3B.8 Calibration of Ensemble Climate Forecasts by Weighted Linear Regression

Friday, 28 July 2017: 3:15 PM
Constellation F (Hyatt Regency Baltimore)
David A. Unger, Innovim/CPC/NCEP/NWS, College Park, MD; and D. C. Collins

One method that the Climate Prediction Center uses to calibrate forecasts from global climate models (GCMs) is a simple linear regression designed for ensemble predictions. The regression is typically trained on several decades of retrospective forecasts (known as hindcasts) generated from the model. Changes in any aspect of the data assimilation system used by the GCM, or even changes in earth's climate over time may impact the performance of the GCM and its calibration.

The use of exponentially weighted linear regression is examined here to assess its potential to adapt model calibration to account for any changes in GCM performance over time. Forecasts of Nino 3.4 sea surface temperatures (SSTs) from the National Centers for Environmental Prediction (NCEP) Climate Forecast System version 2 model (CFSv2) will be used to assess calibration based on weighted regression in relation to one based on equal weighting. Exponentially weighted regression produces calibration statistics that weight recent data more heavily than data in the past, and therefore can adapt more quickly to performance changes than equally weighted regression. This comes at a cost of reducing the effective sample size available for calibration.

Hindcasts from the CFSv2 are available from 1982-2010 while data from recent years are obtained from forecasts collected in real time. Bias characteristics of Nino 3.4 SSTs from the CFSv2 hindcasts are known to have changed around 1999, while less is known about potential changes in performance between the hindcasts and real-time forecasts. CFSv2 data for Nino 3.4 SSTs will be examined to estimate impact of known and potential discontinuities in model performance, and the effective sample size on the two calibration methods.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner