Land use

The results for nutrient concentrations according to various land use types are summarised in Table 5. Across the two sampling events, only seven sites were classified as industrial and four as viticulture. These subsets do not allow for statistically viable analyses and are excluded from the table. As the AGE data are concentrated along major rivers, including it here would likely skew the analysis, no longer being just a function of land use. The presented data must thus be treated as unvalidated for the Pedigree calculation, though as mentioned earlier this does not automatically make it poor—however, for any future measurement campaign an independent validation set would be a recommendation.

For four of the five land use types, the median nitrate value was the 2–5 mg/l interval, which roughly corresponds to the “good but not very good” classification from Table 3. The exception is the urban park setting, where 1–2 mg/l represents the median. It is concerning to report that concentrations above 5 mg/l (no longer classified as good) are common throughout all settings, accounting for 35% all collected samples. These high values are less common in urban settings. The significant amount of these values on agricultural land might be a result of chemicals used in the fields. The percentage of concentrations above 5 mg/l has decreased slightly, from 48% in the 2019 measurements to 36% in 2021. This could be the result of Luxembourg becoming the first country in the European Union to ban the use of glyphosate in 2020 (Government GDL, 2020), with a clear link between glyphosate and the evolution of inorganic anions at the highest oxidation states, that is, phosphate and nitrate (e.g., Manassero et al. 2010). It is furthermore concerning to note that forest sites accounted for the highest fraction of high nitrate concentrations. Nitrate content in forests is a complex parameter, dependant on factors such as forest harvesting methods, forest composition, altitude, and environmental factors (e.g., Mupepele and Dormann 2017), and agroforestry is the subject of ongoing studies at the University of Luxembourg. The phosphate concentration had median values in the 0.05–0.1 mg/l interval in the two types of urban settings, and 0.02–0.05 mg/l in other setting. These all fall within the “very good” and “good” categories. High phosphate concentrations were rare (Figure 2), with 11 samples in 2019 and 15 in 2021 in the highest two intervals. These are discussed later in the manuscript.

Types of water bodies

Different types of water bodies are associated with different biodiversity, and play different roles in processes such as the carbon cycle (e.g., Davies et al. 2008; Tranvik et al. 2009). The type of sampled body was therefore one of the parameters required to make up each data point, with the citizen scientists asked to choose from 6 options: stream, river, pond, lake, wetland, and source. With no definition distinguishing a stream from a river, or a pond from a lake, participants individually decided what they are sampling. These divisions thus cannot be considered exact. During the quality control, it was found that samples from similar locations at the same water body sometimes classified it differently. Nonetheless, all water body types were left as the participants defined them, and their summary is presented in Table 6. As with land use data, the results are not validated through an independent data set.

The data sets from both events were merged for this section, as there were no significant statistical differences between the events according to water body type. The observed nitrate concentration in streams and rivers follow similar distributions, with the median interval being 2–5 mg/l for both types. The percentage of measurements above 5 mg/l was 36% for streams, and 42% for rivers. Considering the non-exact definitions of these two water types, this difference is not statistically significant. However, for the phosphate concentration, there is a clear difference, with the median of the stream data being the 0.02–0.05 mg/l interval, and that of rivers significantly higher at 0.1–0.2 mg/l.

For ponds, the median interval for nitrate concentration was the lowest one available, with values not exceeding 0.2 mg/l. This value applied to 32 of the 52 data points. The median for the phosphate concentration was 0.02–0.05 mg/l, also representing very good water quality. As ponds can be highly susceptible to environmental change and urbanisation due to their small volume, making the effects of urbanisation more profound (e.g., Hassall 2014; Beklioglu et al. 2016), these results suggest the overall quality of ponds in Luxembourg is good, at least regarding nutrient content. Concerning algae on the surface, this was reported in 20 cases, giving a frequency of 38%. For streams and rivers, this value was significantly lower at 9%.

