LETTER • OPEN ACCESS You may also like

- On the key influence of remote climate

On the assessment of the wave modeling variability from Tropical Cyclones, North
and South Atlantic mid-latitude storms on
uncertainty in wave climate projections the Senegalese coast (West Africa)
Rafael Almar, Elodie Kestenare and Julien
Boucharel
To cite this article: Hector Lobeto et al 2023 Environ. Res. Lett. 18 124006 - Contributions to 21st century projections of
extreme sea-level change around the UK
Tom Howard, Matthew D Palmer and Lucy
M Bricheno

- Future changes and seasonal variability of

View the article online for updates and enhancements. the directional wave spectra in the
Mediterranean Sea for the 21st century
Andrea Lira-Loarca and Giovanni Besio

This content was downloaded from IP address 147.161.167.112 on 12/03/2024 at 09:26

 Environ. Res. Lett. 18 (2023) 124006 https://doi.org/10.1088/1748-9326/ad0137

LETTER

On the assessment of the wave modeling uncertainty in wave

OPEN ACCESS
climate projections
RECEIVED
2 May 2023 Hector Lobeto1,∗, Alvaro Semedo2, Melisa Menendez1, Gil Lemos3, Rajesh Kumar4, Adem Akpinar5,
REVISED Mikhail Dobrynin6 and Bahareh Kamranzad7
7 September 2023
1
ACCEPTED FOR PUBLICATION
IHCantabria—Instituto de Hidráulica Ambiental de la Universidad de Cantabria, Santander, Spain
2
8 October 2023 Department of Coastal and Urban Risk & Resilience, IHE Delft Institute for Water Education, Delft, The Netherlands
3
Universidade de Lisboa, Faculdade de Ciências, Instituto Dom Luiz, Lisboa, Portugal
PUBLISHED 4
10 November 2023 Centre for Climate Research Singapore, 36 Kim Chuan Rd, Singapore 537054, Singapore
5
Department of Civil Engineering, Bursa Uludag University, Bursa, Turkey
6
Deutscher Wetterdienst (DWD), Hamburg, Germany
Original content from 7
Department of Civil and Environmental Engineering, University of Strathclyde, Glasgow G11XJ, United Kingdom
this work may be used ∗
under the terms of the Author to whom any correspondence should be addressed.
Creative Commons E-mail: [email protected]
Attribution 4.0 licence.
Any further distribution Keywords: wave climate, climate change, uncertainty, wave modeling
of this work must
maintain attribution to Supplementary material for this article is available online
the author(s) and the title
of the work, journal
citation and DOI.
Abstract
This study investigates the epistemic uncertainty associated with the wave propagation modeling in
wave climate projections. A single-forcing, single-scenario, seven-member global wave climate
projection ensemble is used, developed using three wave models with a consistent numerical
domain. The uncertainty is assessed through projected changes in wave height, wave period, and
wave direction. The relative importance of the wave model used and its internal parameterization
are examined. The former is the dominant source of uncertainty in approximately two-thirds of
the global ocean. The study reveals divergences in projected changes from runs of different models
and runs of the same model with different parameterizations over 75% of the ensemble mean
change in several ocean regions. Projected changes in the wave period shows the most significant
uncertainties, particularly in the Pacific Ocean basin, while the wave height shows the least. Over
30% of global coastlines exhibit significant uncertainties in at least two out of the three wave
climate variables analyzed. The coasts of western North America, the Maritime Continent and the
Arabian Sea show the most significant wave modeling uncertainties.

1. Introduction characterization of the wave climate and its variability

is crucial for a range of applications, including infra-
Ocean wind waves play a key role in the impact the structure design and assessment of coastal impacts,
ocean may have on human activities. Wind waves among others.
transport more than half of the energy propagating Ocean wind waves are projected to change over
across the ocean surface [1, 2], thus conditioning the the twenty-first century under a warming climate
shape and size of the elements confronting them, both [13]. Climate change is affecting the main forcing of
in the open ocean (e.g. offshore structures [3] or wind waves, the surface wind [14, 15], changing the
vessels [4]) and in the coastal zone (e.g. coastal pro- transmitted energy [16] and, hence, the characterist-
tection infrastructures [4, 5]). In line with the latter, ics of the waves. In addition, the ice melting accel-
the energy transported by waves shapes the coastline, eration in high latitudes triggered by the increasing
eroding and moving materials, seeking to reach a nat- temperatures [17] is generating an expansion of wave
ural equilibrium [6]. Extreme events of wind waves generation areas [18, 19], thus inducing an increase
may therefore significantly impact offshore activities in the wave energy propagating from the poles [20].
such as route shipping or the offshore wind industry The assessment of the future behavior of wind
[7, 8], and the coast, through flooding episodes [9, waves under climate change has been a compelling
10] and major erosion events [11, 12]. An accurate subject of analysis for the last two decades [20–30],

