In recent years, the finance sector has made progress in understanding the different representative concentration pathways (RCPs) and shared socio-economic pathways (SSPs) used in Intergovernmental Panel on Climate Change (IPCC) reports. The range of different pathways allow for the consideration of a variety of warmer futures, each with different outcomes and impacts. Financial and macroeconomic modellers have sought to link these broad scenarios to financial and economic factors. This linkage comes from estimates of the overall levels of damage associated with each scenario and the resultant impact. The magnitude of physical hazards and their evolution over time can inform asset-level estimates of risk under a given scenario. Damage functions are a way to provide a probabilistic view of potential harms an asset may face from a physical hazard. Arriving at an overall damage function¹ for an asset requires aggregation of multiple hazards and consideration of the relationship between them (Neumann et al., 2020).

As the impacts of climate change have become clearer (Pörtner et al., 2022), the relationship between climate risk and financial risk has grown more apparent. In 2017, the international Financial Stability Board (FSB)² introduced the Task force on Climate-related Financial Disclosures (TCFD, 2017). The TCFD guidance aims to provide financial markets with the information they need to appropriately price climate risks and ensure the efficient allocation of capital. While many firms³ have endorsed the TCFDs recommendations for disclosure, the usefulness of these disclosures depends on the quality of the data and analyses they include.

In many regions, asset-level data is sparse or non-existent, making it particularly challenging to evaluate performance under changing conditions. Even in locations with good data, most models do not consider interaction effects between hazards or the second-order economic effects of an event (e.g., delinquency rates after hurricanes; Gromowski, 2018). For example, a model may provide the total damages to an asset from flooding but not account for the months of lost revenue post-flood. Another challenge relates to the time horizons under which physical risks materialise. While extreme events can occur at any time, major changes in the prevalence and severity of these events often require years of observational data to be identified, for example the prevalence, frequency, and intensity of hurricanes (Masson-Delmotte et al., 2021). As a result, financial actors can struggle to integrate the insights from their physical risk analyses into current and future strategy.

Nevertheless, there are some areas where NTCPs already offer potential to inform financial decisions. One example is where climate risk can affect lenders' credit risk exposure through its impact on the value of assets used as loan collateral, such as mortgaged properties (Bellrose et al., 2021). In one instance, a meta-analysis of studies on the effect of flood risk on house prices indicates that properties located in the 100-year floodplain have a price discount of −4.6% (Beltrán et al., 2018). Another study of the impacts of winter floods in 2013–14 in the UK indicated that properties affected by prolonged flooding saw decreases in sales prices of between −4.2 and −2.6% (Garbarino and Guin, 2020). The study also found that these decreases were not reflected in lenders' property valuations for mortgage refinancing, and that their valuations were therefore biassed upwards.

Climate risk can also increase financial stress on borrowers, by creating costs to repair damage or reducing the incomes which borrowers rely on to repay their loans (Figure 1). A before/after analysis of delinquency rates (i.e., the percentage of loans within a financial institution's loan portfolio whose payments are delinquent) for mortgages in areas affected by Hurricane Harvey showed that damage to properties caused by the hurricane affected the ability of homeowners to make their mortgage payments in the months following. Mortgages for properties estimated to have hurricane damage saw a 205% increase in 90+ day delinquency compared to delinquency rates 6 months prior to the hurricane (Gromowski, 2018). As we shall further examine in the following section, decadal predictions already offer skilful predictions of multiyear fluctuations in hurricane-related losses.

Hurricane damage in the United States (US) is rising as population and property in coastal regions increase with Atlantic hurricanes now accounting for most of the insured losses in the US. Hurricane activity also varies on decadal timescales (Landsea, 2007; Klotzbach and Gray, 2008) with hurricane damage in the US varying by up to a factor of 5 between recent decades. Skilful decadal predictions of hurricane activity offer the re-insurance sector an opportunity to better manage these varying risks (Lockwood et al., 2021).

Factors controlling hurricane formation include sea surface temperature (SST), temperatures in the upper troposphere, atmospheric stability and vertical wind shear in the upper troposphere (e.g., Emanuel, 2007; Latif et al., 2007; Saunders and Lea, 2008; Garner et al., 2009). On decadal timescales these factors are influenced by North Atlantic SST (Goldenberg et al., 2001; Dunstone et al., 2011) which is highly predictable (Doblas-Reyes et al., 2013; Yeager and Robson, 2017), enabling skilful long-range predictions of hurricane frequency (Smith et al., 2010; Caron et al., 2014, 2018; Hermanson et al., 2014).

