Cyclical Climate Patterns and Flood Risk and the Impact on Reservoir Design and Operation
Prudent flood management and the design of flood control structures (e.g. dams, spillways or culverts) require an accurate estimate of the likelihood of infrequent, extreme events at any given point in time. Failure to assess such risks appropriately could lead to either an overbuilt design that wastes resources or to inadequate design that cannot handle floods that do occur.
Most flood management design for extreme events is based on the “100-year” flood—the flood level that has a 1 percent chance of being reached or exceeded in a given year. Predictions of the level of the 100-year flood assume that on average, flood risk does not vary over time, and therefore the level of risk can be determined from data collected over relatively short periods of time.
Increasing evidence, however, shows that this assumption is incorrect, and that, following climate, flood occurrence exhibits a “regime-like” pattern in which there are long-term, quasi-periodic and cyclical trends of wetter and drier periods. This evidence suggests that if designers of reservoirs fail to anticipate longer-term climate cycles, they are misinformed as to the risk of flood.
On the other hand, by better understanding these cyclical risks, climate forecasters can assess which regime they are in, and the chances of the regional climate undergoing a “regime change” at any given time.
The following studies examine the cyclical effects of climate on the calculation of likelihood of extreme flood events, focusing on two regions, the Pacific Northwest of the United States and Peru. In both cases, results of the study yield significant implications for assumptions about reservoir design and management.
Shaleen Jain, Upmanu Lall, 2001, “Floods in a Changing Climate: Does the Past Represent the Future?” Water Resources Research.
In “Floods in a Changing Climate,” Jain and Lall examine the 88-year long flood record for the Similkameen River, in Washington State.
A correlation analysis suggests a statistically significant relationship between Similkameen River annual maximum floods and both winter NINO3 (an El Nino/La Nina/Southern Oscillation index) and the PDO (Pacific Decadal Oscillation) indices.
ENSO and PDO are quasi-periodic climate patterns that affect long-term weather conditions in many parts of the world, including extreme events such as flooding. ENSO happens irregularly every 2 to 7 years and lasts 9 months to 2 years; the PDO happens over a longer time-scale, around 20 to 30 years. For example, higher floods in the post 1940s period are consistent with the 1948-1976 ENSO and PDO climate regimes identified from records.
Given these results, it would be a mistake to think that the flood patterns on the Similkameen River of the last 30 years will be representative of the next 30—even without taking human-caused climate change into consideration.
Carlos H.R. Lima, Upmanu Lall, 2010, “Spatial Scaling in a Changing Climate: A Hierarchical Bayesian Model for Non-Stationary Multi-Site Maximum and Monthly Streamflow” Journal of Hydrological Sciences, doi: 10.1016/j.jhydrol.2009.12.045.
Estimating the potential magnitude and timing of major flood events requires a significant amount of data (no less than 100 year record of stream flow) to be reliable. However, the desired amount of data is usually not available.
To get around this problem, engineers have typically used information from similar, nearby sites to estimate the likely magnitude and frequency of unusual but extreme flood events. At sites with no streamflow data (ungauged sites), regional analysis is used to estimate the desired variable at sites where data does exist; results are then extrapolated using stastistical scaling to the ungauged site.
In this paper, Lima and Lall attempt to increase the reliability of these approaches by applying hierarchical Bayesian models to account for the potential role of climate variability and change in estimating these extreme events for planning purposes.
The Potential for Better Flood Prediction
Going beyond the impact on infrastructure design, an ever-growing body of research suggests that it may be possible to use climate patterns to predict floods—in some cases from a season to a year in advance.
Research on a the relationship between flood frequency and macro-climate phenomena for specific river basins in North America, China and Brazil among other locations, confirm the thesis that the location and timing of extreme floods are predictable months or seasons in advance as long as underlying climate patterns are understood.
In each case, Columbia researchers looked at the various different climate factors that drive floods in a particular region in order to understand how cyclical climate patterns affect our understanding of flood risk, and how, in many cases, those factors can lead to useful flood prediction on an intermediate time scale (from a season to a year in advance).
While the precise climate indicators for a given region or watershed vary by location, in many cases a careful analysis of flood record data coupled with an understanding of general climate patterns can yield models that predict both timing and magnitude of annual floods.
In the United States intermountain West, the magnitude and timing of spring snowmelt floods reflects a combination of seasonal snow accumulation and spring temperature patterns that are in turn related to low-frequency climate variables such as ENSO and PDO.
In Montana, researchers considered global sea surface temperatures (SSTs), regional snowpack levels and output from the European Community-Hamburg v4.5 General Circulation Model (ECHAM 4.5 GCM), an atmospheric climate circulation model.
For the Three Gorges Dam in China, the prediction model includes a combination of sea-surface temperatures and upland snow cover, both available one season ahead of the prediction period.
Following are a few of the key studies and a brief explanation of their significance for developing better flood prediction models.
Shaleen Jain, Upmanu Lall, 2000, “Magnitude and Timing of Annual Maximum Floods: Trends and Large-scale Climatic Associations for the Blacksmith Fork River, Utah,” Water Resources Research, Vol. 36, No. 12, Pages 3641-3651.
