CW3E Publication Notice: Are AI weather models learning atmospheric physics? A sensitivity analysis of cyclone Xynthia

CW3E Publication Notice

Are AI weather models learning atmospheric physics? A sensitivity analysis of cyclone Xynthia

March 20, 2025

Scientists from the CW3E machine learning team recently published an article titled “Are AI weather models learning atmospheric physics? A sensitivity analysis of cyclone Xynthia” in npj Climate and Atmospheric Science. This study was led by Jorge Baño-Medina (CW3E) and co-authored by Agniv Sengupta (CW3E), James D. Doyle (U.S. Naval Research Laboratory), and Carolyn A. Reynolds (U.S. Naval Research Laboratory), Duncan Watson-Parris (SIO), and Luca Delle Monache (CW3E). This work aligns with CW3E’s goal to develop and leverage emerging artificial intelligence (AI) technologies to improve the prediction and advance our understanding of extreme weather events. The work was supported by the Office of Naval Research (ONR), California Department of Water Resources’ AR Program, and the U.S. Army Corps of Engineers’ Forecast Informed Reservoir Operations.

The central research question of this study is: Are Artificial Intelligence (AI) weather models learning atmospheric physics? Addressing this question is crucial to understanding how these models work and increasing our trust in their estimates. These tools can then be applied with confidence for the prediction of (extreme) weather, as well as for observation network design, probabilistic forecasting, and the study of atmospheric dynamics.

This study presents an innovative methodology to study the sensitivity of forecasts to initial condition uncertainty through AI global weather models, which have recently shown remarkable forecasting skill, rivaling traditional physics-based systems, while offering the advantage of much lower computational real-time demands (on the order of seconds vs hours). These models, typically trained to forecast weather up to 6 hours ahead, can also produce longer forecasts through auto-regression. The study focuses on Cyclone Xynthia (02/2010), an extreme weather event in Western Europe that resulted in significant casualties and damages estimated at several millions of dollars. Gradients of kinetic energy at 36 hours lead time with respect to atmospheric features at the initial time—commonly referred to as sensitivities—were computed leveraging the backpropagation algorithm of the AI model. To understand how well the AI model has learned the physical relationships between different variables, we use physics-based sensitivities derived from an adjoint model as a reference for comparison.

The results indicate that the AI-based sensitivities exhibit spatio-temporal patterns that are consistent with those from physics-based models (see Figure 1). Moreover, when perturbations based on these sensitivities are scaled according to estimates of initial condition uncertainty, the AI-derived response closely resembles that of the physics-based model.

These findings are significant because they demonstrate that AI weather models are not only capable of producing accurate forecasts very quickly but also of learning consistent spatio-temporal atmospheric links in the atmosphere. This has important practical implications for weather services, which traditionally rely on computationally intensive physical models to understand the weather patterns that lead to extreme events. The AI-based methodology presented here could complement or, in some cases, replace these costly physical models, enabling the rapid computation of sensitivities in a fraction of the time and with minimal computational resources. This increased efficiency could enhance our ability to forecast extreme weather events in real-time by generating large ensembles based on these sensitivities to better represent the tail of the distribution. These tools will also allow services to deploy observations in the identified sensitive areas, such as during the Atmospheric River Reconnaissance campaign (Ralph et al., 2020), a project endorsed by the World Weather Research Program (WWRP) for flight-track planning, to improve the initialization of the forecast. They will also be used to improve our understanding of the underlying physical processes.

Figure 1. Physics-based sensitivity fields of the kinetic energy over the Bay of Biscay (green box) at 36 hours of lead time for a) water vapor, b) 700-hPa potential temperature, and c) 700-hPa meridional wind, as computed in D14. AI-based sensitivities of the kinetic energy for d) integrated water vapor, e) 700-hPa air temperature, and c) 7-hPa meridional wind. The bottom row shows the 700-hPa geopotential sensitivity fields at g) 48 hours, h) 36 hours, and i) 24 hours of lead time. Contours in panel d), show the total column water vapor from 10 to 30 every 4 kg/m2 at initial time, February 26th 12 UTC. Similarly, in panel e) the contours show the values of temperature every 4K, while in panels g), h), and i), the 700-hPa geopotential is displayed every 50 m2/s2 from 2800 to 3200. Vectors are used to represent the wind intensity at 700 hPa in panel f). Gray shading indicates the panels representing the AI-based sensitivities at 36 hours of forecast lead time. The sensitivities are represented with color bars with different scales to visualize the spatial structures found across variables.

Baño-Medina, J., Sengupta, A., Doyle, J. D., Reynolds, C. A., Watson-Parris, D., & Delle Monache, L. (2025). Are AI weather models learning atmospheric physics? A sensitivity analysis of cyclone Xynthia. npj Climate and Atmospheric Science, 8(1), 92, https://doi.org/10.1038/s41612-025-00949-6

Ralph, F. M., Cannon, F., Tallapragada, V., Davis, C. A., Doyle, J. D., Pappenberger, F., Subramanian, A., Wilson, A. M., Lavers, D. A., Reynolds, C. A., Haase, J. S., Centurioni, L., Ingleby, B., Rutz, J. J., Cordeira, J. M., Zheng, M., Hecht, C., Kawzenuk, B., & Monache, L. D. (2020). West Coast Forecast Challenges and Development of Atmospheric River Reconnaissance. Bulletin of the American Meteorological Society, 101(8), E1357-E1377. https://doi.org/10.1175/BAMS-D-19-0183.1

CW3E Publication Notice: Changes to Atmospheric River Related Extremes Over the United States West Coast Under Anthropogenic Warming

CW3E Publication Notice

Changes to Atmospheric River Related Extremes Over the United States West Coast Under Anthropogenic Warming

March 17, 2025

A new study, “Changes to Atmospheric River Related Extremes Over the United States West Coast Under Anthropogenic Warming” was recently published in Geophysical Research Letters by CU Boulder PhD Candidate Tim Higgins along with Aneesh Subramanian (CU Boulder), Peter Watson (University of Bristol), and Sarah Sparrow (Oxford University). This research utilized large ensemble simulations from two climate models (CESM2 and HadAM4). Given the immense size of the datasets and the absence of IVT in the HadAM4 runs, a machine learning approach – trained on human expert hand labels, IWV, mean sea level pressure, and wind velocity at 850 mb – was used for atmospheric river (AR) tracking (Higgins et al. 2023). The study focused on extreme ARs that surpass IWV thresholds occurring, on average, less than once per year on the coast and within ARs under early 21st century forcing. This work aligns with CW3E’s goal to advance understanding and predictability of extreme events and was supported by the California Department of Water Resources’ AR Program and the US Army Corps of Engineers’ Forecast Informed Reservoir Operations.