© 2023 The Author(s). Published by IOP Publishing Ltd

Environ. Res. Lett. 18 (2023) 124006 H Lobeto et al

encouraged by the severe implications these changes of these features. Model internal parameterization
may have, especially for extreme events [31, 32]. can also be tuned, leading to variations between runs
The standard approach to conduct these studies is of the same model [43]. In this context, predefined
based on wave climate projections [24–26, 33]. These internal parameterizations, known as source term
products represent future wave climates, for different packages, are available. These source terms packages
scenarios, developed using forcing drivers from global comprise a set of equations that address the wave gen-
climate models (GCMs) or regional climate models eration and dissipation, also including some tunable
(RCMs). Multiple studies on the matter have led to parameters. Nevertheless, these source terms pack-
a community consensus about the projected beha- ages do not encompass the entire model parameteriz-
vior of climatological mean wave conditions in several ation, as some other issues, such as the wave–bottom
ocean regions, such as an increase in significant wave interaction, fall outside of them and can also affect the
height (H s ) in the Southern Ocean and in the tropical wave model outcomes.
Eastern Pacific, and a decrease in the North Atlantic To date, only a very few studies have addressed
Ocean, Northwestern Pacific and Mediterranean Sea the uncertainty associated with wave modeling in
[21, 34]. The projected changes in extremes are, projected changes in wave climate [21, 22]. These
however, still characterized by great uncertainty studies assessed the contribution of wave modeling
[30, 35, 36]. uncertainty to the total uncertainty in the projec-
The uncertainty associated with the projected ted changes, distinguishing its significance from other
changes in wind waves based on wave climate pro- sources such as those associated with the GCMs and
jection ensembles is normally assessed through the the future scenarios. Nevertheless, a specific study
agreement in the sign and magnitude of the changes that focuses on isolating and analyzing in detail
projected for the different ensemble members [24, the epistemic uncertainty related to wave model-
37]. Nevertheless, this integrated assessment is ing has not yet been conducted. Thus, several ques-
unable to unravel the origin of the uncertainties tions still arise and remain unanswered, such as the
found. Several sources of uncertainty are present in actual influence of wave model selection on projec-
assessing projected changes in wave climate condi- ted changes in wave climate, the extent to which the
tions. Uncertainty propagates through all the stages parameterization of the numerical model affects the
involved in this assessment (figure 1), a process changes, and which of these sources of uncertainty is
known as the uncertainty cascade [38, 39]. Lower more significant. The aim of this study is to address
steps within the uncertainty cascade will there- these and related questions by isolating the epistemic
fore accumulate the uncertainty inherited from top uncertainty associated with wave modeling, examin-
sources [40, 41]. ing the relative importance of its main sources in
Beyond the aleatoric uncertainty associated with wave climate projected changes, and quantifying its
the chaotic natural variability of the climate variables magnitude.
involved [38, 39], the uncertainty in wave climate
projected changes also integrates the socio-economic 2. Methods
scenario uncertainty, the uncertainty related to GCMs
and the epistemic uncertainty associated with the 2.1. Wave climate projection ensemble
wave modeling part. These sources of uncertainty are This study uses a wave climate projection ensemble
usually embraced by including representative mem- forced by a single run (r1i1p1f1) of the CMIP6 [44]
bers of different configurations. For example, it is GCM EC-EARTH3 [45], which has been proven to
common practice to include several scenarios and perform well in reproducing climate metrics [46], and
GCM forcings to consider these uncertainties in the a single future climate scenario (SSP5-8.5 [47, 48]).
assessment [30, 42]. The use of different wave models Runs use three-hourly surface wind fields and daily
and/or wave model setups, however, is uncommon in ice coverage fields as forcings (more details in previ-
studies of this kind. ous articles [43, 49]). The time slices 1995–2014 and
This study particularly focuses on the epistemic 2081–2100 are used as baseline and future periods,
uncertainty associated with the wave modeling com- respectively. The wave climate projection ensemble
ponent of the simulations in wave climate projections. is produced using the most popular wave models
Wave models (e.g. SWAN, WAM) reproduce the gen- within the climate community: WaveWatch III v6.07
eration, propagation and dissipation of wind waves [50] (hereinafter WW3), WAM v4.6 [51] and SWAN
through numerical equations, but have inherent sim- v41.20AB [52, 53].
plifications that cause the numerical output to diverge The three models used are third-generation spec-
from reality. Model differences mainly arise from tral wave models that share a similar theoretical back-
the numerical scheme used to solve the governing ground. The main characteristic of this type of mod-
equations, the number of wave propagation features els is not restricting the shape of the wave spectrum as
modeled (e.g. bottom friction, white-capping, ice in previous generations. All of them are based on the
interaction) and the equations used to represent each solution of the action balance equation (equation (1))

2
Environ. Res. Lett. 18 (2023) 124006 H Lobeto et al

∂N ∂cx N ∂cy N ∂cσ N ∂cθ N Stot 2.3. Analysis of variance

+ + + + = (1)
∂t ∂x ∂y ∂σ ∂θ σ The relative contribution to the total uncertainty
between the wave model used and the model
parameterization—i.e. the inter-model and intra-
where N (x, y, σ, θ) is the wave action density, c model uncertainties, respectively, is estimated
is the propagation celerity of the wave energy, σ through a one-way analysis of variance (one-way
is the intrinsic frequency and θ is the propaga- ANOVA), similarly as it has been done in previous
tion direction. Stot is the total sum of source terms studies [22, 40]. ANOVA method is used to compute
of different physical processes parameterized in the the explained variance (EV; equations (2) and (3))
model. of each source of uncertainty based on the sum
WAM and WW3 use explicit numerical propaga- of squares (SS) between individual member runs
tion schemes limited by time steps due to Courant– [68, 69].
Friedrichs–Lewy (CFL) criteria, whereas SWAN uses
an iterative approximation to a fully implicit scheme SSinter
EVinter (%) = × 100, (2)
to avoid such limitations [54, 55]. As a result, WW3 SStotal
and WAM models are more efficient in regional and
global domains, whereas SWAN model is computa- EVintra (%) = 100 − EVinter (3)
tionally more efficient in coastal areas. Models are
also different regarding the processes solved and the where SStotal is the total SS and SSinter is the SS between
parameterizations used to solve them. SWAN model wave models:
differs from WAM and WW3 on including coastal- XX

SStotal = ∆ij − ∆ ¯ 2, (4)
specific parametrizations (e.g. triads, quadruplets) to
i j
solve processes in limited water depths and complex
coastal areas. All this makes WAM and WW3 models X

SSinter = nj ∆ ¯ 2,
¯j−∆ (5)
to be typically used for global [56–61] and regional
j
[42, 62–64] scales, while SWAN model is extensively
used to develop coastal-scale studies [65–67]. where ∆ij is the relative projected change for run i of
Each ensemble member is developed using a wave ¯ j is the mean relative projected change of
model j, ∆
model with a different numerical parameterization. model j runs, ∆¯ is the overall mean projected change
Differences lie in the source term package selected and nj is the number of runs of each propagation
to develop each ensemble member. Default paramet- model.
ers are employed for each simulation. The ensemble
comprises seven members, integrating four WW3 2.4. Quantification of uncertainty
runs developed with the source term packages ST2, The inter-model and intra-model uncertainties are
ST3, ST4 and ST6, two SWAN runs with the source independently quantified by assessing the differ-
term packages ST1 and ST6 and one WAM run ences between the projected changes from different
with the Cycle 4.5 source term package. Each source wave models and different model parameterizations,
term package parameterization implements different respectively. Discrepancies are measured through the
approximations for the wind–wave interaction and relative mean difference (RMD) metric, computed as:
the wave dissipation. A succinct definition of each
source term package is provided in supplementary ∆n − ∆m
RMD (%) = ¯ × 100, (6)
material. ∆
Each ensemble member produces a global three-
where ∆n and ∆m represent the relative change in
hourly time series of significant wave height (H s ), ¯ represent the
runs n and m, respectively; and ∆
mean wave period (T m ) and mean wave direction
ensemble mean relative change.
(θm ), with one-degree spatial resolution. Grid nodes
The inter-model uncertainty (Ie ; equation (4)) is
covered by ice for more than 30% of time are not
quantified by computing, first, the RMDs between
considered in the analysis. A global validation against
runs from each possible combination of wave models
buoy and reanalysis data has been undertaken [49]. A
(i.e. WW3–SWAN, WW3–WAM and WAM–SWAN).
detailed description of the numerical configuration of
Thus, the number of RMDs between two different
the experiments can be found in two previous articles
wave models is equal to the number of runs for the
[43, 49].
first model multiplied by the number of runs for the
second one. Since the number of runs differs between
2.2. Projected changes in wave climate models, so does the number of RMDs for each model
Projected changes are computed as the relative pro- combination. Thus, a weighted mean and a weighted
jected change (in %) between the baseline period and standard deviation are computed, to avoid results
the future period, normalized by the historical value. biasing, as follows:
In the case of wave direction, the relative projected
changes are normalized by 360◦ . Ie = x̄w ± σw (7)

