Improved Climatic Data for Mechanistic-Empirical Pavement Design
The North Carolina Department of Transportation (NCDOT) has adopted the AASHTOWare Pavement ME Design software in an effort to improve pavement design for North Carolina roads. A critical component of the software is the Enhanced Integrated Climatic Model (EICM), which accounts for environmental effects. The EICM requires hourly historical climate records for the entire expected lifespan of the road. NCDOT presently has access only to small 5-year samples of climatological data from select locations. These short records must be repeated to fill in data for long analysis periods, yet studies have shown that repeating small samples of climatic data may adversely affect pavement performance predictions. This research effort will develop long-term, high-quality, hourly data files for multiple locations across North Carolina. Statistical procedures will help to fill in gaps in the hourly data. A sensitivity analysis will assess the impact of the improved climatic data files on pavement performance predictions by the mechanistic-empirical design software. Ultimately, the improved climatic data will boost confidence in pavement performance predictions.
October 2015 — A section that provides guidance on the virtual station feature of the Pavement ME Design software has been added to the final report.
July 2015 — The Pavement ME Design results are available for analysis. The final report is in progress.
May 2015 — The quality control algorithm is complete and full sets of high-quality, internally-consistent, and complete hourly meteorological data files spanning 35 years are available for 41 stations in and around North Carolina. The impact of these files on pavement performance predictions is being tested within the Pavement ME Design software.
February 2015 — The precipitation algorithm is complete and 35 years of continuous hourly precipitation data are now available for the 20 default stations.
A problem exists with an upstream data provider such that a small percentage of the cloud coverage data from NCDC are missing. There are 141 hours when no cloud coverage is available for the entire region. NARR data from the nearest grid point instead provide an estimate of the coverage during these gaps.
Current efforts involve the development of a final quality-control algorithm on the full time series of hourly data.
October 2014 — Continuous hourly data for cloud cover spanning 35 years are now available for the 20 default HCD stations. Detailed analyses show that the NARR-derived cloud cover values are wildly incorrect when compared with observations, so the gap filling procedure cannot and should not use the NARR-derived cloud data. Statistical tests indicate that a nearest-neighbor approach using only ISD observations provides the best match to the actual cloud cover observations at each station.
The hourly precipitation gap-filling procedure will use a combination of spatially dense Global Historical Climatology Network–Daily (GHCN–Daily) observations and three-hourly NARR data. Development of this algorithm is in progress.
August 2014 — Continuous hourly data sets of temperature, dewpoint temperature, and wind speed spanning 35 years are now available for the 20 default HCD stations. Gap-filled cloud cover and precipitation data require further analysis.
July 2014 — Continuous hourly data sets spanning 35 years are now available for 847 NARR grid point locations across North Carolina and surrounding states. View an interactive map of all of the land-based sites with complete HCD files spanning 35 years.
Tests of the kriging estimates indicate that the procedure will work well for estimating meteorological variables over large temporal gaps, even with no adjustment for elevation (see the next paragraph below). This page shows the scalar accuracy measures (no lapse rate correction) after estimating hourly data over the course of an entire year (1985) at both the Asheville Regional Airport (AVL) and Piedmont Triad International Airport in Greensboro (GSO). Data from these two stations were removed from the spatial interpolation procedure so that the estimates remain independent of the observations at these locations. Note that the final estimates for precipitation will most likely derive from a nearest-neighbor interpolation. It is important to recognize that the scalar accuracy measures are compared hour by hour and do not account for small-scale spatial discontinuities, such as fronts or a precipitation swath moving across the state. Therefore, it remains critical to analyze the time series of the observations compared with that of the estimates. Please view the time series for each meteorological variable at both Asheville and Greensboro (no lapse rate correction).
Since the elevation varies considerably across the state and between adjacent observation sites, any spatial interpolation will have difficulty reproducing observed meteorological variables without a correction for the lapse rate. Therefore, a lapse rate correction to the temperature has been applied using the near-surface (first hybrid model level) and 850-mb heights and temperatures at the nearest NARR grid point. First, temperatures are reduced to mean sea level, kriged, then the lapse rate correction is reapplied. This technique dramatically improves the temperature record when compared with observations, particularly for the high and low points of the diurnal cycle. Please see the scalar accuracy measures, and plots for both Asheville and Greensboro using the lapse-rate correction technique and compare them with the plots shown above.
Given these results, the operational interpolation procedure will apply a lapse rate correction using the modeled NARR lapse rates.
June 2014 — Three-hourly North American Regional Reanalysis (NARR) data must be temporally interpolated to hourly data. Choices include a local harmonic analysis, cubic spline interpolation, and linear interpolation. A comparison of each of these methods using a complete year of observations from the Asheville Regional Airport reveals that a simple linear interpolation is the best overall method. View the results of this comparison here, with graphical summaries here.
The code for a kriging system has been built. This geostatistical interpolation technique provides estimates for missing values at any station, based on data from surrounding sites. This is a far better scheme than a simple weighted average. Tests remain ongoing.
May 2014 — The processing of several terabytes of North American Regional Reanalysis data over 35 years is complete. Daily data from all Global Historical Climatology Network–Daily (GHCN–Daily) sites have been collected as well. The latter data will provide upper and lower bounds for temperature and precipitation at many more sites than those available in the database of hourly observations. Maps of these data and groundwater observations appear below.
Quality-control checks reveal that several stations are colocated and others have the wrong geographic coordinates in the metadata from NCDC. These issues have been addressed. New maps appear below (and in the December 2013 update), along with a new PDF document showing the periods of record for each station.
North American Regional Reanalysis grid points over North Carolina and adjacent states. Data are available continuously from 1979 to the present in 3-hour increments. Click for a larger image.
Global Historical Climatology Network–Daily (GHCN–Daily) sites in North Carolina and adjacent states that provide daily observations of quantities that are useful to include in the development of hourly historical climate data files. A total of 2905 unique stations in North Carolina, South Carolina, Tennessee, Georgia, and Virginia provide valid data during the 1979–2013 period. Click for a larger image.
Location of groundwater wells providing irregular field measurements (red; 3542 stations) and daily observations (blue; 92 stations) of groundwater depth between 1979 and 2013. Many sites are colocated. Click for a larger image.
December 2013 — Initial meteorological data processing is complete. A total of 175 North Carolina stations contribute useful information to the historical database of hourly observations between 1979 and 2013. Some stations are colocated, so the number of unique stations is less. Stations in adjacent states provide additional observations. This map shows a total of 381 stations.
Sites in North Carolina, Virginia, Tennessee, Georgia, and South Carolina providing hourly historical data between 1979 and 2013. Click for a larger image.
Sites in North Carolina and adjacent states providing hourly historical data between 1979 and 2013. Colors indicate the number of days with an observation for each site during this period. Sites far from North Carolina have been removed. Click for a larger image.
Periods of record for sites in North Carolina and adjacent states providing hourly historical data between 1979 and 2013. Note that some WBAN station numbers changed in 2006, but these stations have been combined (as of May 2014). Click the image for a PDF showing the complete list of stations. Zoom in on the PDF document to see individual days with at least one hourly observation.