The forced response of rare extreme AR events to increased global temperatures was first quantified in three warming scenarios (Figure 1). In California, the SSP 3-7.0 scenario in CESM2 creates the potential for unprecedented peak AR IWV on the coast, with already destructive events becoming significantly more frequent. Even mild warming scenarios show a strong sensitivity of rare extreme AR events to temperature increases. The HadAM4 simulations indicate that in just a 2°C increase from preindustrial levels, an event with a 100-year return period under early 21st century conditions could occur every 10-20 years. Similar trends also emerge in the Pacific Northwest, where 100-year return period events in the early 21st century become roughly 20-year events in the 2°C warming.

Five distinct weather regimes of mean sea level pressure (MSLP) anomalies were identified using self-organizing maps, an unsupervised machine learning method. On average, MSLP anomalies exhibited minimal change under warming scenarios for all regimes. Changes to frequencies of extreme ARs during the 2° increase scenario in comparison to the early 21st century forcing scenario were then examined under varying weather regimes (Figure 2). The frequencies of extreme ARs increased under all regimes, but there was a trend of disproportionately higher percentage increases during regimes that historically had higher AR activity (Regimes 2, 3, and 5). These regimes were all characterized by anomalously low pressure over the North Pacific that likely facilitated poleward vapor transport near the coast.

Figure 1. Maximum winter severity of (a) California IWV and (b) Pacific Northwest IWV occurring within Atmospheric river events at varying periods. The solid lines represent simulations from HadAM4, and the dashed lines represent simulations from Community Earth System Model version 2 (CESM2). The HIST scenario in HadAM4 represents historical forcing from 2006 to 2015 and the 1.5°C and 2°C scenarios represent model runs from the same years with temperature differences based on the Half a degree Additional warming, Prognosis, and Projected Impacts framework. The HIST scenario in CESM2 represents historical forcing from 2007 to 2014 and the SSP 3‐7.0 scenario represents possible future forcing from 2092 to 2099.

Figure 2. Changes in frequencies of extreme Atmospheric river (AR) events under each winter weather regime (a–e, filled contours, with increments of 0.25 events per year) and mean for all regimes (f). Mean sea level pressure anomalies during each weather regime are shown in line contours with increments of 6 mb. The number of detected events in each scenario and the percentage change in the number of events per season is shown in the corner of each panel. Dotted areas represent locations in which the difference between the number of extreme AR events is statistically significant at the 95% level using bootstrapping.

Higgins, T. B., Subramanian, A. C., Graubner, A., Kapp‐Schwoerer, L., Watson, P. A. G., Sparrow, S., et al. (2023). Using Deep Learning for an Analysis of Atmospheric Rivers in a High‐Resolution Large Ensemble Climate Data Set. Journal of Advances in Modeling Earth Systems, 15(4), e2022MS003495. https://doi.org/10.1029/2022MS003495

Higgins, T. B., Subramanian, A. C.,Watson, P. A. G., & Sparrow, S. (2025).Changes to atmospheric river related extremes over the United States west coast under anthropogenic warming. Geophysical Research Letters, 52 e2024GL112237. https://doi.org/10.1029/2024GL112237

CW3E Publication Notice: Future MJO Change and Its Impact on Extreme Precipitation and Temperature Over the Western US in CMIP6

CW3E Publication Notice

Future MJO Change and Its Impact on Extreme Precipitation and Temperature Over the Western US in CMIP6

March 14, 2025

A new paper entitled “Future MJO change and its impact on extreme precipitation and temperature over the western US in CMIP6” was recently published in the Journal of Geophysical Research: Atmospheres and authored by CW3E researcher Jiabao Wang, Mike DeFlorio (CW3E), Hyemi Kim (Ewha Womans University, Republic of Korea), Kristen Guirguis (CW3E), and Alexander Gershunov (CW3E). As part of CW3E’s 2025-2029 Strategic Plan, CW3E seeks to leverage novel observations, modeling, and innovations in forecast verification to advance the understanding of extreme weather in the West and their “forecasts of opportunity” where subseasonal prediction skill is relatively high to inform current and future resource and risk management. This study (Wang et al. 2025) analyzed 23 Coupled Model Intercomparison Project Phase 6 (CMIP6) models and showed systematic changes in multiple Madden-Julian oscillation (MJO) characteristics in the future climate, which are primarily contributed by the steepening of the mean meridional moisture gradient over the Indo-Pacific warm pool in a warming climate, and revealed the potential changes in MJO impacts on extreme precipitation and temperature over the western US. This research was sponsored by the California Department of Water Resources Atmospheric River Program.

The multi-model mean of the CMIP6 models shows a ∼17% increase in precipitation amplitude, an ∼11%–14% increase in circulation amplitude, a ∼9% increase in propagation speed, a ∼2-day decrease in period, and a ∼5° eastward extension of the MJO in the future climate (2065-2100) compared to current climate (1979-2014). Analysis of the lower tropospheric moisture budget suggests the dominant role of an increased meridional advection of mean moisture caused by the steepening of mean moisture gradient over the Indo-Pacific warm pool in a warming climate in the majority of models. The stronger anticyclonic gyres to the east of the MJO convection center along with the enhanced moisture gradient favor an enhanced export of moisture away from the Equator and a local moistening at the flanks of the MJO convection center. The above change in the moisture advection leads to a more substantial positive moisture tendency to the east of MJO convection and hence an enhanced eastward MJO propagation with strengthened amplitude and faster speed (Fig. 1).

