The Weather App I Built for Myself
If you live in Oklahoma, you already know that the information about a storm is out there. The problem is that it's scattered across a dozen places, run by different agencies with different jobs. The Storm Prediction Center issues the convective outlooks and collects the storm reports. The local National Weather Service forecast offices issue the warnings, and the one in Norman launches the weather balloons. The radar is its own network. The satellites belong to yet another part of NOAA. The forecast models come from somewhere else again. The state runs its own network of weather stations. And private companies each take a slice of it and package it their own way. There are political and commercial reasons it ended up like that, and I don't think anyone did anything wrong. But as someone who just wants to understand the sky over his house, it's all disparate. I wanted one application that brings it together. It didn't exist, so I built it for myself.
One map, many sources
The weather app on luttrell.ai pulls from each of those sources and puts them on a single map of central Oklahoma.
From the Storm Prediction Center it takes the day one, two, and three convective outlooks, including the tornado, wind, and hail probabilities, and every storm report. From the National Weather Service it takes the active tornado, severe thunderstorm, and flash flood warnings and watches from every office that covers Oklahoma. From the Norman office it takes the balloon soundings, twice a day and on special launches.
It reads the raw data from the Twin Lakes radar near Oklahoma City and computes the products itself: reflectivity, storm-relative velocity using the storm motion measured by the latest balloon, correlation coefficient (where debris lofted by a tornado shows up as a hole), and echo tops, which estimate how high the peaks of the storms reach. Satellite imagery comes from the GOES-19 satellite, infrared around the clock and visible during the day.
What's about to happen
For the future, the app uses the HRRR, the High-Resolution Rapid Refresh model, which is rerun every hour. You can scrub forward through the hours and watch where the model thinks storms will form.
Because one model run can be wrong, the app also compares the last six runs and shows where they agree. It computes the significant tornado parameter from the model fields and breaks it into its ingredients, so you can see which one is holding it down. And you can click anywhere on the map to get a model sounding for that exact spot, then flip over to what the real balloon measured.
What's happening right now
On a live day, the same map is a real-time snapshot. The newest radar scan, the current satellite image, the warnings in force, and the reports as they come in are all layered together. One timeline runs across the whole convective day, from 7 AM to 7 AM, with the outlook updates and warnings marked on it.
What actually happened
This is the part I'm most excited about. Every past day can be replayed. The storm reports appear on the map at the moment they happened, and each warning shows how many reports fell inside it. Any day with a tornado report in Oklahoma is kept permanently.
That makes it possible to do a real retrospective on a storm. You can check how accurate the outlook was, and how early the model saw it coming. You can go back and think, that hook echo really was prominent, or it didn't go where we thought it would. You can learn from what actually happened instead of from memory.
On September 15, for example, a line of storms came through northwest Oklahoma in the late afternoon, with wind gusts reported in the sixties and low seventies near Woodward and Alva. I can scrub through that evening, compare the radar to what the model expected an hour or two earlier, and see exactly where the reports landed.
Waiting for spring
It's quiet now. Fall in Oklahoma rarely tests a storm map. I can't wait to try it out next spring, when the bad storms come back.
That's the real point for me. This is exactly the weather product I wanted, and it didn't exist until I built it. No company was going to make it just for the view from my house, and no agency was going to pull everyone else's data into one place. It's one more example of what becomes possible when you build software for yourself.