Stakeholder forecast needs for efficient water management

The invited stakeholder talks highlighted the needs and potential value of reliable prediction of precipitation (J. Jones, CDWR), snowpack, and streamflow (B. Svoma, SRP; J. Lhotak, NOAA CBRFC), and their associated impacts on both short-term and long-term horizons for operational decisions (Fig. 1). Efficient management of water resources involves political and socio-economic considerations, namely, public health and safety, minimum supply requirements during extreme drought conditions, and prioritization of water use for balancing consumption in fisheries, agriculture, municipalities, etc. The talk by J. Jones (CDWR) highlighted that important policy decisions, like initial water supply allocations for the California State Water Project, must be made at the beginning of the wet season in the western United States, which receives ∼75% of its annual precipitation from November through March. In spring, snowmelt runoff forecasts during the snowmelt season (March through July) help water managers in planning strategies to meet the state’s water supply needs. Skillful forecasts, including those of heatwaves and warm rains, can greatly improve water managers’ ability to handle challenges related to capturing and storing runoff induced by spring flows in systems with significant snowmelt runoff.

View Full Size

Lead times necessary for different drought preparedness and response measures vary over daily to monthly time scales during the climatological wet season (winter), demonstrated here as a function of forecast certainty and discretion that water managers have in decision-making. Note that initial water allocations must be made several months in advance when forecast uncertainty is the greatest. (Image taken from J. Jones’s workshop talk.)

Citation: Bulletin of the American Meteorological Society 103, 10; 10.1175/BAMS-D-22-0146.1

• Download Figure
• Download figure as PowerPoint slide

Emerging research-and-operations partnerships and related success stories were highlighted, which provide promising pathways moving forward. One such collaboration involves the CDWR, NASA JPL, and academic institutions including CW3E/Scripps, IRI, and UArizona. This multi-university project led by NASA JPL seeks to provide water resource managers in the western United States with innovative experimental tools for S2S precipitation forecasting (talk given by M. DeFlorio, UCSD; DeFlorio et al. 2021). It is the sole project in the World Weather Programme/World Climate Research Programme S2S Prediction Project’s (Vitart et al. 2012) Real-Time Pilot Initiative (Robbins 2020) (http://s2sprediction.net/xwiki/bin/view/dtbs/RealtimePilot) that focuses on water in the western United States; White et al. (2017, 2022) survey S2S forecast applications across various socioeconomic sectors. The talk by C. Talbot (USACE) drew attention to the Forecast Informed Reservoir Operations (FIRO) initiative, which is another important western U.S. collaboration (Jasperse et al. 2015, 2020; Talbot et al. 2019). FIRO involves the incorporation of weather and hydrological forecast information into water control manuals and water management decisions of retaining or releasing water at reservoir sites in the western United States after careful viability assessments, e.g., at the Russian River basin (Lake Mendocino), the Santa Ana River basin (Prado Dam), and the Yuba–Feather River basins (New Bullards Bar and Oroville Dams).

In addition to the above applied scientific research partnerships, the concept of crowdsourcing competitions to improve S2S forecasts for water management was presented by K. Nowak (USBR). The U.S. Bureau of Reclamation has partnered with the U.S. National Integrated Drought Information System (NIDIS) to organize two real-time S2S Climate Forecast Rodeo competitions (https://www.usbr.gov/research/challenges/forecastrodeo.html), which spanned over 13 months each during 2017–18 (Rodeo I) and 2019–20 (Rodeo II). Their goal was to identify forecast tools that enhance S2S forecasting at lead times ranging from 15 to 42 days that would allow water managers to better prepare for hydrologic regime shifts, e.g., the onset of dry or wet weather extremes.

Workshop recommendations included continued support for research that bridges science and applications and guides the development of decision support tools at S2S lead times, which has historically received less research funding in comparison to weather and climate scales.