Previous studies have shown a significant linkage between MJO characteristics (e.g., location, amplitude, propagation speed) and extratropical circulations and extreme weather (e.g., Liang et al. 2022; Wang et al. 2024). In this study, we examined how the downscaled CMIP6 models capture the observed MJO-extreme relationships and how those relationships may change in the future. In the current climate, wet extremes over California (CA) significantly decrease during MJO phases 2&3 when the enhanced convection is over the Indian Ocean and increase when the MJO is located over the western Pacific in phases 6&7. Model projections suggest that changes in CA precipitation extremes tend to be stronger and more substantial in the future climate in response to MJO activity. On the other hand, MJO phases 1&8 with convection over the Western Hemisphere and Africa generally correspond to an overall decrease in warm spells over the western US, while MJO phases 4&5 with convection over the Maritime Continent are associated with an overall increase. Unlike the wet extremes, MJO-associated changes in warm spells tend to be weaker over most of the western US in the future climate (Fig. 2).

Figure 1. (a)-(b) Multi-model-mean (MMM) filtered precipitation anomalies (shading) and 850-hPa to 700-hPa vertically integrated moisture tendency anomalies (contour, 0.3×10-6kgm-2s-1 interval) in historical and future climates, respectively. (c) Difference between (b) and (a), which represents the future change in MJO precipitation and moisture tendency (0.2×10-6kgm-2s-1 interval). Dots and hatch in (c) denote the significant difference in precipitation and moisture tendency, respectively, at the 0.05 significance level. The red box denotes the region east to the Maritime Continent (MC, day -8 to day 2, 120°E-160°E) where significant changes in moisture tendency are found. (Right top) Filtered 850-hPa to 700-hPa integrated moisture budget anomalies (unit: 10-6kgm-2s-1) averaged over the red box in (c). The residual includes other terms such as precipitation, fluxes, and vertical advection, which is the difference between moisture tendency and horizontal advection. (Right bottom) Future changes in the meridional gradient of mean moisture where dots indicate the difference between future and historical runs at the 0.05 significance level. (Adapted from Figs. 3, 5, 6 in Wang et al. 2025)

The findings in this study indicate a high likelihood of changes in MJO-extreme relationships in the future climate. Given that MJO is considered the dominant source for subseasonal predictability with a time range from 2 weeks to 2 months and this time range currently has much lower skills than the weather and seasonal timescales, this study suggests that the MJO may still provide very useful information for predicting precipitation at the subseasonal timescale in the future but may be less useful in predicting temperature as shown by most CMIP models.

Figure 2. Area-averaged wet extreme and warm spell frequency response over Northern California (CA), Central CA, and Southern CA in historical (blue line) and future (red line) periods after each MJO phase combination. Shading indicates the multi-model standard deviation. (Fig. 10 in Wang et al. 2025)

Wang, J., DeFlorio, M. J., Kim, H., Guirguis, K., & Gershunov, A. (2025). Future MJO change and its impact on extreme precipitation and temperature over the western US in CMIP6. Journal of Geophysical Research: Atmospheres, 130(5), e2024JD042123. https://doi.org/10.1029/2024JD042123

Wang, J., DeFlorio, M. J., Gershunov, A., Guirguis, K., Delle Monache, L., & Ralph, F. M. (2024). Association of western US compound hydrometeorological extremes with Madden-Julian oscillation and ENSO interaction. Communications Earth & Environment, 5, 314. https://doi.org/10.1038/s43247-024-01449-w

Liang, S., Wang, D., Ziegler, A. D., Li, L. Z., & Zeng, Z. (2022). Madden–Julian Oscillation-induced extreme rainfalls constrained by global warming mitigation. npj Climate and Atmospheric Science, 5(1), 67. https://doi.org/10.1038/s41612-022-00291-1

CW3E Publication Notice: Heresy in ENSO teleconnections: Atmospheric Rivers as disruptors of canonical seasonal precipitation anomalies in the Southwestern US

CW3E Publication Notice

Heresy in ENSO teleconnections: Atmospheric Rivers as disruptors of canonical seasonal precipitation anomalies in the Southwestern US

January 31, 2025

This research was recently highlighted by the UC San Diego Today campus-wide newsletter in an article titled Atmospheric Rivers Explain Atypical El Niño and La Niña Years

The paper “Atmospheric Rivers as disruptors of canonical seasonal precipitation anomalies in the Southwestern US” has been recently accepted for publication in Climate Dynamics. The study was conducted by Rosy Luna-Niño (CW3E), Alexander Gershunov (CW3E), F. Martin Ralph (CW3E), Alexander Weyant (UCSD), Kristen Guirguis (CW3E), Michael J. DeFlorio (CW3E), Daniel R. Cayan (CW3E), and Park Williams (UCLA). This research was supported by the California Department of Water Resources Atmospheric River Program Phase III.

The motivation of this paper was the unexpected wetness observed in Southern California (SoCal) and, overall, the Southwestern US during the water year 2023 (WY2023: October 2022- April 2023). Considering the westwide domain, WY2023 precipitation anomalies were opposite to those expected and predicted in the majority of seasonal precipitation prediction systems based on the concurrent La Niña conditions (e.g. DeFlorio et al. 2024). We identified other WYs that didn’t behave as expected with respect to ENSO conditions and term them “heretical”, i.e. behave in a way opposite to ENSO canon. Two types of heretical WYs were defined (Figure 1a): unexpectedly wet or heretical La Niña years (e.g 2011, 2017, and 2023) and unexpectedly dry or heretical El Niño years (e.g. 1964, 1977, 1987, 2007, 2013, and 2015).

One of the main results from this research is that ARs were key wildcards, or agents of heresy, in producing the opposite precipitation anomalies during the heretical WYs: heretical La Niña/El Niño WYs were characterized by anomalously robust/deficient AR activity (Figure 1b). Three out of the five lowest-ranking seasons, with respect to AR landfall days, occurred during heretical El Niño years (WYs 1964, 1977, 1987). While, the record high frequency of AR days occurred during the weak La Niña WY2017 — one of the wettest years on record for California (Gershunov et al. 2017). In SoCal, precipitation totals in heretical Las Niña WYs were comparable to those of canonical El Niño WYs 1998 and 1983. In Northern California, the wettest heretical year was the weak La Niña 2017, with total precipitation comparable to El Niño WY1998, surpassed only by El Niño WY1983. These results emphasize the critical role of AR activity in shaping ENSO-related precipitation anomalies in the Southwestern US. The anomalous bounty or paucity of ARs, their intensity and orientation (Guirguis et al. 2019), can either amplify the expected anomalies based on ENSO conditions or negate them, disrupting seasonal predictions to the extent of generating entirely unexpected — heretical — precipitation anomalies.

