The Atmospheric Rivers Program

  • Half of California’s water supply and nearly all of its heavy precipitation events and floods come from a type of storm called an atmospheric river.
  • In December 2022 and January 2023, California experienced nine back-to-back atmospheric rivers; the longest stretch of continuous atmospheric river conditions in the 70 years that records have been collected.
  • The 2022–2023 family of atmospheric rivers led to impressive rainfall and snowfall totals and record-breaking floods. All that water also put a sizable dent in the state’s multi-year drought, though some areas still have not recovered fully.
  • Recent investments in data collection, analysis, and models and tools helped scientists forecast the location, timing, and strength of the 2022–2023 atmospheric rivers more accurately and with longer lead times. The quality of these forecasts helped authorities warn communities and deploy emergency response resources effectively.
  • Data from the 2022–2023 events will help scientists improve forecasting further and support ongoing efforts to use atmospheric river forecasts to inform water management decisions, such as how much water to release from dams.

About Atmospheric Rivers

What is an Atmospheric River?

Atmospheric rivers (ARs) are a type of storm that produces 50 percent of California’s water supply and are responsible for 90 percent of the state’s floods. As the name implies, they are like rivers in the sky: elongated plumes of moisture that lift up when they interact with mountains and produce large amounts of precipitation, both rain and snow. On average, a single atmospheric river carries about  27 times the amount of water in the Mississippi River . Climate projections show that atmospheric rivers will become increasingly important in the future by producing up to  20 percent more of California’s total precipitation , while also becoming even more extreme, increasing  flooding and flood damages .

The  AR scale developed at CW3E  estimates an atmospheric river’s strength and impact. The scale ranges from a low of 1 (primarily beneficial) to a high of 5 (primarily hazardous), based on the amount of water vapor the storm is transporting and how long the transport lasts. Mild AR1 or AR2 storms bring needed rain or snow to replenish water supplies. Intense AR4 or AR5 storms can cause serious flooding. The AR rank is calculated for a specific location. CW3E’s website hosts an AR scale  forecast page  and anyone can sign up for  AR scale alerts for incoming storms . Figures 1 and 2 show how each of the nine atmospheric rivers ranked on the AR scale as measured at the San Francisco and Los Angeles airports, respectively. The graphics show all atmospheric rivers that affected each airport location: nine at San Francisco (two of which had no break between them, from December 29 to 31) and five at Los Angeles.

Figure 1. Atmospheric River Events Recorded at San Francisco International Airport

Figure 2. Atmospheric River Events Recorded at Los Angeles International Airport

A Family of Nine Atmospheric Rivers

An  atmospheric river family  is a sequence of atmospheric rivers that occur within a relatively short period of time. Atmospheric river families can amplify impacts because they are associated with increasingly wet antecedent conditions (e.g., saturated soil, full reservoirs), which can exacerbate flooding. Between December 26, 2022, and January 17, 2023, California experienced a family of nine events with near continuous atmospheric river conditions for 23 days, which is the longest duration of continuous atmospheric river conditions in the 70 years that records have been collected (Figure 3).

Figure 3. Continuous Atmospheric River Days by Duration and Maximum Intensity Over the Last 70 Years

This figure shows the number of continuous days of atmospheric river conditions along the U.S. west coast, allowing for a one-day gap, and the associated maximum intensity over the duration of the event. The years shown in text are water years for events that exceeded the 95 th  percentile for either duration or intensity, while the black dots are all other events that did not meet these criteria. The nine-member atmospheric river family had the longest atmospheric river conditions in the 70-year record. Figure provided by K. Guirguis and based on the atmospheric river catalog of Gershunov et al., 2017, doi:10.1002/2017GL074175.

The extended atmospheric river family resulted from a strong jet stream that extended from west to east across the entire Pacific Ocean basin, which fed the development of multiple atmospheric rivers that impacted California.  This type of atmospheric pattern has been associated with families of atmospheric rivers before, such as in water years 1983, 1998, 2006, 2011, and 2017 .

During these 23 days, the total amount of integrated water vapor transport (IVT), which is how atmospheric rivers are measured, ranged  from 300 to 500 percent more than normal throughout California, with the highest percentages in the Central Valley .  

