1. Automated Real-Time Operational Rain Gauge Quality-Control Tools in NWS Hydrologic Operations Chandra R. Kondragunta 1 and Kiran Shrestha 2 1 Hydrology Laboratory, Office of Hydrologic Development, National Weather Service, NOAA, Silver Spring, Maryland 2 RSIS Inc., McLean, VA P2.4 Introduction: The Office of Hydrologic Development (OHD), a division of the NWS, supports the NWS hydrological operations by developing, implementing and maintaining hydrological and hydrometeorological models and systems. Inaccurate and inconsistent hydrometeorological data can significantly degrade forecasting processes. Therefore, quality control and development of quality-control tools to support the NWS hydrologic operations have become one of the missions of the OHD. Because of the need to handle high volumes of data to give timely forecasts, it is necessary to provide QC aides and automate these quality-control procedures. In this paper, a conceptual QC model and an automated Spatial Consistency Check (SCC) which is in the current NWS operations are discussed. Spatial Consistency Check: This check is used to identify outliers which are not spatially consistent with the neighboring rain gauges (Kondragunta, 2001). This check flags the outliers in an automated fashion for later manual examination. There are several steps in this check. 1. Outlier Identification: Outliers are identified following a statistical procedure for a small region (1 0 x1 0 lat.-lon. grid box) of the service area. There are four steps involved in the identification of outliers: (i) Determination of median, 25 th and 75 th percentile of the data points under consideration. (ii) Calculation of Mean Absolute Deviation (MAD) as follows: where X med is the median of the data N is the total number of stations X i is the i th value of the data (iii) Calculation of test index (Madsen, 1993) for each station as follows: if (MAD=0) Index=0. Else If (Q 75 Q 25 ) then Index=|X i – Q 50 | / (Q 75 – Q 25 ) Else Index=|X i – Q 50 | / MAD Where Q k is the k th percentile. (iv) The index calculated in the step three is compared to a predefined (user defined) threshold value (typical value is 2). If the index is greater than the predefined threshold value, then the rain gauge data is flagged as an outlier. The procedure described so far is applied to all gauges that fall in a 1 0 x1 0 latitude-longitude grid box. Automation section describes how this check is applied to entire service area. 2. Automation: The first step is repeated in an automated fashion using moving window technique for the entire service area. If a gauge is picked four times as an outlier, then that gauge is finally flagged as an outlier. This procedure ensures that gauges are checked for spatial consistency in all four quadrants. 3. Convective Screening: Since gauges receiving rain fall from a convective system don’t have to be spatially consistent with their neighbors, a convective screening technique is used to make sure that the flagged gauge was not under the influence of a convective system. This step uses lightning data to identify gauges that are under the influence of a convective system. If there exists at least one lightning strike within approximately 10km. Radius from the gauge during the past one hour, then that gauge is removed from the outlier list and the gauge is considered valid. Results and Discussion: The SCC check was applied to two cases (Aug 09, 2005 and Aug 17, 2005) in the Mid Atlantic River Forecast Center (MARFC) region. On average, there were about 560 gauge reports, of which about 20 rain gauges were flagged as outliers by the SCC. The flagged gauge reports were manually edited using the Graphical User Interface tool available in the Multi-sensor Precipitation Estimator (MPE) software (Lawrence, et al. 2003). Two of the MPE outputs Gauge-Only analysis (Seo, 1998a) and Multi-sensor analysis (Seo, 1998b) were validated against daily cooperative rain gauge data. Presented in Figures 1a-b are the scatter plots between validation gauges and gauge only analysis and multi-sensor analysis from MPE before the rain gauge data were quality controlled for the two cases combined. Presented in Figures 2 a-b are same as Figures 1a-b, except for the MPE output after rain gauge data were manually quality controlled by removing gauges flagged by SCC. Bias ratio, Root Mean Square Error (RMSE) and correlation coefficient (COR.) are calculated and presented in the scattered diagrams. These results indicate that RMS error and correlation coefficient have improved considerably for the gauge-only analysis from before QC to after SCC QC. There is also improvement in the multi-sensor analysis, but the improvement is less than gauge only because the multi-sensor field tends to minimize the errors. The bias ratio was not improved because only outlying gauge values were quality controlled from this check. References: Kondragunta C. R., 2001: An outlier detection technique to quality control rain gauge measurements. Eos Trans. Amer. Geophys. Union, 82 (Spring Meeting Suppl.), Abstract H22A-07A. Lawrence, B. A., M. I. Shebsovich, M. J. Glaudemans and P. S. Tilles, 2003: Enhancing precipitation estimation capabilities at National Weather Service field offices using multi- sensor precipitation data mosaics. 83 rd AMS Annual Meeting, 19 th International Conference on Interactive Information Processing Systems for Meteorology, Oceanography and Hydrology, Long Beach, California, February 9-13, 2003. Madsen, H. 1993: Algorithms for corrections of error types in a semi-automatic data collection. Precipitation Measurement and Quality Control. B. Sevruk & M. Lupin (eds.), Proc. of Symp. on Precipitation and Evaporation, Vol. 1, Bratislava, Slovakia, September 20-24, 1993 Seo, D-J., 1998a: Real-time estimation of rainfall fields using rain gauge data under fractional coverage conditions. J. Hydrol. 208, 25-36. Seo, D-J., 1998b: Real-time estimation of rainfall fields using radar rainfall and rain gauge data, J. Hydrol., 208, 37-52 Figures 1a-b. Scatter plot between Co-op daily gauges (Y- axis) and gauge- only analysis (X- axis) (left) and multi-sensor analysis (right) before QC for combined cases Aug 09 and 17, 2005. Figures 2 a-b. Same as Figures 1a-b, except for after QC. ab a b Operational Hydrologic data flow & Conceptual QC Model Conceptual QC Model: Level - I QC: Checks in this level are performed on an individual datum observed by a sensor. Example: Gross Error Check Level - II QC: Checks in this level also involve a single observation. Observations in this level are checked against some boundaries for their validity. Example: Climatological Range Check Level - III QC: Checks in this level are based on an observation being checked against other independent observations of the same type or different type from different source. Examples: Spatial Consistency Check, Multi Sensor Check, Temporal Consistency Check etc. Level - IV QC: Checks in this level are based on human expert judgment. Example: Manual QC checks MARFC Aug 09 & 17, 2005 (No QC) Gauge-OnlyMARFC Aug 09 & 17, 2005 (No QC) Multi-sensor MARFC Aug 09 & 17, 2005 (SCC QC) Gauge-OnlyMARFC Aug 09 & 17, 2005 (SCC QC) Multi-sensor BIAS = 0.83 RMSE = 22.16 COR. = 0.25 BIAS = 1.53 RMSE = 8.04 BIAS = 1.0 RMSE = 11.1 BIAS = 1.23 RMSE = 6.47 COR. = 0.68 COR. = 0.55 COR. = 0.8 Conclusions: On average Spatial Consistency Check flagged 3.6% of gauges as outliers. Validation results indicate considerable improvement in the RMS error and correlation coefficient for both gauge-only analysis and multi-sensor analysis.