Unlike the Similkameen River basin, conventional climate wisdom (as reported by the National Oceanic and Atmospheric Administration) suggests that the climate in the Blacksmith Fork River basin is only weakly affected by ENSO and PDO effects.
However, Jain and Lall found that patterns contained in two large-scale climate indices, NINO3 and PDO, were able to explain 40% of the variance of the annual maximum flood series. Furthermore, selected combinations of the extreme phases of both indices appears to lead to higher flood potential in the Blacksmith Fork River. Thus, even in cases where it appears that larger scale climatic forces have limited effects on local climate, a deeper analysis can illuminate hitherto unseen connections that allow for the potential to predict flood timing and magnitude.
A. Sankarasubramanian, Upmanu Lall, 2003, “Flood Quantiles in a Changing Climate: Seasonal Forecasts and Causal Relations,” Water Resources Research, Vol. 36, No. 12, Pages 3641-3651.
In this paper, Sankarasubramanian and Lall present two methods to develop a better estimation framework for changes in local or regional flood frequency and for forecasting flood risk in the season of its occurrence: a quantile regression analysis and a method for local likelihood estimation. In the paper they discuss the pros and cons of each approach.
The application of these approaches to the Montana flood series demonstrates that these or similar methods may offer the possibility for short-term seasonal forecasting.
Gonzalo Pizarro, 2006 “Instruments for Managing Seasonal Flood Risk Using Climate Forecasts,” Dissertation, Columbia University Graduate School of Arts and Sciences.
In his 2006 dissertation, Gonzalo Pizarro gives an overview of the new climate-precursor-related flood prediction tools along with their potential application for infrastructure design and operational rules and risk management tools such as flood insurance and catastrophe (CAT) bonds as applied to the western United States.
In his study, Pizarro finds that there are “coherent regions of peak flows in response to climate drivers,” the main ones being the ENSO and the PDO. He then presents a model for flood risk estimation based on the current climate state that is capable of issuing peak flow forecasts a season ahead of time.
Pizarro then describes potential improvements in reservoir operations based on these season-ahead predictions of probable inflows. The paper concludes with an analysis of the potential for better financial management of flood risk using the same improved prediction models to help insurance companies reduce the variance of potential losses or claims.
Hyun-Han Kwon, Casey Brown, Upmanu Lall, 2008, “Climate Informed Flood Frequency Analysis and Prediction in Montana Using Hierarchical Baysian Modeling,” Geophysical Research Letters, Vol. 35.
For this paper, Kwon, Brown and Lall looked at climate factors that could lead to better prediction of the annual maximum flood for the Clark Fork River basin in Montana. To conduct this analysis they looked at daily streamflow records available from 1930 to the present.
Predictors considered for this study were global sea surface temperatures (SSTs), regional snowpack levels and output from the European Community-Hamburg v4.5 General Circulation Model (ECHAM 4.5 GCM), an atmospheric climate circulation model.
The results demonstrated statistically significant links to all three predictors and were able to reduce uncertainty accompanying the estimated value of the 100-year flood.
Hyun-Han Kwon, Casey Brown, Kaiqin Xu, Upmanu Lall, June 2009, “Seasonal and Annual Maximum Streamflow Forecasting Using Climate Information: Application to the Three Gorges Dam in the Yangtze River Basin, China,” Journal of Hydrological Sciences, 54(3).
In this paper, Kwon, et al explore the potential for seasonal prediction of hydrological variables that are potentially useful for reservoir operation of the Three Gorges Dam in China.
The study suggests that a skillful prediction model could have a number of potential benefits for management of a reservoir such as the Three Gorges Dam. For example, if climate forecasters can predict excess inflows during an inflow period, extra water can be released for hydroelectricity or other benefits instead of allowing it to spill over the dam. Conversely, if lower inflows are predicted, releases can be curtailed to conserve water for the predicted dry period.
For multipurpose reservoirs, the possibility of flood further complicates operations, as managers typically leave space in the reservoir to capture excess water should a flood occur. There is an opportunity cost, however, for lost potential water storage for the season ahead. Better forecasting could mitigate such opportunity costs by giving the operators the flexibility to make use of that volume when no excess flows are expected.
A combination of specific Pacific and Indian Ocean temperatures combined with an analysis of snow cover yields a model for the Basin that may help reservoir managers in the future.
Use of Index Insurance for Floods
Abedalrazq F. Khalil, Hyun-Han Kwon, Upmanu Lall, Mario J. Miranda and Jerry Skees, July 2007, “El-Nino-Southern Oscillation-based Index Insurance for Floods: Statistical Risk Analyses and Application to Peru,” Water Resources Research.
A fundamental goal of insurance is to pool a large number of uncorrelated or negatively correlated risks. Doing so creates a more robust risk management product for the insurer because it decreases the likelihood that a small number of events will cluster outside of the expected average. A larger total cash flow also makes the insurer more resilient to larger individual payouts.
In this study, Kwan et al. examine how index insurance could be applied to floods in Peru by tying insurance payouts to ENSO climate patterns. Specifically, the paper looks at the reliability of the ENSO index to determine maximum payouts, the effect of sampling uncertainties and how closely the index relates to local outcomes, the potential for clustering of payouts, the potential for predicting the index with some lead time prior to the flood season and the evidence for climate variability over time as determined from the long ENSO record.