With real-time decadal prediction now an operational WMO activity (Smith et al., 2013; Kushnir et al., 2019; Hermanson et al., 2022), forecasts of hurricane activity and total US economic losses for the coming 5 years were made in partnership with Willis Towers Watson to facilitate adjustments to extreme event occurrence rates, inform capital allocation strategies and influence premium pricing. These forecasts also provide opportunities to structure multi-year deals to take advantage of this predictability.

In Figure 2, linear regression of past observations onto an index based on predicted North Atlantic tropical ocean temperatures (Vecchi et al., 2011) is used to provide a probabilistic hurricane activity forecast. US damage forecasts (not shown) use a modified regression to account for the non-linear relationship with damage. Retrospective forecasts reveal skill (rank correlation ~0.7), outperforming persistence forecasts (using the last 5 years to predict the next five). Figure 2 shows that an active period is predicted for 2021–2025.

The El Niño Southern Oscillation (ENSO) has a significant impact on agriculture, water, and health (Zebiak et al., 2015). Some of the nations that experience profound ENSO impacts are geographically small, located in or near the tropics, and have economies dependent on a few climate-sensitive industries. In addition, there are documented relationships between ENSO and movements in commodity prices (Brunner, 2002; Ubilava and Holt, 2009), agricultural production (Gutierrez, 2017), and broader macroeconomic indicators (Cashin et al., 2017). In that regard, Kitsios et al. (2022), have demonstrated that econometric forecasts of certain commodity spot prices can be further improved with the inclusion of climate predictions of the relevant indices of climate variability i.e., exogenous factors. Here the highlighted commodity is coconut oil, and the exogenous factor is the Niño4 index (Rasmusson and Carpenter, 1982) provided by climate simulations. The Niño4 index is chosen as coconut oil is predominantly produced in the tropics where the physical influence of ENSO is most direct.

The commodity prices are accessed from the World Bank database⁴ with real commodity log-returns based on monthly inflation rates calculated using the G7 averaged consumer price index.⁵ ENSO forecast data was generated using the Climate re-Analysis and Forecast Ensemble (CAFE) system (O'Kane et al., 2019, 2020, 2021a,b). The model representation of the log-returns of the commodity price allows for additive seasonality, autoregressive processes, and lagged exogenous ENSO (Niño4) factors (for details see Kitsios et al., 2022). The autoregressive models are built using available data from January 1980 to December 2020, with all combinations of lags assessed up to a lag of 12 months. Combinations of lags that produce autocorrelated residuals or are shown to be heteroskedastic are excluded. Of the remaining combinations, the most parsimonious model was determined as that which minimises the Bayesian Information Criterion (BIC) (Schwarz, 1978) from within a class of models for which there are multiple lag combinations that have similar BIC values.

Out of sample correlations between the actual and forecast log returns produced by this econometric model are illustrated in Figure 3. Three variants are illustrated: one using perfect future ENSO information (perfect-ENSO, blue line) as an upper bound on skill; one with no ENSO information (no-ENSO, red line); and finally, the CAFE forecasts (GCM-ENSO, cyan line). Per start date the perfect-ENSO and GCM-ENSO cases have the same lag structure and the same values for the model coefficients, with the coefficients only learnt using data prior to the start date. In the no-ENSO case the same endogenous lags are adopted, but the model coefficients are re-calculated such that this model best fits the data. For the perfect-ENSO and GCM-ENSO cases, the hollow black circles indicate forecast times at which ENSO is found to be Granger or predictively causal, with a statistically significant reduction (to a 95% confidence) in their residual sum of squares (RSS) as compared to the no-ENSO case. This is after accounting for the increased number of parameters (i.e., F-test).

