Dam Safety Review
Summary Description
- standard based on Dam Safety Guidelines - Canadian Dam Association (CDA)
- risk-informed approach: undesired events in therms of likelihood and potential consequences
- provincial and territorial regulations
- evaluation and systematic review of all aspects (including visual inspections of all components)
There are over 16 000 dams in Canada and 38% are managed by Quebec authorities of which 22% are used for hydroelectricity. 87% of dams can be found in five provinces (Ontario, Quebec, British Columbia, Alberta and Saskatchewan) (Ozkan et al. (2023)). Additionally, half of Canada dams are older than 50 years old, increasing the risk of climate change impacts due to infrastructure aging making them more vulnerable.
Dams present many vulnerabilities linked to climate change which are challenging to quantify. Extreme floods are projected to increase in frequency and severity in many areas across Canada, which directly affect the inflow flood design. Furthermore, there is a projected shift in precipitation pattern and flood mechanisms such as rain-on-snow events or snowmelt-driven floods which also affect the estimated design flood as well as operations and maintenance. Dams located in permafrost regions could be affected by instability due to thawing. Rising temperatures could also create large ice movements. Access to infrastructure could also be limited by more common extreme events such as forest fires (i.e. an Hydro Quebec had an installation evacuated by workers to let firefighters do their work) (MacTavish et al. (2022), Islam et al. (2024)). Similarly, freezing rain, ice storms or extreme snowfall could limit the ability to access a dam or operate flow control equipment such as pulling logs. Reservoir water quality is likely to be affected by climate change due to the rise in temperature (intense stratification leading to anoxic conditions) and switch in precipitation patterns (influx of nutrients and turbidity during snowmelt), among others (Ozkan et al. (2023)). Structural stability of concrete dams might also be challenged due to the increase in total horizontal load caused by precipitations and streamflow (Ozkan et al. (2023)).
The following sections describe the role, some method and models, as well as gaps and recommendations in dam safety review.
Role in the Electricity System
Methods and Models
Many methods and models are used for DSR as they can require many inspections based on dam regulations (Engineers and British Columbia (2023), Natural Resources (2023)) or the practices of utility: - field reviews - interviews with site staff - equipment testing - extensive understanding of the design, construction, operation and maintenance
Among many others. Please refer to local legislation for more information. The following sections only tackle a few methods and models used in DSR as a wide variety of field (structure, geotechnical, hydrological, hydraulic, seismic, mechanical, electrical, etc.) expertise is required to perform a comprehensive review. Therefore, methods presented will focus on hydrology aspects affected by climate change, with some commentary on other types of models and methods.
The figure below briefly resumes decision-making challenges faced by professionals in the dam safety sector of the electricity system in Canada. These challenges were shared during the workshops held in 2025. These examples linked together the methods and models as well as characteristics presented in the following section such as PMP.
For all climate variables listed below, please refer to the climate dataset section for guidance and availability.
Flood frequency analysis (FFA) (England et al. (2019), Islam et al. (2024)) is often used to determine the design flood required by regulations. It is determined by the return period such as 1/100 or 1/1000 years. Therefore, the goal of FFA is to link flow values with probabilities of occurrence (Ouranos (2021)).
\[ Aep = 1/T = 1-CDF_{high flows} \]
where
- \(Aep\) is the annual exceedance probability,
- \(T\) is the return period (years),
- \(CDF_{high flows}\) is the cumulative distribution of high flows.
Flow records rarely go up to 100 years (and, of course, not to 1000). Therefore, a probability density function is fit to either flow annual maxima (AM) or peak-over-threshold (POT). POT allows more events of high flows to be considered as some years could have multiple events over thresholds and others none. However, it adds issues linked to the independence of events.
Moreover, such analysis requires long timeserie of observed inflows (at least 30 years based on World Meteorological Organisation criteria) and different methods exists to determine the distribution parameters (such as generalized extreme value, generalized Pareto or lognormal) (World Meteorological Organization (1989)). Different metrics of goodness of fit can be used to validate the fitted distributions, such as Aikake or Bayesian Information Criteria, Kolmogorov-Sminov, etc. (Islam et al. (2024)).
Common FFA methods are univariate and stationary, meaning that climate change and impacts of other features than flows are not taken into account when calculating the design flood. To evaluate non-stationarity, Ouranos (2021) proposed two techniques to assess non-stationarity using hydrological model fed with a large ensemble of climate projections over 20-year period block.
Other, recent studies have looked into non-stationary extreme value distribution, such as integrating linear trends in function of time or climate variables as covariates in distribution parameters (Islam et al. (2024), Maria et al. (2024), Levin (2025)). Levin (2025) used precipitation, annual snowfall, mean annual temperature as covariates in North-central United States, where 45% of streamgages were predicted with precipitation, 14% temperature and less than 1% with snowfall as causal mechanisms (36% had no climate variables as causal mechanisms). Northern part had the biggest differences between stationary and non-stationary FFA (going up to 20%). The study used measured streamflow and climate data from a monthly water balance model which was itself fed by temperature and precipitation data from monthly NOAA gridded observations.
