Developing a Typical Rainfall Period to Analyze CSOs in San Francisco

The city of San Francisco has used InfoWorks CS as the basis for developing a continuous long-term (five-year) rainfall period, which is being used to evaluate sewer flows in a collection system model.

Feb 1st, 2008
Th 0802uwmdeveloping

By Sharon Tsay, Dominique Brocard, Wallis Lee, and Greg Braswell

The city of San Francisco has used InfoWorks CS as the basis for developing a continuous long-term (five-year) rainfall period, which is being used to evaluate sewer flows in a collection system model.

Ultimately, the city aims to develop a 30-year plan for its combined sewer and wastewater treatment system, a major cornerstone of which is the creation and application of a detailed hydraulic model of the sewer system. This will evaluate the effect of various alternatives on sewer system flows, as well as CSO activation frequencies and volumes.

To undertake this work, a set of design storms was needed to properly evaluate system performance. As well as traditional single-event design storms, a continuous rainfall period is required to evaluate CSO operation over several years.

Climate Challenge

San Francisco’s climate, which is heavily influenced by its geography, posed a major challenge. The city sits on a peninsula that extends into the Pacific Ocean, which keeps temperatures in the city within a range of 50ºF to 70ºF in summer and 45ºF to 60ºF in winter. The city’s varied topography also creates distinct ‘microclimates,’ resulting in considerable weather variations between locations.

San Francisco receives around 21.6 inches of rain per year, with 95% falling between October and April, and 80% concentrated between November and March. Snow is rare.

Rain Gage Data

The city has historical rain data from three primary sources: National Weather Service (NWS) rain gages; San Francisco Hydraulic-Hydrologic Data Acquisition rain gages; and the San Francisco Public Utilities Commission’s rain gage network.

There are two historic NWS gages in San Francisco, one at the Oceanside wastewater treatment plant and the other in Duboce Park. This latter gage, known as the Federal Office Building (FOB) gage, has a reliable recording period dating back to 1907, and consequently data from it was used to select the typical period.

Citywide rain gage data is also available from the two other sources. Since rainfall can vary substantially from area to area, the typical period was chosen so that it fell within one of two periods for which there was citywide gage data.

Defining Storm Events

The minimum inter-event time defines the number of dry hours that have to occur between storms before the rainfall can be considered to be a new storm event. This is often calculated statistically by choosing a value that gives a coefficient of variation of 1 for the inter-event times. This does not work well on the west coast, however, where the resulting inter-event times are frequently 300 hours in duration.


San Francisco’s climate, which is heavily influenced by its geography, posed a major challenge in developing a typical rainfall period for the city.
Click here to enlarge image

To avoid having such a large inter-event time, a value of 24 hours was used to define storm events, based on the length of time needed to empty the CSO storage boxes within the city as simulated in the calibrated InfoWorks CS model.

Analyses were undertaken using the hourly rainfall data from the FOB gage. Rainfall statistics were collated for each storm event during the available record period, from 1907 to 2005, including total storm depth, peak intensity and duration.

Annual rainfall statistics, including total annual rainfall depth and number of storms, were also calculated from 1907 to 2004 (the complete 2005 and 2006 rainfall data were not available at the time). Storm depth is important for CSO control measures that involve storage, while peak intensity controls the occurrence of individual CSO activations.

To develop a typical period, a five-year span that closely resembled the long-term average in terms of number and distribution of storms was chosen. Individual storms were then added and removed so that the resultant five-year period matched average trends. The first step in this process was defining the long-term statistics that the typical period should match.

Long-Term Rainfall Trends

A number of factors affect long-term rainfall averages in San Francisco, and it was important to take these into account. Rainfall can vary spatially, so it was important to consider the fact that one key rain gage had moved twice (though analysis showed that the changes were not significant).

Global climate phenomena also significantly affect San Francisco. El Niño, for example, produces a marked and rapid warming of the Tropical Pacific Ocean that can significantly increase the amount of rainfall the city experiences. Spikes in annual rainfall at the NWS gage (1973, 1983 and 1997) have been proved to coincide with particularly strong El Niño years.

The Pacific Decadal Oscillation can also lead to changes in climate patterns. This longer-term phenomenon has been indicated by research that shows temperatures in the North Pacific follow a pattern of warming and cooling taking place over a number of decades. When the PDO Index (a number calculated based on the spatial average of temperatures in the Pacific Ocean north of 20º) is warm, or positive, the southern half of the US receives higher-than-average rainfall. The last change was in 1997, when the PDO shifted back to warm.

Global climate change also appears to be increasing rainfall in San Francisco, though it is not clear whether this trend will continue.

Based on this analysis of rainfall factors, the city determined that the typical period should not match overall historical averages but should instead match the averaged characteristics of the past 30 years of rainfall. This approach will create a more conservative typical period that takes into account recent increases in the overall rainfall and the warmer PDO period. Because climate change models vary so widely in their predictions, it was decided not to hypothesize any increase in rainfall from this source.

Spatial Variation

Because of the variability in the amount of rainfall experienced during individual storm events, it was important to account for spatial variation when choosing the typical period. To illustrate the potential range of rainfall, factors were created for each rain gage in the city based on the 50 largest storm events in terms of depth and, for a similar number, by peak intensity as recorded by the NWS rain gage.

Storm depths and peak one-hour intensities were calculated for these storms for each city rain gage. Depth and intensity factors were calculated by dividing the storm value recorded at a given rain gage by the value at the NWS gage. It was decided that when spatial variation was taken into account a longer simulation period might be better to account for regional variations – while a year may be considered typical at one gage, it may not be at other points in the system. Selecting a longer period – in this case, five years – minimizes the effects of year-to-year variation.

The years 1977 to 1981 were chosen as the basis of the five-year typical period. These five years fell within the first of two periods containing widespread rain gage data from across the city and closely resembled the long-term period of record.

Although other periods with available data were closer to the 30-year average in terms of overall depth, this period was chosen because it most closely matched the 30-year average in terms of the distribution of storms – in particular intensity and depth ranges. Storms were then added or removed to make it more closely match the typical 30-year average.

Conclusions

The calibrated InfoWorks CS model of the San Francisco collection system was used to simulate the five-year typical period. The results of this simulation were evaluated to determine the frequency of CSO activation. Activations occurring within six hours of another overflow are considered the same discharge event, and any separated by more than six hours were considered to be separate events.

The results of the five-year typical period simulation were compared to historical records for the past 20 years and the results were found to agree very closely, indicating that the typical period is suitable for use in predicting average overflows.

About the Authors:

  • Sharon Tsay is a modeling specialist with Metcalf & Eddy (Wakefield, MA).
  • Dominique Brocard, PhD, PE, is director of Metcalf & Eddy’s hydraulic modeling group.
  • Wallis Lee, PE, and Greg Braswell are members of the Bureau of Engineering at the city of San Francisco’s Department of Public Works.

More in Environmental
Potable Water Management 4.0
Sponsored
Potable Water Management 4.0