The University of South Alabama Mesonet consists of 26 sites across the north-central Gulf of Mexico coast. Although the original purpose of the mesonet was monitoring landfalling tropical systems, meteorological data are collected and disseminated every 5 min year-round to serve a multitude of purposes, including weather forecasting, education, and research. In this paper a statistical analysis and like-sensor comparison demonstrates that variables, measured by different sensor types or by sensors at different heights, correlate well. The benefits of sensor redundancy are twofold, offering 1) backup sensors in the case of sensor failure during severe weather and 2) the ability to perform a large number of internal consistency checks for quality control purposes. An oceanographic compliment to the University of South Alabama Mesonet system, which was deployed by NOAA’s Atlantic Oceanographic and Meteorological Laboratory (AOML) to measure surface waves and ocean currents in an area south of Mobile, Alabama, is described. A preliminary comparison of mesonet wind data and ocean wave data show good agreement, offering promising opportunities for future research.
Not just federal agencies collect surface weather data. As costs decline and needs increase, more state and locally operated surface meteorological networks are becoming available. These so-called mesoscale networks of weather stations, or “mesonets,” typically collect data at finer space and time resolution than that of the traditional national networks, make data available in (near) real time, and record more variables than just temperature and precipitation. Mesonet data serve a multitude of needs, including forecasting, education, agriculture, leisure, hydrology, and research applications in a wide variety of fields. A survey conducted in 1991 (Meyer and Hubbard 1992) identified 100 such networks and 831 stationary stations in the United States and Canada. Since their survey many more mesonets have been established, including the Oklahoma Mesonet, with 120 stations (Brock et al. 1995; McPherson et al. 2007); the West Texas Mesonet, with 56 stations (Schroeder et al. 2005); the Kentucky Mesonet, with 45 stations (information online at http://www.kymesonet.org/index.html); the Missouri Mesonet, with 16 stations (online at http://www.eas.slu.edu/People/CEGraves/Mesonet/mesonet.html); and the Alabama A&M Mesonet, with 14 sites (online at http://wx.aamu.edu/ALMNet.php).
The University of South Alabama (USA) built its first mesonet station in January 2005, with the primary focus being hurricane and tropical storm landfall monitoring. Hence, most stations were installed in coastal counties and the line of counties to the north of the coastal counties. As of April 2010, 26 stations were fully operational. The location of all of the sites is shown in Fig. 1 and listed in Table 1. The mesonet stretches about 325 km in an east–west direction, across three states. The north–south dimension ranges from about 100 km at the western end to about 30 km in southeast Alabama north of the Florida Panhandle. The spacing between the stations ranges from 5.4 to 55.6 km, with an average of about 30 km. The majority of the station hosts are public schools, which provide a safe, well-maintained environment with free access to the Internet. All of the stations are surrounded by a barbed wire fence, which has no doubt contributed to the fact that no vandalism has been encountered to date. Schools also offer the opportunity to integrate weather station data with public education. Near–real time, archived, and metadata are available online (see http://chiliweb.southalabama.edu/). The University of South Alabama Mesonet maintains strong ties with the National Weather Service (NWS) office in Mobile, Alabama, and provides valuable real-time observations in data-sparse regions of Alabama directly inland from the coastal counties.
An oceanographic compliment to the USA Mesonet system was deployed by National Oceanic and Atmospheric Administration’s (NOAA’s) Atlantic Oceanographic and Meteorological Laboratory (AOML) in an area south of Mobile Bay. Three bottom-mounted acoustic wave and current (AWAC) instrument platforms were set out during the 2007–09 hurricane seasons to monitor surface waves, ocean currents, and bottom temperature. Data from the USA Mesonet and these instruments will help form an integrated picture of offshore conditions during severe storms and to help validate storm surge models.
A technical overview of mesonet hardware, instrumentation, communication, and maintenance, as well as AWAC specifications are given in section 2. Section 3 discusses a statistical analysis and like-sensor comparison as well as a preliminary comparison of AWAC and mesonet data for Hurricanes Gustav and Ike of 2008. A summary and look to the future are provided in section 4.
2. Technical overview
a. Station hardware configuration
Figures 2 and 3 show schematics of the USA Mesonet station layout, while Fig. 4 shows a station photograph. Each station is built within a 9.14 m × 9.14 m fenced enclosure. The fence is 2.44 m tall, including three strands of barbed wire at the top of a chain-link base. A 3.05-m-wide gate is located on the north side of the enclosure. In the center of the enclosure is a 1.22 m × 1.22 m concrete foundation that anchors the tower base plate. The concrete foundation is 0.61 m deep. Two smaller concrete pads measuring 0.46 m × 0.46 m are located to the north of the tower foundation and form the base of the rain gauge supports. The tops of the rain gauge funnels are 82.5 cm above the ground. The wiring from the rain gauges is routed to the tower through a conduit buried underground. At a distance of 1.12 m south of the tower foundation there is a “tower lift anchor pipe” that supports a winch used to raise and lower the towers that are hinged at the northern end of the base plate. Two 3-m-long, east–west oriented cross arms are mounted on the tower to support instruments at 1.5-, 2-, 9.5-, and 10-m elevation. Additionally, a 1.5-m-long, north–south-oriented cross arm is mounted at the 10-m level only. A solar panel is mounted on the tower at about 7 m above ground level (AGL). All stations are grounded using a 2.44-m copper ground rod attached to the tower base by a copper wire. The towers are secured with three heavy duty (4.76-mm diameter) guy wires secured into the ground with 15.88-mm-wide, 1.22-m-long earth anchors. This robust construction was chosen because of the relatively frequent occurrence of severe weather in the area, including tropical storms and hurricanes.
