Increases in Trifluoroacetate Concentrations in Surface Waters over Two Decades

Abstract

Trifluoroacetate (TFA) is a persistent perfluorinated alkanoic acid anion that has many anthropogenic sources, with fluorocarbon refrigerants being a major one. After an initial burst of research in the late 1990s and early 2000s, research on this ubiquitous pollutant declined as atmospheric emissions of the precursor compounds grew rapidly. Thus, there is little contemporaneous information about the concentrations of TFA in the environment and how they have changed over time. This research determined the change in TFA concentrations in streams by resampling a transect that was originally sampled in 1998. The transect was designed to determine the regional distribution of TFA both upwind and downwind of major metropolitan areas in Northern California as well as a set of globally remote sites in Alaska. The results showed that TFA concentrations increased by an average of 6-fold over the intervening 23 years, which resulted in a median concentration of 180 ng/L (range 21.3–2790). The highest concentrations were found in streams immediately downwind of the San Francisco Bay Area, while substantially lower concentrations were found in the upwind, regionally remote, and globally remote sites. The C₃ to C₅ perfluorinated alkanoic acids were also investigated, but they were rarely detected with this methodology.

Short abstract

Concentrations of trifluoroacetate, a “forever chemical”, has increased in the environment approximately 6-fold over the last 23 years.

Introduction

Trifluoroacetate (TFA) is an ultrashort-chain perfluorinated alkanoic acid anion that is exceptionally stable in the environment. Initial research focused on the potential for hydrofluorocarbon (HFC) refrigerants to create TFA in the atmosphere,¹⁻³ but it has since been shown that TFA arises from many additional sources including fluorochemical production facilities,⁴⁻⁶ degradation of fluoropolymers,⁷ inhaled anesthetics,⁸ specific pesticides with a trifluoromethyl group,⁹ certain pharmaceuticals,⁵ and fluorochemical-based fire-fighting foams.¹⁰ TFA produced from HFC degradation in the atmosphere was expected to partition into water, ionize, and wash out of the atmosphere. Since the main precursor compounds were globally distributed due to their long atmospheric lifetimes (e.g., 14.4 years for HFC-134a),² TFA was expected to appear in regions with no local sources of TFA. The concern was that TFA could accumulate in certain terminal waterbodies that lack any significant outflow.^11,12 Since TFA is weakly phytotoxic,^13,14 the concentration of TFA could ultimately achieve toxic concentrations in the future. Furthermore, if problems arise, then the effects could persist for extended time periods due to the stability of TFA.

Attention shifted away from TFA to the longer fluorinated chain acids in the late 1990s and early 2000s when the longer chain perfluorinated compounds, such as perfluorooctanesulfonic acid (PFOS), were discovered to be ubiquitous in the environment. Unlike TFA, these longer perfluorinated acids had both high animal toxicity and the potential to bioaccumulate, so they represented a more immediate risk to both humans and the environment. Research into TFA dwindled with only a few recent ambient measurements being reported,^15,16 while most of the remaining research focused on suspected point sources such as fluorochemical production plants⁴⁻⁶ and fire-fighting training sites.¹⁰

However, the use of first-generation fluorocarbon refrigerants, such as HFC-134a, continued to rise and the atmospheric concentrations climbed along with them. Between 1998 and 2021, the atmospheric concentrations of HFC-134a increased approximately14-fold from 8.7 to 121 ppt at the Mona Loa Observatory, Hawaii,¹⁷ so the amount of TFA formed in the atmosphere would be expected to increase by a similar value. Furthermore, HFC-134a is being phased out and its apparent replacement, namely hydrofluoroolefin-1234yf (HFO-1234yf), has a higher product yield for TFA (near 100%)¹⁸ compared to HFC-134a (approximately 30% depending on pressure and temperature¹⁹), which would also suggest that more TFA will be forming in the atmosphere. HFO-1234yf has a shorter lifetime of 11 days,²⁰ so its transformation to TFA is more likely to be on a regional scale rather than a global scale, which could elevate concentrations close to urban regions.²¹⁻²³

Although the atmospheric production of TFA is expected to be much higher than in the late 1990s and early 2000s, there is a paucity of contemporary TFA data in the literature. Only three studies specifically collected data to determine recent temporal trends. The first sampled five urban lakes in Beijing, China, that were first measured 10 years earlier.²⁴ The results showed a 17-fold increase in TFA concentrations over a 10-year period.²⁴ The second project involved the analysis of rainfall in Germany that showed an approximately 3- to 4-fold increase in concentrations compared to the mid-1990s.¹⁶ The third study was a recent pair of ice cores in remote northern Canada that covered the time frame from early 1977 to 2017.²⁵ Both ice cores clearly showed increasing concentrations of TFA, which supports the expectation that atmospheric generation and deposition of TFA is increasing. These three studies highlight that TFA concentrations are increasing, but there are relatively few studies for a compound that is common and exceptionally persistent.

