Home
This Title All WIREs
WIREs RSS Feed
How to cite this WIREs title:
WIREs Clim Change
Impact Factor: 5.124

S2S reboot: An argument for greater inclusion of machine learning in subseasonal to seasonal forecasts

Full article on Wiley Online Library:   HTML PDF

Can't access this content? Tell your librarian.

The discipline of seasonal climate prediction began as an exercise in simple statistical techniques. However, today the large government forecast centers almost exclusively rely on complex fully coupled dynamical forecast systems for their subseasonal to seasonal (S2S) predictions while statistical techniques are mostly neglected and those techniques still in use have not been updated in decades. In this Opinion Article, we argue that new statistical techniques mostly developed outside the field of climate science, collectively referred to as machine learning, can be adopted by climate forecasters to increase the accuracy of S2S predictions. We present an example of where unsupervised learning demonstrates higher accuracy in a seasonal prediction than the state‐of‐the‐art dynamical systems. We also summarize some relevant machine learning methods that are most applicable to climate prediction. Finally, we show by comparing real‐time dynamical model forecasts with observations from winter 2017/2018 that dynamical model forecasts are almost entirely insensitive to polar vortex (PV) variability and the impact on sensible weather. Instead, statistical forecasts more accurately predicted the resultant sensible weather from a mid‐winter PV disruption than the dynamical forecasts. The important implication from the poor dynamical forecasts is that if Arctic change influences mid‐latitude weather through PV variability, then the ability of dynamical models to demonstrate the existence of such a pathway is compromised. We conclude by suggesting that S2S prediction will be most beneficial to the public by incorporating mixed or a hybrid of dynamical forecasts and updated statistical techniques such as machine learning. This article is categorized under: Climate Models and Modeling > Knowledge Generation with Models
Clusters of temperature winter (December, January, and February) anomalies for the European region, ordered by their frequency: Cluster (a) has a frequency of 32%, (b) a frequency of 20%, (c) a frequency of 18%, (d) a frequency of 14%, (e) a frequency of 8%, and (f) a frequency of 8%
[ Normal View | Magnified View ]
Temperature winter (December, January, and February) forecast skill: (a) cross‐validated correlation of the cluster‐based hindcasts with observations for the years 1967–2016. (b) Cross‐validated correlation of the canonical correlation analysis hindcasts with observations for the years 1967–2016. (c) Same (a) but for the years 1982–2010. (d) Correlation of the North American Multi‐Model Ensemble (NMME) hindcasts for the years 1982–2010 with observations. Significant values (p < .05) according to the two‐sided Student's t test are shown in hatches. The cluster‐based forecast performs better than the NMME models according to the cross‐validated correlations
[ Normal View | Magnified View ]
Predicted December, January, and February 2017/18 surface temperature anomalies from (a) North American Multi‐Model Ensemble (NMME) suite of models, (b) International Multi‐Model Ensemble (IMME) suite of models both initialized on November 1, 2017, (c) the observed surface temperature anomalies for December, January, and February 2017/18 and (d) same as (a) but for the AER statistical model initialized on November 8, 2017. Predicted January, February, and March 2018 surface temperature anomalies from the (e) NMME suite of models initialized on December 1, 2017, (f) IMME suite of models both initialized on November 1, 2017, (g) the observed surface temperature anomalies for January, February, and March 2018 and (h) same as (e) but for the AER statistical model initialized on December 1, 2017. Smoothing was applied to the statistical model and observed surface temperature anomalies
[ Normal View | Magnified View ]
Precipitation winter (December, January, and February) forecast skill: (a) cross‐validated correlation of the cluster‐based hindcasts with observations for the years 1967–2016. (b) Same as (a) but for the hindcast skill using canonical correlation analysis. (c) Same as (a) but for the years 1982–2010. (d) Correlation of the North American Multi‐Model Ensemble (NMME) hindcasts for the years 1982–2010 with observations. Significant values (p < .05) according to the two‐sided Student's t test are shown in hatches. The cluster‐based forecast performs better than the NMME models according to the cross‐validated correlations. (Reprinted with permission from Totz et al. (). Copyright 2017 American Geophysical Union)
[ Normal View | Magnified View ]
Predicted February 2018 surface temperature anomalies from (a) North American Multi‐Model Ensemble (NMME) suite of models, (b) International Multi‐Model Ensemble (IMME) suite of models both initialized on January 1, 2018, (c) the observed surface temperature anomalies for February 2018, and (d) same as (a) but for the AER statistical model initialized on November 8, 2017. Predicted March 2018 surface temperature anomalies from (e) NMME suite of models initialized on February 1, 2018, (f) IMME suite of models initialized on February 1, 2018, (g) the observed surface temperature anomalies for March 2018, and (h) same as (e) but for the AER statistical model initialized on December 1, 2017. Smoothing was applied to the statistical model and observed surface temperature anomalies
[ Normal View | Magnified View ]
(a) Regression of November sea level pressure (SLP) anomalies (hPa) onto October monthly mean Eurasian snow cover extent (contouring) and onto December meridional heat flux anomalies at 100 hPa, averaged between 40°N and 80°N (shading). This figure is the same as fig. 4 from Cohen, Furtado, et al. (). (b) Observed mean SLP (contours) and SLP anomalies (shading) for January 2018
[ Normal View | Magnified View ]
Correlation of Niño 3.4 index with surface temperatures across the Northern Hemisphere (contouring). Correlation of 90, 95 and 99% are represented by light, dark and darkest shading, respectively. Red shading represents positive correlations and blue shading represents negative correlations
[ Normal View | Magnified View ]

Browse by Topic

Climate Models and Modeling > Knowledge Generation with Models

Access to this WIREs title is by subscription only.

Recommend to Your
Librarian Now!

The latest WIREs articles in your inbox

Sign Up for Article Alerts