On 29 August 2005 Hurricane Katrina made landfall on the Louisiana–Mississippi state line. Two USA Mesonet stations were operational at the time: Pascagoula and Agricola. Pascagoula, located near the coast, recorded the higher wind speeds of the two sites: a maximum 10-m wind speed of 34.07 m s−1 was measured at 0715 central standard time (CST). Between this time and 1215 CST the wind speeds frequently peaked to around 30 m s−1 and rarely dropped below 25 m s−1. After 1215 CST the winds remained below 25.0 m s−1 until a wind speed of 26.23 m s−1, which was recorded at 1324 CST. Five minutes later, at 1329 CST, several instruments stopped recording data, including the 10-m anemometer. A post storm survey revealed that the tower was leaning sideways by about 5° (Fig. 5); one of the guy wires had snapped, and a bolt in the tower base plate had loosened. The leaning of the tower caused several instrument wires to become disconnected from the datalogger. The leaning of the tower was attributed to the strong winds that occurred over several hours, also evidenced by the roof damage on the apartment building to the northwest of the tower (Fig. 5). The storm surge of Katrina at Pascagoula was not high enough to reach the datalogger enclosure, the bottom of which is 1 m above ground level. While communication was lost to the station, the data were retrieved from the datalogger after the event. After this experience, the station’s foundation, guy wire grade, and earth anchor size were upgraded to the specifications listed above. Stations constructed according to the old specifications were upgraded.
The ability of the new construction to withstand high winds was demonstrated by a high wind gust of 38.25 m s−1 recorded at the Robertsdale site at 0412 CST 27 March 2009. This event was associated with an outbreak of severe weather and a bow echo, which passed directly over the station. Strong west-southwesterly winds tore the temperature–relative humidity (T–RH) probe with its radiation shield off the cross arm at 10-m elevation. The cover of the secondary rain gauge was blown off and later retrieved in front of a line of pine trees about 30 m northeast the tower. The T–RH probe and radiation shield were never found. A house about 1300 m to the west-southwest of the tower incurred significant roof damage with debris strewn to the east and east-northeast of the house. Throughout this event the station tower remained upright and secure.
On 10 November 2009, Hurricane Ida made landfall in Baldwin County, Alabama, as an extratropical storm. Several stations recorded tropical storm–force winds, including the station on Dauphin Island. The tower of this station is located 11.5 m from a then-2.5-m bluff (after the storm the height of the bluff measured only 1.5 m). The distance between the bluff and the water’s edge (between the low and high tide marks) measures about 7.5 m. According to official storm reports, a surge of 0.8 m occurred on top of a coincident high tide, taking the total water height to 1.25 m. The station stopped reporting in real time around 1800 CST 9 November because of power outages on the island. None of the other stations went offline during this event. The site at Dog River, at 3-m elevation, is potentially susceptible to surge, as demonstrated by Hurricane Katrina (2005); the surge on Dog River was just over 4.3 m, which would have impacted the 4-m-high datalogger enclosure at this station had it been installed at the time. However, this was a very rare and extreme event. Hurricanes Georges and Ivan caused a water rise below 4 m on Dog River. Other than Dauphin Island, Dog River, Gasque, and Pascagoula, the USA Mesonet stations are far enough inland or at a high enough elevation where storm surge is not expected to be an issue. Therefore, no specific precautions have been made for storm surge, such as positioning the datalogger enclosure at an elevated height. This would make access difficult for maintenance technicians and was, therefore, considered to be not worthwhile. The bottom of each enclosure is approximately 1 m AGL.
Occasional problems with water intrusion in sensors during heavy rainfall and strong wind events have been identified, for example, during Hurricane Ida, which made landfall in Baldwin County, Alabama, as an extratropical storm at 0600 CST 10 November. At both Leakesville and Elberta the 2-m HMP45C temperature and relative humidity probe encountered problems during the event. At Elberta the temperature suddenly dropped to −40°C at 2000 CST 9 November 2009, while the relative humidity fluctuated rapidly up and down between 0% and 100%. The 2-m winds were around 9.5 m s−1 between 2000 and 2030 CST, while the rainfall rates reached values of up to 75.0 mm h−1 during that time. The temperature values returned to normal around 0600 CST 10 November with the relative humidity still reading values well in excess of 100% at the time; the latter returned to normal around 1400 CST 10 November. In Leakesville during the same event, the 2-m temperature suddenly dropped to −10°C shortly before midnight on 9 November; it resumed reporting normal values around 0900 CST the following morning. The relative humidity probe recorded values well above 100% between 0100 and 0900 CST 10 November. The 2-m wind speed reached values between 8 and 10 m s−1 during this time, while the rainfall rates reached values of 30.0 mm h−1.
Table 2 lists all of the instruments installed on every USA Mesonet station. Instrument mounting heights, measuring range, accuracy, and dynamic response characteristics can be found in Tables 3 and 4. World Meteorological Organization (WMO) instrument siting standards are followed where possible. All of the instrument mounting heights follow WMO standards, 15 sites violate the WMO object distance requirements for wind (distance between object and instrument should be greater than 10 times the height of the object) resulting from nearby trees or buildings, two sites violate object distance requirements for temperature (distance between tower and paved areas should be at least 30 m), six sites violate both wind and temperature object distance requirements, and three sites meet all WMO siting requirements. These compromises were necessary in order to satisfy other, more practical siting requirements: 1) the site should preferably be in a data-sparse region; 2) the land owner needs to be interested in hosting a site; 3) there should be a no-charge lease of land; 4) there should be a clear line of sight to nearby building to allow for cost-free radio communication; 5) there should be free access to the Internet; 6) there should be easy access to the site for maintenance; and 7) the site should be safe from vandalism. Distances and heights of nearby obstructions can be viewed via station panoramic and satellite photographs (online at http://chiliweb.southalabama.edu/).
As Table 2 shows, there is a significant amount of sensor redundancy. This redundancy is intentional given the frequency of severe weather in the area, including thunderstorms all year-round and tropical storms and hurricanes during the Atlantic basin hurricane season.
Data are sampled every 3 s and recorded every minute using a Campbell Scientific CR 3000 datalogger and Campbell Scientific, Inc., Loggernet software. The last 3-s sample of the preceding minute is recorded for all variables. A Campbell Scientific AM16/32 multiplexer is used to accommodate the large suite of sensors. All of the stations are powered by solar energy using a BP Solar SX30U solar panel, which charges a 12-V battery. A fully charged battery will operate the station for approximately 6 days before discharging to the critical dysfunctional level of 10 V.
