Continuous-Monitoring Data

Automatic-monitoring techniques were used to collect continuous measurements of physical parameters at the monitoring stations from October 2018 through September 2021. Each station was equipped with a digital modem to transmit stored data and to enable the monitoring systems to be remotely programmed. Visits to the study sites were generally made on a weekly basis to collect samples and verify the accuracy of the sensor data. Continuous data collected during this study are published in the U.S. Geological Survey (USGS) National Water Information System (NWIS; U.S. Geological Survey, 2023b).

The dataloggers at each monitoring station were programmed to make gage-height (water level) measurements every minute. Baseline data (data that are recorded regardless of the state of runoff), including gage height, water temperature, and specific conductance, were recorded every 10 minutes, but recording increased to a 1-minute basis whenever the gage height exceeded 0.03 ft. Gage height in each weir box (fig. 5A) was measured by a gas-purge bubbler system and a National Institute of Standards and Technology traceable pressure sensor. Measurements made by this system were corrected for density effects that resulted from salt-laden runoff during the winter seasons by using concurrent measurements of specific conductance. Redundant measurements of gage height were made with a submersible pressure sensor and by an ultrasonic sensor. The ultrasonic sensor was recorded by a secondary logging system to reduce the likelihood of data loss during the monitoring period. Secondary gage-height measurements also were used to check the accuracy of the primary measurements. Continuous measurements of gage height were converted to continuous-discharge values by programming the dataloggers with the gage height-discharge relation for the weir (Rantz and others, 1982). A conductivity and water-temperature probe was mounted just below the static water surface in the weir box. The accuracy of measurements of specific conductance depends primarily on the amount of sensor drift and fouling. Corrections for fouling and drift, which were minimal, were applied to the data to improve the accuracy (Wagner and others, 2006). These corrections were made on the basis of the performance of the sensor before and after sensor maintenance and the response of the clean sensor after it was placed in three or more certified conductance standard solutions in the field.

Precipitation, snow-depth, and air-temperature data were measured at the OGFC monitoring station at a recording frequency of 5 minutes. Precipitation was measured with an electronic weighing-bucket gage having a 31-square-inch (in²) collection area and mounted about 10 ft above the ground surface. Snow depth was measured with a temperature-corrected ultrasonic sensor mounted about 7 ft above the ground surface on the south side of the monitoring shelter to shield the measurement area from snow plowing (fig. 5B). Nonfrozen precipitation (rainfall) was measured to estimate the volume of runoff based on the drainage area of each site, and the estimated volumes were used to determine flow-proportional sampling thresholds and runoff coefficients. In July 2020, two-dimensional sonic anemometers were installed to measure wind speed. The instruments were mounted about 1.3 ft from the ground surface at the center point of each trench near the guardrail. The sensors were operated during the summer and fall months and temporarily removed during the winter period. An anemometer also was installed above the roof line on the HMA shelter and was left in place for the duration of the study. Each sensor was mounted to correspond to a true-north bearing. Wind data were collected to better understand the prevailing direction and magnitude of wind as it relates to the transport of sediment from the road surface. Measurements were internally processed by the datalogger every 10 seconds, and daily average wind-vector direction and wind-vector magnitude were recorded for each sensor. The measurement locations for the meteorological sensors were not ideal and were limited by the local site characteristics, such as the limited width of area between the tree line and the highway and the snow-throw zone adjacent to the highway. The meteorological data were used to represent local site conditions and to aid in the interpretation of discharge, sediment transport, and road-condition values.

In November 2019, three soil sensors were installed 3, 9, and 15 ft from the edge of the HMA pavement to measure soil temperature, soil moisture, and bulk electrical conductivity in the highway shoulder. Sensors were buried about 4 in. below the ground surface in shallow pits filled with maintenance sand to provide a homogeneous measurement area for each location. The measurement volume for each sensor was about 0.27 cubic foot per second (ft³/s) with a sensing radius of 0.25 ft around the probe. Sensors were installed to help determine the linear extent of salt broadcast by the physical removal of salt-laden snow and slush by plow trucks, vehicular spray, and wind-blown deposition from the pavement surface. Sensors were not installed adjacent to the OGFC pavement because water often ponded during heavy rain in the shallow gully between the monitoring shelter and the west bank just beyond the shoulder, making the site conditions difficult to measure and too dissimilar for comparison to the HMA highway shoulder.

Noncontact pavement sensors were installed at each monitoring station to measure pavement surface temperature and pavement condition. Sensors were mounted on a mast (fig. 5B) about 16 ft above the pavement and at a 40-degree surface angle, which focused them on the first travel lane just inside the rightmost tire track at each site. Noncontact measurements of pavement temperature were made by an infrared thermometer with a 12-degree field of view, resulting in an elliptical area of measurement of approximately 22 ft². Air temperature and relative humidity were measured within the device housing on the mast. The sensor also monitored the near infrared spectral differences of the roadway surface to determine the pavement surface condition and detected the state of water on the pavement (dry, damp, wet, snow, ice, standing water, deep snow, or black ice; High Sierra Electronics Inc., 2017). These measurements were made within a 1.6-degree field of view, resulting in an elliptical area of measurement of approximately 0.4 ft² at each site. The pavement sensors were calibrated to dry pavement conditions during periodic site visits. In addition to the pavement sensors, high-resolution fixed-image cameras were mounted on the adjacent mast (fig. 5B) at each monitoring station and focused on the general measurement location of each pavement sensor. Images were periodically collected each day and used along with other recorded sensor data to confirm the accuracy of the pavement conditions. For example, precipitation occurring above the freezing point and elevated discharge values were expected for a wet or standing-water pavement condition.

Collection and Analysis of Samples

The conveyance of runoff from the pavement surfaces to the monitoring stations included three primary components: the trench, weir sump, and weir-box outlet (fig. 3). To accurately quantify the total amount of sediment exported from each pavement section, sediment retained in the trenches and weir sumps was measured in addition to sampling the runoff that was conveyed to the weir-box outlet (fig. 3). Samples of highway runoff were collected by using three unrefrigerated automatic samplers during runoff-generating events (rain, mixed precipitation, and snowmelt) and analyzed for concentrations of SS and SS particle-size distribution. For selected events, additional composite samples were measured for concentrations of dissolved major ions, total dissolved nitrogen (TDN), dissolved organic carbon (DOC), particulate nitrogen (PN), particulate carbon (PC), TP, TSS, and selected total-recoverable metals (table 3). The dry mass for selected particle-size classes of sediment retained in the trenches and weir sump (fig. 3) was determined annually, and subsamples for particle-size classes less than 2.0 mm in diameter were analyzed for concentrations of TP and selected total-recoverable metals (table 4). Dry-deposition samples were collected, beginning in the second monitoring year, to estimate the portion of SS collected in the trench (fig. 3) at each site resulting from wind and vehicle turbulence.

Suspended Sediment Samples

Runoff samples for the analysis of SS and particle size were collected on a flow-proportional basis immediately downstream from the weir at the entrance of the 8-in. outlet pipe (weir-box outlet; fig. 3) by using automatic samplers (fig. 5C). Samples were collected for nearly every runoff event from October 2018 through September 2021. For sample collection, runoff events were defined as a function of discharge, where sequential measurements of discharge greater than or equal to 0.005 ft³/s were separated by 6 hours or more of discharge less than 0.005 ft³/s. Runoff events consisted of any form of runoff, including rainfall, mixed precipitation, and snowmelt. A discharge of 0.005 ft³/s was chosen as the cutoff for all automatic sample collection because it was the minimum value that produced a sufficient depth of water in the weir sump to reliably collect runoff samples.

Suspended Sediment Sample Collection

At each station, two (of three) automatic samplers were routinely operated under datalogger control to collect runoff samples. Each autosampler was configured to hold twenty-four 1-liter (L) polyethylene sample bottles into which between one and seven 0.13-L subsamples were composited. These 1-L bottles represent subcomposite samples of event runoff, whereas a composite sample represents the entirety of a runoff event. The first subsample was collected when flow exceeded the minimum discharge threshold (0.005 ft³/s), and subsequent samples were collected at flow-proportional intervals (samples collected at equal volumes of discharge) (fig. 6). After a runoff event was sampled and discharge subsided below 0.005 ft³/s for a minimum period of 6 hours, the datalogger was programmed to instruct the sampler to move the distributor arm to the next sample bottle or, if all bottles in the sampler were full, to the next automatic sampler. This method allowed for the collection of additional samples for subsequent runoff events without compromising the previously collected samples. Similarly, if the system determined sampler errors or three successive failures to collect samples, the datalogger would automatically instruct the secondary or tertiary sampler to resume sampling. Approximately 50 subsamples of runoff were collected for an equivalent runoff of 1 in. of rain, resulting in the same sampling resolution irrespective of storm size. The flow-proportional thresholds generally remained constant during the project; however, these thresholds were increased so the sampler capacity was not exceeded when field work was limited during the COVID–19 outbreak. Periodically, the secondary sampler was used to collect concurrent replicate subsamples. In such cases, both the primary and secondary samplers were triggered simultaneously by the datalogger throughout runoff events.

