Traditional weather prediction tells us what is likely to happen within the next 24 hours and up to two weeks ahead, whereas climate prediction tells us what will likely happen in the coming seasons. Until recently, there was a gap between the weather and climate predictions leaving us unsure about what may happen between two weeks to two months from now. This time window between weather (up to 2 weeks) and climate (a season and longer) is known as “subseasonal.”
The forecasting effort targeting this subseasonal time window and bridging the gap between the weather and climate predictions is known as subseasonal-to-seasonal (S2S) prediction.
Having more accurate information about the Earth system (atmosphere, ocean, land, and sea ice) in this intermediate future would be of tremendous value to society.
S2S prediction is a new but rapidly developing endeavor. It involves understanding the extent to which the Earth system can be predicted (predictability), optimizing the use of available observations and filling observational blind spots, and improving prediction tools, mostly giant computer codes known as numerical predication models.
A special collection, entitled Bridging Weather and Climate: Subseasonal-to-Seasonal (S2S) Prediction, has gathered together articles published in Journal of Geophysical Research: Atmospheres and Geophysical Research Letters, with the latest insights. Here are examples of some of their collective outcomes.
There are various known sources of S2S predictability, including the El Niño–Southern Oscillation (ENSO) in the tropical Pacific, the Madden–Julian Oscillation (MJO) in the tropical Indian and Pacific Oceans, Quasi-biennial Oscillation (QBO) in the stratosphere, polar vortex, sea ice, and soil moisture. The articles in the collection explore their influences on S2S prediction skills, physical processes governing their influences, and the capability of prediction models to capture these influences.
Prediction of many phenomena are subject to these influences [Garfinkel et al., 2018; Jucker and Reichler, 2018; Wang et al., 2018; Albers and Newman, 2019; Butler et al., 2019; Cox et al., 2019; Jenney et al., 2019a; Jenney et al., 2019b; Jia et al., 2019; Lee et al., 2019; Quinting and Vitar, 2019; Zampieri et al., 2019; Zheng and Chang, 2019; Stone et al., 2019; Minami and Takaya, 2020; Rao et al., 2020; Son et al., 2020; Toms et al., 2020], including high-impact events, such as tropical cyclones, tornados, heat waves, and lightning, are foci of some studies (Baggett et al., 2018; Chang and Wang, 2018; Pasquier et al., 2019; Tippett and Koshak, 2018; Yang et al., 2018; Gao et al., 2019; Gensini et al., 2019; Miller and Wang, 2019; Wulff and Domeisen, 2019).
It is clear from the results of this collection that the sources of predictability should not be studied in isolation. It is their combined effect and their interactions with other slowly-varying phenomena (sea surface temperature, the mean flow, monsoons, blocking events) that determines the predictability and prediction skills of many phenomena [Karpechko et al., 2018; Xiang et al., 2019; Barnes et al., 2019; Byrne et al., 2019; Dias and Kiladis, 2019; Feng and Lin, 2019; Garfinkel et al., 2019; Hagos et al., 2019; Karmakar and Misra, 2019; Lee et al. 2019; Zheng et al., 2019; Rao et al. 2019; Gadouali et al., 2020]. It remains a challenge to study both individual and combined effects of the sources of S2S predictability in a systematic and comprehensive way.
Conventional understanding is that as the prediction lead time increases beyond the traditional weather timescale (2 weeks), influences on prediction skills from initial conditions would decrease and yield to influences from models’ capability of representing key physical processes. However, a number of articles in the collection point out that prediction skills sensitively depend on the quality of initial conditions (in soil moisture, ground snow, sea ice, water vapor, stratospheric conditions, the MJO) even on the S2S time scales [Dirmeyer et al., 2018; Zampieri et al., 2018; Choi and Son, 2019; DeFlorio et al., 2019; Li et al., 2019; Wang et al., 2019; Wei et al., 2019; Wu et al., 2019; Martin et. al., 2020; Molod et al., 2020]. This finding calls for more attention to observations and their input to prediction models via data assimilation. This also points out a main caveat of using global reanalysis products to initialize models without their own data assimilation capabilities in studies of S2S prediction.
Articles in the collection illustrate that measuring S2S predictions skills highly depend on the metrics used [Wang et al., 2018; Kim et al., 2019]. There are a large variety of variables, their proxies, and indices that can be selected to measure prediction skills. One metric may demonstrate how good S2S predictions might be (skillful up to 45 days) and another one how poor they are (less than 10 days) for the same model. This arbitrary factor is worsened by subjective choices of measuring “skillful” or “useful” predictions (e.g., 0.5 correlation between prediction and observations) and by low standards (e.g., against persistent prediction or climatology). Ultimately useful prediction should be measured with the perspectives of their end users in consideration.
This collection of articles is just the tip of the iceberg in the study of S2S prediction. It’s a valuable set of research papers that presents new knowledge and tools, and identifies many interesting directions for further research.