The objective of this research is to determine the increase in TFA concentrations in surface waters over 23 years. This will be accomplished by resampling a transect originally sampled in 1998.²⁶ The transect sampled small streams that represent local conditions upwind and downwind of major metropolitan areas. The repeat sampling visited the exact same sites in the same month of the year as the original study to be as consistent as possible. Additionally, five streams were sampled in interior Alaska, USA to estimate the increase in TFA concentrations in a region that is representative of global background conditions.

Materials and Methods

Sampling Locations

The sampling locations originally sampled in 1998 were resampled in 2021 when possible. The locations were spread along a transect both upwind and downwind of the major metropolitan areas of the San Francisco Bay Area and Sacramento, CA, to determine the influence of urban areas on the atmospheric deposition of TFA (Figure 1). During the spring, summer, and fall, the predominant surface wind climatology brings oceanic air through the San Francisco Bay Area which then spreads out into the Central Valley.²⁷ The air mass then moves into the Sierra Nevada Mountains during a daytime upslope flow, although this flow reverses during the night. Winter climatology is dominated by stagnation in the Central Valley punctuated by periodic storms that come from the Pacific Ocean. Precipitation and annual average temperature for sites in the region are given in Table S1.

Figure 1.

Location of sampling sites along a transect upwind and downwind of major metropolitan areas in Northern California. The predominant surface wind climatology is marked with the white dashed lines.²⁷ The open symbols represent streams that were originally sampled in 1998 that were dry in 2021. Satellite photo from Google Earth.

The transect was divided into five segments to represent different conditions. The first segment was the “upwind coastal” sites that represented the clean air mass coming from the Pacific Ocean. The second segment was the “east San Francisco Bay Area” sites that were immediately downwind of the Bay Area and were expected to be impacted by the urban area. The third segment was the “Sierra Nevada foothills” that was downwind of both the bay area and the Sacramento metropolitan area. This segment is also downwind of the extensive agricultural region in the Central Valley. The fourth segment was the high “west slope Sierra Nevada” that was largely above the urban pollution of the Central Valley at an elevation of 1600 m or higher. The last transect segment was the “Lake Tahoe Basin”, which was at a similar elevation as the west slope Sierra Nevada sites but may have more impacts due to urbanization around the lake. The original study had a sixth segment on the east side of the Sierra Nevada Mountains, but this was not sampled in 2021 due to recent wildfire damage.

The streams chosen for sampling sites were selected to represent local conditions, so they were small watersheds that originate within the transect segment. The watersheds were also picked so that atmospheric deposition would be the dominant source of TFA. Ideally, the watershed would have no development in it, but this was simply not possible in the east San Francisco Bay Area and the Sierra Nevada foothills segments. In these cases, low-density residential housing was tolerated, but any significant industry eliminated the stream from use. None of the sites were expected to receive agricultural runoff with only one possible exception.

All the sites that were originally sampled in 1998 were revisited in the same month, namely August, as the prior sampling campaign for consistency. Unfortunately, 2021 was a dry year in Northern California (Table S1), and 25% of the sampling sites were dry when they were revisited, so additional nearby sites were sampled when possible to achieve the desired goal of having 10 sites per transect segment. The east San Francisco Bay Area and the high west slope of the Sierra Nevada Mountains were the segments that had the most dried-up sites (Figure 1). The latitude and longitude were recorded for each site (See Supporting Information) so they could be visited again in the future.

The 1998 study also collected samples along a road transect from British Colombia, Canada to Fairbanks, AK, to estimate the northern hemispheric background concentrations of TFA. Instead of a transect, five sites were sampled in triplicate in the interior of Alaska to obtain an estimate of the TFA concentrations in the absence of any regional inputs. These samples should approximate the global background concentrations of TFA in the northern hemisphere.