Nine Individual Atmospheric Rivers

This extended family consisted of nine atmospheric rivers over three weeks. The orientation and landfall location of the nine atmospheric rivers at maximum intensity is shown in Figure 4. The color of the arrow indicates the maximum AR scale. Each atmospheric river has unique features that determine the amount of precipitation in different locations, periods of intense precipitation, strong winds, and the amount of snow versus rain. For example, characteristics of the second event, December 29–31, included a mesoscale frontal wave (an undulation that can modify the duration, intensity, and direction of the atmospheric river) that extended its duration. This atmospheric river was also a warm one, resulting in rain on top of snow. Learn more about each atmospheric river by exploring the interactive maps in Figure 4.

Figure 4. Characteristics of the Nine Atmospheric Rivers in December 2022 and January 2023


Impacts


Forecast Skill

Subseasonal Forecasts

Subseasonal forecasts are forecasts that extend beyond a two-week lead time and extend out to approximately six weeks. Subseasonal forecasts naturally have quite a bit of uncertainty with the long lead time. Nonetheless, subseasonal forecasts provide important situation awareness. The week three forecast on December 8 indicated a large signal of atmospheric river activity impacting the western U.S. coastline during the week of December 23–29, but the week 3 forecasts did not predict the active period the following week.  The subseasonal forecasts did indicate a drying period  that started on January 19 and continued to be predicted well for the next several weeks. This information helped to indicate when there might be some relief from the active atmospheric river period.

Lead Times

CW3E’s  AR landfall tool  is a widely used forecast product because it provides clear visualization of incoming atmospheric rivers. Figure 5 shows an example of the AR landfall tool on December 28, which forecasted the next seven atmospheric rivers through January 13.

Figure 5. AR Landfall Tool Output on December 28, 2022

Each of the nine atmospheric rivers had their own unique characteristics and posed their own unique forecast skill challenges. Nonetheless, the landfall tool provided a 6-day lead time on average across the nine events, as measured at San Francisco International Airport (Figure 6). The second and third atmospheric rivers in the family, which occurred around New Year’s Day, had nearly a 10-day lead time. Longer lead times are critical to allow prepositioning of emergency management services in anticipation of where the greatest impacts will occur.

Figure 6. Lead Time Prediction of Landfalling Atmospheric Rivers, December 27, 2022–January 17, 2023

Example of Improved Forecast Skill: The January 15 Atmospheric River

The atmospheric river that made landfall on January 15 in San Diego was forecasted exceptionally well. The forecast showed 90 percent probability of an AR 3 hitting San Diego with only 6 hours of uncertainty between the start and stop of the AR (Figure 7). 

Figure 7. Forecast Details for the January 15, 2023, Atmospheric River

AR scale forecast from the U.S. National ensemble forecast (GEFS) issued on January 14 for San Diego. A total of 27 of the 30 ensemble members show an AR 3 making landfall, all within 6 hours of the actual start time.

West-WRF

West-WRF is a weather forecasting model being developed at CW3E that builds on the U.S. and European models to focus on forecasts of extreme precipitation, particularly atmospheric rivers over the western United States. The model was developed in recognition that national models are not tailored to storms unique to the western states. West-WRF runs at higher spatial resolution over the western United States for longer lead times than global and national models. Since CW3E began running West-WRF, atmospheric river forecasts have improved, providing an additional day of lead time (Figure 8). Improving weather forecasts is extremely challenging. Precipitation forecasts across the United States for  1-inch  and  2-inch  thresholds have leveled off since 2008. CW3E has been able to make improvements by focusing on a specific type of storm-atmospheric rivers that affect western states. Beginning in December 2022, CW3E began sharing West-WRF forecast data with DWR for experimental use in their hydrologic models.

Figure 8. West-WRF Forecast Improvement, 2015–2022

Innovative modeling with access to a supercomputer has led to the development of a 200-member ensemble probabilistic forecast, which means that there are 200 different forecasts that vary slightly. All the forecasts represent plausible outcomes. The West-WRF ensembles run at a 9 km resolution and a lead time out to 7 days. By running such a large ensemble with such high resolution, CW3E is able to capture the extremes in the range of possible outcomes that could be missed by the global models and their 30- and 50-member ensembles (Figure 9). Running a model of this size and high spatial resolution is possible thanks to supercomputing capabilities available to CW3E with support from DWR and the U.S. Army Corps of Engineers.