Figure 1. a) Total precipitation (October-April; mm) and b) Number of days of landfalling ARs in Southern California. Solid horizontal lines in a) and b) indicate mean annual precipitation and average landfalling AR days, respectively; dashed lines in a) show thresholds based on the standard deviation (σ). Circles in b) show the October SST anomalies in the Niño3.4 region as indicator of ENSO conditions at the beginning of the WY. Green and brown colors highlight unexpected wet and dry water years, respectively, based on ENSO: unexpectedly wet for La Niña and unexpectedly dry for El Niño.

To understand the implication of ARs and their precipitation in seasonal prediction, we reexamined the relationship between ENSO and seasonal precipitation — separating precipitation into non-AR and AR components. In concordance with previous studies (e.g., Dettinger et al. 2011; DeFlorio et al. 2013), we found that the correlation between ENSO and total seasonal precipitation is positive and significant mostly in the southwestern US (Figure 2a). The correlation of ENSO with non-AR precipitation looks very similar to that with total precipitation (Figure 2b). On the other hand, the overall correlation pattern for AR precipitation is weaker (Figure 2c), with the highest correlations obtained over southern Arizona, and New Mexico (Sonoran) desert. The Desert Southwest may be the region of the Western US where AR landfalling activity produces a precipitation signal synchronized with contemporaneous ENSO activity, especially during late winter (January-March) and early spring (February-April). This topic needs further investigation, focusing on the landfalling ARs in Baja California and their inland penetration into the southwestern desert.

Figure 2. Spearman correlation between Oceanic Niño Index and seasonal a) total precipitation, b) nonAR precipitation, and c) AR precipitation during the WYs 1952-2023. Shaded colors indicate significant correlations at 95% confidence level. Black contour limits areas of positive and negative correlation.

Our results show that the relationship between ENSO and precipitation from ARs is complex and affects seasonal predictability in the Western US. While ENSO remains the main source of seasonal predictability, AR landfalls can disrupt expected ENSO precipitation patterns such as during the heretical wet La Niña years of 2011, 2017, and 2023. This nuance is now expressed as a disclaimer in CW3E CCA-forecast based on Pacific SST (https://cw3e.ucsd.edu/s_and_s_forecasts/), noting that CCA predicts mostly seasonal non-Atmospheric River precipitation.

DeFlorio, M. J., D. W. Pierce, D. R. Cayan, and A. J. Miller. 2013., Western U.S. extreme precipitation events and their relation to ENSO and PDO in CCSM4. Journal of Climate, 26, 4231-4243. doi:10.1175/JCLI-D-12-00257.1. https://doi.org/10.1175/JCLI-D-12-00257.1

DeFlorio, MJ., Sengupta, A., Castellano, CM., Wang, J., Zhang, Z., Gershunov, A., Guirguis, K., Luna-Niño, R., Clemesha, RES., Pan, M., Xiao, M., Kawzenuk, B., Gibson, PB., Scheftic, W., Broxton, PD., Cornuelle, BD., Miller AJ., Kalansky, J., Delle Monache, L., Ralph, FM., Waliser, DE., Robertson, AW., Zeng, X., DeWitt, DG., Jones, J., Anderson, ML. 2024. From California’s extreme drought to major flooding: Evaluating and synthesizing experimental seasonal and subseasonal forecasts of landfalling atmospheric rivers and extreme precipitation during Winter 2022 – 2023. Bulletin of the American Meteorological Society. 105(1), E84-E104 https://doi.org/10.1175/BAMS-D-22-0208.1

Dettinger, M. D., Ralph, F. M., Das, T., Neiman, P. J., and Cayan, D. R. 2011. Atmospheric rivers, floods and the water resources of California. Water, 3(2), 445-478. https://doi.org/10.3390/w3020445

Gershunov, A., Shulgina, T., Ralph, F. M., Lavers, D. A., and Rutz, J. J. 2017. Assessing the climate‐scale variability of atmospheric rivers affecting western North America. Geophysical Research Letters, 44(15), 7900-7908. https://doi.org/10.1002/2017GL074175

Guirguis, K., Gershunov, A., Shulgina, T., Clemesha, R. E., and Ralph, F. M. 2019. Atmospheric rivers impacting Northern California and their modulation by a variable climate. Climate Dynamics, 52, 6569-6583. https://doi.org/10.1007/s00382-018-4532-5

Luna-Niño, R., Gershunov, A., Ralph, F. M., Weyant, A., Guirguis, K., DeFlorio, M. J., Cayan, D. R., and Williams, A. P. 2024. Heresy in ENSO teleconnections: Atmospheric Rivers as disruptors of canonical seasonal precipitation anomalies in the Southwestern US. Preprint Available: https://doi.org/10.21203/rs.3.rs-4583843/v1

CW3E Field Team Installs Telemetered Streamgages

CW3E Field Team Installs Telemetered Streamgages

December 30, 2024

Throughout 2023-2024, the CW3E Field Team installed telemetry at eight existing streamgages and two new streamgages in the Russian River and Yuba River watersheds.

In the Russian River watershed, the six upgraded sites are White Creek (WIC), Perry Creek (PEC), Mill Creek (MCR), Cold Creek (CDC), Boyes Creek (BYS), and Mewhinney Creek (MEW) near Lake Mendocino. The two newly installed sites are Galloway Creek (GAC) and Dry Creek (DAC), which both flow into Lake Sonoma.

In the Yuba River watershed, the two upgraded sites are Upper Dry Creek (UDC) and Little Dry Creek (LDM) on the Dry Creek tributary of the Yuba River.

Real-time stage and discharge data is now available on CDEC’s website for WIC, PEC, MCR, CDC, and stage data for BYS, MEW, GAC, DAC, UDC, and LDM. Discharge data for BYS, MEW, GAC, DAC, UDC, and LDM will be available once the rating curves to calculate discharge are developed for these sites.

The stage and discharge data supports the Russian River Forecast Informed Reservoir Operations (FIRO) project and the Yuba-Feather FIRO project, which research ways to improve the reservoir operations at Lake Mendocino, New Bullards Bar Reservoir, and Lake Oroville by updating the water control manuals to reflect improvements in weather and streamflow forecasts. A list of all of CW3E’s sites on CDEC with links to data, both streamgages and weather stations, can be found on CDEC’s website.

