CW3E Publication Notice: Deep Learning Forecast Uncertainty for Precipitation over Western US

CW3E Publication Notice

Deep Learning Forecast Uncertainty for Precipitation over Western US

March 24, 2023

The growing popularity of deep learning does not miss the numerical weather prediction community. CW3E researcher Weiming Hu, along with the co-authors Mohammadvaghef Ghazvinian, William E. Chapman (NCAR/UCAR Climate & Global Dynamics Lab), Agniv Sengupta, F. Martin Ralph, and Luca Delle Monache, recently published a paper titled “Deep Learning Forecast Uncertainty for Precipitation over Western US” in the Monthly Weather Review. The work contributes to the goals of CW3E’s 2019-2024 Strategic Plan to support Atmospheric River (AR) Research and Applications through the use of Emerging Technologies, namely Deep Learning-based post-processing algorithms.

This study focuses on uncertainty quantification for daily precipitation forecasts using a Deep Learning architecture, Unet, under the model output post-processing framework. For training, it leverages CW3E’s 34-year Reforecast dataset based on the West Weather Research and Forecasting (West-WRF) model available at a 3 km spatial resolution. Precipitation ground truth is obtained from the Parameter Elevation Regression on Independent Slopes Model (PRISM), which is a daily gridded precipitation dataset over the continental US at a 4 km spatial resolution. The Unet is trained to learn the non-linear relationship between the West-WRF and PRISM to calibrate precipitation forecasts. Unet has been tested for four water years conditioned on the state of the ENSO, namely 1997 (an El Niño year), 2011 (a La Niña year), 2013, and 2016 (ENSO neutral years) and compared with other benchmark methods involving parametric methods (censored, shifted Gamma distribution and mixed-type meta Gaussian distribution) and non-parametric methods (Analog Ensemble).

The main contribution of this work is the generation of probabilistic forecasts up to 4-day local lead times from a deterministic model, like West-WRF, with improved accuracy and reliability. Fig. 1 shows the study region (panel a), the deterministic (panels b and c), and probabilistic (panel d) skills of the proposed Unet model. Two metrics are used for deterministic verification – the Root Mean Square Error (RMSE) and the Pearson correlation. The Continuous Ranked Probability Score (CRPS) is used as the probabilistic metric. For all metrics, their skill scores are calculated against climatology with the higher being better. Unet consistently shows outperformance over other benchmark methods across all lead days.

Figure 1: (a) shows the study domain. (b, c, d) show deterministic (RMSE, correlation) and probabilistic (CRPS) verification aggregated from four test years and all grid points from the study region.

Fig. 2 presents the verification over space using the Brier score which is a metric that measures the accuracy of probabilistic forecasts. The skill score is compared with three different benchmark methods. Positive values (in green shading) indicate skill improvement in the Unet compared to the benchmark. Panels (a, b, c) reflect the model skill for detecting rain events (>1 mm). More importantly, panels (d ~ i) illustrate how Unet produces more accurate forecasts for extreme events (the 95-th and 99-th percentiles) over areas that are typically susceptible to strong precipitation.

Figure 2: Brier skill scores of Unet averaged from all lead days for three thresholds, 1 𝑚𝑚 (first row), 95% (second row), and 99% (third row) of the location-specific climatological distribution, against MMGD (first column), CSGD (second column), AnEn (third column); the fourth column is generated from PRISM. (j) shows a map of PoP using 1 𝑚𝑚 as the threshold, whereas (k) and (i) show the precipitation map for the 95-th and 99-th percentiles, respectively.

We also investigated the performance sensitivity to data volume sizes and found that Unet continues to learn non-linear relationships better than traditional methods and improves its performance as more data becomes available.

Hu, W., M. Ghazvinian, W. E. Chapman, A. Sengupta, F. M. Ralph, and L. Delle Monache, 2023: Deep Learning Forecast Uncertainty for Precipitation over Western US. Mon. Wea. Rev., https://doi.org/10.1175/MWR-D-22-0268.1, in press.

CW3E Event Summary: 20-22 March 2023

CW3E Event Summary: 20-22 March 2023

23 March 2023

Click here for a pdf of this information.

