Hydrological model sensitivity to small-scale rainfall forecasts

Verification scores are typically used to evaluate the effects of enhanced data measurements, more advanced equipment, and better user training on precipitation forecasts. Different verification techniques may emphasize disparate storm characteristics, and analysis of several storms with unique values of a single verification measure may highlight the contrast between the storm events. This work evaluates the sensitivity of an urban hydrological model to three separate verification measures. Errors in simulated water levels associated with forecasts with different verification scores are acquired for various precipitation patterns over New York City. The error values allow an advance estimate of expected street water levels, given a forecast, its verification scores, and a simulated hydrologic situation based on the forecast. The water level predictions can be used to allocate maintenance crews to resolve possible steam tunnel leaks.

The steam distribution system in New York City includes 105 miles of steam tunnels in the southern half of Manhattan, which are driven by seven co-generation plants. Steam travels from the co-generation plants to clients through pipes enclosed in concrete housing structures. Steam pipe ruptures can occur when water enters the housing structures and comes into contact with the pipe, which is defined as an impingement. A sufficient amount of cold water may cause the flowing steam to condense and form pools of water. If the water is not removed by steam traps or pumping, waves can occur on the water surface and may be carried forward by the pressurized steam. The water gains momentum and may cause a water hammer, which can damage pipe walls and valves. Steam rising from the ground may indicate critical, intrusive water levels that are almost in contact with the steam pipe. In order to minimize the risk of steam pipe failure in New York City, work crews currently respond to steam sightings throughout the city to remove any water accumulation near the pipe.

External water most commonly approaches the steam system through rainfall (any water on the streets that enters the manholes), groundwater infiltration, and water main leaks. Analysis shows that critical water levels in pipe housing structures and steam traps are not correlated to nearby Hudson River levels or New York Bay levels, thus eliminating the consideration of groundwater intrusion. The Watson Hydrological Model (Cordazzo et al, 2009¹) is used to calculate water accumulation on city streets during rainfall events as an indicator of how much water enters manholes and influences the steam system. The model is run with historical NCEP Stage IV precipitation estimates. Calibration indicates that high water levels occur near locations where steam impingements are observed during three storm events that were evaluated during August through October 2008. Correlations are computed between steam impingement records and street water levels. The point biserial coefficient is used to calculate the correlation between the binary impingement records and continuous water level data, and should be close to 1 for all locations indicating that street water levels are good indicators of impingements.

Watson Hydrologic Model (WHM) calibration to historic storm events yields threshold street water levels that cause impingements. Precipitation forecasts for more recent storm events can be used within WHM to calculate water depths, and given the pre-computed thresholds, can be used to estimate impingements. The WHM is coupled with the mesoscale Weather Research and Forecasting System (WRF). The WRF (ARW core, version 3.1.1) yields data in QPF format for input to WHM. For storms for which both Stage IV observations and WRF forecasts are analyzed, the differences in WHM computed street water levels will be matched with the corresponding verification scores. The Model Evaluation Tools (MET) verification package is used to compare Stage IV rainfall observations and the WRF model forecasts. WRF results are conformed to MET standards with the WRF Post Processor. Reports by the National Climate Data Center (NCDC) indicate that in the past two years, New York City has been subject to multiple severe storms of varying types, including supercell thunderstorms and coastal storms. Diagnostic verification measures are used to analyze a selection of these severe storm types.

Once the verification scores for each type of method are obtained, the differences in simulated street water level between the Stage IV and forecasted WRF rainfall for each storm event are assessed with relative error scores. The verification scores are compared to the water level errors for a variety of storm events and indicate what magnitudes and types of verification errors lead to the largest errors in water level predictions throughout southern Manhattan. These relationships may be superimposed on Watson Hydrological Model outputs to obtain better water level and impingement predictions when WHM is coupled with WRF forecasts. Introducing possible changes to simulated results given certain forecast errors will assist with more effective management of the steam tunnel system. Site visits may be limited to those that will most likely witness critical threshold water levels during a given storm event instead of every steam sighting. It may also be determined that more steam tunnel impingement data for model calibration and water level threshold assessment may be needed before resource allocations can be reliably based on precipitation forecasts.

¹Jonas Cordazzo, U. Mello, E. Novakovskaia, D. Rude, L. Treinish, and A. Praino. Urban Flood Forecasting using an Integrated Hydrometeorological System. Symposium on Urban High Impact Weather. 89th Annual Meeting of the American Meteorological Society, Phoenix, AZ, January 2009.