, land-surface modeling, and climate risk analysis.

The dashboard supports the dataset associated with Gavahi, Foroumandi, and Moradkhani (2023), A deep learning-based framework for multi-source precipitation fusion, Remote Sensing of Environment. The underlying PDFN model integrates multiple precipitation products using a deep learning architecture based on 3D-CNN and ConvLSTM layers to capture spatial and temporal precipitation dependencies and improve quantitative precipitation estimation.

Live dashboard: Multi-Source Precipitation Data Fusion
Source code: GitHub repository
Related dataset: Zenodo data record

My Role

I led the design and implementation of the visualization dashboard, including the client-side map interface, raster loading workflow, aggregation tools, geospatial selection tools, export workflow, static reference overlays, and deployment architecture for serving large precipitation GeoTIFF datasets through a web-accessible interface.

Scientific and Data Context

The platform visualizes daily fused precipitation estimates over the continental United States (CONUS). The public dataset is described as having:

  • Spatial resolution: 0.05 degrees
  • Coordinate reference system: WGS1984
  • Temporal coverage: January 1, 2015 to April 30, 2024
  • Data product: merged daily precipitation estimates generated using the PDFN framework

The dataset was designed to support applications such as land-surface modeling, flood forecasting, drought monitoring, drought prediction, and hydrologic analysis.

Key Dashboard Capabilities

Dynamic precipitation raster visualization

The dashboard allows users to load and visualize precipitation raster data directly on an interactive Leaflet map. Users can inspect precipitation fields spatially, switch units, and explore the dataset through an intuitive browser interface.

Key capabilities include:

  • Daily precipitation loading by selected date
  • Aggregated precipitation loading over a user-defined date range
  • Aggregation methods: Average, Maximum, and Total Sum
  • Dynamic legend updates based on precipitation magnitude
  • Unit switching between millimeters and inches
  • Hover-based map inspection for precipitation values and location context

Interactive geospatial selection tools

The platform supports multiple ways to define a region of interest for analysis and export:

  • Search by state, county, or place
  • Draw custom polygons
  • Draw rectangular selections
  • Upload custom GeoJSON boundaries
  • Upload zipped Shapefile boundaries
  • Clip exports to a selected area or export the full raster

Raster export workflow

The dashboard includes an export workflow that lets users download precipitation data for either the selected region or the entire raster.

Supported output formats:

  • GeoTIFF
  • CSV
  • NetCDF

This makes the dashboard useful not only as a visualization interface, but also as a practical data-access and preprocessing tool for downstream hydrologic modeling and research workflows.

Static geospatial reference overlays

The dashboard also includes a static view mode for displaying U.S. reference layers. Available overlays include:

  • U.S. states
  • U.S. counties
  • U.S. subdivisions
  • National forests
  • Bailey ecoprovinces
  • Major rivers
  • U.S. aquifers
  • HUC8 hydrologic units
  • HUC6 hydrologic units
  • National Park Service boundaries

These overlays help users interpret precipitation patterns in relation to administrative boundaries, hydrologic regions, ecological regions, and major physical geography.

Basemaps and user interface controls

The interface includes multiple basemap options and map controls to support different analysis contexts:

  • OpenStreetMap Standard
  • OSM Humanitarian
  • OpenTopoMap
  • CARTO Positron
  • CARTO DarkMatter
  • Esri World Street Map
  • Esri World Topographic
  • Esri World Imagery
  • Esri World Shaded Relief
  • Esri Light Gray Canvas
  • Esri Dark Gray Canvas
  • Esri National Geographic

The dashboard also includes fullscreen mode, a hide/show control panel, clear/reset controls, and a structured About/Daily/Aggregated workflow to guide users through the application.

Technical Implementation

The application is implemented as a lightweight, single-page geospatial web application. It does not require a frontend build system such as React, Vite, or Node.js. The dashboard relies on browser-based geospatial libraries and a separate raster data server.

Core technologies and libraries include:

  • JavaScript for the client-side dashboard logic
  • Leaflet for interactive web mapping
  • GeoRaster / GeoRasterLayer for browser-based GeoTIFF rendering
  • Turf.js for geospatial operations
  • Leaflet.draw for polygon and rectangle selection
  • shpjs for zipped Shapefile uploads
  • proj4 and related geospatial utilities
  • Python HTTP server for serving precipitation raster files
  • Cloudflare Tunnel for exposing local app and data servers when needed

The system separates the dashboard interface from the large raster data archive. The browser loads the map application and requests precipitation GeoTIFF files from a dedicated data server using a predictable path structure:

/tiff/YYYY/MM/pdfn_YYYY-MM-DD_v1.0.tif

This architecture keeps the web application lightweight while allowing large scientific raster datasets to be served and visualized on demand.

Impact

This dashboard makes a complex deep learning-based precipitation fusion product accessible to users who need fast spatial exploration, date-based analysis, regional clipping, and export-ready data. It bridges the gap between scientific model output and practical use by providing a polished interactive interface for hydrologic research, hazard analysis, and environmental decision support.

Citation

Gavahi, K., Foroumandi, E., and Moradkhani, H. (2023). A deep learning-based framework for multi-source precipitation fusion. Remote Sensing of Environment, 295, 113723. https://doi.org/10.1016/j.rse.2023.113723