3
Environ. Res. Lett. 18 (2023) 124006 H Lobeto et al

where, x̄w is the weighted mean uncertainty and σw can, therefore, only be attributed to the numerical
is the weighted standard deviation of the uncertainty, parameterization of the model—in other words,
estimated as: to the wave modeling epistemic uncertainty. This
is illustrated in figure 1. The colored boxes in the
X
i =N
1 uncertainty cascade depict the sources of uncertainty
x̄w = RMDi · , (8)
N · ni associated with the projected changes assessed in this
i =1
study. In contrast, the gray boxes represent sources of
v
u i =N uncertainty not linked to the discrepancies observed
uX 1
σw = t
2
RMDi − RMDi · (9) among ensemble members.
N · ni The wave modeling epistemic uncertainty can be
i =1
seen as the addition of two sources of uncertainty:
where N is the number of combinations, RMDi is the (i) the selection of the numerical model and (ii) the
mean RMD for the combination i and ni is the num- internal parameterization of the model. Regarding
ber of elements within combination i. the former, each numerical model has some spe-
The intra-model uncertainty (Ia ) quantifica- cific features not shared with the others, thus indu-
tion is analogous to the inter-model uncertainty cing differences in the results. On the other hand,
(equation (4)). RMDs are computed between model despite all model runs sharing most of the numer-
runs from the same wave model with a different ical parameterization, the numerical approximation
numerical parameterization. However, since the dif- of some specific processes may differ. This study
ferent number of model runs would lead to a strong considers both sources of uncertainty by including
imbalance between the number of RMDs for each numerical simulations developed with different wave
wave model (six for WW3, one for SWAN, none for models and with different parameterizations of the
WAM), only the runs for WW3 are considered to same model (see section 2).
quantify Ia . The discrepancies between ensemble members
are addressed in figure 2. Figure 2 shows the regional
2.5. Significance of uncertainty and global (ocean regions are defined in figure SM1)
The relevance of uncertainty is assessed to identify uncertainty cascades [40] for projected changes (see
areas where it may have a greater impact. This section 2) in mean H s , T m and θm (panels (a)–(c),
is achieved by evaluating the magnitude of uncer- respectively). Each cascade is divided in three levels.
tainty, the projected changes, and the discrepan- From top to bottom, each level displays the ensemble
cies among members. Thus, a specific ocean loca- mean projected change, the wave model mean pro-
tion (i.e. ocean grid point) is considered to have sig- jected changes, calculated as the mean change from
nificant uncertainty if the mean uncertainty value all members of a specific model, and the projec-
is greater than 25% (the same approach is applied ted change for each ensemble member, along with
for inter- and intra-model uncertainties). In addi- the 5%–95% range (assuming normal distribution).
tion, uncertainty values are deemed significant if the Results for 99% percentile H s (H s 99) are also assessed
absolute ensemble mean projected changes exceed the and shown in figure SM2. The width of the dis-
absolute global median projected change and/or if played uncertainty cascades reflects the divergence
the standard deviation of individual member pro- between ensemble members (lower level) and wave
jected changes is greater than twice the ensemble models (intermediate level). Projected changes in
mean projected change. The latter two conditions aim mean H s (figure 2(a)) show the greatest differences
to exclude regions exhibiting very high uncertainty in the North Pacific Ocean. In particular, TWNP
values, which arise from low ensemble mean values is the ocean region where the greatest differences
derived from low individual member changes. between member runs (from −9.5% to −5%) and
wave model means (from −8.5% to −4.8%) can be
3. Results seen. On the other hand, TESP shows the lowest dif-
ferences. Note that most regions show an agreement
This study isolates and quantifies the wave modeling between all ensemble members in terms of the sign
epistemic uncertainty. To that end, a single-scenario, of change. The main exceptions are TESP, ETSA and
single-forcing wave climate projection ensemble ETSI, where two out of the seven members diverge in
developed with multiple wave models and para- this change feature. Projected changes in H s 99 (figure
meterizations is used (see section 2). Using a single SM2) show the greatest differences in TWNP and
scenario and a single forcing GCM avoids attrib- TWSP. Additionally, only TNIO shows discrepancies
uting inter-member divergences to the uncertainty in the sign of change between ensemble members.
associated with the scenario and the forcing climate Projected changes in T m (figure 2(b)) show a gen-
model. In the same vein, all numerical propagation eral homogenous behavior between ocean regions
runs are developed, as much as possible, using the as most of them show 5%–95% ranges for indi-
same bathymetry and computational grid [49], hence vidual member runs lower than 2.5%. TWNP is the
avoiding model set-up discrepancies. The differences only exception, showing a 5%–95% range between

4
Environ. Res. Lett. 18 (2023) 124006 H Lobeto et al

Figure 1. General outline of the uncertainty cascade related to the assessment of wave modeling uncertainty in wave climate
projected changes. The left diagram depicts the uncertainty cascade. Colored boxes represent the sources of uncertainty linked to
the projected changes assessed in this study. Right diagram displays the configuration of the experiment in relation to the sources
of uncertainty.