The no-ENSO case has statistically significant positive correlations for only the first 4 months. The perfect-ENSO case has significant correlation and ENSO is found to be Granger causal for the entire 2-year period. This is, however, not a plausible forecast since one cannot know ENSO perfectly into the future. The GCM-ENSO case is a realistic and fair forecast, has significant correlations for the entire 2-year forecast window, and has a lower RSS than the no-ENSO case for most lead times. The associated significant RSS reductions at forecast months 21 and 22 should not be interpreted as a return of skill, but rather the RSS for the no-ENSO case increasing faster than that of the GCM-ENSO variant. This is consistent with the known forecast time dependent predictability of ENSO on inter-annual timescales (Luo et al., 2008; Knight et al., 2014; Dunstone et al., 2020). Over the first 8 months the GCM-ENSO and perfect-ENSO cases have identical statistics. This is because the coconut oil commodity has one exogenous lag of 8 months. This means that up to this point only historical ENSO information is used in all auto-regressive with external factor models, and it is not until a forecast time of 9 months that an actual ENSO forecast needs to be made. It is at this point that the forecast error measures begin to differ.

In summary, all commodity forecasts adopting exogenous ENSO factors out-perform those that did not. This indicates the importance and demonstrates the utility of incorporating NTCP information in a practical application of predicting commodity returns.

The Wupper catchment water board manages water level and quality of the Wupper River in Germany. To improve water management strategies multiyear predictions of the Standardised Precipitation Index (SPI, McKee et al., 1993) are required as this is correlated to dam water levels in the catchment. Multiyear annual and hydrological seasonal means are needed at high spatial resolution to manage 14 dams in an area of 800 km².

Statistical downscaling from 200 km to ~11 km (Kreienkamp et al., 2018) was applied to all ensemble members of the global decadal prediction system MPI-ESM-LR (Mauritsen et al., 2019). Relationships between large-scale temperature, relative humidity, and geopotential height in Central Europe to local-scale temperature and precipitation in Germany have been considered using NCEP reanalyses (Kalnay et al., 1996) and high-resolution HYRAS observational data (Rauthe et al., 2013).

Downscaled precipitation has been recalibrated by applying the Decadal Climate Forecast Recalibration Strategy (DeFoRest, Pasternack et al., 2017) to correct bias, drift and conditional bias and adjust ensemble spread.

The skill of probabilistic decadal SPI predictions in reproducing past HYRAS observations has been evaluated using correlation and ranked probability skill score (RPSS, Ferro et al., 2008) in comparison to the reference prediction observed climatology over 1962–2020. The SPI drought index was assessed for multi-year annual and seasonal mean predictions for the coming 3 years. A significance level of 95% is applied using 1,000 non-parametric block bootstraps and considering autocorrelation.

Results showed that high-resolution statistical downscaling transfers the prediction skill of the global prediction system to higher spatial resolution and that the recalibration enhances SPI prediction skill. Additional skillful predictions were also found for three-year seasonal means, especially so for August to October in the Wupper river catchment. The study of Paxian et al. (2022) further reveals promising 3-year mean SPI skill for several additional regions in Germany.

The feedback of the Wupper catchment water board concluded that the product is understandable and well-structured and of utility during the daily decision-making work of water boards in Germany. A high probability of drought conditions is currently forecast for 2021–2023.

Glaciers are important indicators of climate change. They also store 75% of the earth's available freshwater (Jansson et al., 2003), also acting as buffers preventing precipitation from immediately turning into runoff. They are essential parts of mountain water storage and supply downstream communities and ecosystems, upon which 22% of the global population depends (Immerzeel et al., 2020). Quantifying changes in mass balance is challenging considering glacier response times (Raper and Braithwaite, 2009). In temperate maritime climates this is generally between 15 and 60 years but can exceed a century for high latitudes (Cuffey and Paterson, 2010). Decadal time scales are rarely considered in glacier modelling, even though they are critical for water resource management (Frans et al., 2016; Lane and Nienow, 2019). By predicting glacier mass balance on decadal timescales, it is possible to predict runoff, an important input for water resource decisions.

Here multi-decadal glacier discharge is examined using the Open Global Glacier Model (OGGM; Maussion et al., 2019) forced by monthly temperature and precipitation from the Decadal Climate Prediction Project (DCPP, Boer et al., 2016; Maussion et al., 2019). Results of a 10 member hindcast ensemble of the sixth version of the Model for Interdisciplinary Research on Climate (MIROC6) are analysed where all ensemble members, initialised each year for the period 1960–1999, are bias corrected for model drift as in Boer et al. (2016) and referenced to observations from the Climatic Research Unit gridded Time Series version 4 (Harris et al., 2020). OGGM is forced for 279 reference glaciers, from a global database of 216 000 glaciers, all of which have observational records exceeding 5 years (Van der Laan et al., 2022).

