Short Time-Interval Rainfall Disaggregation for Continuous Hydrologic Simulation
Traditionally design storms have been used to design and analyze urban drainage systems and hydraulic structures. Design storms can be developed with the desired temporal resolution to accommodate urban hydrology needs, but because the temporal distribution is generally arbitrary the application of complex disaggregation techniques is unwarranted. Continuous hydrologic simulation is recommended as an alternative to the traditional design storm approach for the design and analysis of hydrologic and hydraulic structures for reasons discussed in James (1994) and James and Robinson (1982). Continuous simulation models require long-term rainfall records (preferably more than 50 years) to generate the long-term statistical response of the hydrologic system required for accurate design and analysis of engineering systems and the evaluation of ecological and sustainability issues.
Accurate hydrologic simulation of small urban catchments requires the use of a rainfall time series with a fine temporal resolution. Studies have shown that when the response time of a watershed is shorter than the total duration of rainfall excess, the runoff rate is observed to depend on the depth of rainfall and the intensity distribution (Ball 1994; Woolhiser and Goodrich 1988; Hjelmfelt 1981). But for fully developed hydrographs Ball (1994) found the temporal pattern of rainfall excess to have little influence over the peak discharge. Thus, for short duration storms coarse time resolution rainfall data may smooth the high rainfall intensities (especially those observed during convective storms), and runoff could be underestimated. Hernandez and Nachabe (2000) demonstrated that when Hortonian runoff is dominant, infiltration and runoff are very sensitive to time resolution. They observed finer temporal resolution rainfall to produce more runoff than coarser rainfall. In general, hydraulic analysis of drainage systems requires rainfall data in 5- to 15-minute increments to produce hydrographs that accurately predict peak flows (Nix 1994).
The procurement and management of long-term rainfall records is no longer a problem for locations where records are available electronically. Today, the primary difficulties with long-term rainfall records are (1) unavailability at the desired location or (2) not being recorded at the desired temporal resolution. One solution to these problems would be to employ a synthetic rainfall generator to produce long-term rainfall fields with the desired spatial and temporal resolution. A second solution for circumstance (2) (i.e., when a long-term rainfall record exists but has too coarse temporal resolution) is to employ a temporal disaggregation technique to disaggregate the record into a finer temporal resolution. The issue then becomes the selection and application of an appropriate disaggregation method to produce a long-term rainfall record at the desired temporal resolution.
This chapter compares several temporal rainfall disaggregation techniques applicable to continuous hydrologic simulation. The focus is the disaggregation of hourly rainfall records into sub-hourly increments because in North America hourly rain gauges are relatively common and the records often have sufficient lengths for use in long-term continuous simulation. The rainfall disaggregation methods included in the study were selected based on the needs of hydrologic modelers. In general, hydrologic modelers desire techniques that are conceptually intuitive, easily grasped, and sufficiently flexible that they could be applied to any locality and for any desired level of disaggregation (so they could be relatively easily standardized). Based on these criteria, the five methods selected for comparison were the uniform distribution approach (described below), the quadratic spline and quadratic interpolating polynomial approaches (described by Durrans et al. (1999)), the geometric similarity approach (the continuous-deterministic disaggregation model described by Ormsbee (1989)), and the backpropagation ANN approach (described by Burian et al. (2000)). Methods that require the estimation of numerous parameters were not included in this study. The relative performance of the five techniques for disaggregating hourly rainfall records from Alabama into 15-minute increments is reviewed below. Additional evaluation of the uniform distribution, the geometric similarity, and the ANN techniques is reported for 5-minute and 15-minute rainfall in Arkansas.
This paper is only available in PDF Format:
View full text PDF