ensemble members of approximately 3%. Changes clear, as there are regions where the contribution of
in mean θm (figure 2(c)) show the strongest spa- the model parameterization to the total uncertainty
tial heterogeneity across ocean regions among all the is considerably higher than the wave model used (e.g.
wave climate metrics analyzed. In general, except for TNAO, TWSP). Other regions such as TSIO, TNIO
ETNA, the differences are significantly higher in the and TESP show a split dominance between both
Pacific Ocean than in the rest of the ocean basins, sources of uncertainty. The behavior of H s 99 is, in
especially in the tropical region. On the other hand, general terms, very similar to the one for mean H s .
regions such as TNAO, TSAO and TSIO show very Main exceptions can be found in ETNP and TNAO,
good agreements between individual member runs where the intra-model uncertainty clearly dominates
and wave models, with differences lower than 1% in over the inter-model uncertainty for H s 99 and the
both cases. For completeness, the individual member opposite for mean H s . The analysis of the projected
projected changes and the ensemble mean changes changes in mean T m evidence that this parameter is
across the global ocean are included in figures SM3–6. the one in which the selection of the wave model plays
The relative importance between the wave model a more important role in contrast to the model para-
used and its internal parameterization within the total meterization in the total uncertainty found, as more
wave modeling uncertainty in wave climate projected than 75% of the regions show this behavior. Only
changes (i.e. inter-model and intra-model uncer- TENP and TWSP show opposite results, both with
tainty, respectively) is assessed through an ANOVA a relative importance of the intra-model uncertainty
(see section 2). Figure 3 presents the results, illus- above 65%. The analysis of mean θm shows a great
trating that, for example, the uncertainty in global heterogeneity in the Southern Ocean as the main out-
projected changes in mean H s is approximately 80% come. In this regard, while ETSP and ETSA show the
attributable to the chosen model and 20% to the relative importance of the inter-model uncertainty
model setup configuration. Results show an overall higher than 80%, ETSI shows the opposite behavior
higher contribution of the inter-model uncertainty to with less than 10%.
the total uncertainty with respect to the intra-model Nevertheless, results from figure 3 only informs
uncertainty—namely, the use of different models has about the relative importance of each contribut-
a greater influence on the differences found in the ing element and nothing about the total amount of
wave climate projected changes than the use of dif- uncertainty of each source. In order to compare the
ferent model parameterizations. In fact, at least 60% existing uncertainty between regions, a regional (and
of the ocean regions show a higher importance of global) quantification of both sources of uncertainty
the inter-model uncertainty for each metric analyzed (see section 2) for mean H s , H s 99, mean T m and
(69%, 62%, 85% and 62% for mean H s , H s 99, mean mean θm is provided in figure 4. For each metric,
T m and mean θm , respectively). the mean inter-model and intra-model uncertainties,
Across extra-tropical regions, the inter-model along with the confidence intervals (estimated as the
uncertainty for mean H s remains above 60% relative mean ± one standard deviation) are displayed.
to intra-model uncertainty, regardless of the region The highest uncertainties in mean H s are found
analyzed. In tropical regions this pattern is not so in TWSP, exceeding mean values of 100% for both

5
Environ. Res. Lett. 18 (2023) 124006 H Lobeto et al

Figure 2. Uncertainty cascades for (a) mean H s , (b) mean T m and (c) mean θm . Projected changes, per region and globally. Lower
levels of the cascades represent more disaggregated changes: Top level—ensemble mean relative change, intermediate level—wave
model mean relative changes, and lower level—ensemble member relative changes. Outside gray dashed lines represent the
5%–95% range. WAM—Cycle 4.5 is displayed as ST4 for the sake of simplicity.

inter- and intra-model uncertainties. The former also values lower than 15%. H s 99 shows the greatest
shows mean values over 100% in ETSA and ETSI. uncertainties in the tropical latitudes of the Indian
Note that figure 2 shows great discrepancies between Ocean, exceeding mean values of 70% for both the
SWAN and the other two wave models in the lat- inter- and intra-model uncertainties, likely due to the
ter two regions, likely causing the high inter-model higher differences between WAM and the rest of the
uncertainty values found. On the other hand, it wave models in these regions (figure 2).
is also worth noticing the low uncertainties found The inter-model uncertainty for mean T m exceeds
for projected changes in mean H s in the Northern mean values of 50% in 7 out of the 13 regions ana-
Hemisphere, especially in the Atlantic Ocean, where lyzed. This denotes T m to be the parameter for which
the inter- and intra-model uncertainties show mean the selection of the wave model causes the greatest

6
Environ. Res. Lett. 18 (2023) 124006 H Lobeto et al

Figure 3. Relative contribution to the total wave modeling epistemic uncertainty, expressed as the explained variance (in %), for
projected changes in mean H s , H s 99, mean T m and mean θm , per region and globally, between the inter-model (EV inter ,
equation (1)) and intra-model (EV intra , equation (2)) uncertainties.

Figure 4. Quantification of the inter-model and intra-model wave modeling epistemic uncertainty for projected changes in mean
H s , H s 99, mean T m and mean θm , per region and globally. Black arrows indicate values higher than 100%.

differences in the estimated projected changes. On in which low ensemble mean changes are caused by
the other hand, only two regions (ETNP and TWSP) low individual member changes, from ocean areas
show mean values of intra-model uncertainty above where ensemble mean changes are very low due to
60%. Regarding mean θm , as expected from the results the balance between strong individual change sig-
presented in figure 2, sensitive differences can be seen nals of different signs. Figure 5 depicts the ocean
between regions. Inter-model uncertainties in ETNP, areas where the inter- and intra-model uncertainties
TWSP, TESP, ETNA and TNIO exceed mean values are significant for the projected changes in mean H s ,
of 90%, whereas for the rest of the regions, it shows T m and θm (see section 2). It identifies ocean areas
mean values always lower than 40%. The same con- where the high uncertainties found are relevant due
clusions can be extracted for the intra-model uncer- to the magnitude of the projected changes and/or
tainty: while ETNP, TESP and TNIO show mean val- due to the great discrepancies between members.
ues above 60%, the rest of the regions show values Correspondingly, it facilitates the identification of
lower than 30%. ocean regions where the wave modeling uncertainty
Despite figure 4 allowing the identification of the is not critical in the assessment of wave climate pro-
regions showing the highest wave modeling uncer- jected changes. Results for H s 99 are included in figure
tainties, it precludes identifying precisely in which SM7.
areas these uncertainties are more important. The fact Results indicate that inter-model uncertainty is
that RMDs are computed by normalizing with the more important than intra-model uncertainty across
ensemble mean (see section 2), leads to large uncer- the global ocean. In this regard, the proportion of the
tainties where the ensemble mean changes are very global ocean showing a significant inter-model uncer-
low. Thus, it is relevant to distinguish between cases tainty for mean H s , T m and θm is always higher than

7
Environ. Res. Lett. 18 (2023) 124006 H Lobeto et al

Figure 5. Ocean areas showing significant (a) inter-model and (b) intra-model wave modeling uncertainty for mean H s (blue),
mean T m (green) and mean θm (orange) projected changes. Stippling indicates significant uncertainties for the three metrics
analyzed.