1) Wind sensors
Wind speed and direction are measured at 2 and 10 m using the R.M. Young 05103 combined propeller and vane anemometer. This sensor was chosen for its manufacturer-specified measurement range of up to 100 m s−1, given the location of the mesonet in a hurricane-prone region. Vertical wind speed is measured at 10 m only, using the R.M. Young 27106T Gill propeller anemometer, which has a time constant of 2.1 m. The sampling interval for both instruments is 3 s. In addition to recording the last 3-s sample of the preceding minute, the datalogger records the maximum 2- and 10-m wind speeds and the minimum and maximum vertical wind speeds over the preceding minute. For the horizontal winds (at 2 and 10 m) the mean wind speed, mean wind vector magnitude, mean wind vector direction, standard deviation of the wind vector direction, and standard deviation of the wind speed over the preceding minute are calculated and recorded. The datalogger also reports the sum of the twenty 3-s observations of the vertical wind speed in the preceding minute, allowing an average vertical wind speed to be calculated by dividing the recorded variable by 20.
2) Solar radiation sensors
A Li-Cor LI-200S pyranometer and potosynthetically active radiation sensor (Li-Cor LI-190S quantum sensor) are mounted on the south side of the north–south-oriented 1.5-m cross arm in order to avoid fence, tower, and guy wire shadows.
3) Temperature and relative humidity sensors
A Campbell Scientific HMP45C combined temperature–humidity sensor measures relative humidity and temperature at 2 and 10 m. Temperature is measured at two additional heights (1.5 and 9.5 m) using a Campbell Scientific 107 thermistor. The thermistor is placed in a Campbell Scientific 41305-5A six-plate nonaspirated surrounding radiation shield and the HMP45C is protected by a Campbell Scientific 41003-5 10-plate nonaspirated surrounding radiation shield. Surface temperature is measured using an Apogee/Campbell Scientific SI-111 infrared radiometer, placed facing downward on the 2-m cross arm.
4) Rain gauges
Rainfall is measured by two tipping-bucket rain gauges—a Hydrological Services TB3 tipping-bucket rain gauge (TB3) and a Texas Electronics TE525 rain gauge (TE)—placed in the northwest and northeast corners of the enclosed fence. Both gauges measure 0.254 mm of rainfall per bucket tip. Every minute, the USA Mesonet dataloggers report the number of tips that occurred in the preceding minute, as well as the accumulated rainfall since midnight. The accumulated rainfall since midnight is not actually reset precisely at midnight. The reason for this is that any rain falling during the last minute of a given day (i.e., between 2359 CST of that day and 0000 CST of the next day) is logged at 0000 CST of the next day. Because this rain needs to be accounted for in the preceding 24-h time period, the rainfall accumulation cannot be reset to zero at 0000 CST, and it is actually reset at 0001 CST.
The TB3 functions as the primary rain gauge because it includes a siphon tube [to deliver a preset volume of collected water to each bucket, reducing undercatchment during heavy rainfall, e.g., Humphrey et al. (1997)], a built-in level for more precise positioning, dual-reed switches, and a sturdy tipping bucket made of synthetic ceramic-coated brass. The TB3 was chosen because its siphon mechanism will provide more accurate rainfall measurements during heavy rainfall events, such as hurricanes and tropical storms. Dual rain gauges are used at USA Mesonet stations in case of damage resulting from severe weather or outage as a result of accumulated debris or nesting insects. The redundancy also allows for internal consistency checks at each station for quality control (QC) purposes.
Two identical Väisälä PTB101B silicon capacitive barometers are placed side by side in the datalogger enclosure at a height of 1.10 m AGL. Using station elevation data, sea level pressure is calculated and recorded by the datalogger for both sensors.
6) Soil sensors
Currently only some stations monitor soil moisture and temperature. Soil temperature and moisture are measured at 5, 10, 20, 50, and 100 cm below ground level.
c. Data collection and station communication
All weather stations in the USA Mesonet record data at 1-min intervals. Data are stored in the datalogger and sent to a building using line-of-sight transmission via a 900-MHz spread-spectrum radio. An antenna inside the datalogger enclosure communicates with a directional glass-mount dipole antenna positioned on the inside of a window in clear line-of-sight view of the tower. These antennas have a range of 1 mile. At some sites this configuration has proven to be insufficient and two 10-mile range antennas have been installed; an omni directional antenna was placed on the tower and a directional Yagi antenna was mounted at the building. Both of these antennas are placed outside. All antenna manufacturers and types are listed in Table 2.
All of the dataloggers are equipped with a Campbell Scientific CFM100 expandable memory module, which adds 64 Mbytes of memory to the datalogger, allowing data to be stored for up to 35 days. Because a technician downloads all of the data during his monthly routine site maintenance visits, all of the data gaps (resulting from communication glitches) in the data archive are eventually filled.
The data are pushed on to the Internet via a Lantronix UDS1100. The Campbell Scientific, Inc., software package LoggerNet is run on a data server at USA and attempts to poll each station every 5 min. In the absence of any significant communications errors the system follows this schedule. If a station begins to experience communications errors, the polling system will increase the time between polling cycles until proper communications are restored. As soon as the polling system finishes collecting data from a station, a task is started (by the LoggerNet software) that updates a MySQL database and (near) real-time Web display with the data just collected. Minute, hourly, and daily data are stored in the database. The minute data can be viewed in a tabular form by minute, graphed on a 24-h basis, or downloaded in “.csv” format in up to 31-day chunks (online at http://chiliweb.southalabama.edu/).
Station communication performance was monitored for 14 days and revealed that the average delay between data collection in the field and arrival at the server (once a poll has been initiated) was between 17 and 30 s for all of the stations monitored. The maximum delays ranged between 1 min, 26 s and 55 min, 49 s. These statistics include all sites except one site with known communication issues that are being investigated and may require stronger radios than the RF-400 currently in use.
Five minutes were chosen as the polling interval as opposed to polling every minute in order to avoid system congestion. Like most mesonet operators, the USA Mesonet is subject to a stringent budget and, hence, only one server (and a backup) is in use to poll all stations. Furthermore, 5 min is sufficiently close to real time for operational purposes and exceeds the polling interval of many traditional networks that offer data only once per hour. A 5-min polling interval also complies with the high standards set by mesonet providers such as the Oklahoma Mesonet. Once a poll has been initiated the data will be collected to include the most recent data recorded in the field. For educational purposes, research, and postevent analysis, archived data at 1-min intervals are available from the database.