Fig. 1: Left: A map showing the locations of the telemetered streamgages at six sites flowing into Lake Mendocino and two sites flowing into Lake Sonoma in the Russian River Watershed . Right: A map showing the locations of the telemetered streamgages at two sites on the Dry Creek tributary of the Yuba River.

Fig. 2: (left) Garrett McGurk installs the new water level sensor at Cold Creek. (middle left) Ethan Morris installs a staff gauge at Dry Creek. (middle right) Sarah Burnett installs the new water level sensor at Little Dry Creek. (right) Adolfo Lopez-Miranda and Ava Cooper prepare to route the water level sensor cable at Upper Dry Creek.

CW3E AR Update: 23 December 2024 Outlook

CW3E AR Update: 23 December 2024 Outlook

December 23, 2024

Click here for a pdf of this information.

Unsettled Weather Pattern to Continue Over US West Coast Through This Weekend

  • A series of atmospheric rivers (ARs) will continue to propagate across the North Pacific and impact the Pacific Northwest and Northern California over the next 7 days.
  • The first AR is forecast to make landfall early late today and bring AR 2/AR 3 conditions (based on the Ralph et al. 2019 AR Scale) to coastal Southern Oregon.
  • The second and third ARs are forecast to make landfall on Wed 25 Dec and Thu 26 Dec.
  • The fourth and potentially most impactful AR is forecast to make landfall on Fri 27 Dec.
  • As the fourth AR moves onshore, a mesoscale frontal wave may prolong the AR duration and bring a second pulse of stronger moisture transport to the region. There is considerable uncertainty in the evolution of this AR.
  • The NWS Weather Prediction Center is forecasting 7–15 inches of total precipitation over portions of western Washington, western Oregon, and Northern California during the next 7 days.
  • Model differences in the forecast evolution of these ARs is driving differences in forecast precipitation. Overall, EPS is forecasting much higher precipitation totals across western Oregon and Northern California during the next 10 days compared to GEFS.
  • A marginal risk excessive rainfall outlook (ERO) has been issued for the Southern Oregon and Northern California Coast Ranges, as well as the Northern Sierra Nevada foothills today into early Wed 25 Dec.
  • Marginal risk EROs have also been issued for coastal Washington, Oregon, and Northern California for Wed 25 Dec into early Sat 28.
  • Numerous stream gages in western Washington, western Oregon, and Northern California are forecast to rise above action/bankfull stage over the next 10 days, with the greatest potential for flooding in southwestern Oregon.

Click images to see loops of GFS IVT and IWV forecasts

Valid 0000 UTC 23 December 2024 – 0600 UTC 30 December 2024

Summary provided by C. Castellano, S. Bartlett, and M. Steen; 23 December 2024

To sign up for email alerts when CW3E post new AR updates click here.

*Outlook products are considered experimental

For any unfamiliar terms, please refer to the American Meteorological Society Glossary.

AR Recon WY2025 Operations Update

AR Recon WY2025 Operations Update

December 20, 2024

A pallet of three weather buoys parachutes to the Pacific Ocean after being released by the Air Force Reserve’s 53rd Weather Reconnaissance Squadron Hurricane Hunters Dec. 15, 2021. Credit: (U.S. Air Force courtesy photo)

In support of AR Recon, CW3E, in collaboration with the NOAA-funded Scripps Institution of Oceanography’s Global Drifter Program continues to enhance our understanding of atmospheric pressure at the ocean’s surface. This year, satellite-tracked drifters, equipped with barometers, are again being deployed in the North Pacific during the AR Recon season. We utilize two types of drifting buoys: Directional Wave Spectra-Barometer (DWS-B) and Surface Velocity Program – Barometer (SVP-B). The DWS-B buoys, which are un-drogued, provide valuable information on wave spectra, while the drogued SVP-B buoys measure near-surface ocean currents at a depth of 15 meters. Both types deliver critical data on sea-surface temperature and sea-level pressure, essential for numerical weather predictions and climate studies in this data-sparse region. These buoys are meant to increase sampling density in regions in the North Pacific that experience the highest number of Atmospheric River (AR) days annually and are frequently associated with high pressure gradients.

Already this year, the USAF 53rd WRS has conducted two drifter deployment flights, the first on December 7th and the second on December 13th, releasing a total of 30 new drifters. These new buoys join 345 drifters deployed as part of AR Recon since its founding. Additionally, during the second buoy deployment mission, the 53rd WRS was also able to deploy 14 dropsondes over an AR event that brought significant impacts to the Bay Area on Saturday December 14th.

These AR Recon observations and buoy deployments have been shown to improve forecasts, and targeted deployments are coordinated annually in collaboration between CW3E, NOAA, and the US Air Force. More information on AR Recon, our data collection efforts this season, as well as our plans for the rest of 2025 can be found in this recent press release.

(Deployments marked by cyan symbols at approximately 40N and 140W-160W): Drifting buoy locations in the northeast Pacific, with those released during AR Recon 2023 shown in purple. SLP: Sea Level Pressure. SVP-B: Surface Velocity Program – Barometer. DWS – B: Directional Wave Spectra Barometer. Graphic courtesy of B. Kawzenuk, CW3E.

CW3E Director of Research Presents at ADIA Lab Symposium

CW3E Director of Research Presents at ADIA Lab Symposium

December 10, 2024

Luca Delle Monache, CW3E Director of Research, presenting at the ADIA Lab Symposium on Abu Dhabi on 20 November 2024. (link to video recording)

CW3E Director of Research Luca Delle Monache recently attended the second edition of the Abu Dhabi Investment Authority (ADIA) Lab Symposium as an invited speaker. The symposium was held in Abu Dhabi, United Arab Emirates, from November 19 to 21, 2024.

The symposium speakers’ lineup included Nobel Prize winners and leading figures from academia, industry, and government, who discussed the recent developments in data science and artificial intelligence. This year’s Symposium’s critical themes included sustainability, innovation, and trustworthy AI.

During his presentation, titled “Deep Learning and Data-Driven Models for the Prediction of Extreme Weather Events”, Dr. Delle Monache provided an overview of Atmospheric Rivers (ARs) and the associated impacts, a description of the Forecast Informed Reservoir Operations Program, and CW3E’s strategy to improve our ability to predict extreme events associated with ARs based on a tailored physics-based model (West-WRF), a 200-member dynamical ensemble, machine learning postprocessing of the ensemble, and the recently developed artificial intelligence (AI) data-driven models.