Atmospheric River Brings More Rain and Flooding to California and Arizona

  • An atmospheric river (AR) formed over the subtropical Northeast Pacific and made landfall in Southern California on 20 Mar
  • A surface cyclone developed on the cold side of the AR and underwent rapid intensification as it approached the Bay Area
  • AR2 conditions were observed in coastal San Diego County and southern Arizona
  • The AR produced at least 1–3 inches of precipitation across coastal Southern CA and North Central AZ, with the highest amounts (> 4 inches) observed in the San Gabriel Mountains, CA and Coconino County, AZ
  • Rainbands wrapping around the low-pressure center produced > 4 inches of precipitation in a 24-hour period in portions of San Mateo and Santa Cruz Counties
  • Lower snow levels during this storm (compared to the two previous storms) allowed for significant snowfall accumulations (> 12 inches) in the higher terrain of the Transverse Ranges
  • Two weak tornadoes were reported in California during this storm; an EF0 on 21 March in a trailer park near Carpinteria, CA and an EF1 on 22 March in an industrial area of Montebello, CA, both causing structural damage to multiple buildings
  • According to ABC7 News Bay Area, at least 5 people died during this storm, all struck by falling trees. Two fatalities were reported in San Francisco and one in Oakland, Portola Valley, and Rossmoor respectively. (Link to ABC7 News Bay Area Reporting)
  • Damage due to high winds, heavy rain, and flooding were reported across Southern California and in parts of Arizona

Click images to see loops of GFS IVT/IWV analyses

Valid 0000 UTC 19 March – 0000 UTC 22 March 2023

Click image to see loops of NEXRAD Imagery from KMUX

Valid 1800 UTC 21 March – 2345 UTC 21 March 2023


 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Summary provided by S. Roj, C. Castellano, S. Bartlett, and J. Kalansky; 23 Mar 2023

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

*Outlook products are considered experimental

Ranking Atmospheric Rivers: New Study Finds World of Potential

Ranking Atmospheric Rivers: New Study Finds World of Potential

March 21, 2023

In a new study, titled Global Application of the Atmospheric River Scale, CW3E collaborators Bin Guan (UCLA) and Duane E. Waliser (JPL), along with CW3E director Marty Ralph explore a wider geographic application of the atmospheric river (AR) scale for characterizing strength and communicating potential impacts of ARs. This work contributes to the goals of CW3E’s 2019-2024 Strategic Plan to lead AR Research and Applications because the paper proposes the first globally uniform AR scale for facilitating communication about ARs and intercomparing across different regions.

Read more about this study here.

(a) Examples of atmospheric river (AR) ranking (color shading) based on all ARs detected by tARget version 3 (referred to as the algorithm method) at an arbitrary 6-hr time step of the MERRA-2 reanalysis. Each contiguous area indicates a unique AR. Gray shading indicates grid cells that fall within the boundary of ARs detected by the algorithm, but did not receive an AR rank due to failing integrated water vapor transport (IVT) intensity and/or event duration thresholds; see Section 2 for details. The unfilled (filled) circles mark selected coastal (oceanic) areas used in subsequent analysis. (b) A specific AR from panel (a). Color shading shows the maximum IVT of the event at each grid cell, with the color corresponding to the AR rank if the event duration is between 24 and 48 hr. White (black) outlines show locations where the event is <24 hr (≥48 hr), for which the actual AR rank will be decreased (increased) by 1 from that implied by the color shading. Gray shading indicates locations not considered for AR ranking due to not meeting duration (12 hr) and/or IVT (250 kg m−1 s−1) thresholds. (c) The actual AR ranks received at each grid cell for the case shown in panel (b) after the duration-based adjustments mentioned above are made at the locations marked by white/black outlines. The white/black outlines are repeated from panel (b) to facilitate comparisons.

CW3E Event Summary: 9-15 March 2023

CW3E Event Summary: 9-15 March 2023

20 March 2023

Click here for a pdf of this information.

Atmospheric Rivers Produce Heavy Precipitation and Flooding in California

  • Two atmospheric rivers (ARs) made landfall over California during 9–15 Mar
  • These ARs were characterized by the transport of very warm, moist air from the tropical North Pacific into the midlatitudes
  • The first AR brought AR3 conditions (based on the Ralph et al. 2019 AR Scale) and IVT magnitudes > 750 kg m-1 s-1 to Monterey and Santa Cruz Counties
  • The second AR brought AR2 conditions and IVT magnitudes > 500 kg m-1 s-1 to Central and Southern California
  • The heaviest precipitation fell during the first AR in the Central and Southern Sierra Nevada, with some locations recording > 12 inches in a 3-day period and sustained precipitation rates > 0.5 inches/hour
  • High freezing levels limited snowfall accumulations below 7,000 feet in both storms
  • Snow survey stations located above 7,000 feet recorded 7-day SWE increases > 12 inches
  • High reservoir inflows prompted dam operators to open the main spillway and increase releases to > 15,000 cfs at Oroville Dam after the first AR
  • The combination of heavy rainfall and melting snowpack led to widespread riverine flooding across Northern and Central California
  • The most destructive flooding occurred along the Pajaro River in the community of Pajaro, CA, and along the Kern River in Kernville, CA
  • An EF-1 tornado caused structural damage to mobile homes and farms in Tuolumne County