Figure 9. Comparison of Forecast Skill for Precipitation Total

The ensemble forecasts issued on December 29 at 4 pm PT (00 Z) for Sly Park located in the Consumes Basin. The top panel is from the U.S. National model (GEFS), the middle is the European model (ECMWF), and the bottom is from West-WRF. The red line is the observed precipitation while the dark lines are the max and min of the ensemble forecasts and the dash line in the median. By building on both the GEFS and ECMWF, West-WRF’s large ensembles better forecasted the possibility of an extreme precipitation event than either of the other two ensemble forecasts alone.

The skill of the West-WRF model ensemble prediction was 5 to 15 percent higher than the two global forecasting models for precipitation 1 inch or greater for all lead times during the three weeks when this family of atmospheric rivers occurred (Figure 10). Newly developed machine learning methods improved the skill by an additional 3 percent on average. 

Figure 10. Comparison of Forecast Skill With Respect to Lead Time

The line shows the skill (continuous ranked probability skill score) of ensemble based probabilistic forecasts between December 25 and January 18 for precipitation over 1 inch at lead times from 1 day (on the left) to 6 days (on the right). The close the number skill is to 1, the greater the skill. The U.S. national model (GEFS) is shown in blue, the European model (ECMWF) is shown in green, the West-WRF ensemble is shown in red, and the purple line is when machine learning is applied to the West-WRF ensemble after the forecast has been issued.

Recently developed figures using the  West-WRF ensemble forecast  illustrate the probability of precipitation above a certain threshold, such as 1 or 5 inches. This provides information about the likelihood of extreme rainfall at different locations. For example, Figure 11 shows the probability of 24-hour rainfall exceeding 3 inches on January 1 based on the West-WRF ensemble forecast issued on December 30.

Figure 11. Extreme Precipitation Probability Forecast for January 1, 2023


Figure 12. AR Recon Flight Tracks and Dropsonde Locations, January 2023

Flight tracks (lines) and datasets (markers) collected by aircraft during the January 2023 sequence. Colors represent dates flown.

CW3E partners with multiple global centers, including NOAA’s National Center for Environmental Prediction (NCEP) and Europe’s ECMWF, to assess the impact of the AR Recon data and improve data assimilation methods to further improve forecasts. Preliminary analysis shows the AR Recon data collected from the aircraft during the 13 missions during this family of atmospheric rivers improved the representation of key  variables up to 50 percent . This is a first step in assessing how the AR Recon data improved the forecasts. 


Forecast Informed Reservoir Operations (FIRO)

CW3E, in partnership with the U.S. Army Corps of Engineers and Sonoma Water, pioneered  FIRO  at Lake Mendocino, located in the Russian River Basin. Atmospheric river forecasts represent the “F” in FIRO, a flexible water management approach that uses data from watershed monitoring and improved weather forecasting to help water managers selectively retain or release water from reservoirs. It is a key strategy for resilience to droughts and floods. FIRO applies emerging science and technology to optimize water resources and adapt to climate change without costly infrastructure. When dry conditions are predicted, more water can be stored to buffer against droughts, whereas when a series of storms is predicted, water can be released ahead of the storm events to make room for large inflows and reduce flooding. The impetus to pioneer FIRO came from the observation that, had some of the water been saved when atmospheric rivers hit in late December of 2012,  the subsequent nearly three-year drought could have been mitigated .

Through an integrated, multi-agency process, FIRO was eventually implemented based on improved understanding of atmospheric rivers, forecast skill assessment, enhanced observations, and novel new decision support tools. In water year 2020 with FIRO in operation via a major deviation to the Water Control Manual at Lake Mendocino, FIRO  allowed for 20 percent more water to be stored , which helped mitigate the three year drought that began that year.

Toward the end of the atmospheric river family in January 2023, dam operators at Lake Mendocino used FIRO-based operations to allow the reservoir to stay at levels higher than would otherwise be allowed without FIRO, providing about 11,000 acre-feet more water storage (Figure 13). This additional storage will provide important buffer for water supply in case of upcoming dry years. Higher storage levels also mean a deeper pool and colder water temperatures, which is critical for three endangered fish species downstream in the Russian River.

Figure 13. Rainfall and Storage at Lake Mendocino, October 2022–April 2023

Lake Mendocino dam operators used FIRO to maintain storage at the top of the “FIRO Pool” in January 2023, enabling storage of 11,000 more acre-feet of water than would otherwise be allowed. 

In another example, the Orange County Water District (OCWD) was able to fill its water conservation pool multiple times (Figure 14). In an average year, OCWD recharges about 52,000 acre-feet of storm flow. This year, total recharge is likely to be more than 70,000 acre-feet, but the total recharge will not be known until the conservation pool is fully drained, which may not happen until late summer 2023.