Sample Processing

Two grab samples were collected at each site with the exception of the Alaska samples that were collected in triplicate. The samples were collected into 120 mL glass jars with aluminum foil-lined polypropylene caps. Prior to sampling, the glass jars and aluminum foil were baked in a furnace at 550 °C for 8 h to destroy any TFA. The two grab samples were processed separately to determine the consistency of the analytical methodology. Field blanks, which were 18 MΩ·cm resistance water, were also prepared, transported to the field, and stored with the samples. The samples were stored in a refrigerator until analysis, which was less than one month after sample collection.

Sample extraction and derivatization protocols were identical to the prior research although the instrument and its conditions were different.^26,28 A table of reagent sources and purities is given in Table S2. Briefly, 15 mL of water was added to a 20 mL headspace GC vial along with 1 mL of 0.5 M ammonium carbonate solution. The sample was evaporated to dryness in a vacuum oven set at 90 °C and 30 kPa. A low purge flow of charcoal-filtered air removed the moisture from the oven. Once the samples were dry, the process was repeated a second time so the total volume of water evaporated was 30 mL. Laboratory blanks consisting of 30 mL of 18 MΩ·cm resistance water were processed along with each batch of samples. Next, 1.5 mL of 9 M sulfuric acid, 0.5 mL of methanol, and 10 μL of a 0.01% (v/v) chloroform-d in methanol solution were added to the vial. This solution converts the acids into methyl esters that are volatile. The chloroform-d was used as an internal standard for the instrumental analysis. The vials were sealed with caps that had baked aluminum foil liners added to them.

Instrumental Analysis Conditions

The analytical instrument was an Agilent 8890 gas chromatograph coupled to a 5977B mass selective detector. The system used a PAL 3 autosampler in headspace mode. The samples were heated at 80 °C in an agitator for 1 h before removing two mL of vapor from the vial and injecting it. The analytical column was an Agilent PoraPLOT Q column (25 m, 0.25 mm i.d., 8 μm film thickness) using hydrogen as the carrier gas. The oven program was: start at 50 °C and hold for 1 min; then increase the temperature at a rate of 10 °C/min to 240 °C; then hold the temperature for 5 min. The mass spectrometer was operated in electron ionization (EI) using selected ion monitoring (SIM) mode. Three ions were monitored for each analyte, namely the methyl esters of trifluoroacetic acid, pentafluoropropanoic acid (PFPrA), heptafluorobutanoic acid (PFBA), and nonafluoropentanoic acid (PFPeA). Data reduction was conducted by Agilent MassHunter Workstation software version 10.1. In one case, both field replicates from one stream returned a high and unexpected result so these samples were reanalyzed using GC-EI-MS in scan mode to characterize the whole spectrum to ensure the identity of the analyte.

The techniques utilized had only been validated for TFA in the original research.²⁸ To ensure the methods could be expanded to the other three short chain fluorinated acids, a spike-recovery trial was conducted using 18 MΩ·cm resistance water where 50 ng of each chemical was added to three 30 mL water samples while three additional samples were processed without the spike. This spiking concentration of 1670 ng/L was chosen approximately 10-fold higher than the expected detection limit for the least sensitive analyte, namely PFPeA.

Statistical comparisons between the 1998 and 2021 data sets were conducted by Minitab (version 12.21) statistical software. The data was not normally distributed, so the nonparametric Mann–Whitney test was used for comparisons. Data was rounded to three significant digits.

Results and Discussion

Quality Control

The 2021 sampling campaign resulted in sampling 45 surface waters in the Northern California transect and five in the interior of Alaska. The consistency between the separate field replicates for the same site was good with an average relative percent difference of 7.2% for TFA. The consistency of the subarctic Alaska samples was also good with a relative standard deviation between the three replicates averaging 4.1%. The California transect field blanks showed no appreciable contamination. The California transect method detection limit (MDL), which was calculated as the mean field blank plus three standard deviations of the field blanks, was 19.5 ng/L, which was less sensitive than the GC-electron capture detector system that was utilized in the previous study. TFA was detected in all 90 samples from the transect. Unfortunately, one of the field blanks in the Alaska campaign showed some contamination, which caused the MDL for that batch of samples to be higher, namely 95.1 ng/L. However, this elevated MDL did not affect the results since all of the samples had detectable concentrations of TFA above this level. The MDLs for PFPrA, PFBA, and PFPeA were 18.3, 32.1, and 129 ng/L, respectively. The results from the spike-recovery trial were good for all compounds with recoveries of 101%, 98.9%, 101%, and 91.2% for TFA, PFPrA, PFBA, and PFPeA, respectively. Therefore, the method can be expanded to these additional acids, although these three compounds were rarely detected in the samples.