An automated QC system based on that employed by the Oklahoma Mesonet (Shafer et al. 2000) will be designed to match the USA Mesonet sensor suite, data collection time interval, and Gulf Coast climate. The statistical analyses presented in this paper will provide a baseline to set some of the QC test thresholds. The first QC tests will be implemented starting in fall 2010. The QC process will be continuously refined and updated as longer-term information from like-sensor analyses, nearest-neighbor comparisons, and local microclimatology becomes available.
In addition to the Web site, users can access USA Mesonet data via the Meteorological Assimilation Data Ingest System (MADIS; online at http://madis.noaa.gov/). This facility feeds data directly into the NWS Advanced Weather Interactive Processing System (AWIPS) for operational use. Alternatively, users can download archived data from the MADIS site, which also provides data QC flags.
d. Station maintenance
All sites are visited once per month for routine site maintenance and inspection. Flora and fauna thrive in the warm and moist climate of the north-central Gulf Coast, and wasp nests, algae, and other foreign accumulations rapidly appear between the blades of the radiation shields, under the rain gauges, in the rain gauge funnels, and even within the tower trusses. Kudzu vine grows extremely rapidly and needs to be curtailed before it either takes over fences and towers or intrudes into datalogger enclosures. Additional visits take place if instrument and/or communication problems occur. Routine site maintenance consists of mowing, removing debris, cleaning sensors and radiation shields, visually and audibly inspecting (listening for noisy bearings in the anemometers) instruments, and downloading data from the datalogger. The datalogger door is opened at the start of the station maintenance routine to trigger a datalogger-door-open switch, marking data during this period as possibly suspect. All instruments are sent back to their vendors for calibration once per year even if no malfunction has been reported; dataloggers are returned for calibration every 2 yr. Visitation reports are filed after every site visit.
Station metadata collection is an ongoing process. As metadata becomes available it will be added to the USA Mesonet Web site. To date, the Web site contains station panoramic and satellite photographs and instruments specifications, as well as station latitude, longitude, and elevation.
Panoramic photographs are taken in accordance with guidelines set by the American Association of State Climatologists (AASC). Outward-looking photographs are taken from the station in eight directions (every 45° starting north). Inward-facing photographs are taken 10 m from the station, looking back toward the station in the four cardinal directions (every 90°). A Suunto Tandem 360PC/360R compass was used to find directions. Because the compass points toward magnetic north, the magnetic declination must be taken into account. The National Geophysical Data Center’s (NGDC’s) magnetic declination calculator Web site (online at http://www.ngdc.noaa.gov/geomagmodels/Declination.jsp) was used to calculate the declination for the days the photographs were taken. A measuring tape (m) was used to determine the 10-m distance. An STX Pro 62 tripod was used to ensure that the photographs were taken on a level surface. The tripod features two levels: one at the tripod base, and the other on the rotating ball camera base. A Canon PowerShot SX10 IS digital camera was used to take the photographs. Satellite photographs are captured from Google Maps using TechSmith SnagIt 8. Images are captured at zoom levels 2 and 3 on Google Maps. In both cases, the image is captured with the station coordinates in the center. The zoom level 3 captured image measures 513 pixels × 513 pixels (500 m × 500 m); the image is resized in SnagIt to 750 pixels × 750 pixels to display a larger image on the Web site for viewing. Zoom level 2 images display more of the microscale environment near the weather station, measure 596 pixels × 596 pixels (300 m × 300 m), and are resized to 750 pixels × 750 pixels.
Latitude, longitude, and elevation for each site were determined with a combination of a global positioning system (GPS), U.S. Geological Survey (USGS) topography maps, and a digital topography map tool online. We found station coordinates using a Sokkia GIR1600 Differential GPS Receiver paired with an Archer Ultra-Rugged Field PC. These coordinates were verified by taking several readings in the same location. The coordinates were then entered into an online digital topography program (http://www.digital-topo-maps.com/) that paired Google Maps with USGS maps in order to find the elevations.
The AWAC instrument configuration consists of three bottom-mounted Nortek 600-kHz AWAC Doppler profilers that measure the surface height, directional wave spectrum, subsurface ocean current profiles, and bottom temperature. Data from these autonomous instruments are recovered when the instruments are retrieved at the end of the hurricane season. Each instrument is protected by a Mooring Systems Incorporated (MSI) trawl-resistant housing. The MSI housings consist of a rectangular base, a trapezoidal profile, and a flat top that accommodate the instruments well. Each AWAC instrument is mounted on a pedestal and peers through a hole in the top of the trawl-resistant cover. Power for the AWACs, sufficient to last for the duration of hurricane season (June–November), is provided by two external lithium battery packs affixed to the housing platform base. The instruments are lowered to the ocean floor to ensure an upright deployment and permit a differential GPS position fix (which is important for recovery). The deployment position is marked when the assembly first touches the bottom. The entire package is weighted down with 220 kg of lead to prevent the assembly from moving far off site during severe weather.
The AWAC instruments use three independent sensors to record wave measurements. First, a high-resolution piezo-resistive element measures the pressure signal from passing waves. Second, an acoustic transducer directly measures the water’s surface height. This transducer points vertically upward and emits a short acoustic pulse at a rate of 2 Hz. This pulse is reflected from the water’s surface and is then received by the instrument. This measurement enables the determination of the surface height above the instrument with an inherent resolution of 2.3 cm. Curve-fitting algorithms are used to improve the precision of locating the peak reflection to about 0.1 cm. This direct measurement of the surface [called acoustic surface tracking (AST)] allows the direct calculation of nondirectional wave parameters from the time series of these measurements. The AST time series record also allows for the identification of transient wave events. The third sensor measures the wave-induced orbital velocity of the water at a location just below the water’s surface. This measurement is made by sensing the Doppler shift from an acoustic signal transmitted at a rate of 1 Hz from three transducers symmetrically positioned about the instrument’s center and angled 25° off the vertical. The addition of this water velocity information gives the instrument the ability to calculate a directional wave spectrum. To enable the calculation of the directional wave spectrum, a series of 1024 measurements of water velocities in a cell located below the water surface and the pressure at the depth of the instrument are made at a 1-Hz rate. The AST concurrently makes surface height measurements at a rate of 2 Hz. Directional wave spectrum measurements are collected every 20 min. The manufacturer provides two algorithms for calculating the directional wave spectrum during postprocessing.