Maria et al. (2024) used the same covariates (rainfall split in two seasons) for a study across Canada as well as time. Streamflows were acquired from simulations of GEM (1950-2099) which are driven by a member of CanESM2 RCP8.5 (CMIP5). Non-stationary projected higher increases in 2000-year return levels for 451 dams locations than stationary FFA.
Furthermore, other recent studies have also looked into multivariate FFA, that use copulas to create joint distributions highlighting the interdependence between flooding features and have potential for compound flooding (e.g., high precipitation and snowmelt or sea-level rise) (Maria et al. (2024), Bizhanimanzar et al. (2024)).
Extreme consequences dams require design flood (classification), the maximum flow that dam must be able to evacuate, to be estimated with the Probable Maximum Flood (PMF), which itself is estimated with the Probable Maximum Precipitation (PMP). The WMO defines PMP as being “the greatest depth of precipitation for a given duration meteorologically possible for a design watershed or a given storm area at a particular location at a particular time of the year, with no allowance made for long-term climatic trends” (World Meteorological Organization (2009)). The WMO manual proposes six methods to evaluate the PMP (local, transposition, combination, inferential generalized and statistical) with the majority of them based on physical meteorology (Islam et al. (2024)).
PMPs are usually evaluated by meteorologists or hydrologists and are provided with uncertainty bounds as well as seasonal adjustment factors. FFA with values for rainfall, temperature and snow accumulation linked to the storm can be estimated from observations to calibrate the hydrological models. Soil moisture needs to be maximized when PMF is determined in the hydrological model and other factors such as the combination of the spring PMP with the 100-year snowpack (Ouranos (2015)).
The xHydro python package developed by Hydro-Quebec, ETS and Ouranos offer codes to estimate PMPs and PMFs.
One common method to evaluate PMP is moisture maximization, which is advantageous for his lower data requirements (Islam et al. (2024), World Meteorological Organization (2009)). \[ PMP = P_{obs} * W_{max} / W_{storm} \]
where
- \(P_{obs}\) is the maximum observed precipitation (mm),
- \(W_{max}\) is the maximum precipitable water (mm),
- \(W_{storm}\) is actual precipitable water of the observed storm (mm).
\(W_{max}\) can be estimated with dew point temperature.
Another method, storm transposition doesn’t require detailed information on watershed specific storms (Islam et al. (2024), World Meteorological Organization (2009)).
\[ PMP = \overline{X_{n}} + K_{m} * S_{n} \]
where
- \(\overline{X_{n}}\) is the mean maximum annual precipitation for the duration (n) of interest,
- \(S_{n}\) is the standard deviation,
- \(K_{m}\) is the frequency factor.
\(K_{m}\) is a factor of precipitation duration and mean value. The British Columbia MetPortal offers PMP calculated with transposition across the province.
These estimates come with high uncertainties and are left in the choice of professionals to determine the most appropriate methods. Recent studies are looking into improving PMP calculations and more precisely providing the uncertainties linked to each estimate (Martin (2024)). PMP uncertainties can be assessed via a Monte Carlo techniques and sensitivity analysis need to be done for evaluating PMFs uncertainty ((king2022?))
To account for non-stationarity, studies have used climate simulations to estimate the PMP. Ouranos (2015) used regional climate models to project future PMPs with higher resolution (50 km) in three Quebec watersheds which generally showed an increase of 10 to 20% for the mid-century. However, one watershed showed a decrease in PMP. Alternative methods than WMO, have been listed in literature to account for climate change and uncertainties (Martin (2024)). Recently, practioners in OPG have used MRCC5-CMIP6 to estimate the PMP. It was reported that SSP 3-7.0 produced lower PMPs values over Ontario than SSP 2-4.5. Chan et al. (2024) also observed higher values of PMPs with scenario SSP1-2.6 than SSP3-7.0 over Hong Kong.
The figure below highlights the steps taken by Clavet-Gaumont et al. (2017) for considering climate change in PMP estimates (Ouranos (2015)).
The workflow is based on the methodology developed by Rousseau et al. (2014) and moisture maximization. The estimates rquire RCM liquid precipitation, precipitatble water and snow water equivalent (snow on ground). The RCMs data were pre-processed when precipitable water was missing (used summation of specific humidty from surface to top level) and mean temperature was used to estimate rainfall from precipitation. Many choices were made based sensitivity analysis such as the thresholds used to select PMPs, the best fitted statistical distributions, etc. The storm definition was made using rainfall events of 24, 48, 72 and 120 hours with a moving window. P~event represent 10% of largest storm of the year. Rainfall events are maximized with EQ1 presented in the above figure and PWH~100 the monthly return period value of precipitable water (fitted with a GEV and the maximum likelihood). The maximum of the maximum ratios multiplied by the P~event represents the PMP.