25%, whereas for the intra-model uncertainty the changes in mean θm , the Pacific Ocean is again the
percentages are always lower than 25%. Figure 5(a) basin where this source of uncertainty is more relev-
shows that the ocean areas where the inter-model ant, especially in the tropical and the western extra-
uncertainty is significant simultaneously for the three tropical ocean regions. The tropical North Atlantic
metrics analyzed (8% of the ocean surface) are mainly and the Arabian Sea also show significant inter-model
in the Pacific Ocean, particularly at TENP. Other uncertainties.
small ocean areas in the Atlantic Ocean (e.g. tropical The proportion of the global ocean showing sig-
northeast) and Indian Ocean (e.g. western Arabian nificant intra-model uncertainties for all the met-
Sea) also show this behavior. rics analyzed is very low (<1%; figure 5(b)). Besides,
The inter-model uncertainty in projected changes among the three metrics, mean H s shows signific-
in mean H s is notably important in the tropical Pacific ant results in the smallest proportion of the ocean
basin and the Gulf of Alaska. Some dispersed areas in (8% vs. 23% and 20% for mean T m and θm , respect-
the Atlantic and Indian Oceans also show significant ively). Ocean areas showing significant intra-model
results, such as the southernmost part of the Atlantic, uncertainty for mean H s projected changes are mainly
the seas south of Sumatra and Java and the Arabian located in tropical latitudes, in both the Pacific and
Sea. Mean T m presents the largest proportion of the Indian Oceans. Regarding the projected changes in
global ocean showing significant inter-model uncer- mean T m , ocean areas showing significant intra-
tainties (53%). Most of the Pacific basin, with the model uncertainties are sparsely distributed across all
only exception of the western extra-tropical region, ocean basins. Among them, the easternmost part of
shows this behavior. Additionally, a great propor- the Pacific Ocean shows the clearest results. Finally,
tion of the Northwest Atlantic Ocean and the tropical projected changes in mean θm show the most signi-
north Indian Ocean also show significant inter-model ficant results in the extra-tropical and eastern tropical
uncertainties for this metric. Regarding the projected regions of the Pacific Ocean.

8
Environ. Res. Lett. 18 (2023) 124006 H Lobeto et al

Figure 6. Integrated qualitative assessment of the inter-model and intra-model uncertainty significance for the assessment of
projected changes in wave climate along the global coastlines. Two upward arrows indicate that at least two out of the three wave
climate metrics analyzed show significant uncertainty in projected changes. One upward arrow indicates that one or less of the
three wave climate metrics analyzed show significant uncertainty in projected changes. The green color highlights the case where
both sources of uncertainty show no significance (NS) in wave climate projected changes. The wave climate metrics analyzed are
mean H s , mean T m and mean θm .

Significant uncertainties may have severe implic- other sources also present in assessments of this
ations where the wave climate is a key process driver, kind (e.g. GCM-related uncertainty, scenario-related
as in the coastal zone, where waves play a key role in uncertainty). The analysis has been conducted based
coastal processes such as flooding or erosion [31, 70]. on a seven-member, single-scenario, single-forcing
Figure 6 depicts qualitatively the degree of wave mod- wave climate projection ensemble. Three numer-
eling uncertainty along the global coastlines. To that ical wave models have been selected to develop the
end, the number of wave climate variables in which ensemble members (WW3, WAM and SWAN). Two
the uncertainty is found to be significant is computed primary sources of uncertainty within the wave mod-
for both inter-model and intra-model uncertainties. eling uncertainty have been independently analyzed:
Three variables have been deemed in accordance the inter-model uncertainty, which considers the dif-
with the analysis presented in figure 4: wave height ferences between models; and the intra-model uncer-
(through mean H s ), wave period (through mean T m ) tainty, which considers the differences between model
and wave direction (through mean θm ). Results indic- parameterizations. Furthermore, all members share
ate that more than 35% of the global coastlines show a consistent numerical domain with the ultimate
significant uncertainties in at least two out of the objective of reducing to the minimum the differ-
three metrics analyzed for the inter-model and/or ences between members attributable to this factor.
intra-model uncertainties (orange, purple and red Although the findings presented in this research are
in figure 6). On the other hand, 27% of the global intrinsically influenced by the number of members
coastlines does not show significant wave modeling utilized and their distribution between propagation
uncertainties (green in figure 6). The coasts of Oman, models, the ensemble framework encompasses a sub-
Iran, Pakistan and India, the coasts of the Maritime stantial number of members, developed with the
Continent, the western coasts of North America and most prevalent wave models in wave climate pro-
the eastern coasts of Russia and Japan show the most jections and their most common parameterizations.
significant wave modeling epistemic uncertainty. Collectively, this offers comprehensive coverage of the
most probable scenarios encountered in investiga-
4. Conclusions and discussion tions of this nature.
Results have demonstrated that both the selec-
Over the past two decades, significant progress has tion of the wave model and the internal parameter-
been made in examining the effect of climate change ization of the model affect the value of the estimated
on wind waves, largely due to the concerted efforts wave climate projected changes. In general, the dif-
of the Coordinated Ocean Wave Climate Project ferences between wave models exhibit higher uncer-
(COWCLIP) [23, 71, 72]. Despite its inevitable role tainty with respect to the internal parameterization
as a primary source of uncertainty in such stud- of the model. In fact, over 60% of the ocean regions
ies, the epistemic uncertainty associated with wave analyzed (figure 3) have shown a larger contribu-
modeling has been addressed in only a limited num- tion from inter-model uncertainty compared to intra-
ber of researches [21, 22]. This study has specific- model uncertainty for all metrics analyzed, although
ally analyzed this source of uncertainty in wave cli- the latter is always present too. This conclusion is even
mate projected changes (figure 1) by isolating it from more robust when considering that the intra-model

