A systematic review approach of long term metropolitan rainfall data from a large set of rain gauges

Qiuli Lu and Hazem Gheith, ARCADIS U.S., Inc., Columbus, OH, USA, Rob Herr, Greg Barden and Fang Cheng, City of Columbus, Columbus, OH, USA


The City of Columbus, Ohio began collecting continuous 5-minute rainfall data records in 1995. A total of 42 rain gauges (RGs) have been installed to date with data record periods spanning from 2 to 20 years. The City also has access to an additional 12 years of 5-minute rainfall records (starting in 2003) from 30 State of Ohio Rain Monitoring Systems (STORMS) RGs that were spatially disturbed across Franklin County. These 72 RGs are distributed over Columbus’ 430,000 acres of service area and provide a good resolution for the rainfall spatial variability.

ARCADIS developed criteria for a comprehensive, qualitative review process to facilitate analysis of this large rainfall dataset. The RAINSTAT software, developed by Dr. Hazem Gheith, was used to process all RG data into citywide event datasets. The software tool identified all events with at least one gauge in the network that had a 5-minute rainfall recorded value greater than 0.1 inch. Each citywide event was defined by its start and end date using a 3-hour inter event time. At the RG level, individual RG events with their own start and end date were also identified. In addition, total rainfall, number of non-zero time steps, maximum intensity and storm return frequency were summarized into the RAINSTAT output dataset. Approximately 90,000 RG events were identified for the quality review. Missing events due to gauges downtime were also identified.

A systematic approach was developed to accelerate the data quality review process. The data review process was performed by comparing each RG to its five surrounding gauges. Malfunctioned gauges during an event were excluded from the comparison. A stepped approach was applied to screen and flag questionable RG events. An initial screening process was performed as follows:

First, individual RG events where the difference between the event total rainfall and the surrounding gauges average rainfall was greater than 0.2 inches were identified. The RG event was then flagged if one of the following criteria was met:

Similarly, individual RG events where the difference between the RG event non-zero time steps and the surrounding average was greater than 10 time steps were also flagged.

Finally, individual RG events where the maximum intensity was greater than 4 inches per hour were identified. The RG event was then flagged if the following criterion was met:

The flagging process was an automated procedure and resulted in approximately 9,000 questionable individual RG events from the above three step initial screening processes. They were then reviewed in further detail. In most cases, it was visibly apparent that the RG was malfunction and the individual RG event should have been flagged. The remaining RG events were then compared to the rainfall distribution over the City before it was accepted as a legitimate individual RG event record.

The proposed systematic review approach of the large set of rainfall data saves tremendous time. Random checks proved that the set criteria used in the review approach captured almost all events with ambiguous rainfall records. Only the final few hundred questionable individual RG events (of the original 90,000 RG events) needed more detailed reviewing to assure quality. To accelerate the last review step, a GIS based tool was developed to put rainfall data on a color ramp of the interpolated surface. This surface was interpreted from the total rainfall from valid gauges by inverse distance weighing (IDW).  This rainfall visual representation provided further understanding of rainfall spatial distribution and helped to filter out invalid individual RG events.

The processed rainfall data provided a strong confidence for using the extended rainfall records and its spatial distributions for long-term simulations in various Columbus projects.


Permanent link: