In many mountainous countries of the world, where winters are deep and long, the onset of spring brings warmer temperatures and the melting of vast amounts of snow and ice transforming mountain streams into raging torrents and meandering rivers into flooded valleys. It was this mechanism that, in 2018, brought chaos to several communities in southern British Columbia, Canada, and again in 2019 to widespread areas of Ontario and Quebec. In June 2022 heavy rains and snowmelt combined to close Yellowstone National Park for eight days.
The exposure to snow-derived flood risk is perhaps underappreciated globally. In the US for example, a considerable snowpack builds in the Rocky Mountains, Sierra Nevada, Appalachia and across large swathes of the US/Canada border and Alaska. Persistent winter snows extend across almost the entirety of Canada, Scandinavia and Iceland where the melt season dominates the annual hydrograph there. The same is true for countries whose large river systems are fed by major mountain ranges such as the Rhine flowing from the European Alps, the Ganges River flowing from the Himalayas, and several large rivers in China’s Xinjiang province fed by the Tian Shan mountain range.
The magnitude of the spring snowmelt is not uniform each year. Rather, it is determined by the following factors:
- The volume of snow and ice accumulated during the previous winter,
- the intensity of precipitation before and during the melting season; and,
- the rate of transition from freeze to thaw during spring
These interacting factors, which together determine the hazard outcome, make a great candidate for risk analysis. Think of each factor as a number on a die: if we are lucky enough to roll a set of number ones—small snowpack, dry spring, gently rising temperatures—then flooding will be minimal. On the other hand, roll all sixes—deep snowy winter, intense spring rainfall, sudden onset of hot spring temperatures—and conditions are lined up for a flood catastrophe. A good risk model must be able to estimate the annual frequency of each factor and how their frequencies move together (covariance).
For the JBA Global Flood Event Set and Global Climate Change Event Set we modelled the influence of snow at the catchment scale by including a snowpack simulation scheme running in tandem with a rainfall-runoff model. The snow model considers the catchment-averaged daily temperature and precipitation together with catchment altitude and season. The model builds a representation of the state of the catchment’s snowpack on each simulation day: during sub-zero winter days, a proportion of any precipitation is converted to snow and added to the snowpack. As temperatures warm in spring the simulated snowpack releases water at a rate dependent on catchment-specific factors. On each simulation day, any available meltwater is added to any precipitation and the total is routed through the river system by the rainfall-runoff model for the catchment.
So how important is it to include this extra detail in the modelling process? To answer, we can look in some detail at a single illustrative case study.
DELAWARE FLOODING 2005
Between 2-4 April 2005, nearly 125mm of rain fell on already wet catchments in the Pennsylvania and New York regions of the US. This event coincided with an unusually large snowpack that had built up in the Pocono and Catskill Mountains that simultaneously entered thaw conditions. Combined, these events resulted in the worst flooding seen along the Delaware River for 50 years. Figure 1 shows the out-of-bank conditions at Calhoun St. Bridge.
Figure 1: Delaware River in flood April 2005 at Trenton, New Jersey. (Credit: John Jenks, NJDEP/Bureau of Technical Services. Public domain.)
The gauging station at Trenton is one of the sites used to supply data to calibrate rainfall-runoff-snow models used within JBA’s Global Flood Event Set. The site location and its catchment are shown in Figure 2.
Figure 2 (right): The catchment extent for the Trenton river gauge (gauge indicated by circle marker). The Triangle markers show (top) Slide Mountain and (centre) Camelback Mountain, the highest peaks of both the Catskill and Pocono ranges respectively, indicating the mountainous nature of this catchment.
Using data from JBA’s Global Flood Event Set, we can isolate the 2005 flood event at Trenton and investigate the extent to which snowmelt contributed to the modelled flow. This is possible because our rainfall-runoff and snowmelt models have a mechanistic interpretation (i.e., they are not purely empirical) so we can “look inside” the model and see what the snowpack did during the model simulation. The results are shown in Figure 3.
Figure 3: Time series summary of the 2005 flood event at Trenton. The bottom plot shows how the temperature transitioned from freeze to thaw conditions in early April just as two large precipitation events occurred. The centre plot shows the resulting river flood event – both observed (dashed line) and simulated (solid line). The JBA model matches the observations reasonably well thanks to a significant contribution to the total flow from the model’s melting snowpack.
