Evolution of the North American Multi‐Model Ensemble

Abstract

The North American Multi-Model Ensemble (NMME) has become integral scientific infrastructure in subseasonal and seasonal climate prediction, advancing applied prediction and scientific understanding of the mechanisms and predictability of short-term climate. The NMME was designed to evolve as old models were retired and new models joined the ensemble. This study examines the assumption that prediction skill will increase as the system evolves, focusing on 2 m temperature, precipitation rate, and sea surface temperature prediction. The common period of 1982-2010 is studied for four configurations of the NMME, approximately representing the operational model suites of 2011, 2012, 2014-2018, and 2019-present. Substantial improvement in temperature prediction over both land and ocean is observed, with little change in global precipitation prediction. Sea surface temperature prediction at longer leads has improved over much of the globe, with the notable exception of the central-eastern tropical Pacific, where prediction skill has declined.Plain Language Summary Predicting the climate more than a few weeks in advance requires forecasters to bet on more than one computer model, similar to the method of diversifying an investment portfolio. Many earlier studies have shown that the combination of predictions generated by different computer models, known as a multi-model ensemble, almost always achieves better forecast skill than using a single model alone. The North American Multi-Model Ensemble (NMME), a subseasonal and seasonal prediction system combining individual North American state-of-the-art climate prediction models, has become an integral part of subseasonal and seasonal research and applications. The NMME has continually evolved, as newer models replace older ones; it is assumed that this evolution will produce more skillful predictions over time. But, until now, this assumption has not been tested. We examine the skill of NMME predictions in four different model combinations, including the oldest configuration, two transitional suites, and the operational configuration as of early 2020. Temperature prediction over both land and ocean has improved noticeably through the exchange of older models for newer ones, but precipitation prediction has not substantially improved. Overcoming the difficulty in precipitation prediction may require higher-resolution climate models in the NMME.

Southeastern Pacific Error Leads to Failed El Niño Forecasts

“…The available NMME data set covers the period from 1981 to 2019 in phase‐I and 1981 to 2012 in phase‐II models, and the spanning period varies by model. Following Table 1 of Becker et al (2020), the detailed description of each model including model name, valid periods, and lead times has been provided (Table S1 in the supporting information). The climatological monthly mean SST, as a function of both the initial month and lead time, are first calculated based on the entire period available for each model.…”

Seasonal to multi-year soil moisture drought forecasting

“…This is a measure of the potential predictability (referred to as predictability hereafter) in the perfect model world. Actual realized forecast skill is likely to be lower than the predictability due to model biases and uncertain initial conditions 43,49 .…”

“…A cyclic behavior in the forecast skill ( Fig. 2c) can be due to differences in skill across the annual cycle 43 because of the Nov1 initialization date ( Supplementary Fig. 4).…”

“…We also develop an observational constraint on the predictability estimates that can motivate future development of S2D prediction systems. Finally, we aim to bring fundamental advances by (a) developing the predictability estimates on the S2D timescale continuum instead of multi-year only 41,42 or seasonal only timescale 43 and (b) assessing hydroclimate predictability in the mid-latitude regions where the signal to noise ratio is known to be smaller than in equatorial regions 12,44 and therefore where predicting drought is more challenging 13 .…”

Avaliação da previsão sazonal de precipitação do projeto North America Multi-Model Ensemble (NMME) sobre o Brasil

“…Esse projeto é composto por modelos numéricos de diferentes centros de modelagem climática e teve início em 2010, contando com previsões atualizadas mensalmente desde agosto de 2011 até o presente. Além disso, a frequente atualização dos modelos também tende a gerar melhores resultados em relação as previsões (Becker et al, 2020).…”