Temporal Changes in TFA Concentrations

The main objective of the research was to determine the change in TFA concentrations in surface waters over a 23-year period. The results showed that the average increase in TFA concentrations was 6.0-fold with a standard deviation of 3.8 (raw data are given in Tables S3–S9). This calculation was based on all 33 sites where the exact same site was sampled in 1998 and 2021 regardless of transect segment. The increase in median TFA concentrations between 1998 and 2021 in transect segments ranged from a low of 3.1-fold in the “Sierra Nevada foothills” segment to a high of 10.7-fold in the “east San Francisco Bay Area” segment (Table 1). The high increase in the east San Francisco Bay Area transect segment was largely due to a single stream where both replicates had an abnormally high TFA with an increase of approximately 20-fold. While this one site had abnormally high concentrations of TFA in 2021 relative to the rest of the east San Francisco Bay Area transect segment, it had similar concentrations of TFA as the rest of the transect segment in 1998. This site was downwind of oil refineries and was near Travis Air Force Base, so there may have been a local emission event, most likely Aqueous Film Forming Foam (AFFF) fire-fighting foam released on the military base, between 1998 and 2021 that dramatically increased concentrations. The median concentration in each transect segment in 2021 was statistically higher than the corresponding 1998 data (Mann–Whitney test, p < 0.001).

Table 1. Median and Range of Trifluoroacetate Concentrations (ng/L) in Surface Waters and Increase (x-fold) between 1998 and 2021^a.

	upwind coastal streams	east San Francisco Bay Area	Sierra Nevada foothills	west slope Sierra Nevada	Tahoe Basin	interior Alaska
1998	24.2 (14.8–32.8)	145 (82.9–287)	95.5 (68.5–295)	24.0 (17.8–37.8)	28.0 (19.6–36.8)	21.7 (<9.5–63.0)
2021	183 (23.1–232)	714 (288–2790)	372 (168–1880)	148 (66.1–248)	102 (44.9–202)	134 (101–187)
increaseb	7.2 (n = 7)	10.7 (n = 4)	3.1 (n = 5)	6.9 (n = 6)	4.4 (n = 11)	8.7 (n = 1)

^aAll comparisons between 1998 and 2021 data for each segment were statistically different (Mann–Whitney test, P < 0.001).

^bThe increase for each segment was calculated as the pairwise comparison for sites that were sampled in both 1998 and 2021. The sample size is based on the number of streams with paired data. Note that different values for increase are obtained if the ratio of the median values for all sites is used instead of pairwise comparison, but the results are similar.

The results from interior Alaska showed consistent concentrations with a median value of 134 ng/L (range 101–187). Only one site was resampled, and it revealed an increase of 8.7-fold increase in TFA. If the median values of the Alaska samples in 2021 were compared to the northern samples in 1998, then the increase is 6.2-fold.

There is relatively little data in the literature that monitors TFA concentrations over time. A study of five urban lakes in Beijing, China showed that TFA concentrations increased 17-fold over a 10-year period.²⁴ This increase is clearly greater than the average 6.0-fold observed in this study, but China was undergoing rapid industrial development during that time that might have caused other sources of TFA, such as fluorochemical production plants or increased domestic fluorocarbon consumption, to contribute to the increase. More recent research measured TFA in rainfall in Germany and compared it to historical data from the mid-1990s in the same area.¹⁶ This research showed a 3- to 4-fold increase in TFA concentrations based on the data presented, which is slightly lower than the values observed in the current research. Two recent ice cores in northern Canada have shown increasing TFA concentrations.²⁵ The concentrations increased by nearly an order of magnitude between pre-Montreal Protocol time (1977–1989) and post-2000 time (2000–2014).²⁵ While this time frame does not exactly match the time frame in this research, it shows the general trend of upward TFA concentrations in recent years.