The data for lakes are limited, with just 12 data points. This is hardly surprising given that Luxembourg has no natural large lakes. From these points, four were at an artificial lake near Echternach in the east. Three of these recorded the nitrate concentration below 0.2 mg/l, with one reading in the 0.2–0.5 mg/l range. The phosphate readings were all under 0.1 mg/l. Five readings were in the lake upstream from the dam on the Sauer. While the phosphate concentrations were all low, the two readings reported nitrate levels above 5 mg/l. The Sauer is discussed in more detail in a further section. The remaining three lake readings corresponded to one measurement each in a recreational area in Weiswampach in the north, an urban park in Hesperange near Luxembourg City, and a small unnamed water body in the southwest, classified as a lake by a very generous participant. None of these observed values suggested pollution hotpots.

Major river case study 1: Sauer

The Sauer, or Sûre, is one of the major rivers in Luxembourg (Figures 1 and 4). Its source is located in Belgium, about 30 km west of the border village of Martlenage, where it enters Luxembourg (km 0 in this study). Near the village of Bigonville, it transforms into a lake created artificially by the dam near Esch-sur-Sure at km 36. This is the biggest dammed lake in Luxembourg, with a surface area of up to 3.8 km², and it serves as the main drinking water source for the country. It also provides an environment for diverse recreational activities, as well as for hydroelectric power production. Further downstream, the Sauer passes the cities of Ettelbruck and Diekirch (~km 70–80), before reaching the German border at Wallendorf at km 90. It then forms the national border between the two countries, passing through the cities of Echternach and Wasserbillig, where it empties into the Moselle at km 135. The landscape of the Sauer is predominantly rural, mostly characterised by forest and agricultural areas.

The Sauer was well sampled during the Water Blitz events, with 19 samples in 2019 and 15 in 2021. The measured data are shown in Figure 4, using the river course length as the x-axis. The median values for nutrient concentration in 2019 were 2–5 mg/l for nitrates and 0.05–.1 mg/l for phosphates, and in 2021 5–10 mg/l for nitrates and 0.02–0.05 mg/l for phosphates. A direct comparison between the two events is, however, difficult, as in 2019, nine measurements were obtained in the Ettelbruck-Diekirch urban region, compared with just three in 2021. This could explain the drop in the phosphate values, where the higher readings between km 70–80 from 2019 are sparse in 2021. An analysis of the nitrate values presents a completely different picture. Whereas in 2019, the highest concentrations are all in the aforementioned urban region, in 2021 the 5–10 mg/l interval dominates the entire river course, with 9 of the 15 readings.

The Water Blitz results can be compared with official measurements of the AGE, which maintains eight sampling stations along the Sauer (Table 7; Figure 4), where among many other parameters, the nutrient content is measured every month. As discussed earlier, unlike the range estimates in Water Blitz data, these are given as exact values, thus the only tool used to compare the two data sets is visual comparison. While the geographical overlap between the two sets is not perfect, they do provide completely independent validation sets of the same parameters for each other, greatly improving the Pedigree score of both sets. The AGE data clearly show the increasing nutrient content near, and downstream from Ettelbruck. They also confirm the nitrate concentration increase from the September 2019 to May 2022. However, whether this represents a pollution event, possibly from outside of Luxembourg, a seasonal trend, or a weather-related result, cannot be answered with the presently available data.

Major river case study 2: Alzette

Another prominent Luxembourgish river is the Alzette (Figures 1 and 5) with sources in France, a few kilometres from Luxembourg, and entering the country near the city of Esch-sur-Alzette (not to be confused with Esch-sur-Sure) in the south. A high proportion of the landscape it flows through is urban, including industrial regions near Esch, and the urban agglomerations of Luxembourg City, Mersch, and Ettelbruck, where it empties into the Sauer. As a result, many wastewater treatment plants direct their effluents into the Alzette. There are, however, also areas where agriculture dominates the landscape. The course of the Alzette inside Luxembourg is 68 km.