Samplers were installed in a manner to ensure the best possible performance. Vertical distances from fixed sampling points to the sampler-pump heads were about 6 ft (9 ft in the case of the tertiary sampler) and within optimal suction limits (Bent and others, 2000). All sampler lines were mounted in a steep sloping manner to allow for the complete purging and draining of sample water between samples. Sampler intakes were oriented in a horizontal and downstream direction. This configuration minimizes debris accumulation by forming a small eddy that captures sand particles at the intake and thus allows the sampler to collect a more representative sample in regard to particle size (Edwards and Glysson, 1999). The sampler intake lines consisted of 0.5-in.-diameter polyethylene tubing attached to silicon pump-head tubing.

The vertical distribution of SS in pipe discharge is affected by water turbulence. Fine-grained particulates in the silt and clay size range (less than 0.0625 mm in diameter) are typically well mixed within the water column profile (Butler and others, 1996; Smith, 2002; Smith and Granato, 2010; Selbig and Bannerman, 2011), and concentrations should not be biased by sampling methods; however, a large bias can occur when samples are collected near the floor of a stormwater pipe, where concentrations of coarse sediment particles concentrate in the absence of turbulence. In this study, access to the 8-in.-diameter pipes conveying runoff from the trench system to the weir box precluded the installation of static mixers, stacked sampling ports, or various depth-integrated sample devices that are used with automatic samplers to reduce particle-size bias associated with the collection of stormwater samples. The weir box was chosen because it provided an accurate mechanism to measure discharge over the range of runoff conditions and because the sump area behind the weir plate provided a deposition zone where the coarse-grained sediment could periodically be removed and quantified. The collection of samples immediately downstream of the weir, where the sampling area was well mixed and the SS was dominated by fine-grained sediment, was expected to minimize bias in the sample concentration of the SS regardless of discharge rate, and the absence of coarse or sharp particles (gravel and road debris) that can damage pump-head tubing was expected to improve long-term performance and reliability of the sampling system.

Suspended Sediment Sample Processing and Analysis

Sample bottles for SS concentration and particle-size distribution analyses were collected on a weekly basis when storms occurred. Sample bottles were removed from the autosampler methodically to ensure that the bottles were labeled in the same chronological order in which they were collected. The subcomposite samples contained in each 1-L bottle were not split but analyzed in their entirety for concentrations of sediment and particle size to prevent bias or other issues associated with splitting samples. In cases when the last subcomposite sample bottle contained a single subsample, that subsample was added to the previous subcomposite sample bottle if space was available. Subcomposite samples were analyzed for SS and distribution of particle sizes (less than 0.0625 mm in diameter, greater than or equal to 0.0625 to 0.25 mm in diameter, and greater than 0.25 mm in diameter) at the USGS Kentucky Sediment Laboratory (table 3). Concentrations for subcomposite samples of highway runoff are available through the USGS NWIS (U.S. Geological Survey, 2023b).

Dry-Deposition Samples

Sediment is transported to the trenches from the pavement surface in runoff, and it also can be transported to each trench by wind or vehicular turbulence between runoff events. Sediment deposited in the trenches at the HMA and OGFC stations between runoff events was later transported through the conveyance system with subsequent runoff or remained in the trench drains until the trenches were cleaned out, which was done three times during the 3-year study period. The highway trench systems were rated for heavy vehicle loading with secure grating; therefore, routinely unbolting the trench grates to clean the trenches prior to storms was neither practical nor safe without scheduled traffic control. Because the trenches could not be cleaned routinely, they were considered part of the highway system at each study site. Samples of sediment deposition during the dry antecedent period between storms were collected from each trench, beginning in 2020, to estimate the sediment mass associated with dry-transport processes relative to the total sediment loads measured in the study.

Dry-Deposition Sample Collection

To quantify sediment accumulation in the trenches during selected dry antecedent periods, a 0.4- by 1.4-ft plastic tray was installed at the center of each trench (about 36 ft from the end) and below the grate. The trays were installed after runoff events during routine site visits, and the contents of the tray were collected prior to forecasted precipitation. Samples of dry deposition from each site were collected successfully for three separate periods from February 2020 to May 2020 for deployment durations of 4 to 8 days (Spaetzel and others, 2023); however, many other samples were compromised and discarded during this period because weather forecasts were not accurate or changed without sufficient warning, and storms occurred prior to the removal of the trays. To overcome the challenges of manual deployment of the trays, an automatic dry-deposition collection device was designed and installed beneath each trench grate to quantify sediment accumulation during dry antecedent periods (fig. 7). The same 0.4- by 1.4-ft plastic trays were mounted to a linear actuator and housed within a 6-in.-diameter PVC pipe mounted beneath the centermost grate. The dataloggers at each monitoring station were programmed to extend the tray from the PVC housing following the end of each storm and automatically retract the tray when rain was detected. A contact switch within the housing was measured by the datalogger and provided a secondary method to determine the position of the tray. The duration of time when the tray was extended from the PVC housing and exposed for collection of material was recorded by the datalogger. These devices represented a substantial improvement over manual attempts to collect samples of dry deposition between storm events because the trays were automatically deployed following each runoff event, and sequential dry antecedent periods were sampled between manual removal of the samples. These systems were operated from July 2020 through September 2021 and resulted in 11 sampling periods of 1 to 46 days in duration.

Dry-Deposition Sample Processing

Samples of dry deposition were collected from each sample tray and placed into a resealable polyethene bag during routine visits. The material was dry sieved using a 2.0-mm screen to separate the larger gravel and debris from the finer silt and sand size particles (those most likely to be retained in the OGFC pavement). Particles less than 2.0 mm in diameter were dried in pretared stainless-steel trays at 105 °C to a constant weight, or until the weight change from the previous recorded weight was less than 0.5 milligram (mg) (Clesceri and others, 1998) at the USGS laboratory in Northborough, Massachusetts. The resultant sediment mass for particles less than 2.0 mm in diameter, in grams, was recorded for each deployment period (Spaetzel and others, 2023).

Permeameter Tests

Permeability (also referred to as hydraulic conductivity) is a measurement of the rate at which water flows through the porous-pavement layer and is represented by the coefficient of permeability (K). Semiannual permeameter tests were done in the breakdown lane and first travel lane on the OGFC segment within the study area to estimate changes in pavement permeability during the first 3 years of service. The permeameter design and measurement methods were developed by the National Center for Asphalt Technology, and measurements correlate closely with the results of laboratory tests on asphalt cores under saturated conditions (Cooley, 1999). A Gilson AP–1B falling-head permeameter was used to measure the permeability of the asphalt at the OGFC study site (Gilson Company, Inc., 2019). The field device consists of four stacked tiers of graduated clear plastic tubes, decreasing in diameter from bottom to top. The tiered design optimizes the measurement on the basis of the magnitude of the pavement permeability. In this study, only the bottom two tiers were used.

Permeameter tests were done in the center of the breakdown lane, the tire track area of the first travel lane (about 3 ft from the white line separating the first travel lane from the breakdown lane, also referred to as the fog line), and in the center of the first travel lane, with traffic support by MassDOT. The OGFC pavement in the breakdown lane may be the first area where permeability is affected by sediment deposition because traffic exposure is typically limited, and sediment is less likely to be hydraulically removed by tire action during wet conditions. The first travel lane carries the bulk of the traffic, including about 80 percent of the truck traffic, and may be most affected by deterioration or deformation of the pavement. Permeameter tests were generally done in the evening after traffic volume decreased, under dry conditions in the fall prior to frost conditions, and again later in the spring or early summer. In most cases, measurements were made at 13 equal-spaced locations adjacent to the trench in the center of the breakdown lane, in the tire-track area of the first travel lane, and in 4 to 13 locations near the center of the first travel lane. The number of overall measurements was limited by the amount of water brought to the site and by the duration of lane closures provided by traffic-control support.

Permeameter tests were made by removing any loose particles in the test area with a stiff brush to provide the best seal possible to the device. A moldable sealant was uniformly applied to the outer base of the permeameter to create a watertight seal with the pavement surface. The permeameter was placed on the test area and gently seated to the pavement surface to force the sealant material into the pavement texture. Weights were applied on each side of the device to compensate for head pressure resulting from the water column during testing. The device was filled with tap water at a steady rate up to the selected measurement line. As the water drained into the pavement, the time that elapsed for the water level to reach each specific graduation line was recorded (fig. 8). Pavement surface temperature data also were recorded with a handheld infrared thermometer to document pavement condition at the time of the tests.

