Subseasonal forecasting and MJO teleconnections in machine learning weather prediction models

2025-07-09 · 1 citations

In recent years, machine-learning (ML) models trained on reanalysis data have equaled physics-based forecast models in terms of performance skill for global weather forecasting. With increased rollout stability, the question of how these models perform for subseasonal to seasonal (S2S, week 3 to 8) forecasting has emerged. In this study we run a large set of subseasonal hindcasts over 2004-2023 to evaluate two ML weather forecast models at the S2S time scale, SFNO-HENS (Nvidia, fully ML) and NeuralGCM (Google Research, hybrid). Corresponding hindcasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) are used as a baseline for comparison to a physics-based model. Because our focus is on predicting moisture transport over the Western United states between October and March, we evaluate the models’ prediction skill for the Madden-Julian Oscillation (MJO) and its associated teleconnections in the North Pacific. We find that both ML models are competitive with the ECWMF model, with comparable skill in predicting the North Pacific large-scale circulation and the MJO at week 3 and beyond. Even though overall the mid-latitude subseasonal prediction skill remains low, the ML models exhibit interesting behavior such as a realistic propagation of the MJO across the Maritime Continent (especially for SFNO-HENS) and realistic teleconnections when compared to ECMWF. A SFNO-HENS sensitivity experiment with altered initial conditions in the tropics demonstrates the stability of the model, and it illustrates the capability of ML models to represent important physical processes of the atmosphere at the S2S time scale.

Seasonality in the Predictive Skill of Southwest United States Precipitation in the Seasonal Forecast Systems

Geophysical Research Letters · 2025-06-19 · 2 citations

Abstract We examine seasonality in the skill of 8 seasonal forecast systems to predict the Southwest United States (SWUS) precipitation between October and April. The models share similarities in the large seasonality they exhibit in their prediction skill over the 1993–2016 period: from October to December, the skill is low if not non‐existent, before it gets higher in all the models, with a peak in February‐March in the model with an extended hindcast period (European Centre for Medium‐Range Weather Forecasts, ECMWF, 1981–2024 period). Using ECMWF, further analyses of daily precipitation illustrate how the model is able to better predict the SWUS precipitation over the February‐March forecast target window, versus November‐December. Higher precipitation in early February compared to early November is a contributing factor, as is the annual cycle in tropical eastern Pacific sea surface temperature, and the associated El Niño Southern Oscillation teleconnection with the extratropics.

We need to simulate more northern ITCZs and less southern ITCZs over the east Pacific Ocean in coupled climate models

2025-11-18 · 1 citations

Tropical precipitation biases have persisted since the very first generations of climate models. These biases are highly sensitive to the region and/or season of interest, with commonalities in the east Pacific and Atlantic Ocean basins, possibly due to their similar observed climatological northern hemisphere intertropical convergence zone (ITCZ). However, the colloquial name "double ITCZ bias" comes from a time and/or zonal mean, and may be missing important information about errors at smaller time and space scales. In this study, we explore daily characteristics of the ITCZ in observations, reanalyses, and 25 Coupled Model Intercomparison Project 6 (CMIP6) models over the east Pacific Ocean. We devise and apply an algorithm that determines a region's dominant daily ITCZ configuration, its "ITCZ state," based on the daily mean precipitation field. The five ITCZ states include: northern hemisphere (nITCZ), southern hemisphere (sITCZ), double (dITCZ), equatorial (eITCZ), and absent (aITCZ). We find that nearly all CMIP6 models gravely underestimate nITCZs and overestimate sITCZs during January through May, in contrast with what "double ITCZ bias" suggests. Surprisingly, all reanalyses also underestimate nITCZs and overestimate eITCZs. Errors in ITCZ state interannual variability are consistent with mean errors in reanalyses, while sITCZ interannual variability is far too low relative to the mean in most CMIP6 models. Lastly, all reanalyses and CMIP6 models overestimate precipitation rates in the southern hemisphere ITCZ band for dITCZs and sITCZs, suggestive of errors with atmospheric origins.

We need to simulate more double ITCZs and less southern ITCZs in reanalyses and coupled climate models

2025-02-21 · 1 citations

Model biases in tropical precipitation have persisted for nearly 30 years. These biases are highly sensitive to the region and/or season of interest, with commonalities in the east Pacific and Atlantic Ocean basins, possibly due to their similar observed climatological northern hemisphere intertropical convergence zone (ITCZ). However, the colloquial name ”double ITCZ bias” comes from a time and/or zonal mean, and many studies do not acknowledge that these model biases are most prominent during boreal winter and spring when models overestimate a southern hemisphere ITCZ, not a double ITCZ. In this study, we explore high-resolution characteristics of the ITCZ in observations, reanalyses, and 25 Coupled Model Intercomparison Project 6 (CMIP6) models over the east Pacific Ocean. We devise and apply an algorithm that determines a region’s dominant daily ITCZ configuration, its ”ITCZ state,” based on the daily mean precipitation field. The five ITCZ states include: northern hemisphere (nITCZ), southern hemisphere (sITCZ), double (dITCZ), equatorial (eITCZ), and absent (aITCZ). We find that nearly all CMIP6 models gravely underestimate dITCZs and nITCZs and overestimate sITCZs during January through May, in contrast with what ”double ITCZ bias” suggests. Surprisingly, all reanalyses also underestimate dITCZs and overestimate sITCZs. Errors in ITCZ state interannual variability are consistent with mean errors in reanalyses, while sITCZ interannual variability is far too low relative to the mean in most CMIP6 models. Lastly, all reanalyses and the seven CMIP6 models that can produce dITCZs overestimate precipitation rates in dITCZs, while nearly all CMIP6 models overestimate precipitation rates in sITCZs.