Click images to see loops of GFS IVT/IWV analyses

Valid 0000 UTC 9 March – 0000 UTC 16 March 2023


 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Summary provided by C. Castellano, S. Bartlett, S. Roj, and J. Kalansky; 20 Mar 2023

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

*Outlook products are

CW3E Welcomes Ali Wolman

CW3E Welcomes Ali Wolman

March 20, 2023

Ali Wolman joined CW3E’s Field Team as a Lab Assistant in January 2023. She travels to field sites and launches weather balloons to collect profiles of temperature, pressure, wind & moisture data, which is used to better understand and predict atmospheric rivers (ARs). When she isn’t in the field, Ali is helping organize the field team’s equipment & spaces. Ali also is working to update hands-on climate science programming for K-12 classes.

Prior to SIO, Ali worked in ocean research and science education. After earning a BS in Biology at Cal Poly San Luis Obispo focused on marine biology & conservation, she continued to work around the Central Coast. Ali stayed involved at Cal Poly in research as a SCUBA tech and worked to develop video-data collection technology using artificial intelligence to identify organisms in videos of the ocean floor. Ali also worked for NOAA, teaching marine biology at an informal science teaching center in San Simeon, CA. Before joining CW3E, she taught hands-on, classroom science at the Elementary Institute of Science in San Diego, CA.

CW3E AR Update: 17 March 2023 Outlook

CW3E AR Update: 17 March 2023 Outlook

March 17, 2023

Click here for a pdf of this information.

Atmospheric River Forecast to Bring Precipitation to Southern California

  • An atmospheric river (AR) developing in the Eastern Pacific is forecast to make landfall in Southern California on 20 March and persist in the region until 21 March
  • AR2 conditions (based on the Ralph et al. 2019 AR Scale) are forecast near San Diego, CA with AR1 conditions possible northward to San Luis Obispo, CA
  • AR3 conditions are forecast in Southern Arizona as a result of the inland penetration of this AR
  • There is still uncertainty between the GEFS and ECMWF EPS in the timing and intensity of AR conditions, including direction of IVT, which is leading to uncertainty in watershed precipitation forecasts over the next 7 days
  • The 00Z ECMWF is forecasting 3.10 inches of precipitation over the Santa Ana watershed over the next 7 days, while the 00Z GFS is forecasting 1.81 inches; a difference of 1.29 inches
  • The NWS Weather Prediction Center (WPC) is forecasting >1.5 inches of precipitation for Southern California coastal regions, >3 inches in the San Gabriel Mountains, and 1 inch over Central Arizona
  • Experimental excessive rainfall outlooks have been issued by the WPC for 21 March with a marginal risk (at least 5%) of rainfall exceeding flash flood guidance for much of Southern California and most of the Central Valley as well as Central Arizona

Click images to see loops of GFS IVT and IWV forecasts

Valid 0000 UTC 17 March – 1200 UTC 22 March 2023


 

 

 

 

 

 

Summary provided by M. Steen, S. Roj, and S. Bartlett; 17 March 2023

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

*Outlook products are considered experimental

CW3E AR Update: 13 March 2023 Outlook

CW3E AR Update: 13 March 2023 Outlook

March 13, 2023

Click here for a pdf of this information.

Strong Atmospheric River to Bring Additional Heavy Precipitation and Flooding to California

  • Another atmospheric river (AR) will impact California this evening into Wednesday
  • AR3 conditions (based on the Ralph et al. 2019 AR Scale) are forecast near Santa Cruz, CA, while AR2 conditions are forecast over much of the remainder of coastal California between Sonoma County and San Diego County
  • Strong upslope moisture flux will support heavy precipitation over the Sierra Nevada, Central California Coast Ranges, and Transverse Ranges
  • The NWS Weather Prediction Center (WPC) is forecasting at least 3–7 inches of precipitation in these areas over the next 3 days, as well as 1–3 inches of precipitation in the lower elevations of coastal Southern California
  • Moderate precipitation amounts of 1–3 inches are also forecast in the higher terrain of the Colorado River Basin due to inland penetration of the AR
  • Heavy rain falling in areas with saturated soil conditions and high river/stream levels will likely produce major hydrologic impacts throughout California
  • The NWS WPC has issued a moderate-to-high risk of rainfall exceeding flash flood guidance in the vicinity of the Sierra Nevada, Coast Ranges, and Transverse Ranges through Wednesday morning
  • The California–Nevada River Forecast Center is currently forecasting 13 stream gages to rise above flood stage over the next 5 days
  • Similar to the previous storm, freezing levels will be relatively high, and a majority of the precipitation is forecast to fall as rain in watersheds along the Sierra Nevada
  • Extreme winter storm impacts are expected in the higher terrain of the Sierra Nevada, where up to 5 feet of snow are possible