Discharge

The theoretical relation between gage height and discharge (which is based on the weir dimensions) was tested at each station by simultaneously measuring the flow rate of a 9,600-gallon-per-hour centrifugal pump with an inline flowmeter and measuring gage height in the weir box. To assess the low end of the relation (discharge near 0.003 ft³/s), a small electric pump also was used to simulate low discharge values. Pump-flow values were within the range of flow estimated from a water level of plus or minus 0.01 ft around the theoretical gage height-discharge relation (U.S. Geological Survey, 2023b). The low-end measurement (less than 0.003 ft³/s) resulted in a larger error about the theoretical relation because of the nappe clinging to the downstream weir face; therefore, the shifted-control method (Kennedy, 1984) was used to adjust the gage height-discharge relation on the basis of this measurement. A range of plus or minus 0.01 ft in gage height represents the typical error for the measurements of the water level in the weir box, as determined by using the adjusted gage height-discharge relation. In general, discharge values less than 0.003 ft³/s may be less accurate where the sensitivity of the gage height-discharge relation was poor, and gage-height values often were affected by the formation of slush during extreme cold periods. These gage-height pump-flow tests indicate that the theoretical relation and adjusted low-end relation for each weir provided accurate estimates for discharge greater than or equal to about 0.003 ft³/s.

Suspended Sediment and Water-Quality Samples

QC samples, including field blank, concurrent replicate, and split replicate samples, were collected to identify potential bias in sampling and processing methods and any contamination resulting from the sampling equipment or the sample-collection, processing, and analysis processes. The QC data are summarized in the following sections and available in the companion data release (Spaetzel and others, 2023).

Sediment Chemistry Source-Blank Samples and Replicates

To identify potential sources of contamination in the collection and processing of weir-sump sediment samples that were analyzed for concentrations of phosphorus and total-recoverable metals (table 4), clean silica well-packing sand was used to prepare source-blank QC samples. Source-blank samples were prepared directly from the source-blank material without being dried or sieved. Equipment blanks were samples of source-blank material that were sieved and dried in an oven for 2 days, similarly to the preparation of the environmental weir-sump sediment samples. One pair of replicate source-blank samples also was submitted for analysis at the beginning of the study (November 2019). Because the well-screen sand is not a true blank in the same sense that deionized water is a blank for water samples, the blank samples were evaluated by comparing the concentrations measured in the source-blank and equipment-blank samples as opposed to evaluating the blank samples on the basis of the presence and magnitude of detections of the relevant constituents.

The comparison of source-blank and equipment-blank sample pairs (table 8) does not indicate that sample processing introduced contamination; however, comparisons were complicated by the presence of multiple detection and reporting levels in the data, which were the result of analytical parameters used in the laboratory for each batch of samples. RPDs associated with positive arithmetic differences, indicating that the concentration in the equipment-blank sample was higher, ranged from 2 to 67 percent (median of 22 percent). Even the highest of these RPDs corresponded to an arithmetic difference of only 0.1 milligram per kilogram (mg/kg; environmental results range from 12 to 38 mg/kg, so this is not a substantial difference). The greatest RPDs and arithmetic differences were associated with samples that had higher concentrations of a constituent in the source-blank sample than in the equipment-blank sample. For cadmium (p67880) and zinc (p64180), these results are the consequence of the varying reporting limits (even within a single analytical run). The RPDs observed between source-blank and equipment-blank samples are similar to those observed in replicate environmental results, which are described next.

Replicate samples of sediment were collected by withdrawing two subsamples from the dried and sieved fraction of sediment collected from the weir sump. Six replicate pairs were collected during the study period; however, no results were returned for the concentration of total-recoverable metals for one of the replicates submitted for analysis (sample counts in table 9). The following were collected: two replicate sample pairs sieved to a particle size less than 0.25 mm in diameter (one of which was only analyzed for phosphorus), two replicate sample pairs sieved to greater than or equal to 0.25 to 0.50 mm, one replicate sample pair sieved to greater than or equal to 0.50 to 1.0 mm, and one replicate sample pair sieved to greater than or equal to 1.0 to 2.0 mm. The median RPD for each constituent ranged from 16 to 77 percent (table 9). RPDs were not consistently higher or lower in samples of a particular grain size, although concentrations tended to be higher in the smaller grain-size portions. A single replicate pair accounts for the highest RPDs observed for 9 of the 11 constituents. The variability observed does not show consistent evidence of bias.

Runoff and Precipitation Volumes

Continuous records of discharge and precipitation were used to define runoff-event periods and event-precipitation totals. Due to the minimum volume of water necessary to submerse the intake for the automatic samplers, the minimum discharge that resulted in a sampled runoff event was 0.005 ft³/s, as described in the “Collection and Analysis of Samples” section above. Runoff events were also characterized by using minimum discharge thresholds of 0.003, 0.002, and 0.001 ft³/s to estimate the volume of runoff below 0.005 ft³/s that was not sampled and to assess the sensitivity of salt-load estimates to discharge. In each iteration to determine runoff-event durations based on these three discharge thresholds, the start of a runoff event was defined as the time when discharge equaled or exceeded the threshold value. The end of the runoff event was defined as the time when the discharge decreased below the threshold and remained less than that value for 6 hours. The 6 hours of discharge below the threshold were not included in the runoff-event duration. The total discharge of each event was determined by summing the 1-minute frequency measurements of discharge (converted to volume, in cubic feet [ft³]) over the runoff-event duration. Unless otherwise noted, the loads of sediment and chemical constituents in this report represent the total loads associated with runoff greater than or equal to 0.003 ft³/s because discharge values less than this have substantial uncertainty, as described in the “Discharge” section. Precipitation totals for each event were determined by summing the instantaneous measurements (collected every 5 minutes) of precipitation over the runoff-event duration, including the 6 hours prior to the start of runoff. The precipitation totals multiplied by the drainage area resulted in the precipitation volume (in cubic feet); these event-precipitation volumes were subsequently used in the determination of runoff coefficients.

Runoff Coefficients

A runoff coefficient is the ratio of the volume of runoff to the volume of rainfall. Runoff coefficients for rainfall events between April 1 and October 31 (nonwinter periods) were calculated by dividing the runoff total by the product of the measured rain total and the drainage area for each site. These runoff coefficients were used during the study to select appropriate flow thresholds for triggering the automatic samplers and to identify changes in the pavement contributing area. The ratio of the volume of runoff to the volume of rainfall for a given area should range from zero (no runoff) to one (100 percent of the precipitation is measured in the runoff); however, if a runoff coefficient exceeds a value of one, there likely is an error in the measurement of precipitation, discharge, or contributing area (Church and others, 1999). Changes in the contributing area can result when discharge from an upgradient drainage area is diverted into the drainage area of interest, such as when the inlet of a catch basin is partially or completely blocked by deposits of sediment or debris. Runoff coefficients estimated in this study also were compared to a selection of runoff coefficients available in version 1.1.0 of the Highway-Runoff Database (HRDB; Granato, 2019). The HRDB data selection was limited to the months of April through October, precipitation events greater than 0.05 in., and sites with 100 percent impervious area to represent characteristics similar to the runoff coefficients examined in this study.

Event-Mean Suspended Sediment Concentrations and Loads

EMCs of SS were determined for all sampled runoff events (n=226 at the HMA station; n=168 at the OGFC station) by using the concentrations of SS measured in more than 1,000 subcomposite samples at each monitoring station. Each event was characterized by 1 to 46 subcomposite samples, and each subcomposite sample included 1 to 15 (but most commonly 7) subsamples (fig. 6). The EMC was computed as a flow-weighted average of the subcomposite concentrations collected during a single runoff event by using the following equation:

Individual subcomposite concentrations of SS are published in the USGS NWIS database (U.S. Geological Survey, 2023b), and computed EMCs are published in the associated data release (Spaetzel and others, 2023).

In a few cases, subsamples were lost as a result of errors with the sampling equipment or issues during sample shipment or analysis. When a subcomposite sample collected during the event was lost, the subcomposite concentration was estimated by averaging the concentrations of the samples collected before and after the missing subcomposite. If the missing sample was either the first or last one collected, the most recent subcomposite concentration available was repeated to fill in the missing concentration. For 9 events at the HMA site and 2 events at the OGFC site, autosampler or laboratory errors resulted in the loss of all subsamples. SS EMCs were estimated for these events either by using concentrations from the events immediately before or after or, if a composite sample was being collected concurrently with the subcomposite samples, by using the measured EMC from the paired composite. The SS EMCs measured in composite samples (n=15 at each site) were compared to the calculated EMC values in the “Concurrent and Split Replicate Runoff Samples” section above. Filling in missing data with estimates, although done infrequently, was important to provide the best representation of the sediment load exported from each type of pavement over the 3-year study period. The calculated EMCs were paired with runoff volume to compute a sediment load for each runoff event.

