| CPC G01W 1/10 (2013.01) [G06N 3/045 (2023.01); G06N 3/08 (2013.01); G01W 1/00 (2013.01); G01W 2001/006 (2013.01); G01W 1/18 (2013.01); G01W 2203/00 (2013.01); G06F 11/00 (2013.01); G06N 3/02 (2013.01); G06N 3/044 (2023.01); G06N 3/088 (2013.01); G06N 20/00 (2019.01); Y02A 90/10 (2018.01)] | 20 Claims |
|
1. A computer-implemented method for forecasting weather, the method comprising:
training a machine learning model comprising:
generating a hindcast of a first historical weather state for a selected geographical region and a selected timeframe;
generating a reforecast of a second historical weather state for the selected geographical region, using a weather forecast model and a set of historical conditions prior to the selected timeframe; and
training the machine learning model to estimate a predicted forecasting error in the weather forecast model, given the first hindcast and the reforecast as inputs;
using the trained machine learning model to determine the predicted forecasting error in the weather forecast model, wherein the machine learning model is configured to determine the predicted forecasting error given a weather forecast of the weather forecast model and a set of real-time metocean conditions; and
adjusting the weather forecast model using the predicted forecasting error to produce an augmented weather forecast.
|