Fluoropolymer production is the most important source for many PFAS and thus receives special attention in POPE. The process is based on the most extensively researched fluropolymer production plant of the company Dupont in Parkersburg, West Virginia and the extensive data published by .

This data helps to relate available fluoropolymer production volume V found for different plants in economic datasets to actual emissions of PFAS. This is done with two additional factors: a usage rate and a loss fraction. The usage rates c describes how much of the corresponding PFAS is needed per ton of produced Fluoropolymer V. The loss fraction f describes how much of the used PFAS is expected to be lost in the process and to which environmental compartment (air or water) it is emitted.

Building on this, the expected PFAS emission E [kg] is calculated depending on the medium as 1Emedium=V⋅c⋅fmedium, with the production volume V [kg], loss-fraction f [–] to the corresponding compartment and the usage rate c [–]. With respect to fluoropolymer production, the individual production volumes of sites are based on multiple sources. Several approximated production volumes for different years are taken from , , , and . This allows POPE to consider the production of Polytetrafluoroethylene (PTFE), fluorinated ethylene propylene (FEP), Polyvinylidene fluoride (PVDF), and perfluoroalkoxy alkane (PFA) as point sources. PFAS are typically not used as monomers itself but rather polymerization aids. The loss fraction and usage rate are specific to the PFAS used in the production process and the produced fluropolymer. Ususally both are not fixed and depend on the process used, which is not necessarily publicly known for each plant. , , and provide ranges for the usage rate which POPE takes into account for the upper and lower bound emission scenarios.

For the loss fractions different measurements are available , with the most extensive study published by based on purchasing records. Differences in estimated loss fractions are again used to define higher and lower bound estimates.

Smaller fluoropolymer production plants and other industries are also considered in POPE. Unlike the bigger plants, these are disaggregated via a top down approach. Based on production and emission estimates, PFAS emissions are disaggregated for the following industrial sectors:

Apparel manufacturing

Each sector is of varying importance, specific for each compound. The individual estimated contributions with their upper and lower bounds are found in the are found in the uploaded data (10.5281/zenodo.12783504, ).

The contribution of each sector to the total emissions is based on the estimations by and .

As HFPO-DA and ADONA are mainly used as direct replacements , their totals are related to the substituted volume of processing aid.

This is done by using fixed replacement factors based on , , , and feeding into the upper and lower bound estimates with the arithmetic mean used as the best guess estimate.

It is assumed that the production of related facilities carries on with identical growth factors and loss fractions and with complete substitution with the appropriate replacement. This corresponds to multiplication of the emissions by the replacement factors. The appropriate replacement is chosen based on the associated company. GenX is produced by Chemours (formerly DuPont), whereas ADONA is a 3M/Dyneon product. They can be used as a replacement for PFOA in PTFE production or PFNA in PVDF production.

FOSA is handled differently to all other compounds as no global totals and no production data is available. The total emission is approximated by using measured concentrations and their relation to PFOS as a proxy. This is due to FOSA being a precoursor to PFOS and their uses being very closely related . For this purpose water samples and samples in biota are available (; ). The ratios in biota are corrected by dividing them by the bioaccumulation factors calculated by Soerensen and Faxneld . All measurements are averaged over the respective regions. This gives estimated total FOSA emissions of 2.5 % and 1.1 % of PFOS emissions in Europe and the United States respectively. This would result for example in emissions of 575 kg in Europe or 473 kg in the United States in the best case scenario in 2010.

The second largest contributor to total emissions of PFAS is the release by PFAS containing products during their life cycle . The range of such products, from outdoor jackets and furniture to floor wax and dental floss, is very broad and there is only limited data available on the total usage of these products, their PFAS content, and the associated PFAS loss over time.

The total emissions during lifetime of PFAS containig products in a given area is related to the population of that area (top-down), which in turn gives an estimate of the emissions by product use of one individual. This is eventually used for the local emissions of other areas (bottom-up). Assuming sufficient locality and only slow changes over time, this can be achieved with the total emission load of rivers and the population in their watershed. Similar approaches have been used before, for example by . The relation of the river emission load to a corresponding atmospheric load has to be assessed in a second step afterwards.