As a case study, we consider the Hintereisferner alpine glacier in Austria (Strasser et al., 2018). Based on the coupled model presented by Stoll et al. (2020), mass loss of Hintereisferner contributed 9.4% of discharge over 1991–2010. The remaining discharge is fed by the melting of other glaciers in the basin, snowmelt, and rain (Weber et al., 2010). The total catchment glaciation decreased from 44% in 1969 to 38% in 2009, leading to a decrease in contribution to annual streamflow (Müller et al., 2009). The glaciation of the catchment is expected to approach zero within the 21^st century, significantly altering the hydrological balance in the catchment (Strasser et al., 2018). Decadal scale forecasts in this basin would allow insight into the deglaciation patterns, streamflow contribution and water resources for the downstream Inntal community.

Figure 4 compares observed and hindcast mass balance for the Hintereisferner. The temperature and precipitation amplitudes and trends correspond well to observations, showing the merit of using the decadal scale ensemble mean. The utility of this approach and skill of decadal hindcasts for glacier and runoff modelling is especially evident in the well-captured negative cumulative mass balance (Figure 4D) and in the overall Pearson correlation coefficient between the annual cumulative time series of 0.98, MME of 457.4 mm w.e.⁶ and MAE of 1,373 mm w.e.

The current results are promising for skilful glacier mass balance decadal predictions. The method could also be expanded to all glaciers in an essential water tower region such as the Himalayan mountain range (Pfeffer et al., 2014; Immerzeel et al., 2020). This promises to provide necessary information for the planning of water resource allocation for hydropower, agriculture and drinking water on decadal timescales.

Wheat is the most harvested cereal crop in the world, contributing to the food security of many countries (Toreti et al., 2019). The usefulness of NTCPs for decision-making in the wheat-sector has recently been demonstrated (Solaraju-Murali et al., 2021). Unfavourable hot or dry conditions can induce water stress affecting both grain quality and yield, especially when occurring during sensitive development stages of the plant. NTCPs have been shown to skilfully anticipate multiannual drought and heat stress conditions in several areas with user-relevant agro-climatic indices such as the Standardised Precipitation Evapotranspiration Index (SPEI6) and Heat Magnitude Day Index (HMDI3) prior to wheat harvesting. Figure 5 presents the predicted likelihood of the most likely tercile category of multiannual averaged SPEI6 over global wheat harvesting regions for 2015–2019 from forecasts initialised in November 2014. The predictions show an increase in drought (below-normal category) events over most of the wheat-growing regions for the period (Figure 5A) and there is good agreement with observed SPEI6. This implies opportunities to help stakeholders in the wheat sector during their decision-making processes on multi-annual timescales.

These encouraging results have led to the development of tailored forecast products. A prototype product for multi-year drought forecasts for the coming 5 years over the global wheat harvesting regions has been made available for the agriculture user community (https://climate.copernicus.eu/decadal-predictions-agriculture). Such an initiative aims to establish transdisciplinary partnerships based on decadal predictions to help agricultural users during their decision-making process.

Grapes are grown under open-field conditions in vineyards to supply fruit with characteristics for wine production with market-demanded sensory profiles (aroma, taste, and flavour). A vineyard is meant to last several decades but vineyard longevity is strongly conditioned by climate, making decadal predictions potentially very relevant. Thousands of vine varieties and landraces are used, each with its own climatic suitability, supporting viticulture's global spread. Products of the grapevine account for close to 4.5% of global food trade (Van Nieuwkoop, 2019; Fortune Business Insights, 2022) while international trade of wine in 2019 reached 31.8 bn € (OIV, 2020)⁷.

Choice of grape varieties is highly dependent on climate. Plants need 2–4 years until they become fully productive and climate in this period is critical for correct rooting and aerial development (Webb et al., 2011; Sgubin et al., 2019). Despite irrigation, depleted water reserves left vineyards vulnerable to water stress. Loss of leaves and arrest of secondary metabolism responsible for maturity is indicated by colourless berries. Grapevines close their stomata during daytime temperatures above 35°C, stopping photosynthesis, causing lower sugar accumulation in berries, malic acid breakdown from nocturnal respiration, and dehydration and desiccation, decreasing yields. These grapes are unfit for quality wine. An ability to predict temperature and rainfall would greatly enhance risk management for such investment. The capacity to estimate probabilities for the frequency of extreme events that cause, for example, water stress in grapes, would improve farm management and mitigate risks.