Because the event sampling was based on a minimum discharge value of 0.005 ft³/s, the portion of SS load associated with discharge less than 0.005 ft³/s and greater than or equal to 0.003 ft³/s was estimated. As noted previously, the uncertainty of discharge data down to 0.003 ft³/s is well constrained, but the gage height-discharge relation is poor below 0.003 ft³/s. To estimate the unsampled SS load, the difference between the total annual volume of discharge determined from a minimum discharge threshold of 0.003 ft³/s and the total annual discharge determined from a minimum discharge threshold of 0.005 ft³/s was multiplied by the 10th percentile value of the annual subcomposite SS concentrations (for each site and for each water year).

To compare the magnitudes of SS EMCs between the two pavement types, two statistical tests were done to compare distributions of concentrations. A one-sided Wilcoxon rank-sum test (also referred to as a Mann-Whitney test) was done for the complete EMC populations for sampled runoff events (n=226 at the HMA station and n=168 at the OGFC station) as well as a subset of samples that corresponded to 165 approximately concurrent runoff-generating events at both stations. This subset (n=165 for each station) excludes events where discharge decreased below 0.005 ft³/s at the HMA station between periods of precipitation but persisted at the OGFC station, resulting in a single runoff-generating event at OGFC but two independent runoff-generating events at HMA station. The Wilcoxon rank-sum test is a nonparametric statistical test for comparing two independent groups of data of any given distributional shape; specifically, it determines if there is evidence that the groups are from the same population (Helsel and others, 2020). For the comparison of SS EMCs measured at HMA and OGFC monitoring stations, the null hypothesis was that SS EMCs measured at the HMA station were equal to or less than EMCs measured at the OGFC station. The null hypothesis was rejected if p-values were less than 0.05.

Total loads of sediment were calculated by summing the computed loads of SS in runoff and direct measurements of the mass of sediment retained in the weir sumps and trenches. The SS EMCs in runoff and the weir-sump sediment masses may be discretized into water-year totals because of the sampling intervals; however, the three trench masses represent 22-, 2-, and 12-month periods, respectively; therefore, they cannot be distributed by water year (see “Data Collection Methods” for description of trench sampling). For a representative comparison of total sediment load between water years, the total load of sediment (equal to the sum of SS in runoff collected from the weir-box outlet, retained weir-sump sediment, and retained trench sediment [fig. 3]) for water years 2019 and 2020 was averaged.

Estimated Chloride and Sodium Concentrations

Specific-conductance measurements are commonly used to estimate concentrations of dissolved major ions (Hem, 1982, 1985; Miller and others, 1988; Church and others, 1996; Granato and Smith, 1999; Smith and Granato, 2010; Smith, 2015; Spaetzel and Smith, 2022). Concentrations of dissolved Cl and Na, along with specific conductance, were measured in composite samples and discrete runoff samples (samples that represent a single point in time) at HMA and OGFC monitoring stations to determine a regression equation relating specific conductance to Cl and Na concentrations. Continuous specific-conductance data recorded in baseline 10-minute increments and event-based 1-minute increments were available at both stations nearly every day of the study period. Using these continuous records with the regression equations produced continuous estimates of Cl and Na concentrations that were summarized over storm events to determine EMCs of Cl and Na.

Cl and Na concentrations were estimated from continuous measurements of specific conductance on the basis of equations developed to relate specific conductance to concentrations of the respective ions (see eq. 2 below). These regression equations were developed with the Maintenance of Variance Extension Type 1 (MOVE.1) technique on the basis of concurrent measurements of log-transformed values of specific conductance and log-transformed concentrations of Cl and Na in runoff samples. Because the equation was developed with log-transformed values of specific conductance and concentration, the bias correction factor (BCF) also was determined (Duan, 1983) to correct for the bias introduced by retransforming values (taking the antilog) to estimate event-mean concentrations and loads of Cl and Na. The MOVE.1 technique was chosen for regression analysis because it minimizes errors in both the x and y directions. This technique generates a unique equation, which can then be used to estimate either variable from the other; the estimations reflect the variance of the measured concentrations (Helsel and others, 2020). The MOVE.1 equation is given as

Two multisegment regressions were developed using pooled sample data from the stations (table 10). The ratios of Na to Cl, magnesium to Cl, calcium to Cl, and sulfate to Cl in samples collected from both sites were similar, which was expected because deicing compounds and the frequency and magnitude of application were the same at both stations (due to the proximity of the stations along Interstate 95, see fig. 1). The ratios of the four other ions to Cl were more variable in samples with specific-conductance values less than 400 μS/cm, but for most samples above 400 μS/cm, the ratio of Na to Cl was generally constant, ranging only between 0.57 and 0.62, whereas the ratios of magnesium, calcium, and sulfate to Cl dropped to near 0. Because of the variability of these ion ratios in samples within the lower range of specific-conductance values (generally less than 400 μS/cm), two-part regressions were selected to better represent the large range in specific-conductance values and ion concentrations. In the sample data (n=71), specific conductance ranged from 7 to 119,000 μS/cm, Cl concentrations ranged from 0.96 to 57,700 mg/L, and Na concentrations ranged from 0.72 to 34,600 mg/L. For specific-conductance values less than 400 μS/cm, part 1 of the equation was applied to estimate Cl or Na. For specific-conductance values greater than or equal to 400 μS/cm, part 2 of the equation was applied. Although applied to discrete ranges of specific conductance, the ranges of data used to develop each segment overlap to produce a continuous function (table 10) (Granato, 2006). The standard errors of regressions, also referred to as the root mean square errors, are 3.2 and 3.6 percent, respectively, for the Cl and Na equations for specific-conductance values greater than or equal to 400 μS/cm. The standard error of regressions was 16 percent for both ions at the lower specific-conductance values (less than 400 μS/cm) due to the variability in Cl and Na concentrations (fig. 9).

Loads of Total Phosphorus and Total-Recoverable Metals

Loads of TP and selected total-recoverable metals exported from the two highway pavement sections were estimated for the 3-year study period. To account for the total load of these constituents in highway runoff, data from three monitoring sublocations at each station were summed. These monitoring sublocations are represented in the monitoring-site diagram (fig. 3) and are the trench, weir sump, and weir-box outlet. Concentrations of TP and selected total-recoverable metals were measured in composite samples of runoff (n=15 at each site) collected at the weir-box outlet and in annual samples of the retained weir-sump sediment (fig. 3). The masses of sediment-bound TP and selected total-recoverable metals in the weir-sump and trench sediment were computed by multiplying the concentrations (in milligrams per kilogram) by the mass of sediment (in kilograms) corresponding to each particle-size class. Although concentrations of TP and selected total-recoverable metals were not measured in samples of sediment retained in the trench, the particle-size distributions of the trench and weir-sump sediment were similar; therefore, it was reasonable to use the concentrations measured on weir-sump sediment for estimating the constituent loads retained in the trench. The loads of TP and metals associated with the runoff collected at the weir-box outlet were estimated at each site by multiplying the average concentration measured in the composites (n=15 at each site) by the total volume of flow for the study period. Forty percent of the cadmium concentrations measured at each site were reported as less than the detection limit (0.03 microgram per liter [µg/L], except for one diluted sample reported as less than 0.06 µg/L). Due to the presence of censored data and to avoid biasing the average by substituting half the detection limit for censored results, the nonparametric Kaplan-Meier method (Kaplan and Meier, 1958; Helsel, 2011) was used to estimate the mean concentration of cadmium in the 15 composite runoff samples collected at each site. Ninety-five percent confidence intervals about the means were computed by using a nonparametric bootstrapping method with 10,000 replicates. The Kaplan-Meier estimate of mean concentrations (for cadmium) and 95-percent bootstrap confidence intervals (for all constituents) were computed in the R programming language (R Core Team, 2018) using packages EnvStats (Millard and Kowarik, 2022) and boot (Canty and Ripley, 2022).

Dry Deposition

Sediment was collected from sampling trays in each trench during dry antecedent periods from February 2020 through September 2021. The mass of sediment in these samples and the duration of the collection period were used to estimate rates of sediment entering the trench during dry periods. Sediment collected from the trays (1.4 ft in length) was assumed to represent the typical deposition rate throughout the 72-ft trench. The mass of sediment less than 2.0 mm in diameter collected in the 1.4-ft trays was multiplied by 51.4 (to scale up to the full length of the trench) and divided by the duration of the collection period (in days) to estimate a daily rate of sediment deposition in the trench for each deployment period (in grams per day). The average daily rate for WY 2021 was multiplied by the sum of dry antecedent periods (in days) to estimate a WY 2021 load of sediment less than 2.0 mm in diameter that was deposited in the trench during dry periods. Sediment greater than or equal to 2.0 mm in diameter generally consisted of gravel, woody debris, and litter; therefore, rates were not estimated for this fraction.

