Model Comparison Heatmap Atlas

How to Read This Atlas

What this atlas shows

This page compares how different weather models forecast precipitation for the same four locations, side by side. It is a model-comparison and quality-check tool built for watching model behavior over time — not a precipitation forecast in its own right. Each image below is a heatmap: a grid of colored cells where color encodes a forecast value, letting patterns across many forecast runs and days show up at a glance.

QPF valid-vs-init heatmaps (GFS, IFS, NBM, ICON)

“QPF” is Quantitative Precipitation Forecast — a model’s predicted rainfall amount. Each heatmap has two time axes: one is the forecast’s initialization time (when that model run started, i.e. how many days ahead of time the forecast was made), and the other is the valid date (the actual calendar day being forecast for). Reading across a row shows how the forecast for one specific day changed as it got closer — whether a model was consistent about a coming rain event from run to run, or whether its guess for that day swung around a lot before settling down.

Bilinear vs. nearest interpolation

Weather models compute forecasts on a grid of points spaced miles apart, not at exact addresses. To estimate a value at one specific spot, this pipeline uses one of two methods: nearest simply takes the value from the single closest grid point, while bilinear blends the four surrounding grid points, weighted by distance, for a smoothed estimate. In steep or mixed terrain — like the transition from desert floor to forested peaks around these sites — the two methods can disagree noticeably. Showing both side by side is a way to see how sensitive a forecast is to that choice, not a signal that one method is “right” and the other “wrong.” ICON heatmaps use a single extraction method rather than both.

NBM PoP06 heatmaps

“PoP06” is Probability of Precipitation over a 6-hour window — NOAA’s National Blend of Models’ percentage likelihood that measurable rain falls in that window. These heatmaps roll several 6-hour windows into one daily value (using a 7 AM local cutoff, matching a full local day) in three ways: Daily Mean averages the probability across the day’s windows; Daily Max shows the single highest window probability that day; and Persistence Ratio (mean ÷ max) shows how consistent that probability was across the day — close to 1 means the chance of rain stayed fairly steady all day, while a low ratio means only part of the day carried real risk while the rest looked dry.