The total load of a river is calculated by the river runoff and a reliable PFAS concentration measurement at the estuary. This has for example been done for PFOA and PFOS by as well as for major European rivers. Rivers with known major contributors such as PTFE production sites are ruled out, so that the emissions from these are not taken into account twice. Performing a linear and a log-linear regression between population in the watershed and annual PFOA load leads to a PFOA Emission E [µg d⁻¹], depending on the population p according to 2E=19.2µgperson⋅d⋅p for a linear regression and 3E=5.1µgperson⋅d×10-4⋅p1.2841 for a log-linear regression. Performing the same for PFOS Emissions E [µg d⁻¹] yields 4E=6.4µgperson⋅d⋅p for a linear regression and 5E=3.5µgperson⋅d×10-4⋅p1.0115

The linear regression suggests lower emission values than the exponential regression, therefore POPE uses the linear regression for the lower bound, whereas the log-linear regression is used for the upper bound. As this data is not available for other compounds than PFOA and PFOS, it is assumed that the product use of PFCAs and their replacements scale similarily with poulation as PFOA, whereas PFSAs and their replacement follow the trend of PFOS. In both cases the prescribed total is conserved.

Aqueous Film Forming Foams are relevant for firefighting training and operational use on airports. As airport emissions play a minor role in the total inventory, they are typically considered for global emissions but not spatially resolved. On a regional/local scale on the other hand, airport firefighting emissions might still dominate PFAS emissions . This is mainly due to the fact, that especially in remote regions and low population density areas air transportation of passengers and goods becomes relatively more important, as it is for example the case in Sweden .

The POPE model focuses on bottom-up emission estimations of individual airports. The 60 biggest airports worldwide are handled individually while all other relevant airports are categorized in 5 categories where each category gets assigned a single emission load.

Available Data regarding the PFAS emissions of airports includes measurements at Arlanda Airport, Stockholm, Sweden by , measurements near a military airport near Stockholm by , and the estimated total PFOA emission of airports in Sweden, based on the PFAS-content and their total use of PFAS-based firefighting foam .

The regulations of the International Civil Aviation Organization (ICAO) requires the number of fire-departments, firefighting trucks or firefighters to scale among other factors with the number of flights an airport receives . Therefore the amount of firefighting foam used, and with that, the volume of PFAS emitted, is assumed to scale with the number of firefighters employed using AFFF for their regular exercises. Thus, the PFAS emission of the 60 biggest airports worldwide is scaled in relation to the number of flights per year, relative to Arlanda Airport.

Data on the yearly number of flights is not available for every airport worldwide. For the estimation of emissions of smaller airports, this work makes use of the locations of airports according to the IATA and ICAO Databases and the report by the Nordic Council of Ministers . The ICAO dataset categorizes airports in 3 different size classes. Airports are differentiated as of primarily civilian or primarily military use by designation, since military airports are shown to reach nearly 3 times the emissions of civilian airports of a similar size . All small civilian airports are dropped due to their insignificance. As large civilian and military airports are handled individually, this leaves estimates for medium civilian, small military and medium military airports to be considered. Out of circa 60 000 airports, heliports and landing fields in the data set circa 6500 are included.

Following the medium civilian airport emissions are obtained by subtracting the emissions of Arlanda Airport from the total PFAS emissions attributed to civilian airports in Sweden as estimated by and dividing the remainder by the number of medium airports. The same approach is taken for the military airports based on , where three size classes add an additional degree of freedom. It is assumed, that the fraction of emissions between medium and small military airports is the same as between medium and large military airports. This leads to the approximated emission factors per airport shown in Table .

This allows each individual airport to be assigned an emission based on its size and usage.

The majority of industrial sources in Country Group I such as the historically very important sites in Europe, Japan and the United States are implemented as point sources. Moreover, the most important fluoropolymer production sites in country Group II, primarily located in China, can be modelled as individual point sources. All other industrial emissions are distributed as area sources based on socio-economic data which will be discussed in Sect. .