Pavement Permeability

Permeability measurements were made on Interstate 95 near USGS station 421652071120601 on the OGFC pavement shortly after the road was resurfaced in the fall of 2018 and semiannually through October 2021. Measurements were made in the center of the breakdown lane, the tire-track area of the first travel lane, and the center of the first travel lane by using a falling-head permeameter. Pavement cores were not taken at the study site, so the depth of the OGFC layer at each test location was assumed to be similar to the 0.1-ft design specifications, which is consistent with the OGFC pavement layer depth measured near the edge of the highway next to the trench (fig. 2). The coefficient of permeability was calculated for each measurement by using equation 3, as follows:

The coefficients of permeability determined for the permeameter tests are published in the companion data release (Spaetzel and others, 2023).

Runoff Coefficients

Runoff coefficients, the ratio of runoff to rainfall volume, were computed for runoff events with a minimum discharge value greater than or equal to 0.003 ft³/s and minimum rainfall greater than or equal to 0.05 in. to evaluate the accuracy of drainage-area estimates. Events that occurred in winter months (November through March) were excluded from runoff coefficient summaries because the presence of snow and ice impacts the timing and volume of runoff associated with precipitation events. For example, the addition of snowmelt to runoff volume during a winter rain event will result in a runoff coefficient greater than 1. As described previously, the theoretical bounds of runoff coefficients are 0 to 1, but these bounds may be violated due to site-specific and event-specific conditions that impact the drainage area. Granato and Cazenas (2009) noted that among stations compiled for the HRDB, 46 percent had runoff coefficients that varied by more than one order of magnitude, which is true of the two highway sites in this study. Although the range of runoff coefficients observed in the selected HRDB dataset reflects variability across a far greater number of events (n=697) and sites (n=60), the median is not substantially higher than those observed at the HMA and OGFC stations in this study (fig. 10).

The distributions of runoff coefficients were similar at the HMA and OGFC stations and over time during the study period (fig. 11) but reflect a wide range of values as a result of event characteristics such as rainfall intensity, traffic conditions, and wind speed or direction. Nonwinter event runoff coefficients over the 3-year study period varied from less than 0.03 to 2.0 for HMA and 0.01 to 1.7 for OGFC, with medians equal to 0.66 and 0.70, respectively. Runoff coefficients greater than 1 were computed for 3 percent of events at HMA and 11 percent of events at OGFC, which indicates that under certain storm conditions, the drainage area increases. These increases may result from a shift in the drainage pathways due to standing-water conditions or spillover from adjacent drainage areas. For example, there are catch basins in the inner breakdown lane opposite the south end of both trenches (fig. 1). Based on elevation data and design specifications, the inner breakdown lane drains parallel to the road from north to south; however, the elevation differences within the inner breakdown lane are subtle (less than 0.05 ft in some cases; fig. 12). If the catch basin is obstructed due to leaves or debris, water may back up in the inner breakdown lane, raising the surface elevation and causing water to flow perpendicular to the roadway (from east to west) into the lanes that drain to the trench. Similarly, the water elevation in the inner breakdown lane may equal or exceed the adjacent travel lanes during high-intensity storms when standing water is likely to occur (fig. 12). These kinds of shifts in drainage areas can substantially affect runoff volume and have been observed in other highway studies (Church and others, 1999; California Department of Transportation, 2000; Smith and others, 2018). Because the runoff coefficients greater than 1 are associated with events that span a range of rainfall amounts, intensities, and durations, it is not possible to identify a single causative factor that leads to excess runoff. Although runoff coefficients greater than 1 indicate that drainage area fluctuated under some conditions, the limited frequency of runoff coefficients greater than 1 (3 and 11 percent at HMA and OGFC stations, respectively), as well as the site-to-site similarities, indicate that the drainage-area estimates (table 1) are valid. Because the drainage areas are the same for each pavement section, loads of sediment and constituents discussed in this report may be directly compared between the two sites.

Runoff losses due to splash and spray from vehicle turbulence and wind also affect runoff coefficients; however, the degree to which these mechanisms affect runoff volume is subject to many factors, from the traffic speed and traffic conditions during rainfall to wind speed and direction. These mechanisms may result in net losses or gains if rainfall from the northbound lanes is deposited on the southbound lanes draining to the study sites. In addition to the natural factors related to rainfall losses that vary from event to event, the error in precipitation measurements also varies with storm conditions. Smith and Granato (2010) observed that rainfall totals computed with colocated gages may differ by up to 0.10 in. but are commonly (95 percent of the time) within 0.04 in. of one another. Larger differences between the amounts of rain measured by precipitation gages could be the result of the physical mounting location and the direction and speed of wind during the storms.

Open-Graded Friction Course Pavement Permeability

The permeability of the OGFC layer is a function of the pavement design criteria, asphalt content, aggregate gradation, density of interconnected aggregate voids, and sediment loading. Permeability values of 164 in/hr are expected for properly functioning OGFC (Mallick and others, 2000; Kandhal, 2002). Mallick and others (2000) reported that the permeability coefficients for OGFC mixes consisting of more than 30 percent aggregates less than 4.75 mm in diameter are less than 46 in/hr, but the permeability coefficients increase substantially for mixes containing 25 to 15 percent aggregates (144 to 192 in/hr) in the same aggregate-size range. In this study, between 20 and 40 percent of the aggregates in the OGFC pavement are less than 4.75 mm in diameter (table 2), and this pavement layer has a design permeability of 64 in/hr (Kevin Fitzgerald, MassDOT, written commun., 2018). The permeability of the pavement layer affects the movement of water to the pavement edge and, in part, prescribes the thickness of the pavement mat that is required to prevent runoff from flowing across the pavement surface under most rainfall conditions. When runoff breaches the surface of the porous-pavement layer, the desired safety benefits of OGFC (including reductions in vehicular spray, nighttime surface glare, undercarriage wash off, and hydroplaning) are not achieved.

Initial permeameter measurements on Interstate 95 were made shortly after the OGFC pavement layer was installed in September 2018. Median permeability coefficients for the center of the breakdown lane and the center of the first travel lane were 210 and 82 in/hr, respectively (fig. 13). The median of permeability coefficients for measurements made in the breakdown lane after the first 10 months was the same (210 in/hr). However, the median of permeability coefficients for measurements made in the center of the first travel lane increased to 96 in/hr, but only four measurements were made at this time because additional measurements were made in the tire track of the first travel lane (median of 140 in/hr). Median values for measurements made in the breakdown and in the tire-track area of the first travel lane decreased throughout the study period (fig. 13). In the series of measurements taken at the end of the study in October 2021, the median permeability coefficients for the center of the breakdown lane and the center of the first travel lane decreased to 3.1 and 38 in/hr, respectively. The individual measurements made in the center of the first travel lane were more variable, in part because fewer measurements were made in this location (fig. 13).

The average variabilities of measurements (interquartile range divided by median of each test set) made in the center of the breakdown lane, the center of the first travel lane, and the inner tire path of the first travel lane were about 29, 14, and 13 percent, respectively. Some of the variability could be due to variations in the thickness of the OGFC layer because permeability measurements are sensitive to the pavement-mat depth; a difference of 0.01 ft could affect the permeability coefficient by 10 percent or more. Pavement permeability in the breakdown lane decreased by nearly two orders of magnitude from September 2018 through October 2021, with the greatest change occurring in 2021 (fig. 13). Pavement permeability in the first travel lane decreased by about 53 percent in the center of the lane and by about 83 percent in the inner tire pathway during the study, although the first measurements in the inner tire pathway were not made until July 2019. These findings indicate that the void structure of the OGFC pavement was filling with sediment over the 3-year study period. The results also indicate that OGFC pavement permeability is maintained within the travel lanes longer than in the breakdown lane because of the cleaning pressure or suction caused by tires traveling over the layer (Isenring and others, 1990). The traffic volume and rate of speed also tend to promote this cleaning action (Van Heystraeten and Moraux, 1990; Putman, 2012).

Road-Weather Conditions

Road-weather conditions determined by noncontact measurements of the first travel lane at each monitoring station were collected during the study period to understand the differences in runoff behavior on each pavement types and to identify changes in runoff behavior related to decreases in OGFC pavement permeability over time. The frequencies of like measurements during nonwinter runoff periods are summarized for comparison (table 11). During the summer of 2019, the noncontact pavement sensor failed and was removed from the HMA monitoring station, and data were unavailable for several months while it was being repaired. As a result, only the data collected when both noncontact sensors were operating are listed in table 11. Pavement-condition data may be considered a semiquantitative measurement: a damp pavement condition implies that the road is moist but has insufficient water to cause tire spray, a wet pavement condition implies that the road has sufficient water to cause tire spray, and a standing-water condition implies that the depth of water on the pavement results in a wake behind the tires.