9
Environ. Res. Lett. 18 (2023) 124006 H Lobeto et al

uncertainty is estimated from four WW3 models and resources, so including multiple wave models
runs, of which two are outdated (i.e. ST2 and ST3) and/or parameterization may imply a considerable
with respect to the remaining two (i.e. ST4 and ST6). increase in the demands. Hence, until computational
However, while the dominance of the inter-model resources allow for such an approach, it is strongly
uncertainty with respect to the intra-model uncer- recommended to perform extensive calibration and
tainty is clear in the extra-tropical region, it is not validation of the simulations to select the most appro-
as clear in tropical latitudes. Uncertainty has also priate model and parameterization, and minimize the
been quantified by assessing the divergences between discrepancies with the real ocean surface.
members (figure 4). The inter-model uncertainty has Results presented here serve as a basis to under-
shown mean values exceeding 50% in 31%, 23%, 54% stand the scope of the wave modeling uncertainty in
and 38% of the ocean regions for mean H s , H s 99, wave climate projections. They underscore the need
mean T m and mean θm , respectively. In contrast, val- for additional investigation into the origin of the
ues for intra-model uncertainties are 8%, 23%, 15% observed uncertainties. The parameterization of pro-
and 23%. It is important to note that, on average, for cesses such as the energy transfer from the wind to the
both cases, the projected changes in mean T m exhibit ocean and the wave energy dissipation are examples of
the greatest uncertainty, particularly in the Pacific likely causes for the differences found among the pro-
Ocean. jected changes of ensemble members. Specific studies
A more detailed analysis has determined in which that isolate these processes are required to elucidate
ocean areas the wave modeling epistemic uncertainty the distinct contribution of such processes to wave
is significant (figure 5). To that end, the uncertainty modeling uncertainty. Such insights will ultimately
values have been analyzed together with the mag- help to provide a more rigorous description of the
nitude of the projected changes and the deviations projected changes and their robustness.
between members. The period of the waves has been
found to be the wave climate variable showing the Data availability statement
greatest uncertainties across the ocean (53% and 23%
of the ocean surface for inter- and intra-model uncer- The data that support the findings of this study are
tainties, respectively). After the wave period, the dir- available upon reasonable request from the authors.
ection is the wave characteristic showing signific-
ant uncertainties in a larger ocean area (42% and Acknowledgments
20%) and, finally, the wave height, which shows the
lowest proportion (29% and 8%). Particularly, the The authors greatly appreciate the valuable com-
Pacific Ocean stands out as the basin where signi- ments from Mark Hemer and Jean Bidlot, which
ficant uncertainties have been found in larger areas. have helped to improve this manuscript. HL and
On the contrary, the Tropical South Indian Ocean MM acknowledge financial support by CoCliCo
and extra-tropical southern regions of the Atlantic project, which received funding from the European
and Indian Oceans exhibit the least significant uncer- Union’s Horizon 2020 research and innovation pro-
tainties. Additionally, figure 6 shows that a high pro- gram under Grant Agreement No. 101003598, and
portion of the global coastlines is affected by sig- the ThinkInAzul programme, with funding from
nificant wave modeling uncertainties. In fact, 80% European Union NextGenerationEU/PRTR-C17.I1
of them show significant inter-model and/or intra- and the Comunidad de Cantabria. GL acknowledges
model uncertainties in at least one out of the three financial support of Portuguese Fundação para a
wave climate variables analyzed, and over 30% in at Ciência e a Tecnologia I.P./MCTES through national
least two of them. Thus, using one model or another funds (PIDDAC)—UIDB/50019/2020—Instituto
leads to results with differences that cannot be neg- Dom Luiz.
lected for processes where these variables are involved.
This study has demonstrated that the assessment
ORCID iD
of projected changes in wave climate based on a
single wave model with a unique configuration—
Hector Lobeto  https://orcid.org/0000-0002-8380-
which is also the most common approach—may be
6311
affected by relevant wave modeling uncertainties, and
eventually bias the results. These uncertainties cas-
cade and become critical to study changes in pro- References
cesses that use the wave climate as a driver, such [1] Semedo A, Sušelj K, Rutgersson A and Sterl A 2011 A global
as coastal erosion [32, 73] and flooding [31, 74, view on the wind sea and swell climate and variability from
75]. Using multiple models with different configur- ERA-40 J. Clim. 24 1461–79
ations may be a suitable approach to address the epi- [2] Kinsman B 1984 Wind Waves: Their Generation and
Propagation on the Ocean Surface (Courier Corporation)
stemic uncertainty in wave climate projection assess- [3] Haritos N 2007 Introduction to the analysis and design of
ments. However, developing wave climate projec- offshore structures–an overview Electron. J. Struct. Eng.
tion ensembles requires extensive computational time 1 55–65

