API

API Update: ANN Bias Correction For Temperature Forecasts

We have integrated an Artificial Neural Network (ANN) into our API to improve temperature forecasts.

API Update: ANN Bias Correction For Temperature Forecasts
Simplified Artificial Neural Network (ANN) diagram

As of February 21st - all temperature values returned from the API include updated model bias corrections using an Artificial Neural Network (ANN).

This methodology allows for much more accurate near term (1-3 day) temperature forecasts, especially in areas which models persistently forecast either too warm, or too cold temperatures (bias).

A 48th hour forecast bias field - warm colors indicate areas where model forecast was too warm (warm bias), and cool areas indicate areas where model forecast was too cold (cold bias). Black = No/minimal bias

In most areas, forecast accuracy improvements are small (1-2 degrees F). However, in some areas with more obvious model biases - forecast improvements are dramatic. Such as Phoenix, Arizona (USA), and Miami, Florida (USA).

Average low temperature forecast improvements of ~3 Degrees Fahrenheit for Phoenix AZ (USA) 
Average low temperature forecast improvements of ~2.5 Degrees Fahrenheit for Miami FL (USA)

Although, notably - the improvements are smaller in areas with little to no obvious model bias (See Raleigh, NC and Boston, MA).

Average high temperature forecast improvements of ~1 Degree Fahrenheit for Raleigh, NC (USA)
Average low temperature forecast improvements of ~1 Degree Fahrenheit for Boston, MA (USA)

These bias corrections impact any temperature field returned by the Weather Forecast API.

As always we are open to any feedback our users may have about the API. For any questions about the API - please contact support@weatherbit.io or contact us via our secure message form.

Thank you for using our API.

The Weatherbit API Team

Read next