During nonwinter months (April–October) of WYs 2019–20, a standing-water condition occurred in 0.25 and 0.24 percent of the measurements, respectively, at the OGFC site (table 11). Measurements of standing-water conditions for the same period were about five times more frequent at the HMA site, indicating that the runoff was draining into the OGFC layer as designed. Similarly, a wet condition occurred about five times more frequently at the HMA site than the OGFC site during WY 2019, but the frequency of a wet condition was about the same on each pavement type in WY 2020. Isenring and others (1990) suggest that porous asphalt layers remain wet longer after rain compared to dense-graded pavement layers because the water within the OGFC layer is pressured out by tire action. In WY 2021, wet and standing-water conditions were recorded more frequently on the OGFC pavement than the HMA pavement. It is unclear why these conditions occurred at a greater frequency on the OGFC pavement during this period. The increase in wet and standing-water conditions may be related to the substantial decrease in permeability coefficients in the breakdown lane at the OGFC site during WY 2021 (fig. 13). Because the OGFC permeability was decreased by sediment entrapment, the thickness of the OGFC layer at the measurement area was insufficient to accommodate the same amount of lateral flow, resulting in flow breaching the pavement surface. This condition may have been further increased by the greater precipitation volumes in the summer of WY 2021, which were substantially higher than in the prior WYs (table 11). Conversely, although the noncontact pavement sensors were routinely calibrated and verified, small changes in the macrotexture of the pavement could result in the shift of values from one condition to the next and in increased frequency of the wet or standing-water condition.

The effects of winter weather conditions on OGFC pavement, in particular the formation of ice and snow, are not fully understood. The permeability of the OGFC layer results in lower thermal conductivity than HMA pavement because the interconnected voids in the OGFC layer allow air movement throughout the pavement layer (Putman, 2012). As a result, the surface temperature of OGFC can be 1 to 2 °C lower than that of adjacent HMA (Isenring and others, 1990; G. Lefebvre, 1993, as presented in NASEM [2009]), which can lead to more frequent frost and ice formation. In this study, monthly median overnight (6 p.m. to 6 a.m.) temperatures of the OGFC pavement tended to be cooler than HMA temperatures by 0.1 to 0.5 °C in WYs 2019–20 and warmer than HMA temperatures by up to 0.3 °C in WY 2020; however, these differences were within the instrument accuracy (plus or minus 0.5 °C) (High Sierra Electronics Inc., 2017).

In this study, the two pavement sections were subject to identical deicing treatments, so differences in the frequency of various wet and icy road-weather conditions should reflect distinct properties of HMA and OGFC pavement. Pavement conditions reflecting winter weather, particularly “ice” conditions, were measured nearly twice as often on the HMA pavement as on the OGFC pavement (fig. 14A). The number of chemically active measurements, when the salinity of pavement water was sufficient to reduce the freezing point below 0.0 °C and maintain a damp or wet condition on the pavement surface, was marginally greater on the HMA pavement than on the OGFC pavement during WYs 2019 and 2020. In WY 2021, about 20 percent more chemically active measurements were recorded for the OGFC pavement than the HMA site (fig. 14B). Although salt crystals are likely to move into the porous structure of the OGFC pavement, much like sediment, research has shown that the salt solution can be pumped in and out of the void structure by the tire suction and high traffic volumes (A.P. Greibe, 2002, as presented in NASEM [2009]). It is unclear how the change in the permeability coefficients from winter to winter or the clogging of the breakdown lane at the OGFC site affected the pavement surface conditions, but pavement-condition measurements during the final winter indicate that damp and wet conditions on the first travel lane were more frequent at the OGFC site than the HMA site.

Suspended Sediment Concentrations

During the study period (October 1, 2018, to September 30, 2021), 2,040 flow-proportional subcomposite samples (fig. 15) were collected at the highway monitoring stations to characterize 226 and 168 sampling events at the HMA and OGFC stations, respectively. These subcomposite samples consist of about 6,700 (HMA station) and 6,500 (OGFC station) discrete subsamples collected on a flow-proportional basis. Although the stations experienced the same precipitation conditions, the number of sampled runoff events was greater at the HMA station. Runoff from some small runoff events did not exceed the discharge threshold (0.005 ft³/s) to trigger sampling at OGFC, and some events at OGFC encompassed multiple sampling events at HMA because the duration of runoff was often extended on the OGFC pavement. The median and average concentrations of SS in subcomposite samples over the study period were 20 and 34 mg/L in samples collected from HMA and 13 and 23 mg/L in samples collected from OGFC (fig. 15). In the particle-size distributions of SS in samples from both stations, generally 90 percent of the material was less than 0.0625 mm in diameter, and the median percentages of material less than 0.25 mm in diameter were 98 and 97 percent, respectively, at HMA and OGFC. Sediment greater than 0.25 mm in diameter was generally retained in the trench or weir box, where settling of larger material occurred. The SS EMCs computed for all events ranged from less than 0.5 to 677 mg/L at the HMA station and from 2.0 to 192 mg/L at the OGFC station, with respective median EMCs equal to 29 and 15 mg/L (fig. 16). The minimum SS EMC at the HMA station occurred during a small (0.05 in.) precipitation event that did not generate enough runoff to be sampled at the OGFC station; a single subcomposite was collected during the 1-hour event, and the concentration of SS was censored at less than 0.5 mg/L. Annual (WY) comparisons of the SS EMCs indicate that concentrations were consistently lower at the OGFC station during the study period (fig. 17). EMCs of SS were significantly higher in HMA samples, based on the Wilcoxon rank-sum tests for the full populations of EMCs and the subset of paired events (p-value less than 0.05; table 12), when considered over the entire study period and when grouped by each water year.

Sediment Loads

The total sediment load from each study site represents the mass of material conveyed off the highway section via runoff and material deposited by wind and vehicle turbulence in the trenches between runoff events. Sediment measured in the runoff samples, weir sump, and trench were summed over the 3-year study period. The total load of sediment (particle sizes up to 6.0 mm in diameter) was 202 kg from the HMA site and 120 kg from the OGFC site, a difference of 41 percent relative to the HMA site for particle sizes up to 6.0 mm. The proportions of total sediment load measured in the runoff samples were 17 and 19 percent at the HMA and OGFC monitoring stations, respectively. The largest proportion of the load was measured in the sediment retained in the weir sump, which accounted for 61 percent of the HMA load and 62 percent of the OGFC sediment load (table 13). The comparison of the study-period loads by particle-size class (fig. 18) indicates that the greatest differences in loads between the two stations were in the particle-size ranges less than 0.0625 mm to 2.0 mm in diameter, indicating that particles in this size range are readily retained by the voids in the OGFC pavement. Particles greater than 2.0 mm in diameter are in the gravel-size range and are similar to the smaller aggregates used in the OGFC pavement (table 2), so these particles are not trapped in the OGFC voids. The 3-year total load of sediment up to 2.0 mm in diameter at the OGFC site was 49 percent less than at the HMA site (loads equal to 85 and 168 kg, respectively). Of the total load of sediment particles less than 2.0 mm in diameter, 33.8 and 23.7 kg (20 and 27 percent) were estimated for runoff discharged from the HMA and OGFC sites, respectively, and sediment retained in the weir sump accounted for 58 percent of the HMA monitoring station load and 55 percent of the OGFC site load (table 13). The additional SS load that was estimated for unsampled discharges between 0.003 ft³/s and 0.005 ft³/s equaled less than 0.5 kg per year at both stations. This additional estimated SS load in runoff accounted for 3 to 12 percent of the annual SS loads and 0.9 percent (HMA station) and 1.8 percent (OGFC station) of the 3-year total sediment loads.