Recovery of the instruments is facilitated by a “pop up” float attached to the housing platform base. During recovery operations, the float is released by a Benthos 867-A acoustic release through a hole in the top of the trawl-resistant cover. The release command is sent from a recovery ship. Divers also provide assistance to ensure the recovery of the instruments. Upon recovery, the instruments are cleaned and the data are extracted. The data are then converted to wave parameters using the manufacturer’s software. This software produces a number of quality control parameters for each data ensemble and indicates if there are any potential problems with the data.
3. Sensor comparisons and statistical analyses
Sensor redundancy at the USA Mesonet stations provides a unique opportunity to perform like-sensor comparisons. Statistical analyses were performed to acquire QC baseline information, to evaluate sensor performance, and to assess the usefulness of lesser-quality, but more affordable, sensors. Statistical analyses were performed on 2007 and 2008 data from the four USA Mesonet stations that operated during both of those years (i.e., Agricola, Bay Minette, Mount Vernon, and Pascagoula). The geographic details of these stations are listed in Table 1 and their instrumentation details (identical for each) are in Tables 2 –4. Statistical analyses were performed on the archived, 1-min data, which were manually quality controlled beforehand. These analyses lead to the identification of occasional data gaps and instances of instrument failure. The former is a result of communication errors and datalogger outages. During the first years of the project instrument failure was not always immediately detected because of understaffing. A part-time weather station technician performed site maintenance on an irregular basis, while no QC meteorologist had been hired yet. Additionally, the lack of a safe and efficient tower-lowering mechanism during that time made it extremely difficult to access instruments at the 10-m level. In spite of having to discard some data, large enough quantities of data remained to conduct statistical testing and to show significant differences. All of the statistical analyses were done using the JMP software package. To study the relationships between two variables, scatterplots and Pearson correlation coefficients were used. Time series were used to study hourly, daily, and monthly trends. Patterns in bivariate data were revealed using nonlinear regression techniques. Histograms revealed the nature of the distributions of the variables. Because of the large number of observations, the distribution of the sample mean is approximately normal. With the population variance unknown and the sample mean normally distributed, the test statistic follows the t distribution, allowing the paired t test to be employed to compare average temperatures, average wind speeds at different heights, and other averages. All of the results presented were significant at the 5% level.
The large degree of instrument redundancy at USA Mesonet stations provides a unique opportunity to perform internal consistency checks by comparing 1) different types of instruments collecting the same variable and/or 2) instruments collecting the same variable, but at different elevations.
a. Comparison of different makes of instruments
1) Rain gauges
Each USA Mesonet station has two different models of tipping-bucket rain gauges (Table 2), referred to as the TB3 and TE. The TB3 is considered the better-quality gauge because of the following attributes that are not featured on the TE: 1) a siphon that allows the rain to flow at a steady rate to the tipping-bucket mechanism regardless of rainfall intensity, 2) a built-in level for precise positioning, 3) sturdy tipping buckets made of synthetic ceramic-coated brass instead of plastic, and 4) a finer filtering mechanism. A photograph of both gauges is shown in Fig. 6. In the current station design, the gauges are positioned 3.05 m apart (Fig. 4) from one another. Prior to the summer of 2008, the rain gauges were less than 50 cm apart. Both gauges measure 0.254 mm of rainfall per bucket tip. Every minute, the USA Mesonet dataloggers report the number of tips that occurred in the preceding minute (converted to millimeters), as well as the accumulated rainfall since local midnight. In this document, the following three rainfall products will be discussed: rain rate (number of tips per minute), rainfall accumulation since local midnight (accumulated rainfall since local midnight for a given minute), and the 24-h rainfall total (accumulated rainfall at midnight of the following day).
Table 5 lists the frequency (and percentage) of the number of bucket tips per minute, or precipitation rates. Each bucket holds the equivalent of 0.254 mm, so the number of bucket tips per minute can be converted to precipitation rates (mm min−1 or mm h−1). Most of the time (99.3%–99.4%), the buckets sit idle because rainfall events are relatively rare. These percentages seem high and were verified using data from a nearby Cooperative Observer Program (COOP) site available from the National Climatic Data Center (NCDC; available online at http://www.ncdc.noaa.gov/oa/ncdc.html). The least amount of missing minutes occurred at Bay Minette (Table 5), so this site was chosen for the verification. The TB3 at the USA Mesonet site at Bay Minette collected a total of 1202.5 mm of rain in 2007 and 1442.25 mm in 2008. In 2007 it had 12 min of missing data, while in 2008 it had 11 512 missing minutes. The Bay Minette COOP site collected 1389.38 (2007) and 1600.96 (2008) mm of rain with 1–9 days missing data from both years. This amounts to about 10%–13% less rainfall collected at the USA Mesonet site. Some of this can be accounted for by the missing minutes of the TB3; however, the COOP site also encountered missing data. The discrepancy may be attributed to the different types of rain gauges used; COOP sites use collection gauges instead of tipping-bucket gauges. Tipping-bucket gauges are known to undercollect during both very heavy and very light rainfall (e.g., Humphrey et al. 1997). However, they offer the advantage of automated collection of rainfall rates per minute, hour, or other time interval.
