Wind Speed Variability and Adaptation Strategies in
Coastal Regions of the Pacific Northwest
Bradford Griffin1, Karen Kohfeld1, Andrew Cooper1, and Gerhard Boenisch2
1School
of Resource and Environmental Management
Management, Simon Fraser University,
University Burnaby,
Burnaby B.C.
BC
2Max Planck Institute for Biogeochemistry, Jena, Germany
Research Objectives
Methods (continued)
Results
1. Determine if correlations exist between Pacific Northwest wind speed distributions
(i.e., quantiles), Pacific Ocean climate indices, and monitoring station-specific
attributes (e.g., elevation, geographic location, data source);
2 Assess
2.
A
the
h robustness
b
off relationships
l i hi for
f forecasting
f
i wind
i d speeds
d within
i hi the
h study
d
area; and
3. Identify adaptation strategies to reduce wind damage and convey in a manner easily
understandable by a wide (potentially non-technical) audience.
Hierarchical Cluster Analysis (HCA) – Looks for commonalities among time series
and groups the respective stations accordingly. Useful explanatory tool for early data
exploration. However, HCA does not provide physical process explanations of
groupings; left to the researcher to determine
determine.
Increased speed and variability of coast – Future development of coastal infrastructure,
especially that susceptible to wind damage (e.g., transmission lines or residential
development), will need to consider a wide (possibly changing) range of wind speeds.
Timing of studies measuring wind speeds (e.g.,
(e g wind power) will be important given
apparent cyclic non-seasonal pattern in Coast group wind speeds.
Pacific Ocean climate indices do not correlate well with wind speeds and cannot explain
the range of variation observed. Lack of fit reduces forecasting skill of the LME models.
Mainland
Fixed Effects
Q95 ~ Year + Elevation + Coast/Mainland
+ Data Source + ENSO + PNA
+ PDO + NOI + AO +
Year : (Elevation + C/M + ENSO
+ PNA + PDO + NOI + AO) +
Elevation : (C/M + PNA + PDO
+ AO) +
C/M : (PNA + NOI + AO) +
ENSO : (PDO + NOI + AO) +
PNA : (PDO + NOI + AO) +
PDO : (NOI + AO)
Random Effects
Q95 ~ Coast/Mainland | Station
Variance ~ (Fitted Wind Speed Value)Power
Power = 0.7826
Correlation ~ AR(1) | Station
φ = 0.3433
Context
As coastal communities expand their vulnerability to severe windstorms may increase if
changing climate conditions and risks are ignored. Previous attempts to forecast wind
speeds in the Pacific Northwest have been
limited in scope and had varying results.
Tens of thousands still
However, a link between Pacific Ocean climate
in the dark after
indices (e.g., PDO & PNA) and wind speeds has
massive B.C. storm
been proposed. This type of connection could
December 15, 2006 – www.cbc.ca
allow forecasts from dynamic climate models
(GCMs) to be locally applied and result in better
decision making and adaptation actions.
Photo credit: www.cbc.ca, Nov. 30, 2006
Figure 2 – Study Area showing meteorological
monitoring station locations. Colours refer to
hierarchical clusters of 95% quantile (Coast –
blue, Mainland – orange) and size of dots
indicate elevation of mean sea level.
Methods
Quantiles instead of Mean or Max/Min – A focus solely on mean or max/min values
may underestimate, overestimate, or fail to distinguish real nonzero changes in responseexplanatory relationships. This is especially true for distributions, such as those typical of
wind speeds, which display strong skew and/or exhibit unequal variance across the
magnitudes of observations
observations. Quantiles can better summarize strongly skewed (versus
normally distributed) data to provide a more robust statistical starting point.
Coast
Figure 6 – Fitted and Observed values for 95th quantile
wind speeds on the Pacific Northwest. Representative
LME Model structure shown.
“:” – factor interaction; “|” – grouped by
Next Steps
p
Figure 3 – Hierarchical Clustering Dendogram
identifying geographical clusters of 95% quantile
wind speeds (Coast – blue, Mainland – orange).
Management Actions
Action A
...
Action
... B
Action D
...
Resulting Outcomes
(Reduced Damage Costs)
...$1
Probability
...$2
0
...$3
5
...$4
10
.
15
.
20
.
25
...$m
...
Uncertain Future States
...
Probability
-4
-2
2
0
2
6
10
Wind Speeds
Action C
...
SST
Linear Mixed-Effects (LME) Model
• Fixed (population) effects – e.g., influence of PDO on wind speeds across the region;
• Random (location-specific) effects – e.g., how a valley location may have a different
wind speed response to PDO versus a mountain location;
• Appropriate for climate data because LME can incorporate repeated measurements
over time at permanent monitoring stations, temporal correlation (e.g., time-series),
h
heterogeneous
variance
i
(non-normal
(
l distributions),
di ib i ) or non-stationary
i
(trending)
(
di ) data.
d
Additional fixed effect factors, such as re-analysis SST and SLP calculated values,
upper air measurements, or GIS topographic features could be used to better forecast
PNW wind speeds. A Quantitative Decision Analysis focusing on wind damage
adaptation options available to resource managers would likely help future planning.
...
Figure 7 – Proposed decision analysis tree showing possible management actions, future states of
nature (i.e., wind speeds) with associated probabilities, and the subsequent reductions in damage.
Acknowledgements
Funding and support for this research was generously provided by:
Figure 1 – Wind Speed Quantiles Averaged Across Groups (Coast – blue,
Mainland – orange). Note the decreasing trend of mainland values through
time and the cyclic pattern of coast locations.
Figure 4 – Linear mixed-effects model
for Pacific Northwest wind speeds.
Figure 5 – Wind speeds, seasonal variation, deseasonalized
trend, and residuals for Abbotsford Airport.
School of
Resource and
Environmental
Management