The locations of industries and manufacturing sites in POPE are a collection of several data sets of individual sources as shown in Sect. . Because of their importance POPE considers 26 fluoropolymer production sites individually. Depending on the time frame, these correspond to up to 90 % of the total fluoropolymer production . These production sites are situated in Europe, the USA, and Japan with newer ones found in east Asia, mainly China.

The emissions of all other industries are spatially distributed top-down based on their relative importance. For this, several datasets are available. In Europe POPE makes use of the European Pollutant Release and Transfer Register (E-PRTR) and the statistique des activités économiques dans la Communauté européenne (NACE) Rev. 2 . For the United States POPE includes the Facility Registry Service (FRS) by the United States Environmental Protection Agency .

The E-PRTR contains 60 000 sites from 65 economic activities. Based on the usage of PFAS as reported by , , , and the corresponding OECD Report .

24 of them where deemed relevant for PFAS emissions.The E-PRTR sites are listed with precise coordinates but include no consistent data regarding their emissions or production volume.

The Nace Rev. 2 dataset complements this, as it provides sector specific industry data based on the Nomenclature des unités territoriales statistiques (NUTS) levels. This data relates to wider regions rather than individual points but includes not only the number of relevant sites, but also size related data such as economic activity and number of employees.

POPE uses a two tracked approach incorporating the E-PRTR and NACE distributing half of the relevant emission fraction based on individual locations an the other half by the total number of employees, assuming a homogeneous distribution inside the corresponding NUTS region.

The FRS covers 95 individual sites in the United States categorised by sector, similar to the E-PRTR. Out of the 95 sectors, 15 were deemed relevant to the emission of the included PFAS based on the known usage of PFAS . Building on the per country estimates by and , emissions are again distributed top-down equally between sites, based on the individual contribution of a given sector.

The spatial distribution of emissions by product disposal in country group I is handled similar to the industrial manufacturing as distribution on a per site basis. In country groups II and III the spatial distribution of disposal emissions is identical to the product use emissions.

Data on disposal sites is available for Europe via the E-PRTR and Nace Rev. 2 and for the United States via a dataset by and the FRS . POPE distinguishes between waste water treatment plants, which emit exclusively to the water, waste incineration sites, which emit to the atmosphere, and landfills, where both compartments are important for transport .

The Nace Rev. 2 and E-PRTR datasets contain individual categories for landfillls and waste water treatment plants and the emissions are distributed accordingly. Nace is again operating on a coarser spatial scale while providing the more extensive dataset, in contrast to the E-PRTR, which contains less sites but with the exact geographical location. Both are weighted equally for the emission distribution.

For the United States POPE includes the corresponding categories in the US EPA FRS as well as an extensive dataset on waste water treatment plants by . While the FRS only considers geographic locations, which are therefore treated identically, the dataset by contains outflows for each waste water treatment plant allowing to scale the emissions by the amount of processed waste water. Again, both datasets are weighted equally for emission distribution.

Area sources are used for the emission distribution for industrial sources for countries in Country Group I which are not yet taken into account, as well as smaller sites in Country Group II. Without further information on the distribution over the corresponding sites, emissions are distributed by population and GDP as a proxy for industrialisation .

The population data set is published by and is available at half degree resolution providing yearly numbers. It combines datasets from ISIMIP Histsoc from 1950 to 1999 and the GPWv4 dataset from 2000 to 2020. The GDP data set of is available for 1990 to 2015 at a resolution of 5 arcmin, but effectively accounts for annual GDP per capita on a state by state level. The GDP is adjusted for inflation and aggregated to the resolution of the population dataset.

The approximated total emissions of the sector Etot are distributed over all corresponding grid cells according to their GDP (GDPi,j) as a fraction of the total accumulated GDP (∑GDPi,j):

6Ei,j=Etot⋅GDPi,j⋅(∑GDPi,j)-1.

This process excludes the territory of the United states and the European Union as these are already taken into account as well as all countries assigned to Country Group III, due to their minimal PFAS production capacities.

The emissions caused by product use are diffuse sources, which are calculated on a per grid cell basis. The grid resolution of 0.5 × 0.5° is based on the input data sets.