10
Environ. Res. Lett. 18 (2023) 124006 H Lobeto et al

[4] Vidal C, Medina R and Lomónaco P 2006 Wave height [25] Mori N, Yasuda T, Mase H, Tom T and Oku Y 2010
parameter for damage description of rubble-mound Projection of extreme wave climate change under global
breakwaters Coast. Eng. 53 711–22 warming Hydrol. Res. Lett. 4 15–19
[5] Van der Meer J W 1987 Stability of breakwater armour [26] Semedo A, Weisse R, Behrens A, Sterl A, Bengtsson L and
layers—design formulae Coast. Eng. 11 219–39 Günther H 2013 Projection of global wave climate change
[6] Dean R G 1991 Equilibrium beach profiles: characteristics toward the end of the twenty-first century J. Clim.
and applications J. Coast. Res. 7 53–84 26 8269–88
[7] Cruz A M and Krausmann E 2008 Damage to offshore oil [27] Lemos G, Semedo A, Hemer M, Menendez M and
and gas facilities following hurricanes Katrina and Rita: an Miranda P M A 2021 Remote climate change propagation
overview J. Loss Prev. Process Ind. 21 620–6 across the oceans—the directional swell signature Environ.
[8] Jimenez-Martinez M 2020 Fatigue of offshore structures: a Res. Lett. 16 064080
review of statistical fatigue damage assessment for stochastic [28] Fan Y, Lin S J, Griffies S M and Hemer M A 2014 Simulated
loadings Int. J. Fatigue 132 105327 global swell and wind-sea climate and their responses to
[9] Kennedy A, Rogers S, Sallenger A, Gravois U, Zachry B, anthropogenic climate change at the end of the twenty-first
Dosa M and Zarama F 2011 Building destruction from waves century J. Clim. 27 3516–36
and surge on the Bolivar Peninsula during hurricane ike J. [29] Lobeto H, Menendez M and Losada I J 2021 Projections of
Waterw. Port Coast. Ocean Eng. 137 132–41 directional spectra help to unravel the future behavior of
[10] Staneva J, Wahle K, Koch W, Behrens A, Fenoglio-Marc L wind waves Front. Mar. Sci. 8 655490
and Stanev E V 2016 Coastal flooding: impact of waves on [30] Lobeto H, Menendez M and Losada I J 2021 Future behavior
storm surge during extremes—a case study for the German of wind wave extremes due to climate change Sci. Rep.
Bight Nat. Hazards Earth Syst. Sci. 16 2373–89 11 1–12
[11] Senechal N, Coco G, Castelle B and Marieu V 2015 Storm [31] Vousdoukas M I, Mentaschi L, Voukouvalas E, Verlaan M,
impact on the seasonal shoreline dynamics of a meso- to Jevrejeva S, Jackson L P and Feyen L 2018 Global
macrotidal open sandy beach (Biscarrosse, France) probabilistic projections of extreme sea levels show
Geomorphology 228 448–61 intensification of coastal flood hazard Nat. Commun.
[12] Castelle B, Marieu V, Bujan S, Splinter K D, Robinet A, 9 2360
Sénéchal N and Ferreira S 2015 Impact of the winter [32] Alvarez-Cuesta M, Toimil A and Losada I J 2021 Modelling
2013–2014 series of severe Western Europe storms on a long-term shoreline evolution in highly anthropized coastal
double-barred sandy coast: beach and dune erosion and areas. Part 2: assessing the response to climate change Coast.
megacusp embayments Geomorphology 238 135–48 Eng. 168 103961
[13] Fox-Kemper B et al 2021 Ocean, cryosphere and sea level [33] Dobrynin M, Murawsky J and Yang S 2012 Evolution of the
change Climate Change 2021: The Physical Science Basis. global wind wave climate in CMIP5 experiments Geophys.
Contribution of Working Group I to the Sixth Assessment Res. Lett. 39 2–7
Report of the Intergovernmental Panel on Climate [34] Oppenheimer M et al 2019 Sea level rise and implications for
ChangeScience Basis vol 2018 (Contribution of Working low lying islands, coasts and communities IPCC Special
Group I to the Sixth Assessment Report of the Intergover) Report on the Ocean and Cryosphere in a Changing Climate
ch 9 vol 355
[14] Casas-Prat M, Wang X L and Swart N 2018 CMIP5-based [35] Meucci A, Young I R, Hemer M, Kirezci E and Ranasinghe R
global wave climate projections including the entire Arctic 2020 Projected 21st century changes in extreme wind-wave
Ocean Ocean Modelling 123 66–85 events Sci. Adv. 6 7295–305
[15] Lemos G, Semedo A, Dobrynin M, Behrens A, Staneva J, [36] O’Grady J G, Hemer M A, McInnes K L, Trenham C E and
Bidlot J-R and Miranda P M A 2019 Mid-twenty-first Stephenson A G 2021 Projected incremental changes to
century global wave climate projections: results from a extreme wind-driven wave heights for the twenty-first
dynamic CMIP5 based ensemble Glob. Planet. Change century Sci. Rep. 11 1–8
172 69–87 [37] Lemos G, Menendez M, Semedo A, Miranda P M A and
[16] Reguero B G, Losada I J and Méndez F J 2019 A recent Hemer M 2021 On the decreases in North Atlantic
increase in global wave power as a consequence of oceanic significant wave heights from climate projections Clim. Dyn.
warming Nat. Commun. 10 205 57 2301–24
[17] Gulev S K et al 2021 Changing state of the climate system [38] van der Keur P et al 2016 Identification and analysis of
Climate Change 2021: The Physical Science Basis. uncertainty in disaster risk reduction and climate change
Contribution of Working Group I to the Sixth Assessment adaptation in South and Southeast Asia Int. J. Disaster Risk
Report of the Intergovernmental Panel on Climate Change ed Reduct. 16 208–14
V Masson-Delmotte et al (Cambridge University Press) ch 2, [39] Gong W, Gupta H V, Yang D, Sricharan K and Hero A O
pp 287–422 2013 Estimating epistemic and aleatory uncertainties during
[18] Lantuit H et al 2012 The Arctic coastal dynamics database: a hydrologic modeling: an information theoretic approach
new classification scheme and statistics on arctic permafrost Water Resour. Res. 49 2253–73
coastlines Estuaries Coasts 35 383–400 [40] Toimil A, Camus P, Losada I J and Alvarez-Cuesta M 2021
[19] Thomson J and Rogers W E 2014 Swell and sea in the Visualising the uncertainty cascade in multi-ensemble
emerging Arctic Ocean Geophys. Res. Lett. 41 3136–40 probabilistic coastal erosion projections Front. Mar. Sci.
[20] Lobeto H, Menendez M, Losada I J and Hemer M 2022 The 8 1–19
effect of climate change on wind-wave directional spectra [41] Toimil A, Camus P, Losada I J, Le Cozannet G, Nicholls R J,
Glob. Planet. Change 213 103820 Idier D and Maspataud A 2020 Climate change-driven
[21] Morim J, Hemer M, Cartwright N, Strauss D and Andutta F coastal erosion modelling in temperate sandy beaches:
2018 On the concordance of 21st century wind-wave climate methods and uncertainty treatment Earth Sci. Rev.
projections Glob. Planet. Change 167 160–71 202 103110
[22] Morim J et al 2019 Robustness and uncertainties in global [42] Bricheno L M and Wolf J 2018 Future wave conditions of
multivariate wind-wave climate projections Nat. Clim. Europe, in response to high-end climate change scenarios J.
Change 9 711–8 Geophys. Res. Oceans 123 8762–91
[23] Morim J et al 2020 A global ensemble of ocean wave climate [43] Kumar R, Lemos G, Semedo A and Alsaaq F 2022
projections from CMIP5-driven models Sci. Data 7 1–10 Parameterization-driven uncertainties in
[24] Hemer M A, Fan Y, Mori N, Semedo A and Wang X L 2013 single-forcing, single-model wave climate projections
Projected changes in wave climate from a multi-model from a CMIP6-derived dynamic ensemble Climate
ensemble Nat. Clim. Change 3 471–6 10 51