Sediment and debris often accumulate on paved surfaces between runoff events. Highway sediments can be a mix of materials, including pavement particles, atmospheric dust, adjacent natural soils, vehicle residuals (rust, tire particles), and plant and leaf materials. On curbed roadways, between 75 and 90 percent of the street sediment often is deposited near the curb (Sartor and Boyd, 1972; Pitt, 1979; Selbig and Bannerman, 2007); however, the study sites on Interstate 95 lack curbing, and sediment is likely blown from the pavement surface to the highway shoulder by wind and vehicle turbulence. In this study, material deposited in the trenches was included in the measurement of total sediment loads because it was not feasible to clean the trenches between all events. Wind data were collected near the road surface adjacent to the guardrails at each site; wind was predominantly from the north, with a mean wind-vector speed of about 2 miles per hour. The prevailing wind direction and constant wind source indicate that sediment was likely blown off the highway surface rather than onto the highway surface from the adjacent shoulders; therefore, although this material was not conveyed to the trenches by runoff, it is an important component of the total load from the highway surface. Average extrapolated rates of sediment less than 2.0 mm in diameter deposited in the trenches during dry periods between February 2020 and September 2021 were 39.8 and 12.7 grams per day (g/d) for the HMA and OGFC trenches, respectively. The lower rate of deposition to the OGFC trench indicates that the sediment is retained in the void structure of the OGFC even under wind and vehicle turbulence. In WY 2021, the average deposition rates of sediment (less than 2.0 mm in diameter) in the HMA and OGFC trenches were 38.1 g/d and 14.5 g/d, respectively. The resultant annual dry-deposition sediment loads of 10.8 and 4.12 kg for the HMA and OGFC trenches represent about 17 and 8 percent, respectively, of the WY 2021 total sediment loads less than 2.0 mm in diameter. The addition of sediment in the trenches during the dry antecedent periods had a minimal effect on the total study-period loads.

The annual (water year) sediment loads indicate that OGFC pavement retention of sediment decreased during the study period. The average total load of sediment (particle sizes up to 2.0 mm in diameter) in WYs 2019–20 was 68 percent lower at the OGFC site than the HMA site, but this gap narrowed to 19 percent in WY 2021. Furthermore, in WY 2021, the load of sediment between 1.0 and 6.0 mm in diameter at the OGFC site exceeded the HMA sediment load in these particle-size classes (fig. 19). The raveling of OGFC pavement material may be contributing to the increased sediment load of particles greater than 1.0 mm in diameter because, as previously noted, the OGFC pavement is largely composed of particles greater than 2.36 mm (table 2). These observations in loads also are consistent with the decrease of pavement permeability and increases in standing-water conditions that were observed over the study period.

Chloride and Sodium Concentrations

EMCs of Cl and Na were calculated from estimated dissolved concentrations of Cl and Na derived from regression equations between specific conductance and sample concentrations. Data representing each measurement interval (event duration) were summed to estimate an EMC for each runoff event based on a discharge threshold of 0.003 ft³/s. The Cl and Na EMCs are published in the companion data release (Spaetzel and others, 2023). The application of deicing compounds on both pavement sections was identical, with the maintenance strategy focused on the treatment of the primary OGFC pavement surface, yet the distribution of the EMCs from the OGFC pavement was slightly greater than the distribution of the EMCs from the HMA pavement (fig. 20). Median EMCs for Cl and Na were 26 and 20 mg/L from the HMA pavement, compared to 36 and 27 mg/L from the OGFC pavement. Although drainage behavior on the two types of pavement differs during runoff events, the cumulative effect is not fully accounted for over time. Much like the differences observed in the dry deposition between the two pavement sites, the loss of crystalline salt during dry conditions is likely limited, and as a result, a greater fraction of the salt remains on or in the OGFC pavement void structure rather than being blown from the roadway. Isenring and others (1990) suggest that salt tracking, where vehicle tires drag the salt down the road, occurs less on OGFC than HMA pavement for similar reasons.

Chloride and Sodium Loads

Loads of Cl and Na were computed on an annual basis by summing the runoff-event loads, which are the products of the estimated EMCs and the event discharge volumes. The sums of the Cl and Na loads are referred to herein as salt loads. Salt loads are reported for three discharge thresholds (see the “Analysis Methods” section) to demonstrate the sensitivity of salt-load estimates to discharge. Loads associated with discharges greater than or equal to 0.003 ft³/s are consistent with the loads of sediment and chemical constituents given in this study because discharge values below 0.003 ft³/s are subject to greater uncertainty (see the “Discharge” section, within the larger section “Data Quality”).

Loads of salt estimated for discharge greater than or equal to 0.003 ft³/s were highest in WY 2019 (789 and 1,480 kg at HMA and OGFC, respectively) and lowest in WY 2020 (530 and 523 kg at HMA and OGFC, respectively) (fig. 21). The total snowfall in WY 2020 was 26.6 in., whereas the total snowfall amounts in WYs 2019 and 2020 (45.9 and 54.3 in., respectively; table 14) were more similar to the long-term average snowfall of 51.4 in. (computed for WYs 1973–2021) (National Oceanic and Atmospheric Administration, 2023). MassDOT (2020) identified the winter of WY 2020 as the fourth mildest winter between 2001 and 2020. Accordingly, MassDOT applied deicing material to the highway on fewer days in WY 2020 than in WYs 2019 and 2021 (table 14), and at the statewide level, this difference corresponded to a 41-percent decrease in salt usage compared to the long-term (20 year) average (Massachusetts Department of Transportation, 2020).

The relative magnitudes of annual salt loads were largely driven by winter conditions and deicing practices; however, the variability between sites was affected by the relative proportion of salt export by three primary mechanisms. These transport mechanisms are physical removal of salt-laden frozen precipitation by snowplow, export of salt crystals by wind or vehicle turbulence, and direct runoff of dissolved salt. In this study, loads were measured on the basis of the concentrations of Cl and Na in runoff and do not account for salt export off the highway surface by snowplow or wind. Because the application of deicing compounds was the same on each pavement section (due to proximity, see fig. 1), differences in salt loads between the two sites were driven by pavement characteristics that affect the relative magnitudes of the three transport mechanisms. For example, snow and salt is trapped in the OGFC voids, which removes the salt and snow from the surface where it is available for export off the pavement via wind, vehicle turbulence, or snowplow. The retention of snow and salt in OGFC pavement results in higher salt loads and longer periods of discharge (specifically, longer runoff-event tails). Figure 22 illustrates these differences by showing the timing and magnitude of the Cl load associated with a winter runoff event in WY 2020. Discharge exceeded 0.001 ft³/s at the OGFC station about 2 hours after the HMA station because discharge was attenuated in the porous-pavement layer. Runoff from the HMA site reached the total Cl load around 14:00, but the Cl load in runoff from the OGFC site continued to accumulate and exceed the HMA load as snowmelt from the voids continued to discharge highly concentrated (with respect to Cl) runoff.

Transport of salt from the highway surface by wind, vehicle turbulence, or plow is evident in records of soil conductivity measured at 3, 9, and 15 ft from the edge of the HMA pavement (fig. 23). Increases in soil conductivity (up to about 4.1 decisiemens per meter at 25 degrees Celsius) during wet precipitation and snowmelt indicated that salt-laden water was being mobilized. These spikes in conductivity occurred most frequently at 9 and 15 ft from the pavement, where the bulk of snow was deposited from plow trucks. The mass of salt transported off the highway pavement by wind and plow cannot be determined from the available data. The consistent relation between increases in soil conductivity adjacent to the highway and discharge (due to precipitation and snowmelt) indicates that it may be a substantial component of salt loading that is not accounted for in the measurements of highway runoff from the weir-box outlet alone.

The relative proportion of salt export by wind and vehicle turbulence is also affected by the duration of dry antecedent periods. After shorter dry antecedent periods, during which salt can crystalize on the road surface and become entrained by wind and turbulence, a higher proportion of the salt load is available for export in direct runoff. The average time between runoff events in WY 2020 (88 hours; table 14) was less than the average time between runoff events in WYs 2019 and 2021 (95 and 129 hours, respectively). The combination of the shorter antecedent conditions and the lower snowfall amounts (26.6 in.) in WY 2020 resulted in similarly low-magnitude salt loads at the two pavement study sites (530 and 523 kg at the HMA and OGFC sites, respectively, based on 0.003 ft³/s runoff).

Based on the data available in this study, under the same treatment of deicing materials, salt loads in runoff from the OGFC pavement were about two times greater than loads from the HMA pavement during years with average snowfall amounts but were approximately equal in an atypically mild year. Load estimates were between 1.2 and 2.6 times greater when loads associated with flows less than 0.003 ft³/s also were estimated because winter discharges in this low range are often highly concentrated with respect to salt. Although explanations are offered for the roles that the three salt-transport mechanisms play in the salt-load estimates, it is not possible to quantify the effect of variability in the relative significance of each mechanism from year to year and between the pavement sections.

Concentrations of Nutrients and Total-Recoverable Metals

Although the primary focus of this study was to quantify the sediment loads from HMA and OGFC pavement sections, concentrations and loads of chemical constituents from highway runoff also may adversely affect the quality of receiving waters. Therefore, the concentrations of nutrients and total-recoverable metals were measured over the 3-year study period at each site to estimate loads of these constituents. Concentrations were measured in composite stormwater samples collected from the outlet of the weir box and in samples of sediment retained in the weir sump (fig. 3). The weir-sump sediment samples represent the concentrations of constituents bound to sediment particles, whereas the runoff concentrations represent both dissolved (unfiltered) and particulate-bound fractions.