To spatially represent the product use related emissions, the datasets for the population and GDP are utilised, again. As the used river concentration data by only considers European rivers, it is expected, that applying this data globally overestimates the product related emissions. Therefore, it is assumed that the use of PFAS-related products scales with GDP per capita. The mean GDP of the European Union for the year 2008, when the data of applies, is used as a baseline. This leads to an emission of PFAS per person and US Dollar economic performance.

Taking the introduced datasets and assumptions into account, the product related emissions of a grid cell Ei,j are computed as 7Ei,j=Pi,j⋅GDPi,jGDP‾Europe⋅f, with the corresponding grid cell population Pi,j [inhabitants], the grid cells Gross-Domestic-Product GDPi,j [USD], the average Gross-Domestic-Product of the European Union GDP‾Europe [USD] and the calculated per person emissions f [µg inhabitant⁻¹ d⁻¹].

As the main sources are industrial sites, there is no expected seasonality in the grand total of emissions. The minimal seasonality of small emission contributors, for example the use and disposal of outdoor clothing, is assumed to be much smaller than industrial trends or seasonal variation due to environmental conditions.

Temporal changes in PFAS emissions from industrial production are primarily influenced by three factors. Firstly, the global increase in PFAS production volume, secondly, the impact of changes to capture and reprocessing technology on the loss fraction; and finally, changes in factory operations including, but not limited to, the impact of national and international regulations. In addition to the overall total, there are discernible regional variations that may align with or oppose the global pattern, thereby resulting in a spatial redistribution.

The main reason for temporal variability in the emission of a single site is changing production volume. Modeling the changes in production volume is challenging, as most companies do not publish exact site specific production volumes.

Contrary to that, published the total amount of used APFO for PTFE production in the Dupont industrial site in Parkersburg West Virginia from 1951 to 2003 based on purchasing records. This is shown in Fig. . This dataset will be used as a basis for the model.

Observable in the timeseries of used APFO in Parkersburg is the exponential increase up to the year 2000. As there is no known change in APFO usage rates this increase is attributed fully to an increase in production volume. An exponential fit to this part of the curve yields a yearly growth factor of 0.074 with R2=0.96.

Based on this example it can be assumed that without regulation or fade out an exponential increase in fluoropolymer production volume is reasonable.

Consequently, an exponential fit of the yearly fluoropolymer production volume is used for the remaining 25 production sites. For 11 of these, several years with approximated production volumes were available so that individual growth factors could be calculated. For the remaining 14 production sites only the production volume of one year is known. In these cases the average growth factor of plants where more data is available within the region was averaged and applied. These growth factors can be found in Table . The corresponding fits are available in the uploaded data (10.5281/zenodo.12783504, ).

For all fluoropolymer production sites, the opening and closing of factories is considered, specifically closings due to country specific regulations and bans. The production volume before the opening of a factory and after closure is set to zero accordingly. This is a simplification as after closure of a production site significant residual emission is expected, but already accounted for beforehand. Depending on the used transport model and circumstances, a retardation of emissions may be needed.

This allows to obtain an approximate production volume V for all included sites over the covered timespan to calculate the emissions according to Eq. ().

As a second source of temporal change, the loss fraction according to Eq. () is included. The changes in loss fraction were subject of the study of . The authors differentiated between five periods with estimated loss fractions to water and air. Since the main concerns for these shifts were of economical nature, as for the early shifts the worrying effects on the environment were not yet known, it was assumed that country specific regulations played a minor role. Therefore no country specific scaling was performed here. The loss fractions were smoothed out over 10 years each to represent the expected gradual transition due to the adaptation of new technology. The temporal development of the loss fraction is depicted in Fig. 

The emissions in a specific scenario s into medium m at time t Es,m,t are therefore approximated by 8Es,m,t=Vref⋅eks⋅(t-tref)⋅cs⋅fm,t, with the growth factor k depending on the scenario s, the reference production Vref in the reference year tref the usage rate cs, depending on the scenario and the loss fraction fm,t at time t to the corresponding medium m.