Low rainfall rates of 15 mm h−1 are most common when it does rain, constituting 74%–78% of all nonzero observations. As the rainfall rate increases, the likelihood of that rainfall rate occurring decreases. Occasionally the two gauges recorded different rainfall rates; this was examined more closely using the correlation coefficients between the two types of tipping buckets for the three rainfall products mentioned above (Table 6). The correlation coefficients ranged from 0.868 to 0.9996, indicating a very high degree of agreement between the two gauges. This is no surprise considering no rainfall was recorded about 99% of the time. After removing observations where no rainfall was recorded by both gauges, the correlation coefficients decreased only slightly (ranging from 0.787 to 0.999) for all three products. Among the three rainfall products, the rainfall rate showed the lowest degree of agreement (at 0.777–0.809), while the other two products (rainfall accumulation and daily totals) showed about the same level of (almost perfect) agreement (at 0.999). Table 7 shows the difference in rainfall rates between the two gauges (TE rate − TB3 rate). Around 30% of the time both rain gauges collected the same rainfall rate per minute. Between 67% and 68% of the time the collection rate differed by just one bucket tip per minute. Out of all one-bucket tip discrepancies, the TB3 more often collected one tip more than the TE. A difference of two tips occurred 1.5% or less of the time, with the TB3 usually collecting more than the TE. A difference of three tips was rare, occurring only 7 times overall. In six out of these seven extreme cases, TB3 recorded more rain than TE.
The above discussion shows that both rainfall rates and accumulations correlate well (all of the correlation coefficients are statistically significant at the 5% level). About a third of the time the gauges collect the same amount of rainfall, while the remaining two-thirds of the time they differ by just one tip. Rainfall accumulations show higher correlation coefficients than rainfall rates (number of tips per minute). When adding rainfall rates together, a one-tip deficiency (TB3 records one tip more than the TE) during 1 min can be offset by a one-tip excess (TB3 records one tip less than the TE) in a following minute, causing the rainfall accumulations of both gauges to be similar.
As indicated above, there is a high degree of correlation in rainfall collection between the two tipping buckets, with the TB3 recording slightly higher values occasionally. It has been shown that tipping-bucket rain gauges suffer from inaccuracies during very low rainfall rates and during very high rainfall rates. Undercatchment during heavy rainfall events (buckets cannot reposition themselves fast enough after a tip to collect all of the rainfall entering the outer funnel) and underestimation during light drizzle (water evaporates before the bucket gets a chance to tip) are typical errors associated with tipping-bucket rain gauges (Humphrey et al. 1997). Because the TB3 is equipped with a siphoning mechanism to reduce undercatchment during heavy rainfall events, this gauge may record higher rainfall rates during heavy rain events. This was investigated by plotting mean differences in rainfall rates (TE − TB3) as a function of TB3 rainfall rates (Fig. 7). At each station the negative mean differences indicate that larger TB3 rainfall rates occur on average. For small TB3 rainfall rates, smaller differences between the two gauges are observed. A distinct increase (at an increasing rate) in mean rainfall-rate differences is seen as the TB3 rainfall rate increases. Table 8 further confirms that the TB3 collects more at high rainfall rates. TB3 rainfall rates in excess of 150 mm h−1 and corresponding TE rainfall rates are shown. The TB3 collected more rainfall than the TE for 38 out of the 42 min when TB3 rainfall rates in excess of 150 mm h−1 were recorded. This was by using one tip (24 times), two tips (11 times), or three tips (3 times), and the two buckets collected the same number of tips 4 times. This analysis provides confidence that the siphoning mechanism is working well in reducing the underestimation of TB3 rainfall at higher rain rates.
Because the TB3 mostly exceeds the TE in rainfall-rate recording, a similar trend is expected in rainfall totals. Figure 8 shows the relative frequency distribution of the difference (TE − TB3) between the two gauges in midnight precipitation totals. For Agricola the difference ranges from −8.50 to 0.25 mm; at Bay Minette the differences range from −9.5 to 1.0 mm; for Mount Vernon the range is from −5.8 to 1.25 mm, while for Pascagoula the range is from −8.7 to 0.75 mm. The TB3 recorded smaller midnight totals only between 1.0% and 7.0% of the time. The TB3 recorded more 24-h rainfall between 72% and 80% of the time, while the two gauges were in agreement between 18.0% and 21.0% of the time.
2) Solar radiation sensors
The photosynthetically active radiation (PAR) sensors experienced problems for a large part of 2007 and 2008 at all stations except Mount Vernon. Figure 9 shows a scatterplot of incoming solar radiation in the daylight spectrum (wavelengths between 400 and 1100 nm, also known as total solar radiation) versus PAR (wavelengths between 400 and 700 nm) at Mount Vernon in 2007 and 2008. A very strong agreement between total solar radiation and PAR is indicated. This is corroborated by an extremely strong positive correlation (0.995 806, p < 0.0001) between the two variables. Because the radiation values vary significantly by month and hour of day, separate correlations coefficients were evaluated for each; the monthly correlation coefficient varied between 0.993 34 and 0.998 98, while the hourly correlation coefficient varies between 0.984 82 and 0.996 29.
The maximum total radiation value of 1381W m−2 was measured at 1403 CST 14 April 2008, while the maximum PAR value was 2649 microeinsteins (1119 CST 15 April 2007). As expected for locations east of the standard meridian for the central time zone (90°W), maximum values of solar radiation were recorded between 1100 and 1200 CST. The reason the maximum total radiation value exceeds the solar constant (1367 W m−2) is that the LI200X pyranometer measures the total of both the sun and sky radiation.
b. Comparison of identical of instruments at identical locations
Table 9 summarizes the differences between the two identical barometers in 2007 and 2008, located at each mesonet site at the same height AGL. The pressure values used are station pressures and not sea level pressures, explaining the differences in the mean values between stations. Pascagoula, at the lowest elevation (see Table 1), records the highest mean pressure value, followed by Mount Vernon, Agricola, and Bay Minette. At all four stations the correlation between the two barometers is extremely high (0.996 82 to r = 0.998 27, p < 0.000 01), indicating a nearly perfect agreement between the two sensors. Figure 10 shows the frequency distribution of the difference in pressure (barometer 1 minus barometer 2) for each site, while the range of the pressure difference is listed in Table 9. The largest difference of −2.20 hPa occurred at Bay Minette, which was also the only site where one sensor always recorded a larger value than the other. This is not related to the sensor location because barometer 1 is located on the left side of the datalogger enclosure and barometer 2 is to the right of barometer 1 at all sites. Brock and Richardson (2001) reported that aneroid barometers are very sensitive to temperature. They mention that the temperature error function 1) is nonlinear, 2) depends on ambient pressure, and 3) changes in functional form from one sensor to the next, even among sensors of the same type from the same manufacturer. The two sensors at Bay Minette may, therefore, have extremely different temperature error functions leading to errors of opposing signs and a large difference in pressure. All of the pressure differences observed fall within the manufacturer-specified range of accuracy of ±2 hPa for temperatures between 0° and 40°C. Scatterplots of the difference in pressure (barometer 1 − barometer 2) versus 2-m temperature (not shown) suggest a decreasing trend in the pressure difference as the temperature increases, but the correlation coefficients are weak (ranging from 0.308 499 to 0.596 747 for the four stations).