In the opening section, I stated that serious flood hazard results from the unfortunate alignment of multiple contributing factors. The sequence of events at Trenton in April 2005 provides a concrete example of this. According to the simulation model – which is founded on scientific principles and empirical data—approximately 50% of the flood peak was the result of water introduced to the catchment from melting snow with the remainder a result of two unfortunately-timed intense rainfall events. In total, the volume of water was too much for the banks to contain and considerable inundation ensued.
IMPACT OF CLIMATE CHANGE AND SNOWMELT ON FUTURE RISK
JBA’s Global Flood Event Set generates a synthetic set of thousands of years of daily precipitation and temperature incorporating a wide range of plausible intensity and timing combinations so that a large portfolio of floods—like the Trenton flood—have been created. However, the Global Flood Event Set is designed to produce events with a frequency, extent, and intensity representative of current climatic conditions. To address the need for event sets representative of future conditions JBA have developed a Global Climate Change Event Set. (See Ashleigh Massam’s blog about validation of climate change factors for more insight).
One of the big advantages of using simulation models to generate the river flow within an event set is the ability to change the model inputs (precipitation and temperature) using, for example, scenarios generated by global climate simulation models, and watch how these changes impact the river flow. In extratropical regions the climate change trend is typically for increases in both precipitation and temperature. Revisiting the Trenton catchment, Figure 4 shows how the NCAR CESM4.0 model estimates of median seasonal temperature and precipitation have increased from the baseline 1980-2005 period to the 2036-2065 scenario period.
Figure 4: Plots comparing the median temperature and precipitation by season for the baseline (1980-2005) and future (2036-2065) time slices based on the response of the CESM4.0 global climate model to the RCP4.5 emissions scenario at all locations enclosed by the Trenton catchment shown in Figure 2.
These two factors compete when it comes to producing extreme river flows—sometimes drier conditions make it difficult for a catchment to saturate so that heavy rains are mostly absorbed by soils or, at less permeable catchments, higher rainfalls tend to dominate and bring higher river flows. When we include snowmelt there is yet more complexity to deal with. Figure 5 shows the simulation timeseries plots for key variables at the Trenton site taken from the Global Climate Change Event Set results.
Figure 5: Timeseries of key input and output variables from the Trenton rainfall-runoff plus snowmelt model for the baseline and RCP4.5 future scenario conditions. Note the clear reduction in the average size of the modelled snowpack during the future period.
Thinking back to Figure 3’s message, we know any interference to snow will have a knock-on effect through to flood levels. Indeed, data from our RCP4.5 climate change scenario at Trenton shows that while average precipitation is higher, the reduced average snowpack size results in lower annual maximum flow rates during the scenario period. This is summarised in Figure 6.
Figure 6 (right): Summary of annual maximum flow rates for the baseline (1980-2005) and RCP4.5 future (2036-2065) scenario. The future scenario contains less variation and has reduced by approximately 180 m3s-1.
What are the ramifications of these results? At least according to model data, with all the caveats that implies, it appears mid-21st-century warming may remove one source of flood risk, namely large winter snowpacks that can be thawed and released with damaging consequences in spring. Of course, flood risk is only one factor in the larger catchment system and interference in the existing equilibrium may have untold consequences on a catchment’s ecological health and its socio-economic management, i.e. this may be a very tiny silver lining on a very large cloud.
CLOSING REMARKS
I’ve emphasised two points in this post:
- The importance of using models with the capacity to represent change when simulating future scenarios (it’s not possible to simply rely on an empirical set of data that only represents current conditions); and,
- the way total risk is formed from several underlying components that may or may not be connected, but which can come together resulting in significant damage (in the case study presented here it was the simultaneous melting of a large winter snowpack onto an already-wet catchment during a significant rainfall event)
Mostly, climate change conditions indicate an increase in the chances of unfortunate conjunctions of underlying risk components. In this blog I have presented one of the, almost certainly rare, cases where climate change reduces the probability of an unfortunate union of factors. We leave it to economists to speculate how best to redistribute gains should there be a reduction in snowmelt related flood losses.
ACKNOWLEDGEMENTS
We acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modelling groups such as NCAR for their production of the Community Earth System Model (CESM) used to produce the RCP4.5 pathway data used in this blog post. For CMIP the U.S. Department of Energy’s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals.