Beyond Markers: a new approach to interactive realtime weather maps

Markers must go! At least when things concern continuous-space processes such as weather. Could anyone care less about a sensor with its exact measurement unless it is mounted exactly where the person is? Why not move beyond markers and point measurements and instead show values everywhere as surface overlays? Markers are interactive and can link to a pop-up timeline - but so are the surface overlays! This is now possible thanks to the availability of fast spatial predictors, efficient storage mechanisms and fast client side rendering.

This talk will outline an open source web application for realtime modelling and visualisation of time-varying continuous-space fields. We will discuss the technical challenges of building this application from the server side perspective of data collection, modelling and storage and from the client side perspective of visualisation and interactivity.

We will describe our approach to an integrated spatio-temporal modelling and visualization of temperatures and rainfall. The predicted temperature surfaces and rain radar data are displayed as overlays in OpenLayers using a custom layer which renders the data on the client side using the HTML5 Canvas Element. The timeline for arbitrary spatial location is built upon the Flot plotting library and is rendered the same way. These elements are backed by a server side raster cube, implemented as a georeferenced multidemensional memory mapped Numpy array. We developed this data structure specifically for storing timestamped raster data in a way that allows for very efficient spatial and temporal slicing. And for those curious ones interested in sub-raster resolutions there is always a live server-side stream oracle ready to compute predictions with MapReduce on multicore.