Ohter industries than fluoropolymer production are handled methodically different in POPE. In this case loss fraction and usage rates are not applicable as neither the produced product quantities, nor the used PFAS mass per product volume are known. The temporal dynamics of emissions is therefore solely determined by the total estimated production volume of a site, assessed according to Sect. . The lower bound scenario assumes only linear growth forward in time from the corresponding reference period, as well as exponential growth backwards in time. The upper bound assumes exponential growth forward in time as well as linear growth backwards in time. The prescribed sector total is conserved in both cases.

Note that regarding the location of industrial sites no temporal change has been applied. The corresponding datasets (US EPA FRS, E-PRTR, and Nace Rev.2) are only available for recent years. It is therefore assumed that outside of fluoropolymer production no spatial shift of emissions over time apart from varying production growth rates and different regulatory impact takes place.

POPE assesses the temporal change in emissions due to product use and product disposal together. It is assumed that the emissions of waste water treatment plants and disposal sites follow the temporal trends of the related product emissions. This implies an included implicit time-lag. As disposed products are expected to slowly leach PFAS, depending on the product and environmental condition, the actual time-lag may be much larger than modeled by POPE. As this is hard to accurately assess, due to the limited data and wide range of products, POPE factors this time-lag into the upper and lower bounds.

The emissions of PFAS related products are changing over time via two different factors. On the one hand shift occurs implicitly with changes of population and GDP via their respective datasets. On the other hand, explicit scaling over time has to be applied to represent the growth in use of PFAS related products. Both factors will be explained subsequently.

As shown in the dataset by the population mostly follows exponential growth, although the growth factors are varying and are generally lower in country group I, and higher in country group II and III. Although the World's population approximately tripled from 1950 to 2020, the population in country group I grew on average about 50 %. Regarding the GDP, provided data from 1990 to 2015. The GDP before and after this time period is extrapolated assuming exponential growth, using the average annual growth over the provided period as annual growth factor. This results in a global real GDP growth which is approximately five fold, with the same distributional trends as the population. The PFAS emissions grow implicitly with population and GDP.

With technological advancement, PFAS containing products grow more prevalent as a share of all consumer expenses. As an approximation, it is assumed that the use of these products grows proportionally with the industrial production volumes. Even though not all industrial production is exclusively for the consumer market, this is a proxy for the prevalence of PFAS containing products and forces zero product related emissions, before their first production.

The estimated production volume of PFOA and PFOS in the year 2009, when the study by was performed, serves as a benchmark production for these compounds. The product use and disposal emissions of a year are scaled by the fraction that the corresponding year's production volume represents of the benchmark production volume. This also implicitly accounts for a part of the emission time lag. This scaling is performed crosswise for the higher and lower bounds of the production volume and the emission per person, respectively.

One additional influence that POPE considers is the conflation of both factors, as for example production increases because there is more economic power demanding PFAS containing products.

This conflation is considered for the upper and lower bounds. Since the majority of the data set is extrapolated into the past, the conflation leads to an underestimation of the total global emissions. Therefore, two different approaches have to be taken for the upper and lower bounds, differing for extrapolation forward in time or backwards in time. Backwards in time the lower bound estimate assumes minimal overlap and neglects this effect, while the higher bound estimate eliminates the scaling with GDP and population by normalizing with the global population and GDP of the current year. This means, that the total emissions grow parallel to the corresponding industrial production volume. Changes in GDP and Population on the other hand redistribute the total emissions, depending if the region underperforms or overperforms compared to the global average growth of population and GDP. As the growth with total production volume far outpaces population and GDP growth, this rarely leads to a decline in local product related emissions over time, but the effect still has to be noted. While extrapolating forwards in time these different approaches for lower bound and higher bound estimate invert, now ignoring the overlap in the higher bound estimate and overcompensating by scaling with the global real GDP and population in the lower bound.

POPE considers two factors for the temporal development of AFFF emissions: increasing prevalence of PFAS as well as supra-national regulations and fadeouts. The latter are handled by country group and comprise regional total bans. The effects of upper limits for PFAS content of firefighting foams and changes in frequency of mandatory firefighting exercises are neglected.