One of the many sources of TFA is the atmospheric oxidation of HFC-134a, so investigating the trend in the atmospheric concentration of this chemical may provide insight into the possible causes for the increase in TFA concentrations. The concentration of HFC-134a increased from 8.7 ppt in 1998 to 121 ppt in 2021 at the Mona Loa Observatory, HI, which represents a 14-fold increase.¹⁷ This increase is greater than the 6-fold observed increase in this study, but this comparison assumes that HFC-134a was the only source of TFA. However, there are many additional sources of TFA, such as pesticides, pharmaceuticals, industrial chemicals and fluorinated fire-fighting foams, that likely elevated the baseline TFA concentrations upon which the HFC-134a contributions were added. Additionally, research conducted in Beijing, China, estimated that only 14% (range 6% to 33%) of the TFA deposition was the result of HFC-134a oxidation.²⁹ This suggests that changes in TFA concentrations over time will not directly match increases in atmospheric concentrations of HFC-134a, which was observed in the current research.

Apart from studies near point sources (e.g., refs (4−6) and (10)), there is little contemporary data (2015 or later) on TFA concentrations in surface waters. A recent, detailed PFAA inventory for Lake Vättern, Sweden showed the input streams ranging from 30 ng/L to 820 ng/L,¹⁵ which is comparable to the values observed in the current study (median of 180 ng/L, range 21.3–2790 ng/L). The lake concentrations were relatively low at 34 ± 5.2 ng/L. Another study of surface water samples in Germany reported that 50% of the samples fell between 480 and 1200 ng/L.³⁰ These results are higher than the current study, but some of the samples, such as the Rhine River, had industrial inputs. The study by Björnsdotter (15) and the current study are apparently the only recent surface water analysis for TFA not associated with known point sources. Since wet deposition is a large TFA contributor to watersheds, it would be expected that local streams lacking point sources would have similar order-of-magnitude concentrations as rainfall, which allows for the comparison to recent precipitation studies. TFA concentrations in German rainfall had a median value of 210 ng/L (n = 976).¹⁶ Similarly, rain samples collected in 2016 in different cities in China had a reported TFA median of 150 ng/L with a range of 8.8–1800 ng/L.³¹ Both of these precipitation studies are in good agreement with the stream samples in Northern California. Two recent ice cores in northern Canada showed TFA concentrations of 40.3 ng/L for the Devon Ice Cap in 2014 and 137 ng/L at Mt. Oxford icefield in 2016.²⁵ The 2016 Mt. Oxford icefield value corresponds well with the remote sites in interior Alaska which had a median concentration of 134 ng/L (range 101–187).

One possible limitation in the current study was that 2021 was a dry year in Northern California (Table S1) which caused several historical sites to be dry and probably reduced the flow in the remaining streams. The lower stream slow may have concentrated the TFA in a lower volume of water, which would result in higher observed concentrations. However, this did not seem to be a consistent pattern in the data. An investigation of the raw data (presented in the Supporting Information) showed that a stream marked as “low flow” or “only a trickle” did not consistently have higher concentrations of TFA compared to the geographically closest site. When 12 “low flow” streams were compared to each stream’s closest site, then six pairs had higher TFA concentrations in the “low flow” creek and six pairs had lower TFA concentrations in the “low flow” streams. Additionally, only two of the transect segments had a “low flow” stream as the highest TFA concentration in the segment while the other three segments had the highest concentrations in other creeks. Overall, it appears that streamflow was not systematically related to TFA concentrations. Furthermore, increased evaporation theory cannot explain the increase in TFA concentrations in interior Alaska, which had significant increases in TFA concentrations between 1998 and 2021 despite being a wetter than average year (Table S1). Therefore, evapoconcentration of TFA may be occurring, but it is not a significant driver of the observed patterns of TFA concentrations in surface waters.

Spatial Distribution of TFA

The distribution of TFA upwind and downwind of major metropolitan areas was also determined (Figure 2). The distribution of TFA followed the same pattern as in 1998. The highest concentrations were immediately downwind of the San Francisco Bay Area. Concentrations were still elevated in the Sierra Nevada foothills, which were downwind of both the San Francisco and Sacramento metropolitan areas, but the San Francisco contribution is likely lower due to dilution of the air mass. The remaining three transect segments, namely the “upwind coastal”, “west slope Sierra Nevada” and “Tahoe Basin” sites had comparable concentrations (Mann–Whitney tests, p > 0.05), which was also observed in 1998.