Click images to see loops of GFS IVT and IWV forecasts

Valid 1200 UTC 13 March – 1200 UTC 16 March 2023


 

 

 

 

 

 

 

 

 

Summary provided by C. Castellano, S. Bartlett, J. Kalansky, and S. Roj; 13 March 2023

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

*Outlook products are considered experimental

CW3E AR Update: 10 March 2023 Outlook

CW3E AR Update: 10 March 2023 Outlook

March 10, 2023

Click here for a pdf of this information.

Atmospheric River Continues to Bring Moisture and High Snow Levels to California with another Atmospheric River Forecast to Make Landfall in Northern California Shortly After

  • Current atmospheric river (AR) made landfall in Northern California around 10am PT 9 March and will continue to push moisture into California over the next couple days
  • IVT will values remain above 250 kg m -1 s -1 off the US West Coast ahead of the next AR which is forecast to make landfall on 14 March and bring AR2/3 conditions (based on the Ralph et al. 2019 AR Scale) to the majority of coastal Central California
  • There is substantial uncertainty between the GFS and ECMWF global models in the timing and location of AR conditions, including direction of IVT, which is mostly due to how each model handles the location of a pair of mid-level troughs
  • As compared to the GEFS, the ECMWF EPS has more confidence in a broader area of AR3 conditions for coastal areas around the San Francisco Bay area
  • The NWS Weather Prediction Center (WPC) has forecast 48-hour precipitation totals >2 inches over the Coast Ranges of Northern and Central California, >3 inches for the Western Transverse Range of Southern California and >5 inches over the Sierra
  • Experimental excessive rainfall outlooks have been issued by the WPC for Monday through Wednesday with a moderate risk (at least 40%) of rainfall exceeding flash flood guidance in the Northern Sierra and parts of the Northern California coast for Monday-Tuesday, and a moderate risk for the Sierra Nevada and Big Sur coast for Tuesday-Wednesday
  • The 00Z GFS is forecasting 11.31 inches of mean areal precipitation in the Upper Yuba watershed over the next 7 days, while the 00Z ECMWF is forecasting 8.11 inches over the same watershed, however, these totals include precipitation forecasted to fall during the current AR
  • Forecast freezing levels are expected to remain quite high (~2,000m) in the Sierra Nevada. This may lead to high fractions of most Sierra watersheds exposed to rain-on-snow

Click images to see loops of GFS IVT and IWV forecasts

Valid 0000 UTC 10 March – 1200 UTC 16 March 2023


 

 

 

 

 

 

 

Summary provided by S. Roj, S. Bartlett, J. Cordeira, and J. Kalansky; 10 March 2023

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

*Outlook products are considered experimental

CW3E Winter 2022-23 Recap

CW3E 2022-23 Winter (Dec – Feb) Recap

March 8, 2023

Click here for a pdf of this information.

Wet Winter Helped Alleviate Drought Conditions and Produced Massive Snowpack

  • A family of 9 ARs made landfall between 26 Dec and 19 Jan and brought about half the of California’s precipitation during these three weeks.
  • Parts of the Central and Southern Sierras and some transverse ranges received 60+ inches.
  • Much of Central California received 100-150% of water year normal precipitation. Northeast and southeast California have received 50% or less of water year normal precipitation.
  • According to the US drought monitor there has been 3-5 class improvement in drought conditions over Central California and at least a class drought improvement over much of California.
  • Winter precipitation eliminated the 3-year drought deficit in 13% of the state.
  • SWE in the Southern Sierra went from 14% of April 1st at the beginning of December to 198% of April 1st SWE at the end of February.
  • By Mar 1, 9 reservoirs exceeded the historical average capacity, up from 1 on Dec 1 2022.
  • Water storage (reservoir +snowpack) in the Western Sierras increased from four-fold, with the greatest gains in the Southern Sierras.
  • As part of AR Recon there were 33 Intensive Observing Period (IOPs) which included 25 C130 flights, 18 GIV flights and the release of 1,186 dropsondes.
  • Precipitation and SWE totals during this winter are similar to Dec 2016– Feb 2017, with differences in the spatial patterns. In particular, northern CA received more precipitation in 2017 than this current winter.
  •