th 90 AMS Annual Meeting 20th Conf on Probability and Statistics 17–21 January 2010, Atlanta, Georgia 219 REAL-TIME OBJECTIVE ANALYSIS OF SURFACE DATA AT THE METEOROLOGICAL DEVELOPMENT LABORATORY Jung-Sun Im*, Bob Glahn, and Judy E. Ghirardelli Meteorological Development Laboratory Office of Science and Technology National Weather Service, NOAA Silver Spring, Maryland 1. INTRODUCTION analyses of the surface observations, we found various issues such as inconsistent site informa- As part of the Localized Aviation Model Output tion for stationary stations, redundant stations re- Statistics (MOS; Glahn and Lowry 1972) Program porting data at the same locations with different (LAMP; Ghirardelli 2005), the Meteorological De- station names and types, simultaneous multiple velopment Laboratory (MDL) is analyzing surface reports with different station types, stations re- data reports on an hourly basis. The analysis peatedly reporting the same values, and spatial scheme used by MDL for gridding MOS forecasts and temporal discontinuities in the analyses. (Glahn et al. 2009) has been tailored to analyze surface observations. MDL is making the analy- In this paper, we describe the intensive effort ses to assess the accuracy of gridded MOS and needed to 1) assure the metadata are correct for LAMP forecasts. In addition to providing verifica- each location, 2) develop efficient quality control tion grids for gridded MOS and LAMP forecasts, it procedures, 3) assign a representative land, is our goal to add gridded LAMP nowcasts to the ocean, or inland water flag to each station, and gridded LAMP forecast suite. These accurate 4) alleviate spatial and temporal discontinuities in high-resolution analyses will eventually help fore- the analyses. This paper focuses on the analyses casters create and verify the National Digital Fore- of temperature and dewpoint over the contermi- cast Database (NDFD; Glahn and Ruth 2003). nous United States (CONUS) on the 5-km resolu- tion NDFD grid with surface observations archived Real-time and retrospective analyses at both a since August 2007. high spatial and temporal resolution are required to establish an Analysis of Record (AOR; Horel 2. THE BCDG ANALYSIS METHOD and Colman 2005), and to create the NDFD fore- casts as well as to verify their accuracy. As a first In support of NDFD, MDL has produced grid- step, a prototype Real-Time Mesoscale Analysis ded MOS forecasts since 2006 (Glahn et al. (RTMA; De Pondeca et al. 2007a; Benjamin et al. 2009). The objective analysis scheme used to 2007) was produced at the National Centers for produce gridded MOS is based on the successive Environmental Prediction (NCEP) in collaboration correction technique called Bergthorssen- with the Earth System Research Laboratory Cressman-Doos (BCD; Glahn et al. 1985; Cress- (ESRL). It represents a fast-track, proof-of-concept man 1959; Bergthorssen and Doos 1955). This of the AOR program and establishes a benchmark successive correction technique consists of mak- for future AOR efforts (De Pondeca et al. 2007b). ing multiple passes over the data, correcting each In addition to RTMA, MDL analyses can be used grid point on each pass by comparing with the to judge the quality of an AOR. data in the immediate vicinity. For gridded MOS, this BCD technique was modified by implementing High quality surface weather observations and the following specific features: effective quality control processes are critical to generate high-resolution objective analyses. The 1) separate analysis systems for land, inland hourly surface observations for the analyses are water, and ocean combined into one, to ac- obtained from NCEP in real time and are addition- commodate the different characteristics as- ally quality controlled at MDL. While performing sociated with land and water, 2) computation on-the-fly of vertical change of a weather element with elevation, so that the * Corresponding author address: Jung-Sun Im, vertical change varies with the location, time National Weather Service, Meteorological Development of day, day of the year, and synoptic situa- Laboratory, 1325 East-West Highway, W/OST21, Silver Spring, MD 20910; email: [email protected] tion, 1 3) a variable radius of influence (R) for land threshold, S is accepted. Otherwise, S is not used and for water points for each specific correc- on this pass. tive pass to account for highly varying data densities, More detailed information on the BCDG tech- 4) error detection which employs a buddy nique such as the gridpoint correction algorithm, check when a datum is in serious question, determination of vertical change with elevation, and and accommodation for land and water can be 5) a terrain-following smoother. found in Glahn et al. (2009). Based on extensive experimentation performed at MDL, we adopted With theses major extensions, the BCD scheme the BCDG options used in gridded MOS, which was thereafter