Project Proposal · DOST PAGASA · Numerical Weather Prediction

Modernizing PAGASA's operational forecast visualization

A browser-native forecast viewer that ends the old tradeoff — fast static maps or slow interactive ones — by being both at once: fully dynamic, yet instant, loading only the data each view needs.


Model WRF · 12 km + 3 km nest Data loaded on demand Viewer server none required Runs on PAGASA hardware

The tradeoff this ends

Operational forecast visualization has always forced a choice. Static pre-rendered maps open instantly but are frozen — a fixed set of pictures you can only look at. A dynamic server-driven viewer is interactive but slow: it needs a dedicated viewer server, moves whole model-output files, and makes forecasters wait. The modernized WRF Dynamic Viewer ends that tradeoff. It is dynamic and fast at the same time — a forecaster explores any product, hour, and location freely, yet each view is near-instant because the viewer loads only the sliver of data it needs for what's on screen, not an entire model file, and renders it right in the browser. And there is no separate viewer server to run: the work moves to the forecaster's own machine, so PAGASA operates less infrastructure, not more.

Why it matters operationally: during a landfalling typhoon, a duty forecaster interrogates the run directly — any field, any hour, any accumulation window, zoomed to any coast — without waiting on a server to render it or pre-generate the one image they happen to need.

Three ways to show a forecast

Static images
Server-rendered viewer
WRF Dynamic Viewer
Interactivity
Look-only
Interactive
Fully interactive
Speed to view
Instant
Slow · server round-trip
Near-instant
Data moved per view
Pre-baked image
Whole model files
Only what's on screen
Separate viewer server
None
Required
None
Renders
Ahead of time
On a server
In the browser

Static is fast but frozen. Server-rendered is dynamic but heavy and slow. The Dynamic Viewer takes the strengths of both columns and none of the costs.

Svelte front-end WebGL field layer PMTiles vector + satellite basemaps Zarr data loaded on demand On-prem, no viewer server

What it looks like — Typhoon Bavi

Live · 48-hour forecast
Fig 110 m wind · Typhoon Bavi · d01

The 48-hour wind forecast, animated over a satellite basemap. Flow particles trace the circulation while the storm tracks toward Luzon — this is one field over time, streamed frame-by-frame from the model, not a pre-made movie. The forecaster can pause on any hour, switch units, or zoom to any coast.

Animated
Fig 2Simulated reflectivity · Bavi

Model-derived radar reflectivity on the dBZ scale, animating the storm's rainbands wrapping and sweeping across the forecast. Convective structure that a single still can't convey — initiation, organization, and landfall — reads at a glance in motion.

Animated
Fig 3Hourly precipitation · Bavi

Hourly rainfall rate, animated as the typhoon's rainbands rotate over the archipelago. Any accumulation window — 1 h to 72 h, or a custom span — is a live control, computed on the fly from the cumulative field.

Animated
Fig 42 m temperature · Bavi

Surface temperature across the same run, recolored live in the browser from the same grid the other products draw on. Choosing a different product doesn't fetch a new picture — it re-reads the data already on hand.