To capture the gradual implementation, increased air travel volumes, as well as the increasing air travel security standards developing over the years, a linear approximation is used from the start date to the previously approximated yearly emission values for the corresponding year. Further than that, there was not sufficient data available to substantiate a modelled decline or increase in use or PFAS-content of firefighting foams and both were therefore set constant. Note that also no opening or closing of airports is implemented, as the expected influence is expected to be minor, especially since even closed airports may still act as significant pollution sources .

Figure shows PFOA concentrations in a selection of rivers including rivers close to former and current fluropolymer-production sites, to capture trends along part of the river's length. Pictured are six rivers, with two of them in the United States, Europe and China, respectively. These are the Rhine in the year 2013, the Elbe in the year 2014, the Cape Fear River in 2013, the Ohio River in the year 2016, the Yangtze in 2003 and the Xiaoquing in 2014 . The rightmost values of each individual river are closest to the sea. Note, that the distance between two measurements may be less than the resolution of the emission-model and therefore, for both the same upper and lower bound may apply. Also note, that the measurements by were mostly under the quantification limit, whereas the measurements of the Xiaoqing by yielded much higher concentrations than the calibration range, leading to potentially higher uncertainty.

POPE and the measurements show similar trends over the considered length of the rivers. Deviations seem to be mostly in form of a consistent bias over the whole river length for all rivers. The expected changes due to an increase in emission load relative to the total watermass follow the expectation. This is very much visible for the Rhine, Elbe, Ohio River and Yangtze. Individual extreme values as for example directly at the plant in the Ohio River, or specific high points at the Elbe or Cape-Fear River are smoothed out due to the nature of the model. Differences in the trend seem comparatively lower to the overall bias. This may suggests large scale distributional errors, affecting the attribution of emissions to whole regions or sectors rather than large local fluctuations due to inaccurately placed point sources or wrong attribution of production volume.

Large relative and the largest absolute deviation for the Xiaoquing River suggest that the modelled emissions in its watershed are underestimated in the POPE emission inventory. In the regions of highest concentration near the plant, the upper bound falls short by an order of magnitude. Downstream the concentration in the model falls to a much lower level than the measurements suggest. One possible explanation may be the influence of atmospheric deposition of emissions to air, which increase the river concentration as a whole and may exend the influence of the production plant downstream. Other explanations may be due to the small length of the river, production variation influencing the single measurement day, or an underestimation of the total emissions in Asia or country group II respectively. Relative overestimation as with the Yangtze river on the other hand suggest again rather distributional errors. Based on the distribution method in China this may hint to the fact, that the relative impact of population as emittent, combined with population and GDP as a proxy underestimates the impact of few high emission point sources. Furthermore, this can be impacted by the assumption that the use of recovery methods during the production process is on the same level as reported for e.g. the United States. This assumption was made due to a supposed strong economic motivation in reusing processing aids. Assuming loss fractions as reported for the United States in the 1950s and 1960s as in Fig.  could increase PTFE-production emissions by a factor of about 10 and therefore explain part of the missing emissions and deviation from observations.

Shown in Table are the RMSNE and the mean difference between measured concentration and the modelled concentration for the best guess estimate. Additionally shown are the Pearson correlation coefficients r to assess how well the model captures the trend along the river, for all rivers where multiple measuring points are available.

The results show varied picture. This is attributed to a multitude of factors, which may be related to the emission inventory but also due to the comparison of a snapshot in time that the river measurements provide to averaged model results, or processes in the river not accounted for, intra-annual variations in production emissions as well as a general environmental retardation. This also becomes visible in the general better agreement for larger rivers for which the relative influence of these variations is smaller. Additionally, there is a better agreement for rivers which are well researched as the Rhine and the Ohio River.

The model tends to overestimate emissions to rivers without major production sites, with positive biases and RMNSEs significantly larger than one, as for example at the Haw River or Daugava. There, the emissions are mostly impacted by the smaller manufacturing industry point sources for which size data is not avaialable but may hint at smaller production volumes and emissions.