We Need to Simulate More Northern ITCZs and Less Southern ITCZs Over the East Pacific Ocean in Coupled Climate Models

Journal of Geophysical Research Atmospheres · 2025-12-17

Abstract Tropical precipitation biases have persisted since the very first generations of climate models. These biases are highly sensitive to the region and/or season of interest, with commonalities in the east Pacific and Atlantic Ocean basins, possibly due to their similar observed climatological northern hemisphere intertropical convergence zone (ITCZ). However, the colloquial name “double ITCZ bias” comes from a time and/or zonal mean, and may be missing important information about errors at smaller time and space scales. In this study, we explore daily characteristics of the ITCZ in observations, reanalyses, and 25 Coupled Model Intercomparison Project 6 (CMIP6) models over the east Pacific Ocean. We devise and apply an algorithm that determines a region's dominant daily ITCZ configuration, its “ITCZ state,” based on the daily mean precipitation field. The five ITCZ states include: northern hemisphere (nITCZ), southern hemisphere (sITCZ), double (dITCZ), equatorial (eITCZ), and absent (aITCZ). We find that nearly all CMIP6 models gravely underestimate nITCZs and overestimate sITCZs during January through May, in contrast with what “double ITCZ bias” suggests. Surprisingly, all reanalyses also underestimate nITCZs and overestimate eITCZs. Errors in ITCZ state interannual variability are consistent with mean errors in reanalyses, while sITCZ interannual variability is far too low relative to the mean in most CMIP6 models. Lastly, all reanalyses and CMIP6 models overestimate precipitation rates in the southern hemisphere ITCZ band for dITCZs and sITCZs, suggestive of errors with atmospheric origins.

Thank You to Our 2024 Reviewers

Geophysical Research Letters · 2025-02-22

Abstract On behalf of the journal, AGU, and the scientific community, the editors of Geophysical Research Letters would like to sincerely thank those who reviewed manuscripts in 2024. The hours reading and commenting on manuscripts not only improve the manuscripts but also increase the scientific rigor of future research in the field. With the advent of AGU's data policy, many reviewers have also helped immensely to evaluate the accessibility and availability of data, and many have provided insightful comments that helped to improve the data presentation and quality. We greatly appreciate the assistance of the reviewers in advancing open science, which is a key objective of AGU's data policy. We particularly appreciate the timely reviews in light of the demands imposed by the rapid review process at Geophysical Research Letters. We received 5,225 submissions in 2024, and 5,597 reviewers contributed to their evaluation by providing 9,697 reviews in total. We deeply appreciate their contributions. We would also like to acknowledge the passing of our beloved colleague, Harihar Rajaram. An AGU Fellow and longtime affiliate of AGU's Hydrology Section, Hari was the Editor‐in‐Chief of Geophysical Research Letters since 2019, a former editor on Water Resources Research, and served on the AGU Publications Committee.

Decadal Variability of the MJO and Implications for Southwestern United States Wintertime Precipitation Predictability

Geophysical Research Letters · 2025-04-02 · 2 citations

Abstract The Madden‐Julian Oscillation (MJO) is the dominant source of tropical convective activity on intraseasonal timescales and a significant influence on extratropical weather through remote teleconnections. Here, we investigate decadal variability of the MJO and its boreal winter teleconnections with the Southwestern United States (SWUS) using an ensemble of historical climate simulations, where tropical atmospheric variability is nudged toward reanalysis. Since the beginning of the 21st century, the MJO has been associated with a wavenumber‐5 zonal teleconnection pattern that propagates along the jet stream waveguide. This coupling is absent in the prior two decades, causing significant differences in the Southwestern United States precipitation response. These changes also result in vastly different windows of opportunity for S2S predictability during each period. This decadal variability is potentially associated with the change from a positive to negative Interdecadal Pacific Oscillation, as well as the global warming signal.

Contrasting Arctic Amplification Response in the Community Earth System Model Large Ensembles and Implications for the North Atlantic Region

Journal of Geophysical Research Atmospheres · 2025-03-22 · 3 citations

Abstract The response of the polar jet to climate warming and rapid Arctic change is a leading uncertainty in climate projections and critical to the future of mid‐latitude surface weather. Previous studies suggest that CMIP5‐6 model projections fall into two groups of either Arctic‐ or tropically‐driven climate change, especially in the North Atlantic. Here, we present distinct warming patterns emerging by the late 21st century between the first two generations of the Community Earth System Model Large Ensemble (CESM‐LE) and use daily diagnostics to assess associated changes in mid‐latitude circulation. We show that the subsequent versions of CESM represent categorically different storylines of North Atlantic climate change. The first version of CESM‐LE (CESM1‐LE, hereafter LENS1) exhibits severe Arctic amplification (AA) along with minor reductions in jet waviness. In contrast, CESM2‐LE (hereafter LENS2) presents subdued AA, a more pronounced North Atlantic warming hole, and a late‐century climate dominated by upper‐tropospheric tropical warming. Uniquely, in LENS2 during winter, the North Atlantic sector projects less warming in the Arctic than in the mid‐latitude mid‐troposphere. The projected North Atlantic jet is reinforced and poleward‐shifted with reduced sinuosity, blocking, and synoptic variability. The surface weather response includes greater precipitation over northern Europe, more intense drying in the eastern Mediterranean, and a lesser decline in cold extremes by late century compared to LENS1.