Figure 2.

TFA concentrations during the two sampling periods and different transect segments. The sample size represents the number of streams sampled with duplicate samples being collected for each stream. A version of this plot with an enlarged 1998 plot is presented in the Supporting Information as Figure S1. Data to generate the 1998 plot was taken from ref (26).

The high concentrations downwind of the San Francisco Bay Area would appear to be the result of primary emissions of TFA or chemicals that rapidly convert to form TFA (e.g., chemicals emitted from the pyrolysis of PTFE). The increase in concentrations downwind of the San Francisco Bay Area was 569 ng/L between 1998 and 2021 while the increase in the least impacted sites in subarctic Alaska and upwind coastal regions were 112 ng/L and 159 ng/L, respectively. Since HFC-134a is globally distributed due to its long atmospheric lifetime, increases in TFA concentrations due to HFC-134a degradation would likely be similar between remote and urban areas. However, the observed data do not support this explanation. Therefore, it appears that additional sources of TFA from urban areas are currently more important than the atmospheric oxidation of HFCs/HFOs, which agrees with prior studies.²⁹ However, this may change in the future as use of short-lived HFOs increases and leads to more impacts near urban areas. Another potential source of TFA is pesticides that are used in the Central Valley. These chemicals may undergo atmospheric transport to the small watersheds before being converted to TFA. While this may be occurring, it is unlikely a dominant driver of the observed TFA concentrations since the concentrations from agriculture would be expected to highest downwind of the Central Valley, which was not observed in this study.

Other Short-Chain Perfluorinated Alkanoic Acids

TFA was the main analyte in this research, but PFPrA, PFBA, and PFPeA were also investigated to both aid in the identification of potential sources of TFA and to obtain baseline concentration data for future studies in the region. However, these acids were rarely detected. PFPrA was detected in two streams in the Sierra Nevada foothills. One of these streams had very high concentrations of 4990 ng/L (average of two replicates), while the other was 84.8 ng/L. The stream with high concentration would appear to have a local source in the watershed, but there was no obvious industrial source. The two samples from this stream were reanalyzed using GC-EI-MS to obtain a full spectrum for the compound, which was confirmed to be the methyl ester of PFPrA. PFBA was not detected in any transect samples. PFPeA was detected in two streams (average concentrations of 204 and 208 ng/L) in the east San Francisco Bay Area streams. Both of these streams were nearby and downwind of oil refineries, which could be a source if fluorocarbon fire-fighting foams were utilized at the installations. One of the sites was also near Travis Air Force base, which might have used fluorochemical fire-fighting materials such as AFFF. PFPeA was also detected a stream in the Sierra Nevada foothills, but it was only detected in one replicate and the value (168 ng/L) was near the detection limit of 129 ng/L, so this value was not as reliable. None of the short-chain PFAAs were detected in the Alaska samples. The pattern of high TFA concentrations relative to the other short-chain PFAAs has been observed in prior research.¹⁵ The reported deposition of PFPrA was recently shown to be about two orders of magnitude lower than TFA. In the five samples where PFPeA was detected, the concentration averaged 10.3-fold lower than TFA.

Implications of Findings

This research along with recent articles suggests that additional attention is needed to monitor the concentration of TFA over time in the environment as fluorocarbon refrigerant usage increases. The few contemporaneous articles providing information on TFA changes over time have all shown large increases in TFA concentrations in the range of 3- to 17-fold, so the literature is in agreement that TFA concentrations are increasing. However, the recent studies have monitored the increase in the deposition of TFA as precipitation^16,25 or surface water concentrations²⁴ where TFA is expected to be washed out to sea in rivers and streams. There has been no recent sampling in susceptible ecosystems and terminal lakes where TFA is expected to accumulate as was done in the work of Wujcik et al.³² and Cahill and Seiber¹² over 20 years ago. However, these are the ecosystems that are the most likely to experience the impacts of TFA.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.2c01826.

• Tables of all of the raw data for each sample, a table of reagent sources and purities, a table of 2021 precipitation and temperature records along with climatic averages, and a larger version of Figure 2 that has the 1998 data enlarged to better show the regional distribution of TFA in that time period (PDF)

The author declares no competing financial interest.