The talk was well received with questions on the specific choice of the weather neural operator, how the new AI data-driven models may impact geoengineering, and what the impact of climate change may be on AR intensity and frequency.

CW3E Publication Notice: Seasonality and climate modes influence the temporal clustering of unique atmospheric rivers in the Western U.S.

CW3E Publication Notice

Seasonality and climate modes influence the temporal clustering of unique atmospheric rivers in the Western U.S.

November 27, 2024

A recent study has provided new insights into the temporal clustering of atmospheric rivers (ARs) in the Western United States, highlighting the significant role of climate modes and seasonality in shaping these impactful weather phenomena–ARs clustering and orientation. This research is detailed in the paper titled “Seasonality and Climate Modes Influence the Temporal Clustering of Unique Atmospheric Rivers in the Western U.S.”, recently published in Nature Communications Earth & Environment. The study was conducted by Zhiqi Yang (CW3E), Michael J. DeFlorio (CW3E), Agniv Sengupta (CW3E), Jiabao Wang (CW3E), Christopher M. Castellano (CW3E), Alexander Gershunov (CW3E), Kristen Guirguis (CW3E), Emily Slinskey (CW3E), Bin Guan (UCLA/JPL), Luca Delle Monache (CW3E), and F. Martin Ralph (CW3E). This research was supported by the California Department of Water Resources (CA DWR) Atmospheric River Program and aligns with CW3E’S 2019-2024 Strategic Plan by seeking to improve subseasonal and seasonal predictability of extreme hydroclimate variables over the western U.S. region.

The study was motivated by the occurrence of nine consecutive atmospheric rivers within a mere three-week span in California during winter 2022/2023 (DeFlorio et al. 2024), as well as the significant economic and hydrological impacts of AR temporal clustering, which can triple damages compared to isolated events (Bowers et al. 2024). Temporal clustering, coupled with the orientation of ARs, modulates their interaction with terrain and influences precipitation and flooding. Despite its importance, the drivers behind AR clustering remain unclear, as well as differences in location and magnitude of clustering between early-winter and late-winter.

In this new article, the authors identified unique ARs over the North Pacific and Western U.S. and utilized Cox regression and composite analyses to examine the influence of six major climate modes—the Arctic Oscillation (AO), Quasi-Biennial Oscillation (QBO), El Niño–Southern Oscillation (ENSO), Madden-Julian Oscillation (MJO), Pacific Decadal Oscillation (PDO), and Pacific-North American pattern (PNA) on temporal clustering and AR orientation during the extended boreal winter (November–March). This study also characterizes differences in location and magnitude of AR clustering in early and late winter periods.

The results reveal that climate modes significantly condition the temporal clustering of ARs. The PNA pattern strongly influences clustering from early to late winter, while the AO dominates early winter clustering. The QBO and PDO emerge as key modulators of late winter clustering (Figure 1). Furthermore, ENSO has a profound effect on AR orientation, particularly influencing the orientation of temporally clustered ARs (Figure 2).

The discovery of relationships between factors with high subseasonal predictability and AR clustering/orientation would lay a foundation to evaluate the subseasonal (2–6 week lead) and seasonal (3–6 month lead) predictability of AR sequences occurring within a short time period. Water resource managers and other applied end users across the Western U.S. stand to reap numerous benefits, such as minimization of flood risks and enhanced planning and decision-making, from improved predictability of hydroclimate phenomena at these extended lead times.

Figure 1. Cox regression coefficients showing climate modes modulations on temporal clustering of unique ARs. Cox regression coefficients between AO, QBO, ENSO, MJO-RMM1, MJO-RMM2, PDO, PNA (removed ENSO signal), and the occurrence of unique ARs based on the ERA-5 AR dataset from 1940/1941 to 2017/2018 (for PDO analysis) and MERRA-2 AR dataset from 1982/1983 to 2020/2021 (for other climate modes analysis) during extended winter (NDJFM, left column), early winter (NDJ, middle column), and late winter (JFM, right column). Note that the coefficients from the regression of AO, QBO, ENSO, MJO, and PNA cannot directly be comparable to those from PDO (see Methods). The results show coefficients that are statistically significant at the 5% level. (Figure 2 in Yang et al., 2024)

Figure 2. Climate modes modulate IVT orientation of temporally clustered unique ARs. Composite analysis in anomalies of the life-cycle IVT orientation of temporally clustered unique ARs during early winter (NDJ) and late winter (JFM). The analysis based on positive (columns 1,3) and negative phases (columns 2,4) of AO, QBO, ENSO, MJO-RMM1, MJO-RMM2, PDO, PNA (removed ENSO signal), using the ERA-5 AR dataset from 1940/1941 to 2017/2018 (for PDO analysis) and MERRA-2 AR dataset from 1982/1983 to 2020/2021 (for other climate modes analysis). The anomalies represent a composite analysis of temporally clustered AR orientation based on the positive and negative phases of climate modes, minus the climatology of temporally clustered AR orientation. Dots indicate statistical significance at the 5% level. Unit: degree. (Figure 10 in Yang et al., 2024)

Bowers, C., Serafin, K. A. and Baker, J. W.(2024). Temporal compounding increases economic impacts of atmospheric rivers in California. Science Advances, 10(3), adi7905. https://doi.org/10.1126/sciadv.adi7905

DeFlorio, M. J., A. Sengupta, C. M. Castellano, J. Wang, Z. Zhang, A. Gershunov, K. Guirguis, R. Luna Niño, R. E. Clemesha, M. Pan, M. Xiao, B. Kawzenuk, P. B. Gibson, W. Scheftic, P. D. Broxton, M. B. Switanek, J. Yuan, M. D. Dettinger, C. W. Hecht, D. R. Cayan, B. D. Cornuelle, A. J. Miller, J. Kalansky, L. Delle Monache, F. M. Ralph, D. E. Waliser, A. W. Robertson, X. Zeng, D. G. DeWitt, J. Jones, and M. L. Anderson. (2024). From California’s extreme drought to major flooding: Evaluating and synthesizing experimental seasonal and subseasonal forecasts of landfalling atmospheric rivers and extreme precipitation during winter 2022/23. Bulletin of the American Meteorological Society, 105(1), E84-E104. https://doi.org/10.1175/BAMS-D-22-0208.1