On the other hand, PFAS concentrations in rivers which are situated near industries which are not included in the model may deviate significantly. This is for example visible for the Danube, similar to the results of . This is exacerbated by the broad approach of industry sectors where corresponding sectors with high average PFAS emissions like apparel manufacturing may in reality be dominated by a few sites with relevant production of e.g. outdoor apparel, while most sites emit little to no PFAS at all .

The correlation coefficients r of the measured concentrations to the POPE best guess emission estimates measures the agreement in dilution and emission input for both. These are mostly positive in the range of r=0.25 to 0.75. Most of the correlation coefficients show, that the general trend of increasing concentration with increasing emission sources and decreasing concentration with increasing runoff is captured. The lowest value of -0.14 is for the Xiaoquing River. Here not only the absolute value is not captured, but also the trend reverts. This may suggest differing positions of the actual emission input and the modelled input.

The calculated biases are mostly in the order of magnitude of the observed concentrations. They hint at a distribution problem with an underestimation of Chinese single point sources and an overestimation of European and North-American emissions. As POPE is neglecting emission sources, as for example several possible precursors, and showed lower estimates for diffuse emissions than previous emission estimations a negative bias would have been expected. This may mean, that other factors such as loss fractions or the used average emission per person are too high. A fraction of the emissions may also be wrongly distributed, for example from country group II to country group I, which is mainly based on population and GDP Asia might be wrongly attributed to Europe and the USA. Factors like a bias in certain input data, as for example the per person emission, might also cause this trend.

For most environmental concerns the sum of PFAS is the relevant quantity. Figure shows how the sum of PFAS in POPE, transported with the HD Modell, relates to the measured sum of PFAS in the studies from Table .

The shown ranges are spatio-temporal ranges covering all measurements over the whole river length. All model data relates to the best guess scenario and the range over all corresponding gridcells with the median indicated by the black bar.

There is high overlap between the measured ranges and the ranges based on the POPE emission inventory. As the correlation factors indicate measured and calculated values do not always agree in space and time. The high temporal variation of the river loads is not well represented in the model. The shown bias is not identical for all rivers. Concentrations in heavily industrialised river basins are underestimated while concentrations in basins with less industry are overestimated. The general concentration range is captured adequately, although the coarse resolution of POPE tends to smooth local high emissions. At the same time the used gridcell mask is in total bigger than the corresponding rivers leading to the inclusion of a wider range of values. Adding all substances improves the agreement suggesting non-systematic errors canceling each other out. which is why the relative contribution of individual PFAS will be assessed later.

Apart from the total concentration in any given river it is of interest to compare the relative PFAS fractions of the POPE emission inventory with concentrations measurements in rivers. Figure shows this fraction of the total PFAS load averaged over all measurements, or the corresponding gridcells respectively. These fractions consider only substances found in both POPE and the study at hand meaning the shown substances do not necessarily represent the most important contributors to the total PFAS load in each river.

The best agreement is again found for the most extensively studied rivers, the Ohio River and the Rhine and the most extensively studied PFAS of PFOA and PFOS. The Rhine is especially of interest as it is influenced heavily by the replacement of PFOA by HFPO-DA. Even though the absolute PFAS concentrations in the Rhine tend to exceed the upper bound the fraction of HFPO-DA and therefore the replacement factor used in POPE is compatible with the measurements. The relative fraction of PFBA on the other hand may hint at a distribution error between countries. The Danube and Yangtze show large differences especially for PFBA and PFNA. The latter was already observable in the totals and is most likely associated with an inaccurate split between the two associated Fluoropolymers PTFE and PVDF. Similar distributional deviations as for PFBA can also be observed for PFHpA which is assumed to be dominantly used in Asia while measurements show an unknown source near the Danube. The two chinese Rivers Yangtze and Xiaoquing suffer again from the insufficient data on used PFAS mainly due to the simplification that country group II uses mostly legacy PFAS while the measurements show that the usage is broader than anticipated. Additionally, some unknown industries not associated with PFSA usage in POPE seem to have a big influence on the Yangtze basin, which is not considered in POPE.

These gaps pose opportunities for improvement of POPE with more accurate industrial data in country group II.