The ability to foresee the evolution of the climate for the impending 1–10 years is key to timely implementation of adaptation measures or change of viticulture business models (Gishen et al., 2016; Graça and Gishen, 2022). Grape and wine production would become more resilient, decreasing the use of plant pesticides and irrigation requirements. Other benefits include sizing of long-term facilities such as wineries and warehouses, managing wastewater and recycling and transport logistics (Graça, 2019).

Useful seasonal forecasts have already been demonstrated for wine production (Santos et al., 2020). Simulations of using seasonal forecasts to inform viticulture decisions have shown the need to consider the context of users and how they use forecasts, highlighting the high value of co-creating forecast services (Vigo et al., 2021). Demand for decadal predictions will be fostered by wider use of seasonal predictions with user-centric communication (Calmanti et al., 2021; Khosravi et al., 2021). Relevant temporal and spatial granularity will also be key for adoption of multi-year forecast information as well as the transparent disclosure of uncertainty to inform user-relevant decisions (Sgubin et al., 2019).

For more than a decade, physical information about the future ocean has been used to produce ecological forecasts at both short (Hobday and Hartmann, 2006; Hazen et al., 2017) and multidecadal timescales (Tommasi et al., 2017a). Ecological forecasts exploit relationships between physical variables and biological responses to predict ecologically relevant information. This includes species' spatial distribution, abundance, and phenology (Payne et al., 2017). In recent years, ecological forecasts at seasonal time scales have also been developed for fisheries (Hobday et al., 2011; Payne et al., 2017), aquaculture (Spillman and Hobday, 2014) and conservation (corals, Spillman, 2011; Liu et al., 2019; Spillman and Smith, 2021) applications. Decision-makers such as fishers/farmers, managers and policymakers in these sectors are now including forecast information in their planning (Hobday et al., 2016; Kaplan et al., 2016). These plans can be based on forecasts about a species':

While In theory decision-making can benefit from prior knowledge of all these factors, in practise most forecast systems to date have only focused on species spatial distributions (Payne et al., 2017).

The high skill of oceanic decadal-scale predictions has been apparent from early studies of climate prediction (IPCC AR5 WG1, Chapter 11, Kirtman et al., 2013). Oceanic predictability is typically higher than predictions for either the land surface or the atmosphere (Merryfield et al., 2020), and therefore represents one of the most promising areas for the application of decadal forecasts. Significant decadal-forecast skill is seen throughout much of the global ocean (Keenlyside et al., 2008; Power et al., 2021) and particularly in the North Atlantic (Figure 6). A majority of work has focussed on the open ocean rather than the coastal shelf-seas where the most of human interactions with the ocean take place. Nevertheless, studies that have examined the predictability of shelf seas and coastal marine ecosystems have shown potentially useful skill at interannual to multi-year timescales (Stock et al., 2015; Tommasi et al., 2017b; Koul et al., 2021, 2022; Miesner et al., 2022).

Due to the length of the available observational record and relatively high predictability, SST is one of the most utilised variables in models of marine ecological responses (Salinger et al., 2016; Merryfield et al., 2020). However, predictability is not limited to SST: multi-annual predictability is also seen in ocean heat content (Yeager et al., 2012) and salinity (Bethke et al., 2021; Figure 6) and large-scale indices such as Atlantic multidecadal overturning (Matei et al., 2012; Hermanson et al., 2014), the dynamics of the North Atlantic sub-polar gyre (Wouters et al., 2013), Atlantic multidecadal variability (Smith et al., 2020) and ENSO (Luo et al., 2008; Barnston et al., 2012, 2019; Knight et al., 2014; Dunstone et al., 2020). This physical predictability also extends into biogeochemical components including primary productivity and nutrient concentrations (Séférian et al., 2014; Gehlen et al., 2015; Park et al., 2019): indeed, most marine ecosystem drivers can potentially be predicted on at least the multi-year time-scale (Fransner et al., 2020; Frölicher et al., 2020).

The potential for operational marine ecological decadal forecast products builds on those already developed at seasonal time scales (Payne et al., 2017). Most of the operational products at seasonal timescales consider changes in the spatial distribution of fish species, with mounting evidence that such predictability could be extended to multi-year timescales (Payne et al., 2022) as shown in forecasts of Mackerel (Scomber scrombus) habitat around Greenland (Figure 7). Decadal predictability of the total catches of fish species at the large-marine ecosystem level has also been shown (Park et al., 2019) including some stock-specific cases (Årthun et al., 2018).