Yang, Z., DeFlorio, M.J., Sengupta, A., Wang, J., Castellano, C.M., Gershunov, A., Guirguis, K., Slinskey, E., Guan, B., Delle Monache, L. and Ralph, F.M.,. (2024). Seasonality and climate modes influence the temporal clustering of unique atmospheric rivers in the Western US. Nature Communications Earth & Environment, 5(1), p.734. https://doi.org/10.1038/s43247-024-01890-x

CW3E at the American Geophysical Union (AGU) 2024 Annual Meeting

CW3E at the American Geophysical Union (AGU) 2024 Annual Meeting

CW3E will be participating in the upcoming 2024 American Geophysical Union (AGU) annual fall meeting this December. Several CW3E graduate students, post docs and researchers will travel to Washington, D.C. or connect remotely to present their work at this year AGU’s meeting, themed “What’s Next for Science.” More information about AGU 2024 annual meeting here.

Each year AGU’s annual meeting boasts the largest gathering of Earth and space scientists, convening over 25,000 attendees from over 100 countries to share research and connect with friends and colleagues. Joining scientists in attendance at AGU every year are educators, policymakers, journalists and communicators interested in better understanding and communicating earth science within their fields. CW3E will be represented by 16 presenters this year and has 36 authors represented in total across 39 submitted talks and poster presentations.

All CW3E submissions are linked below in alphabetical order of first author (last name). CW3E researchers, post-docs and graduate students names are bolded, CW3E funded research collaborators or alumni names are italicized.

Alipour R.S., P. Nikrou, A.M. Nemnem, S. Balachandran, E. Goharian, J. Imran, S.J. Burian, D. Weiss, J. Kastens, X. Li, S. Cohen. H53M-1264 Comparative Analysis of Low-Complexity Flood Inundation Models for Dam Failure Scenarios: OWP HAND-FIM vs FLDPLN.

Balachandran S., A.M. Nemnem, P. Nikrou, R.S. Alipour, E. Goharian, J. Imran, S.J. Burian, X. Li, S. Cohen. H53M-1281 Rapid Inundation Mapping for Catastrophic Floods Due to Dam Failures and Operations.

Battula S.B., J.M. Cordeira. A53G-2167 Analyses and Model Evaluation of Extreme Precipitation Caused by Atmospheric Rivers and Mesoscale Convective Systems over Southeast United States.

Brandt T., K. Haleakala, G. Lewis, R. Weihs, A. Cooper, A.M. Wilson, T. Dixon. H32C-07 Is it rain or snow—what observations are required for mountain precipitation phase determination?.

Chen X., T. Yang, K.M. Mahoney, M. Pan, N.L. Ciwro. A51L & A54B Advancing Precipitation Predictions with Physical Models and Artificial Intelligence.

Clemesha, R.E.S., A. Gershunov, R.L. Niño, C. Poulsen, E. Fleishman, G.L. Rochelle, S. Shanahan, B.H. Udall. A13L-05 History and Future Projections of the Diverse Seasonalities of Precipitation within the Colorado River Basin and Southwestern United States.

DeMott C.A., A. Subramanian, C.A. Davis, J.D. Doyle, S.A. Henderson, H. Kim, D. Lavers, A. Levine, C.A. Reynolds, V. Tallapragada, A.M. Wilson. A53G-2274 Teleconnecting process studies: How can we enhance synergies across observing campaigns of teleconnected phenomena?.

Erfani M., J. Kalansky, L. Su, C. Delaney, J. Mendoza. H43T-04 Evaluating Forecast-Informed Reservoir Operations (FIRO) for Climate Change Adaptation: A Case Study on Lake Mendocino in Northern California.

Ge G., M. Zheng, A. Jensen, H. Li, W. Mayfield, A. Back. A42G Numerical Modeling, Data Assimilation, and Research to Operations (R2O) for Better Forecasting of High-Impact Weather Events.

Ge G., M. Zheng, A. Jensen, H. Li, W. Mayfield, J. Wang. A43M Numerical Modeling, Data Assimilation, and Research to Operations (R2O) for Better Forecasting of High-Impact Weather Events.

Ghazvinian M., T. Dixon, L. Delle Monache, A. Sengupta, M. Pan, P. Mulrooney, B. Whitin, P. Fickensher, F.M. Ralph. H31D-06 Deep Learning-Based Ensemble meteorological Forcings for Probabilistic Hydrologic Forecasting.

Gilmore T.E., F. Birgand, E. Goharian, C. Terry. H13C Cameras as Sensors: Application of Images, Videos, and Artificial Intelligence for Hydrological Monitoring Poster.

Goharian E., S.N. Young, J.S. Horsburgh, S. Khan, R. Isa, S. Neupane, M.H. Goloujeh. H13C-1039 Advancing Camera-Based Monitoring for Operational Hydrologic Applications.

Goharian, E., S. Steinschneider, C. Delaney, X. Cai, K.S. Raney. H41T & H43T Water and Society: Forecast-Informed Reservoir Operations (FIRO).

Gorooh V.A., Z. Zhang, D. Axisa, L. Delle Monache. A21G-1805 Development and Application of LSTM for Nowcasting Seedable Cloud Features Using Geostationary Satellite Information over the United Arab Emirates.

Higgins T., W. Chapman, A. Subramanian, C. Pedersen, L.Delle Monache. GC21C-06 Using Generative AI to Improve Data-driven Weather Model Forecasts.

Hsu T.Y., M.R. Mazloff, R. Sun, S.T. Gille, B.D. Cornuelle, M. Zheng, L. Delle Monache. A21D-1750 The Modification of Air-sea Fluxes in High-Wind Environments by Mesoscale Sea Surface Temperature Structure.

Karakan I., J.F. Jamal, E. Goharian, H. Chaudhry, J. Imran. H51T-0985A Diffusive Wave Routing Model Using Meshless Radial Basis Function Collocation Method for Operational Forecasting.

Kim E.Y., V. Chandrasekar, K.D. Waal, L. Delle Monache. A21G-1807 Dual-Polarization Radar Characterization of Extreme Rainfall Event Over the UAE Region.