The manufacturer recommends calibrating the instrument every year. To optimize the advantage of employing two barometers per station, we recommend that barometer annual replacement should be staggered by 6 months.
c. Comparison of identical of instruments at different heights
Three out of the four sites experienced problems with one temperature sensor or another during large portions of 2007 and 2008. Only at Mount Vernon did all four temperature sensors function throughout both years. Figure 11 shows scatterplots and frequency distributions (on the diagonal of the figure) for all four temperature sensors at Mount Vernon for these 2 yr. The scatterplots on either side of the diagonal are identical. The temperature probes at 2 and 10 m are identical resistance temperature detectors and the probes at 1.5 and 9.5 m are identical thermistors (Table 2). Nearly perfect positive correlations were seen between the probes with the smallest elevation differences in spite of the probes being different types (0.9996 for 1.5 versus 2 m and r = 0.9998 for 9.5 versus 10 m). The correlation coefficient decreased as the vertical distance between a pair of sensors increased, but remained high. The lowest correlation observed was between 1.5 and 10 m (0.9911). The same trends were seen at Agricola for the 1.5-, 9.5-, and 10-m temperature sensors, at Pascagoula for the 1.5-, 2-, and 9.5-m temperature sensors, and at Bay Minette for the 1.5- and 2-m temperature sensors.
The temperature frequency distributions (diagonal of Fig. 11) were all left skewed; this indicates that more observations of higher temperatures than lower temperatures were recorded at all elevations. This is confirmed by the respective medians shown in Table 10 and is as expected with the lengthy, hot Gulf Coast summers. The largest temperature extremes (both hot and cold) are observed at the 1.5-m temperature sensor at all four sites (shown in Table 10 for Mount Vernon) because radiational cooling–warming of the ground below controls the temperature of the air above, to a large degree.
Figure 12 shows the relative frequency distributions of temperature differences at Mount Vernon for sensors mounted at different levels. Not surprisingly, the mean temperature differences between sensors of different types mounted only 0.5 m apart (Figs. 12a,b) are smaller (−0.11° and 0.11°C, respectively) than those of the same type mounted 8 m apart (Figs. 12c,d; 0.56° and 0.43°C, respectively). Interestingly, the mean temperature difference between the lowest two sensors (Fig. 12a) is negative, while the mean temperature difference between the highest two sensors is positive. The 1.5-m temperature is more often (71%) higher than the 2-m temperature than vice versa, as would be expected during daytime. The exact opposite occurs at the two highest sensors (Fig. 12b) where the 10-m temperature is more often (74%) higher than the 9.5-m temperature. The frequency distributions of temperature differences 8 m apart (Figs. 12c,d) indicate different patterns. Although the 10-m sensor recorded higher temperatures than the 2-m sensor 60% of the time, the 9.5-m sensor recorded higher temperatures than the 1.5-m sensor only 47% of the time. The temperature differences between sensors placed 8 m apart display more large, positive values (i.e., the temperature at the higher elevation is warmer) than large, negative values. This suggests that during nocturnal inversions the temperature differences are larger than during the daytime. This is explained by boundary layer mixing reducing the temperature differences at different elevations during the daytime. The distributions of the various temperature differences were similar at the three other mesonet sites.
Each station has two identical hygrometers at 2 and 10 m. The frequency distributions of the RH difference (2 − 10 m) are shown in Fig. 13, and various statistics are presented in Table 11. Strong positive correlation coefficients ranging between 0.772 39 and 0.986 72 (all with p < 0.0001) were observed. On average, the 2-m RH was higher between 60.94% and 83.81% of the time. This is as expected because of the controlling influence of the soil and vegetation below. The frequency distribution (2–10 m) of RH at Pascagoula (Fig. 13d) looks distinctly different from the other three sites; it is right skewed while the others are left skewed. This indicates that at Pascagoula a large amount of negative values are observed; meaning that on many occasions the 10-m RH is larger than the 2-m RH unlike at the other sites. It is suspected this is due to a sensor problem not identified by the manual QC process, because the surface and surroundings at Pascagoula are not significantly different from those at the other sites.
Not considering Pascagoula, the difference between the two RH values ranges between −7.16% and −20.07% on the negative end (10-m RH is larger than 2-m RH) and between 38.14% and 53.90% on the positive end (2-m RH is larger than 10-m RH). The mean difference ranges between 1.24% and 3.82%, with standard deviations between 3.33% and 4.75%.
Table 12 shows the correlation coefficients between the 2- and 10-m wind speeds at all stations, while the frequency distributions of the difference (10 − 2 m) are shown in Fig. 14. At all four sites, the 10-m wind speed exceeded the 2-m wind speed between 69.75% and 82.55% of the time. At all four sites, the 2- and 10-m wind speeds were significantly different (paired t test, p < 0.0001, for each site) on average by about 0.5 m s−1 (Table 12). At each site, larger wind speeds were recorded more often at 10 than at 2 m. The lowest mean wind speeds at 2 and 10 m were recorded at Mount Vernon, which may be related to the location of this site in a river valley. Bay Minette, on the other hand, sits on the edge of a hilltop and is therefore more exposed, while Pascagoula can attribute its relatively high mean wind observations to its proximity to the coast.