Figure 14. Storage Levels at Prado Dam, July 2022–March 2023

Water captured in the Water Conservation Pool (20,000 acre-feet) is drained slowly to promote capture by OCWD for groundwater recharge. This figure shows multiple opportunities for stormwater capture during and immediately after the atmospheric river family. Water in storage above the Water Conservation Pool is managed by the U.S. Army Corps of Engineers according to flood risk management rules. Data source: U.S. Army Corps of Engineers.

In addition, Turlock Irrigation District increased releases on January 3 even though they were 45 feet below flood stage based on hydrology, snowpack, and forecasts. These releases led to better outcomes for hydroelectric power, flood control, environmental flows, and achievement of goals under the Sustainable Groundwater Management Act (SGMA).


Unique Observations

The Advanced Quantitative Precipitation Information (AQPI) Program

The  AQPI Program  fills gaps in radar data in the Bay Area and just recently, CW3E has taken leadership of the Program through support from the AR Program. Just prior to the 2022/2023 atmospheric river family, AQPI had brought online several additional X-band radars (Sonoma County Water, Santa Clara, Santa Cruz) (Figure 15).

Figure 15. New AQPI Radar Locations

Approximate locations of aerial coverage for new AQPI radars.

In addition to filling radar gaps in existing coverage, the new locations provide data at a higher spatial and temporal resolution than what was previously available. This higher resolution helps identify locations where precipitation is occurring, particularly bands of intense rain (Figure 16). The California State Operations Center used AQPI resources to improve situational awareness of storm activity in the Bay Area in its efforts to incorporate information and forecasts to guide deployment of emergency response resources.

Figure 16. Improved Accuracy from the AQPI X-band Radar

The new AQPI Santa Clara X-band radar image from the January 9–10 atmospheric river is shown on the left and the previous KMUX NEXRAD radar image is shown on the right. Warmer colors indicate higher intensity rainfall. The new AQPI radar image shows more accurately and in more detail which areas around San Jose are experiencing high rainfall rates.

Soil Moisture

Soil moisture is an important indication of how the watershed will respond to the precipitation and what the streamflow response will be. For example, during the first atmospheric river in this family, soil moisture was at approximately 83 percent of capacity in the Lake Mendocino watershed. The resulting runoff efficiency, an indication of how much runoff was produced from the precipitation, was only 24 percent near the entry point to Lake Mendocino. After this event, soil moisture increased to 97 percent. As a result, runoff efficiency rose to 45 percent in the December 29–31 atmospheric river event and stayed mostly above 40 percent in subsequent events. 

Several stations throughout the state measure soil moisture (Figure 17), and CW3E is working with DWR and FIRO water agencies to augment the soil moisture network to improve information about watershed state. This information will also help to improve streamflow forecasts.

Figure 17. Soil Moisture Measurement Sites

A current map of the soil moisture sites in California, color coded by who manages them.

What about La Niña?

La Niña is a natural climate fluctuation with strengthened Pacific trade winds and equatorial upwelling and cooler than normal surface and subsurface waters occupying the eastern half of the tropical Pacific. Importantly, La Niña usually produces a set of global teleconnections, typically including an underdeveloped Gulf-of-Alaska/Aleutian Low and a less-than-active branch of North Pacific storms that track into the southern portion of California and across the southwestern United States (Figure 18). La Niña conditions have persisted for the third successive year, and the previous two winters, 2020–2021 and 2021–2022, exhibited this pattern with fewer storms and below average precipitation across the southwestern United States. Although the 2022–2023 winter had an underdeveloped Aleutian Low, a spell of vigorous Pacific storminess, in the form of atmospheric rivers, was well developed with low pressure extending from offshore of California well inland across the western United States.

Interestingly, from a historical perspective, La Niña periods are often associated with peak streamflows in Central and Northern California (Figure 18).

Figure 18. Historical Relationship Between Peak Streamflow and El Niño/La Niña State

The circles on the map indicate the size of the largest flow at the gauge locations relative to the average peak flow. The color is an indication of the El Niño or La Niña state, with darker blues representing stronger La Niñas and darker red representing stronger El Niños. Several of the largest flows in Central and Northern California have occurred during weak La Niña events, while Southern California’s largest flows are associated with El Niños.


What's Next?