called Bergthorssen-Cressman- incorporate a first-guess grid composed of the av- Doos-Glahn (BCDG; Glahn et al. 2009). erage value of the element, a four-pass setup to capture the desired detail in the analysis, limitation The BCDG analysis system has many options of unusual lapse rates when the computed vertical that can be used to tune the system based on data change is of the opposite sign than expected, and density relative to gridpoint density, variation in a terrain-following smoother. data density over the grid, choice of first-guess field, number of corrective passes, smoothness 3. DATA COLLECTION AND INITIAL QUALITY versus detail desired in the analysis, and error CONTROL characteristics of the data. In analyzing surface observation data, BCDG’s error checking capabil- Hourly surface observations are obtained from ity is an essential part of the analysis of the data. NCEP in real time and are additionally quality con- The BCDG software performs this error checking trolled at MDL. The first set of quality control on each pass based on an acceptable difference checks at MDL ensures that all temperature and (threshold) between the station value and the dewpoint observations are in an acceptable range value interpolated from the analysis. Based on for the station’s geographical area, and the tem- considerable testing and meteorological judgment, peratures are greater than or equal to the dew- we have determined the threshold values for each points (Glahn and Dallavalle 2000; Allen 2001). In pass. The procedures of BCDG’s error checking preparing input observation data to be used in the are summarized in Fig. 1. On each pass, the dif- hourly analyses, we collect data observed be- ference between a station’s value (S) and the tween 15 minutes prior and subsequent to the value interpolated from the 1st guess or previous analysis hour. If more than one observation is pass analysis (ΙS) is computed before making an reported for a station, we select the report closest analysis. If the difference is less than or equal to to 10 minutes prior to the analysis hour. The the threshold (Th) specified for each pass, S is analysis system for temperature and dewpoint as- accepted for that pass, but if it exceeds 1.5 times similates six types of observations obtained from the threshold, S is discarded; if it exceeds the METAR (roughly translated as Aviation Routine threshold, but is less than or equal to 1.5 times the Weather Report; OFCM 1995), mesonet, synoptic, threshold, then the two closest neighbors’ values moored buoy, Coastal-Automated Marine Network to S (N1 and N2) are found to perform buddy (C-MAN), and tide gauge stations. checks before S is discarded. The differences of N1 from its interpolated value (ΙN1) and N2 from its 3.1 METAR interpolated value (ΙN2) are computed. If either one of the two neighbors’ differences is greater METAR reports typically come from airports or than 0.6 times the threshold, and the differences permanent weather observation stations. Obser- of both S and its neighbor are of the same sign, vations are taken by automated devices or trained then S is accepted. If not accepted, one more personnel. Some stations have automated obser- check is performed. If either one of the two vations augmented by human observers. METAR neighbors’ differences is less than or equal to 0.6 reports are of high quality, and we have found that times the threshold and the difference between S they are more reliable than all other observational and the neighbor’s value adjusted for terrain data sets. (AN1 or AN2 accordingly) is within 0.6 times the 2 0 0.6∗∗∗Th Th 1.5∗∗∗Th |S-ΙΙΙS| ≤≤≤ Th Th < |S-ΙS| ≤≤≤ 1.5∗∗∗Th |S-ΙΙΙS| > 1.5∗∗∗Th Accept S Buddy Check Discard S Two closest neighbors (N1, N2) are found to perform “buddy checks” before S is discarded. |N1- ΙΙΙN1| > 0.6∗∗∗Th & (S-ΙΙΙS)(N1- ΙΙΙN1) > 0 (Errors in the same direction) Or |N2- ΙΙΙN2| > 0.6∗∗∗Th & (S-ΙΙΙS)(N2- ΙΙΙN2) > 0 (Errors in the same direction) Accept S If not accepted |N1- ΙΙΙN1| ≤≤≤ 0.6∗∗∗Th & |S-AN1| ≤≤≤ 0.6∗∗∗Th Or |N2- ΙΙΙN2| ≤≤≤ 0.6∗∗∗Th & |S-AN2| ≤≤≤ 0.6∗∗∗Th Accept S FIG. 1. BCDG’s error checking procedures. 3.2 Mesonet 3.3 Synoptic Mesonet observations are obtained from local, Synoptic data are comprised of manual and state, and federal agencies and private mesonet automatic observations, and are available every sites. These sites are quite dense compared to 3 or 6 hours. In many cases, these data are re- METAR sites. In fact, over 80% of the stations dundant to the METAR data at the same location used in the BCDG analysis consist of mesonet (this issue will be discussed in section 4.2). type stations. 3 3.4 Buoy, C-MAN, and tide gauge tion that 1°F is an allowable error range in a tem- perature analysis (1°F corresponds to a change in We use observations obtained from moored elevation of 280 ft in the standard atmosphere of buoy, C-MAN, and tide gauge stations.