Composite samples of runoff were collected during 15 storms at the HMA and OGFC monitoring stations to characterize the runoff quality with respect to EMCs of SS, TSS, TDN, PN, total nitrogen, TP, carbon (dissolved and particulate), and total-recoverable metals. The selected events (table 5) represent runoff volumes in the upper 50 percent of the distribution of all runoff-event volumes (fig. 24A) and were collected primarily in the second half of the study period. Although the SS EMCs in the composite samples range from less than the 10th to greater than the 90th percentiles of SS EMCs in all runoff events (fig. 24B), the median and mean EMCs for the 15 storms at HMA and OGFC sites (table 15) were more similar to each other than to those observed in the overall population of SS EMCs. The median SS EMC of all runoff events was 29 mg/L at the HMA site and 15 mg/L at the OGFC site, whereas the medians of the composite runoff samples were 13 and 11 mg/L, respectively. TSS concentrations also reflect similar ranges of values but were typically censored at less than 15 mg/L. Of the HMA pavement results, 47 percent of samples were less than 15 mg/L, and uncensored values ranged from 17 to 128 mg/L. Seventy-three percent of samples from the OGFC pavement were less than 15 mg/L and had uncensored values ranging from 19 to 80 mg/L. Because the composite runoff samples (n=15 storms) represent narrower ranges of concentrations and lower central tendencies than the EMCs of all runoff events, the ranges and magnitudes of nutrient and total-recoverable metal concentrations and loads may be similarly affected.

The concentrations of total-recoverable metals in composite runoff samples varied by an order of magnitude or more for many constituents and were not normally distributed (fig. 25). These distributions are common in highway and urban-runoff data because they are characterized by high variability and often include high-concentration outliers (Smith, 2002; Smith and Granato, 2010; Smith and others, 2018; Granato and Jones, 2019). The mean concentrations at the HMA and OGFC sites differed by less than 30 percent (computed as RPD) for all total-recoverable metals, except for arsenic (40 percent) and zinc (35 percent), although the mean concentrations at the OGFC site were commonly higher (in 6 out of 10 total-recoverable metals) (table 15). Higher concentrations of arsenic and zinc from the OGFC site may be the result of the constituents leaching from the asphalt-rubber binder or the pavement material. Leachate results from this study indicate that concentrations of dissolved arsenic (2.7 and 5.5 µg/L, USGS parameter code 01000) and dissolved zinc (4.5 and 10.0 µg/L, USGS parameter code 01090) leached from samples of HMA and OGFC pavement were 104 and 122 percent higher, respectively, in the OGFC pavement sample (Spaetzel and others, 2023). Other mean concentrations of stormwater composite samples differed by less than 20 percent for barium, chromium, copper, lead, and nickel. Even among the total-recoverable metals that differed by greater than 20 percent, the 95-percent confidence intervals about the means overlap substantially (table 15), which reflects both the high variability of the EMCs and the similarity in EMC magnitudes between the two pavement types.

The concentrations of carbon and nutrients also were similar in samples of runoff from the HMA and OGFC monitoring stations, with means and medians differing by less than 40 percent for most constituents (fig. 26). However, DOC was a notable exception, with distinctly higher concentrations in runoff samples from the OGFC pavement. The mean concentration in composite samples collected at the OGFC site (8.7 mg/L) was approximately two times greater than the mean of DOC concentrations in composite samples collected at the HMA site (4.5 mg/L) (fig. 26). The higher concentrations of DOC at the OGFC site may reflect the dissolution of organic carbon from the asphalt binder in the pavement material or from decomposition of organic matter (leaves and sticks) retained in the OGFC pavement layer. Laboratory studies of leachate from various permeable and open-graded pavement materials indicate that concentrations of DOC are less than 5 mg/L (Kayhanian and others, 2019). In this study, DOC leachate was about 31 percent higher (normalized to sample mass) in the OGFC pavement sample (5.5 mg/L) than the HMA pavement sample (4.2 mg/L) (Spaetzel and others, 2023). In part, this may explain some of the difference in DOC concentration; however, dissolution of organic matter retained in the OGFC pavement may be a greater source of DOC than the pavement itself. Because runoff from OGFC pavement interacts with more sediment and pavement surface area, greater differences in the concentrations of dissolved constituents (such as TDN) may have been expected (Roseen and others, 2012; Moores and others, 2013); however, this effect may have been obscured either by the high variability of runoff EMCs or because the composite samples represented only the upper 50 percent of runoff-event volumes.

The concentrations of TP and total-recoverable metals measured in the annual sediment samples collected from the weir sump were similar at the HMA and OGFC sites. Among all paired concentrations, the average RPD between sites was 38 percent, and although RPDs varied from 0 to 150 percent, concentrations were not consistently higher or lower for a particular site. Concentrations also did not vary substantially year to year, except for TP, which varied by one to two orders of magnitude between WYs 2019 and 2020 (table 16). The low TP concentrations observed in WY 2019 (17 to 37 mg/kg across both sites and particle-size classes; table 16) were consistent with low-end ranges observed in Massachusetts bridge-deck sediment samples (particle sizes greater than 0.0625 mm) (Smith and others, 2018) and may reflect the natural variability of highway-sediment composition with respect to the proportions of organic plant matter, mineral sediment, and pavement material. Concentrations of TP and most of the total-recoverable metals were up to six times greater in the particle-size classes less than 0.25 mm in diameter than in the coarser material (between 1.0 and 2.0 mm in diameter), partly because of the inverse relation between sediment surface area and sediment grain size (Smith and Granato, 2010). Therefore, it was important to account for the effect of grain size on concentration to accurately estimate the loads of these constituents, especially because sediment grain sizes were distributed unequally across particle-size classes in samples from each site (fig. 27; Förstner and Wittmann, 1981; Horowitz and Elrick, 1987; Horowitz, 1991; Smith, 2002, 2005; Breault and others, 2005; Smith and Granato, 2010).

Nutrient and Total-Recoverable Metal Loads

Total loads of TP and metals over the 3-year study period represent the sums of the loads measured in runoff composite samples collected from the weir-box outlet, sediment retained in the weir sump, and sediment retained in the trench (fig. 3). Although five particle-size classes are represented in figure 27, concentrations were measured in samples representing the four particle-size classes in table 16; therefore, the total constituent loads are associated with sediment less than 2.0 mm in diameter. The materials greater than 2.0 mm in diameter primarily consisted of gravels, leaves, and debris. The total constituent loads were between 7 and 64 percent higher at the HMA site for most constituents, except arsenic, cadmium, and zinc (fig. 28). Of these constituents, the load from the OGFC site was higher but only by a maximum of 23 percent (zinc). The loads estimated for the weir and trench sediment reflect greater differences between the two monitoring stations than the total loads, where the HMA constituent loads in the retained sediment were two to five times greater than those at OGFC site. Although similar concentrations were observed in samples of sediment at the two sites (table 16), the distributions of sediment particle sizes were distinct (fig. 27). The sediment masses for the HMA samples were evenly distributed among the four size classes (25 percent in each of four particle-size classes on average). In comparison, OGFC samples had higher proportions of the larger diameter particles (41 to 54 percent of weir-sump sediment particles less than 2.0 mm in diameter were between 1.0 and 2.0 mm in diameter), which resulted in lower loads than those from the HMA site. The difference in particle-size distribution at HMA and OGFC sites is discussed in the “Sediment Loads” section and reflects the retention of sediment in OGFC pavement that diminished over time.

The relative proportions of the loads in runoff varied by constituent and monitoring station from 34 percent (aluminum at the HMA station) to 90 percent (barium at the OGFC station); however, for all constituents, the proportion of runoff load was higher at the OGFC site than at the HMA site (fig. 28A). The average concentrations of TP and total-recoverable metals were no more than 50 percent different between the two sites (although the OGFC load was higher for 7 out of the 11 constituents) in the runoff samples. The total discharge greater than or equal to 0.003 ft³/s over the 3-year study period was 14 percent higher at the OGFC site (39,900 and 45,500 ft³ from the HMA and OGFC sites, respectively). Consequently, the estimated runoff loads are higher despite similar constituent concentrations. As discussed previously, the 15 composite samples for each site represent a lower range of SS EMCs compared to the overall EMC population of events. Therefore, the constituent loads in runoff may be underestimated for the HMA station, based on the concentration data of the runoff composite samples. The sediment load estimated for the HMA site from EMCs of SS measured in all events was 31 percent greater (34 kg) than the estimate of load computed with the average SS EMC of the 15 runoff composite samples (26 kg), whereas the estimates for the OGFC site differed by no more than 2 kg (25 kg based on runoff composite samples and 23 kg based on all SS EMCs).