Exploratory studies have also shown that the use of ecological forecasts can generate economic value for users. One such example examines the use of multi-annual forecasts in the management of Pacific sardine (Sardinops sagax) in the California Current system. Tommasi et al. (2017c) showed that the use of forecasts, in conjunction with established management systems, could improve both the long-term yield and sustainability of the fishery. Economic valuation tools have also shown a potential net-positive yield of annual productivity forecasts for sandeel (Ammodytes sp.) in the North Sea (Kiaer et al., 2021). However, there is currently only limited use of this information by decision makers of forecast products irrespective of timescale (Skern-Mauritzen et al., 2016; Payne et al., 2017), in contrast to other fields (e.g., agriculture, Asseng et al., 2012).

The “northern” cod (Gadus morhua) inhabits northwest Atlantic waters off the coast of Canada's Newfoundland and Labrador (NL) province. These shelf and upper slope waters are dominated by the cold Labrador Current, which varies on interannual to decadal timescales (Han et al., 2019) and is linked to the North Atlantic subpolar gyre which is relatively predictable on multi-year timescales (e.g., Robson et al., 2018). Because Labrador Current temperatures influence cod populations (Lehodey et al., 2006), this species could be a promising candidate for the application of seasonal to decadal forecasts to living marine resource management (e.g., Tommasi et al., 2017b). Of particular note is that especially cold ocean temperatures in the region appear to have been a factor in the economically damaging collapse of the northern cod population in the 1990s (Mullowney et al., 2019), recovery from which is still incomplete (Rose and Walters, 2019).

The possible utility of multi-year climate forecasts for management of this species has been explored using hindcasts from the CanCM4 model (Merryfield et al., 2013). Seasonal forecasts from this model have previously been shown to be relatively skilful at ranges of up to 12 months in predicting SST anomalies in the Labrador-Newfoundland large marine ecosystem that is home to the northern cod habitat (Hervieux et al., 2019).⁸

Ocean potential temperatures are considered along a series of transects spaced approximately every two degrees in longitude from 56°N to 46°N, extending eastward from the NL coast and encompassing the southward component of upper ocean flow associated with the Labrador Current over the shelf and upper slope to a depth of 661 m. Northern cod populations generally winter on the upper slope beneath the cold sub-zero core of the Labrador Current, and migrate in late spring westward into cooler shelf waters to feed (Lilly et al., 2013). Because the extent of coldest Labrador Current waters influences both the available optimal habitat for growth and recruitment and the availability of prey (Rose and Rowe, 2015), predictions of the anomalous cross-sectional area at temperatures >0°C in each of the transects are examined.

Figure 8A shows annual time series from the ORAS4 ocean reanalysis, the CanCM4 analysis, and first-year (lead 0) forecast values for the 50°N transect, near the latitudinal centre of what had been the prime region for this fishery. The multi-year to decadal timescale variations of these three timeseries align reasonably well, particularly in the earlier decades, and a prolonged negative anomaly is evident in each of the time series during the early 1990s, indicative of generally colder temperatures and reduced favourable habitat above 0°C. Figure 8B shows anomaly correlation (AC) as a function of lead year for the decadal forecasts, as well as for persistence of both verification datasets and their lagged correlations with each other. Forecast anomaly correlations (AC) based on verification with the CanCM4 analysis (blue) remain statistically significant and generally higher than persistence of the analysis through lead times of 5 years, whereas skill based on verification with ORAS4 is significant only in the 1^st year.

These results indicate the potential utility of decadal forecasts for predicting subsurface ocean conditions relevant to the management of the northern cod population out to at least 1 year ahead and possibly longer, although further studies are needed to understand the differences between skill measures based on ORAS4 and the CanCM4 analysis.

Realising the full potential of NTCP in the marine sector will require overcoming numerous barriers within, and external to, the field of climate prediction. The most important of these is a need to raise awareness in the user and marine science communities about the current and constantly improving skill of climate predictions, as is being done for seasonal forecasts (Hobday et al., 2016). Considerable investment is still needed to support decadal forecasting and integration of those forecasts into decision support tools.⁹ Overall, collaboration between forecast developers and the marine resource sector is needed to support forecast-based tactical and strategic decisions that reduce environmental risk over annual to decadal time scales.