The wind speeds at the USA Mesonet are recorded at every last 0.3-s interval of the preceding minute, making them essentially wind gusts and not 1-min averages. Time series plots of these wind speeds are extremely noisy (not shown). This is because the wind speeds change rapidly due to boundary layer turbulence. Turbulent eddies are also the cause of the 2-m wind speed occasionally exceeding the 10-m wind speed. However, the largest negative wind speed difference (10- minus 2-m wind) recorded is 4.83 m s−1 while the largest positive difference is more than twice as large at 11.4 m s−1.
d. Comparison of mesonet and AWAC data for Hurricanes Gustav and Ike
Table 13 lists the positions and depths for the 2008 deployment of the three AWAC instruments. The first AWAC instrument (pod 1) was deployed approximately 31.2 km southeast of the Mobile Entrance Lighted Horn M (30.1253°N, −88.0687°W) at the entrance to the Mobile ship channel. The second AWAC (pod 2) was positioned about 10.5 km to the east of pod 1 and the third AWAC (pod 3) is 17.7 km south of pod 1. The water depth was 25 m for the shallow shoreward AWACs and 32 m for the deeper offshore instrument.
During the passage of Hurricanes Gustav and Ike, some of the reported values of wave height were significantly higher than the adjacent measurements. Because of the close proximity of the three AWAC sensors, the wave-height data could be quality controlled via simple intercomparisons. If any sensors reported values greater than 1 m different than its neighbor, the data were replaced with the mean value of the adjacent measurements. Figure 15 displays the significant wave heights of all three AWAC instruments as well as the maximum 20-min wind speed recorded at the USA Mesonet station on Dauphin Island (30.2467°N, −88.0771°W). Very good agreement can be seen between the wave heights and the wind speed, and the passage of Hurricanes Gustav and Ike clearly stand out. As both storms approached the region from the southeast, large waves first arrived at the southeastern-most AWAC (pod 3), followed by pod 1 (placed slightly northwest of pod 3) and then pod 2. The strong wind peak at Dauphin Island occurs a little after the maximum waves pass the AWACs, resulting from its location farther northwest of pod 1.
4. Summary and a look to the future
The University of South Alabama (USA) maintains a high-quality mesonet and collects a large suite of observations at 1-min intervals, allowing cutting edge research of landfalling hurricanes and coastal interactions, frontal passages, sea breezes, and other severe weather events. These coastal observing stations can provide much-needed ground truth data for the development of models that can predict the effects of severe storms on coastal regions. The data are used daily by local National Weather Service forecasting offices and television stations, and for educational and leisure purposes. It is expected that usage of mesonet data will soon extend into additional areas such as renewable energy, agriculture, hydrology, biology and ecosystem protection, the insurance industry, dispersion modeling for several local chemical manufacturers, and even our legal system.
Maintaining a high-quality mesonet requires dedicated and constant attention. It is crucial that expert staff members are on the project from its inception. For various administrative reasons, this did not occur at the USA Mesonet until several years into the project. Since the summer of 2008, four full-time, highly qualified staff members keep the USA Mesonet, the collected data, and the Web site operational and in good working order. The highest-priority task in the very near future for the USA Mesonet will be to design and implement an automated QC system. Archived data will be used to establish QC thresholds. To specify refined QC range, internal consistency, and buddy tests, data will be grouped by season and/or time of day. These thresholds will be reviewed and updated continuously. Another task of high priority will be the continued augmentation of station and network metadata.
The USA Mesonet is equipped with several redundant sensors. Sensor redundancy allows for 1) the availability of extra sensors in the case of sensor failure during severe weather and 2) the ability to perform a large number of internal consistency checks for quality control (QC) purposes. The mean differences between like sensors presented in this paper, as well as their standard deviations, will provide initial threshold values for internal consistency tests, tailored to the USA Mesonet sensor suite and climate conditions. Additionally, the statistical analyses and sensor comparisons show that mesonet data comply with reasonable meteorological relationships. Mean temperature differences of sensors mounted at different heights range between 0.11°C for sensors mounted 0.5 m apart and 0.56°C for sensors mounted 8 m apart. Larger differences between sensors 8 m apart are recorded during nocturnal inversions than during well-mixed daytime hours. Minimum, mean, and maximum differences between wind speeds at different heights are fairly consistent across the four stations examined (Table 12) and show a predominance of weaker 2-m wind speeds than 10-m wind speeds as a result of surface friction. Turbulent eddies may occasionally (about 12%–16% of the time) cause the 2-m wind speed to be stronger, but not by as large a difference. Similarly, the relative humidity sensors mounted 8 m apart only display a mean difference of between 1.24% and 3.82%, with the 2-m RH sensor reading larger values the majority of the time (other than Pascagoula where an unidentified sensor problem likely occurred), as expected because of the controlling influence of the soil and vegetation below.
The like-sensor comparisons also offered the opportunity to assess the usefulness of lesser-quality but cheaper sensors. For example, while the TB3 tipping-bucket rain gauge is a better-quality gauge equipped with a siphon that allows the gauge to perform more accurately in heavy rainfall events, the more affordable TE525 only slightly undercollected the TB3 during such events. This can make the TE525 a good choice for a tipping-bucket rain gauge for certain applications and for operations working under a restricted budget. Likewise, it inspired a way to optimize the advantage of employing two barometers per station: instead of calibrating both barometers annually at the same time, barometer annual calibration should be staggered by 6 months.
A preliminary analysis of the combined mesonet and AWAC datasets hold the promise of an integrated picture of offshore severe storms and exciting opportunities for the future of hurricane landfall research.
The major challenge faced by the USA Mesonet is sustained and continued funding. Several avenues of funding including traditional grants, and private and state sources are being explored. Meanwhile data-derived customer-specific products, including near-real-time data, data archives, data summaries, maps, educational material, and climate data continue to be available to our stakeholders in local, state, and federal governments, private enterprise, and the people of the north-central Gulf Coast.
This project is supported by NOAA Award NA07NWS4680011 and in part by NSF Award ATM-0239492NSF. The authors thank David Brown, Sean Huber, Ivory Reinert, and Russell White for their dedicated and professional service in maintaining and operating the University of South Alabama Mesonet. This project would not have been possible without our numerous site hosts graciously offering a safe location for our weather stations with free access to their property and the Internet. Finally, we extend our gratitude to three anonymous reviewers whose insightful and constructive comments significantly improved the quality of this document.
Corresponding author address: Sytske Kimball, LSCB 136, Department of Earth Sciences, University of South Alabama, Mobile, AL 36688. Email: email@example.com