Original analysis
Forecast accuracy research
Four in-depth reports use ModelSpread’s forecast archive to examine temperature accuracy, NBM range width, daily forecast changes, and precipitation probabilities. Each article connects the measured result to a question a forecast user may actually face.
How accurate are temperature forecasts?
A two-year analysis by forecast day, season, region, daytime high, and overnight low.
Read the analysisUncertaintyDoes a wider NBM range warn of larger errors?
Tests how the width of the NBM temperature range relates to routine and large forecast misses.
Read the analysisRevisionsHow forecasts change as the day approaches
Tracks daily revisions and tests what a large late change says about the error that remains.
Read the analysisPrecipitationDo precipitation probabilities match observed frequencies?
Tests the chance of more than 0.25 inches and the accuracy of 24-hour amount ranges.
Read the analysisData used in these reports
The temperature reports use archived raw National Blend of Models percentile forecasts and observed station highs and lows. The precipitation report uses radar- and gauge-based gridded precipitation estimates. Results are not comparisons of weather apps or scores of the final official National Weather Service forecast.
Product construction and calibration validation are documented separately in Methodology. Download the local accuracy dataset or the research metrics with source hashes.