Lewis G., T. Dixon, B. Whitin, P. Fickensher, A. Reising, D. Rizzardo, R. Hartman, M. Pan, M. Xiao. H06-118 Partnership Approach to Assessing and Improving Operational Water Supply Forecasting Across the Sierra Nevada.

Martens H.R., Q. Cao, A.A. Borsa, M. Pan, A.M. Wilson, E.Knappe, M. Ralph. G11A-04 Water-Storage Gains Driven by Atmospheric Rivers in the Western United States.

Nash D.. EP51C-1348 Differentiating between impactful and non-impactful Atmospheric River events in Southeast Alaska.

Nash D., J. Rutz, J.M. Cordeira, F.M. Ralph, K. Sanders, E. Walter. A52B-06 Inland- penetrating Atmospheric Rivers and Hydrometeorological Impacts in Colorado.

Nemnem A.M., A.H. Tanim, E. Goharian, S. Khan, J. Imran. H12C-03 Modeling Dam Break Flooding under Rainfall Uncertainty: A Case Study of Storm Daniel in Derna, Libya.

Pan M., M. Xiao, Y. Yang, G. Lewis, T. Dixon, C. Shen, Y. Song, D. Feng. H13R-04 Some Lessons Learned in Process-based and Data-driven Modeling for Applied Hydrology.

Patton A.I., L. Luna, A.B. Jacobs, E. Lawrence, B.B. Mirus, D. Nash, J.J. Roering. EP53G-02 Climate Controls on Rainfall Thresholds for Landslide Warning in Southeast Alaska.

Perez A., M. Erfani, E. Goharian. H41T-0795 Integrating Ensemble Hydrologic Forecasts and Interpretable ML Models for FIRO-Based Reservoir Management.

Poulsen C., R.E.S. Clemesha, A. Gershunov, M. D. Dettinger, Z. Zhang, I. Howard, D.W. Stahle, F.M. Ralph. A53J-2230 Influence of Past and Present Spring Precipitation on Western US Drought: A Story of the 1991 California “Miracle March”.

Ralph F.M.. 51D-01 Atmospheric Rivers: From Science to Solutions.

Ralph F.M. H43T-02 Putting the “F” in FIRO: The Challenges and Opportunities of Better Predicting Extreme Precipitation and Stream Flow and the Storms That Produce Them.

Raney K.S., M. Erfani, E. Goharian. H43T-06 Forecast Informed Reservoir Operations: Leveraging Ensemble Forecasts and Advanced Model Predictive Control Methods for Improved Water Storage and Flood Control.

Rush W., J.M. Lora, C.B. Skinner, S. Menemenlis, C.A. Shields, P. Ullrich, T.A. O’Brien, S. Brands, B. Guan, K.S. Mattingly, E.E. McClenny, B.D. Mundhenk, A.B. Nellikkattil, E. Shearer, K. Reid, A. Ramos, R. Tomé, R. Leung, J, Wille, F.M. Ralph, J. Rutz, M.F. Wehner. A34C Atmospheric Rivers: Processes, Impacts, Observations and Uncertainties.

Rutz J., M. Steen, R. Vilela, V. Chandrasekar, B. Garcia. A43A-1944 Comparison of AQPI X-Band and NEXRAD S-Band Radar-Estimated Rain Rates during an Extreme Atmospheric River Event.

Rutz J., V. Chandrasekar, R. Vilela, M. Steen, D. Roberts, F.M. Ralph. A06-03 Updates on the Advanced Quantitative Precipitation Information (AQPI) Project: Improved Radar-based Observations and Forecasts for the San Francisco Bay Area.

Slinskey E.A., J. Rutz, F.M. Ralph. A53G-2164 The NCEI Climate Data Record for Atmospheric Rivers: Initial Results over the Western United States.

Tallapragada V., X. Wu, K. Wu, A.M. Wilson, F.M. Ralph. A51C-01 Advancing Aircraft Reconnaissance Observations, Modeling and Data Assimilation for Improved Atmospheric River Forecasts.

Wang J., M. Zheng, J. Rutz, L.Delle Monache, J. Kalansky, F.M. Ralph. A43M-2159 Assimilating NEXRAD into a Regional Model for Improved Precipitation Forecasts during Two Landfalling Atmospheric Rivers.

Wilson A.M., F.M. Ralph, L.Delle Monache, A. Sengupta, M. Pan, E. Goharian, J.M. Cordeira, V. Tallapragada, C. Talbot, J. Forbis. NH34B-02 Supporting Hazard Preparedness for Extreme Precipitation Associated with Atmospheric Rivers.

Webb M., C. Albano, D. Bozkurt, R. Garreaud, G. Yu, A.M. Wilson. A53J-2316 Drivers of Enhanced Flooding Risk During Atmospheric Rivers in the Western U.S. and Central Chile.

Yang Y., D. Feng, H. Beck, W. Hu, A. Sengupta, L. Delle Monache, R. Hartman, P. Lin, C. Shen, M. Pan. H54D-01 Global Daily Discharge Estimation Based on Grid-Scale Long Short-Term Memory (LSTM) Model and River Routing.

Yang Y., M. Pan, C. Shen, M. Xiao, L. Delle Monache, A. Sengupta, R. Hartman. H53Z-05 Augmentation for Western US Streamflow Prediction Using LSTM-powered Data Fusion Approach.

Yang Z., W. Hu, A. Sengupta, L. Delle Monache, M.J. Deflorio, M. Ghazvinian, M. Pan, J. Kalansky. A41K-1723 Enhancing Weeks 1-2 Forecasts of 2-m Temperature in the Sierra Nevada California through Analog Ensemble post-processing.

Zhang Z., F.M. Ralph, J. Rutz, D. Nash. A53G-2176 The Role of Atmospheric Rivers in the Snowpack over the Upper Colorado River Basin during Water Year 2023.

Zhang Z., F.M. Ralph, X. Zou, B. Kawzenuk, M. Zheng, I. Gorodetskaya, P.M. Rowe, D.H. Bromwich. A41J-1709 Extending the CW3E Atmospheric River Scale to the Polar Regions.

Zheng M., L. Delle Monache, X. Wu, B. Kawzenuk, F.M. Ralph, Y. Zhu, R. Tom, V. Tallapragada, Z. Zhang, K. Wu, J. Wang. A43M-2169 Impact of Atmospheric River Reconnaissance Dropsonde Data on the Assimilation of Satellite Radiance Data in GFS.