An Intercomparison of Precipitation Values from the OneRain Corporation Algorithm and the National W

1
An Intercomparison of Precipitation Values from
the OneRain Corporation Algorithm and the
National Weather Service Procedure Steven M.
Martinaitis, Henry E. Fuelberg, John L.
Sullivan Jr. , and Chandra S. Pathak Departmen
t of Meteorology, Florida State University
South Florida Water Management District
Introduction
Results Statistics for 2005
Results Independent Gauges

• Scatter plot for 2005 calendar year supports the
  spatial maps of OneRain having overall greater
  rainfall totals than MPE.
• The majority of the points (each point represents
  a pixel) lie above the one-to-one line (dashed)
  except for the more extreme precipitation values
  where MPE gt OneRain in most cases.
• Difference histograms shows a bimodal
  distribution with 52.1 of pixels showing
  OneRain 2.0 in. to 16.0 in. greater than MPE.
• Mean Areal Precipitation (MAP) MPE 57.025
  in. OneRain 59.317 in.
• MAP Difference (MPE OneRain) -2.292
  in.Percent Difference of MAP -4.018Standard
  Deviation of Differences 6.064 in.Correlation
  0.796

• To better determine which algorithm provided
  better rainfall estimates, both were placed
  against a set of 13 NCDC daily COOP gauges
  (Right Map of gauge locations).
• Each gauge is compared with the respected HRAP
  44 km grid cell for MPE and OneRain from 2
  January 2005 to 31 December 2005. All days that
  showed the gauge and the algorithm equaled to
  zero were removed for statistical analysis.
• The table below gives the mean daily rainfall
  difference (Gauge Algorithm), the standard
  deviation of those differences, and the
  correlation for all gauges for the entire time
  period, the cool season (November to April), and
  the warm season (May to October) versus MPE and
  OneRain.

• Accurate, reliable rainfall data are needed for
  flood forecasting and regulatory decisions.
• To counteract the sparseness of rain gauges,
  radars provide continuous high spatial (1 km,
  1-degree azimuth) and temporal (5-10 min.)
  resolution using the National Weather Service
  (NWS) WSR-88D Doppler radar network.
• An optimum combination of rain gauge and radar
  data maximizes the strengths of each component
  while minimizing their inherent limitations. We
  compare results from two such algorithms NWS and
  OneRain Corp.
• We consider the 2005 calendar year over southern
  Florida. The datasets provided are compared on a
  pixel-by-pixel basis, within selected watersheds,
  and against an independent gauge dataset.

Data and Methodology
South Florida Water Management District

• The South Florida Water Management District
  (SFWMD) consists of 140 watersheds covering
  42,082 km2. Our study area encompasses the
  entire SFWMD and extends outward to include an
  35 mile buffer.
• Left The SFWMD with watersheds selected for
  study in dark shading. Those watersheds are I.
  Boggy Creek, II. North St. Lucie, III. East
  Caloosahatchee, and IV. Tamiami East.

• Below Monthly MAP values for the Boggy Creek and
  Tamiami East watersheds.

• Below Scatter plots for all gauges for the
  entire time period, the cool season, and the warm
  season versus MPE and OneRain. Each point
  represents a daily pairing of a gauge and the
  respected algorithm. The anomalous points on the
  warm season and overall scatter plots occurred
  during two land-falling tropical cyclones
  Hurricane Katrina and Hurricane Wilma.

NWS/FSU Multi-sensor Precipitation Estimator

• Historical database created at The Florida State
  University (FSU) combines rain gauge data
  provided by the five water management districts
  (WMDs) and the National Climatic Data Center
  (NCDC) with hourly digital precipitation arrays
  (DPAs) provided by the NWS Southeast River
  Forecast Center (SERFC). All gauges undergo
  quality control described by Marzen and Fuelberg
  (2005).
• The final hourly FSU product is created using the
  NWS Multi-sensor Precipitation Estimator (MPE)
  code placed on the Hydrologic Rainfall Analysis
  Project (HRAP) 44 km grid and is used by the
  Florida Department of Environmental Protection
  (FDEP).

Case Study Hurricane Wilma

• Time Period Under Consideration 23 October 2005
  at 1700 UTC to 24 October 2005 at 2000 UTC
  (elapsed time is 27 hours)
• Right Accumulated precipitation during study
  time period for both algorithms
• Below Line graphs of hourly MAP for both
  algorithms and the difference between the two
  (vertical line represents landfall on 24 October
  2005 at 1030 UTC)

OneRain Corporation Algorithm

• The OneRain Corporation provides the five Florida
  WMDs with a near real-time product and an
  end-of-month edited product. This study uses only
  the edited product.
• The algorithm is proprietary, and its
  characteristics are unknown. The final OneRain
  product is placed on a 22 km Cartesian grid, and
  the data are provided at 15 min. intervals.

Methodology

• Time series plots of annual summation of rainfall
  are shown below for the gauges Fort Lauderdale,
  Ortana Lock 2, and Perrine 4W. The gauge is
  represented by the green line while MPE and
  OneRain are represented by the red and blue lines.

• To accurately perform the inter-comparison, both
  datasets were placed on a common grid at a common
  time interval with minimum loss of data quality.
• For a pixel-by-pixel comparison, the OneRain 22
  km Cartesian data were mapped onto the HRAP 44
  km grid using an area weighted averaging
  technique within ArcGIS based on the percentage
  of coverage of 22 km cell. The same technique is
  used to calculate precipitation within the
  selected watersheds.

Results Entire Grid for 2005

• From left to right OneRain precipitation for
  2005, NWS/FSU MPE precipitation for 2005, and
  annual difference (MPE OneRain Difference).
• For annual difference, OneRain gt MPE is
  brown/yellow shading, and MPE gt OneRain is
  green/blue shading.

Conclusions

• In the comparison of the NWS/FSU MPE product and
  the OneRain Algorithm, OneRain shows greater
  rainfall totals over the majority of the study
  area. NWS/FSU MPE tends to show greater values
  with more extreme totals.
• Hurricane Wilma case study shows a change in
  precipitation estimation upon landfall. It is
  inconclusive whether this was due to a Z-R
  relationship change or how each algorithm handles
  tropical events.
• With independent gauges, OneRain has a smaller
  overall mean daily difference while NWS/FSU MPE
  have a smaller standard deviation of these
  differences and better gauge-to-pixel
  correlation. Time series plots of accumulating
  rainfall display how both algorithms vary versus
  individual gauges.
• Neither one of the algorithms stand out as a
  better performing product.
• Future Work Compare over 2004 calendar year and
  WAM hydrologic modeling
• Reference Marzen, J., and Fuelberg, H.E. (2005).
  Developing a high resolution precipitation
  dataset for Florida hydrological
  studies. 19th Conf. on Hydrology, AMS, San
  Diego, CA.

• OneRain produces greater rainfall totals prior to
  landfall while MPE yields greater rainfall totals
  while the eye is over land. However, we were
  unable to prove whether this is due to the
  WSR-88Ds being switched over to the tropical Z-R
  relationship.
• Below Example hourly scatter plots prior to,
  during, and after landfall