Subseasonal to seasonal predictability assessments

Forecasting at lead times of weeks to months remains a challenging scientific problem. The success in predicting the state of the major driver of interannual variability—El Niño–Southern Oscillation (ENSO), most notably for the 1997/98 El Niño and 1998/99 La Niña episodes—has often led to heightened expectations for skillful forecasts of precipitation at longer lead times (noted by D. DeWitt, NOAA CPC). But ENSO explains only a fraction of the interannual variation in precipitation over California, and its signal-to-noise ratio is such that unexpected outcomes can often occur (Wang et al. 2017; Cash and Burls 2019). This was evident during the major El Niño event of 2015/16, which failed to provide the expected hydrologic relief to the Southwest, and a devastating drought continued instead. The skill of available precipitation forecasts degrades beyond a 2-week lead time, rendering them suboptimal for decision-making. S2S forecasting, thus, remains far from being a solved problem.

Research on better understanding and representing influential atmospheric, oceanic, and land surface processes in our Earth system that give rise to predictability remains fundamental for generating seamless forecasts that connect weather and climate scales. This session covered various key physical phenomena, from landfalling atmospheric rivers (ARs) to the tropical–extratropical teleconnections associated with the Madden–Julian oscillation (MJO), interactions between the stratosphere and the troposphere, teleconnections tied to the state of the tropical oceans, along with land–atmosphere interactions governing predictability.

ARs, which are long and narrow plumes of enhanced horizontal water vapor transport in the lower troposphere, were shown to be influential for winter precipitation in the U.S. West with a multitude of hydrometeorological impacts. These storms often act as “drought busters,” alleviating existing drought conditions and replenishing reservoir levels. Recent progress in the domain of AR research was presented at the workshop (Z. Zhang, UCSD; D. Nash, UCSB), including assessments of subseasonal skill in hindcast systems of state-of-the-art dynamic models. Encouraging results indicate useful model skills in predicting AR occurrences up to 2–3 weeks in advance. For subseasonal forecasts, the MJO—the dominant component of tropical intraseasonal variability—provides a source for extended range predictability in the U.S. West in the boreal winter by exciting atmospheric Rossby wave trains giving rise to extratropical teleconnections. While recent investigations have shown that forecast models can skillfully predict the MJO up to 3–4 weeks ahead, specific advances were presented in this session related to changes in intensity and frequency of extremes over this region under different phases of MJO (J. Wang, UCSD). For example, during phases 6 and 7 of the MJO associated with enhanced convection over the western Pacific, there is substantially more AR activity in the earlier part of the wet season (October–December) compared to the latter half (January–March). MJO impacts are also strongly modulated by the stratospheric quasi-biennial oscillation (QBO). Under the easterly phase of the QBO, an eastward-extended trough tends to be found over the West Coast, facilitating more organized moisture transport and a higher frequency of landfalling ARs. In this regard, the role of the stratosphere and, in particular, sudden stratospheric warming events (SSW) of the Northern Hemisphere winter, which have lagged influence on North American hydroclimate, is increasingly being recognized. More research is needed to better understand the mechanistic processes governing the stratosphere-troposphere coupling and the causes and variability of SSW events. Similarly, the modulation of wet/dry extremes by different phase combinations of the MJO and QBO has not been explored fully with the potential for extending predictability out to months.

On seasonal time scales, the climate system derives predictability from slowly varying, influential components with large thermal inertia, e.g., sea surface temperatures (SSTs). Anomalous SST conditions exert remote influences through a variety of teleconnections associated with changes in the atmospheric circulation patterns, as well as by inducing low-level temperature and moisture advection by climatological winds. In recent years, the marginal success in predicting the deviations of winter seasonal rainfall over California has primarily been associated with the success in predicting the state of the ENSO. In addition, the role of the Indian Ocean in mediating the ENSO-related precipitation teleconnection to North America has also been noted. E. LaJoie (NOAA CPC) presented an experimental water year outlook tool conditioned on the Oceanic Niño Index and demonstrated the value of capturing forecast skill as a function of the direction of the forecasted anomalies relative to the observed. Yet, as was highlighted in this workshop by X. Jiang (UCLA), even a skillful ENSO prediction would only lead to an explanation of about 25% of California’s precipitation variability. The low prediction skill of winter precipitation is attributed largely to model shortcomings in simulating the ENSO-independent, anomalous Pacific circulation, which is an important driver of year-to-year California precipitation variability. Promising results are nonetheless starting to emerge from a variety of methodologies, including nudging experiments (W.-T. Hsiao, CSU) that correct tropical and high-latitude fields toward the observed state. A more comprehensive understanding of the processes driving the anomalous atmospheric circulation patterns associated with California precipitation needs to be obtained, and related investigations into the precursors and their detection are warranted. Workshop recommendations emphasized the development of forecast tools that consider cross-time-scale interactions among different climate drivers (also noted in the talk by Á. Muñoz, IRI). The need to develop consistent process-oriented diagnostics/metrics that target the underlying physical mechanisms was also endorsed.

Emerging forecast methodologies

Another focus of the workshop was on the emerging technical methods, including empirical, dynamical, hybrid, and machine learning approaches. This session highlighted a suite of experimental subseasonal and seasonal forecast products (both probabilistic and deterministic) with potential for uptake as decision support tools by the applications community.

Probabilistic forecasts derived from dynamical models are based on the spread of the ensemble members, which relate to the small perturbations in model initializations that grow rapidly with time. Dynamical forecasting talks included IRI’s calibrated subseasonal forecasts based on the Subseasonal Experiment (SubX; Pegion et al. 2019) and calibrated seasonal forecasts based on the North American Multi-Model Experiment (NMME; Kirtman et al. 2014), which provide seamless prediction capabilities starting from 1 day to 6 months in advance. In addition, weather regimes that seek to capture the recurrent patterns of atmospheric circulation (dominant ridges/troughs) provide further temporal context and daily guidance with opportunities for action based on seamless days (talks given by B. Singh and Á. Muñoz, IRI). On the other hand, statistical forecasts have historically relied on linear relationships between lagged spatial fields of predictors and the predictand. Presentations highlighted advances in statistical forecasting beyond the canonical methods, including approaches that attempt to capture the recurrent patterns, including lower-frequency sources, governing predictability in space and time (via the use of extended-empirical orthogonal function analysis; A. Sengupta, UCSD). Innovative hybrid techniques (dynamical models combined with empirical strategies) were also presented. One project, led by P. Gibson (NIWA), focused on the viability of training various machine learning algorithms, including deep neural networks, on large ensembles of climate models to circumvent the issue of limited data record length and then to test the model (trained in model space) on historical observed data for predicting widespread spatial clusters of precipitation. Another talk, given by W. Scheftic (UArizona), presented a two-stage hybrid post-processing system for combining seasonal climate model data with empirical climate relationships (oceanic and atmospheric teleconnection patterns) to create ensemble forecasts of snow mass, temperature, and precipitation. A machine learning framework was also proposed by A. Altinok (NASA JPL) for localized processing of precipitation data to train models at the individual grid cells and consider the effect of local topography.

Many of the forecast tools discussed at this workshop (developed using hindcast skill assessments) routinely generate and communicate winter precipitation outlooks at the beginning and midway through each wet season (at venues like CDWR’s late fall Winter Outlook Workshop) and support decisions on efforts needed for statewide water conservation campaigns and drought preparedness activities.

Concluding remarks

As the weather and climate across the western United States become more variable and droughts more frequent and widespread, water managers will increasingly need longer-lead-time information on potential future water supply conditions to better prepare for S2S extremes. The S2S time scale represents a frontier that shows promise for providing reliable outlooks weeks to months in advance for supporting efficient water management. The workshop offered the opportunity for climate forecasters and researchers to interact directly with water management professionals, identify forecast priorities, and collaboratively take stock of the challenges associated with improving predictions. Together with success stories of science-based research and operations partnerships highlighted at the workshop, the new forecasting approaches will hopefully foster further investigation of underlying processes using Earth science data and ultimately culminate in the development of decision support tools.