  • Each atmospheric river has distinct characteristics that determine the landfall location, timing, and strength, which affect the associated precipitation. CW3E’s West-WRF model will continue to improve forecasting by integrating enhanced scientific understanding and applying novel technological advancements, including supercomputing and machine learning.
  • Analyzing the transition from dry to wet at the onset of this atmospheric river family will improve future subseasonal (2–4 week) forecasts.
  • The AR Recon season will begin earlier and needs to ensure coverage during extreme atmospheric rivers, such as the family that began during the 2022 holiday season.
  • CW3E and DWR are exploring integration of atmospheric river forecasts and forecast tools into hydrologic modeling and emergency operations to better support water management and flood hazard mitigation.
  • 73 California reservoirs are being screened for FIRO suitability to improve resilience to drought and floods. Improving forecast skill will enable greater adoption of FIRO.
  • Continued analysis of the 2023 water year will help researchers determine the impact of AR Recon and the influence of large-scale to mesoscale dynamics on forecasts at varied lead times. In addition, CW3E will assess hydrologic models as the snowpack melts.

This is the first time the State Operations Center used advanced forecasting, including CW3E’s atmospheric river forecast information, in planning for allocation of emergency resources. The Joint Operations Center benefitted from an embedded CW3E meteorologist who provided on-site expertise throughout the atmospheric river events.

This figure shows the number of continuous days of atmospheric river conditions along the U.S. west coast, allowing for a one-day gap, and the associated maximum intensity over the duration of the event. The years shown in text are water years for events that exceeded the 95 th  percentile for either duration or intensity, while the black dots are all other events that did not meet these criteria. The nine-member atmospheric river family had the longest atmospheric river conditions in the 70-year record. Figure provided by K. Guirguis and based on the atmospheric river catalog of Gershunov et al., 2017, doi:10.1002/2017GL074175.

AR scale forecast from the U.S. National ensemble forecast (GEFS) issued on January 14 for San Diego. A total of 27 of the 30 ensemble members show an AR 3 making landfall, all within 6 hours of the actual start time.

The ensemble forecasts issued on December 29 at 4 pm PT (00 Z) for Sly Park located in the Consumes Basin. The top panel is from the U.S. National model (GEFS), the middle is the European model (ECMWF), and the bottom is from West-WRF. The red line is the observed precipitation while the dark lines are the max and min of the ensemble forecasts and the dash line in the median. By building on both the GEFS and ECMWF, West-WRF’s large ensembles better forecasted the possibility of an extreme precipitation event than either of the other two ensemble forecasts alone.

The line shows the skill (continuous ranked probability skill score) of ensemble based probabilistic forecasts between December 25 and January 18 for precipitation over 1 inch at lead times from 1 day (on the left) to 6 days (on the right). The close the number skill is to 1, the greater the skill. The U.S. national model (GEFS) is shown in blue, the European model (ECMWF) is shown in green, the West-WRF ensemble is shown in red, and the purple line is when machine learning is applied to the West-WRF ensemble after the forecast has been issued.

Flight tracks (lines) and datasets (markers) collected by aircraft during the January 2023 sequence. Colors represent dates flown.

Lake Mendocino dam operators used FIRO to maintain storage at the top of the “FIRO Pool” in January 2023, enabling storage of 11,000 more acre-feet of water than would otherwise be allowed. 

Water captured in the Water Conservation Pool (20,000 acre-feet) is drained slowly to promote capture by OCWD for groundwater recharge. This figure shows multiple opportunities for stormwater capture during and immediately after the atmospheric river family. Water in storage above the Water Conservation Pool is managed by the U.S. Army Corps of Engineers according to flood risk management rules. Data source: U.S. Army Corps of Engineers.

Approximate locations of aerial coverage for new AQPI radars.

The new AQPI Santa Clara X-band radar image from the January 9–10 atmospheric river is shown on the left and the previous KMUX NEXRAD radar image is shown on the right. Warmer colors indicate higher intensity rainfall. The new AQPI radar image shows more accurately and in more detail which areas around San Jose are experiencing high rainfall rates.

A current map of the soil moisture sites in California, color coded by who manages them.

The circles on the map indicate the size of the largest flow at the gauge locations relative to the average peak flow. The color is an indication of the El Niño or La Niña state, with darker blues representing stronger La Niñas and darker red representing stronger El Niños. Several of the largest flows in Central and Northern California have occurred during weak La Niña events, while Southern California’s largest flows are associated with El Niños.