11
Environ. Res. Lett. 18 (2023) 124006 H Lobeto et al

[44] Eyring V, Bony S, Meehl G A, Senior C A, Stevens B, and Pacific Island focus: from past to present Geosci. Data J.
Stouffer R J and Taylor K E 2016 Overview of the coupled 8 24–33
model intercomparison project phase 6 (CMIP6) [61] Dee D P et al 2011 The ERA-interim reanalysis:
experimental design and organization Geosci. Model Dev. configuration and performance of the data assimilation
9 1937–58 system Q. J. R. Meteorol. Soc. 137 553–97
[45] Döscher R et al 2022 The EC-Earth3 Earth system model for [62] Elshinnawy A I and Antolínez J A Á 2023 A changing wave
the coupled model intercomparison project 6 Geosci. Model climate in the Mediterranean Sea during 58-years using
Dev. 15 2973–3020 UERRA-MESCAN-SURFEX high-resolution wind fields
[46] Meucci A, Young I R, Hemer M, Trenham C and Ocean Eng. 271 113689
Watterson I G 40 years of global Ocean wind-wave climate [63] Lira-Loarca A and Besio G 2022 Future changes and seasonal
derived from CMIP6 ACCESS-CM2 and EC-Earth3 GCMs: variability of the directional wave spectra in the
global trends, regional changes, and future projections Mediterranean Sea for the 21st century Environ. Res. Lett.
(https://doi.org/10.1175/JCLI-D-21) 17 104015
[47] Lee J-Y et al 2021 Future global climate: scenario-based [64] Soran M B, Amarouche K and Akpınar A 2022 Spatial
projections and near-term information Climate Change 2021: calibration of WAVEWATCH III model against satellite
The Physical Science Basis. Contribution of Working Group I to observations using different input and dissipation
the Sixth Assessment Report of the Intergovernmental Panel on parameterizations in the Black Sea Ocean Eng.
Climate Change ed V Masson-Delmotte et al (Cambridge 257 111627
University Press) ch 4, pp 553–672 [65] Akpınar A, Bingölbali B and Van Vledder G P 2016 Wind
[48] O’Neill B C et al 2016 The scenario model intercomparison and wave characteristics in the Black Sea based on the SWAN
project (ScenarioMIP) for CMIP6 Geosci. Model Dev. wave model forced with the CFSR winds Ocean Eng.
9 3461–82 126 276–98
[49] Lemos G et al 2023 Performance evaluation of a global [66] Abu Zed A A, Kansoh R M, Iskander M M and Elkholy M
CMIP6 single forcing, multi wave model ensemble of wave 2022 Wind and wave climate southeastern of the
climate simulations Ocean Modellimg 184 102237 Mediterranean Sea based on a high-resolution SWAN model
[50] Group, T. W. I. D. User manual and system documentation Dyn. Atmos. Oceans 99 101311
of WaveWatch III version 6.07 [67] Kutupoğlu V, Çakmak R E, Akpınar A and van Vledder G P
[51] Group T W 1988 The WAM model—A third generation 2018 Setup and evaluation of a SWAN wind wave model for
ocean wave prediction model J. Phys. Oceanogr. 18 1775–810 the Sea of Marmara Ocean Eng. 165 450–64
(available at: wamdi_group_1988) [68] Wu C F J and Hamada M S 2011 Experiments: Planning,
[52] Booij N, Ris R C and Holthuijsen L H 1999 A Analysis, and Optimization (Wiley)
third-generation wave model for coastal regions 1. Model [69] Neter J, Kutner M H, Nachtsheim C J and Wasserman W
description and validation J. Geophys. Res. Oceans (others) 1996 Applied linear statistical models
104 7649–66 [70] Alvarez-Cuesta M, Toimil A and Losada I J 2021 Modelling
[53] Ris R C, Holthuijsen L H and Booij N 1999 A long-term shoreline evolution in highly anthropized coastal
third-generation wave model for coastal regions: 2, areas. Part 1: model description and validation Coast. Eng.
verification J. Geophys. Res. 104 7667–81 169 103960
[54] Björkqvist J V, Vähä-Piikkiö O, Alari V, Kuznetsova A and [71] Hemer M A, Wang X L, Weisssse R and Swail V R 2012
Tuomi L 2020 WAM, SWAN and WAVEWATCH III in the Advancing wind-waves climate science: the COWCLIP
Finnish archipelago–the effect of spectral performance on project Bull. Am. Meteorol. Soc. 93 791–6
bulk wave parameters J. Oper. Oceanogr. 13 55–70 [72] Morim J et al 2022 A global ensemble of ocean wave climate
[55] Padilla-Hernández R, Perrie W, Toulany B and Smith P C statistics from contemporary wave reanalysis and hindcasts
2007 Modeling of two northwest Atlantic storms with Sci. Data 9 358
third-generation wave models Weather Forecast 22 1229–42 [73] Toimil A, Álvarez-Cuesta M and Losada I J 2023 Neglecting
[56] Perez J, Menendez M and Losada I J 2017 GOW2: a global the effect of long- and short-term erosion can lead to
wave hindcast for coastal applications Coast. Eng. 124 1–11 spurious coastal flood risk projections and maladaptation
[57] Grigorieva V G, Gulev S K and Sharmar V D 2020 Validating Coast. Eng. 179 104248
Ocean wind wave global hindcast with visual observations [74] Kirezci E, Young I R, Ranasinghe R, Muis S, Nicholls R J,
from VOS Oceanology 60 9–19 Lincke D and Hinkel J 2020 Projections of global-scale
[58] Reguero B G, Menéndez M, Méndez F J, Mínguez R and extreme sea levels and resulting episodic coastal flooding
Losada I J 2012 A global ocean wave (GOW) calibrated over the 21st century Sci. Rep. 10 11629
reanalysis from 1948 onwards Coast. Eng. 65 38–55 [75] Tebaldi C, Ranasinghe R, Vousdoukas M, Rasmussen D J,
[59] Hersbach H et al 2020 The ERA5 global reanalysis Q. J. R. Vega-Westhoff B, Kirezci E, Kopp R E, Sriver R and
Meteorol. Soc. 146 1999–2049 Mentaschi L 2021 Extreme sea levels at different
[60] Smith G A, Hemer M, Greenslade D, Trenham C, Zieger S global warming levels Nat. Clim. Change
and Durrant T 2021 Global wave hindcast with Australian 11 746–51

12