The Alzette was one of the best-sampled water bodies during both Water Blitz events, with 18 samples recorded in 2019, and 26 in 2021 (Figure 5). For the nitrate concentration, in 2019 the median interval was 5–10 mg/l, at which level the water condition can no longer be classed as good, with three samples between Walferdange and Shieren measured in the highest available interval, > 10 mg/l. That these three readings are not separated by any lower measurements strongly suggests the presence of a pollution hotspot. While many factors could have led to these measurements, the presented results certainly justify further investigation. As it is, a few days before the Water Blitz 19, the wastewater treatment plant of Beggen, located between Eich and Walferdange at ~km 35, was subject to a malfunction, with wastewater leaked from the plant to the river. The malfunctioning was only noticed 20 hours later as dead fish were spotted in the Alzette.

As with the Sauer, the Water Blitz results can be compared to official measurements of the AGE, which maintains nine sampling stations along the Alzette (Table 8; Figure 5). Unfortunately, only the 2019 AGE data were available. These show an increase in nitrate concentration around Luxembourg City, with the three stations upstream of it containing around 2 mg/l, increasing downstream to around 5 mg/l. While all measurements fall within the classification of “good,” some of them are nonetheless very close to 5 mg/l, which separated intervals in the Water Blitz. In this case, it would have been nearly impossible for citizen scientists to match the colour correctly, which might explain the high frequency of 5–10 mg/l results. That none of the official stations picked up anomalously high values corresponding to the Beggen spill can be explained by the timing of the measurements (Table 8). It is not possible to attribute the high concentrations to the spill with absolute certainty; they could be the result of other, less significant, events, or even the result of the complex relation between leaching and rainfall (Wang et al. 2015). Furthermore, the signature of the spill would need to be confirmed by measuring other parameters, such as the full spectrum of data as measured by the AGE (Table 1; AGE 2015). Nonetheless, we put forward that this case study particularly strongly demonstrates how citizen science initiatives have the potential to complement official measurements in meaningful ways.

For the data collected during Water Blitz 2021, the median value of nitrate concentrations was the interval 2–5 mg/l. This was also the most commonly observed interval, with 11 from the 26 samples. While eight samples recorded values higher than the 5 mg/l corresponding to the reference threshold for eutrophication, none of the records were in the highest available interval, above 10 mg/l. All the readings above 5 mg/l are downstream from Luxembourg City (~km 30), suggesting that concentrations of nitrate are relatively high in this section. However, several lower measurements also exist, and when these are taken into account, the overall picture is not negative.

For the phosphate concentration, the median interval in the 2019 was 0.2–0.5 mg/l, and in 2021 0.1–0.2 mg/l. The 2019 AGE measurements varied from 0.14 to 0.31, with just two of the nine values classed as good, and the others fitting within the 0.326 mg/l threshold for “moderate.” From the plot, there is no clear trend in either the Water Blitz or the AGE sets, suggesting that if pollution sources exist, they are very localised.

Detecting potential pollution hotspots

The data collected during the Water Blitz events also provide insight into smaller water bodies, which often are not sampled for official statistics, even though they can play an important role in catchment biodiversity. While a solitary high reading in a stream can be explained by measurement errors, especially when lower readings were recorded in the vicinity, multiple high readings do at least invite further investigation. One such stream is the Gander, a tributary of the Moselle in the southeast, partially forming the border between Luxembourg and France. From the 10 readings sampling it, three recorded nitrate concentration in the highest available interval, > 10 mg/l, and five further ones in the 5–10 mg/l interval. These readings were collected in both sampling events, and by several independent participants. Although errors in readings can never be completely ruled out, it is unlikely that different users obtained the same result by mistake. While the records do not conclusively illustrate the state of the Gander, they should be taken seriously, and warrant further investigation.

Another region demanding attention is the Mullerthal, just south of where the Sauer becomes the German border. In an area about 10 km across, four readings of nitrate above 10 mg/l and eight between 5–10 mg/l were observed, with just three readings in other intervals. These high concentrations were measured by different users in several different streams, very strongly suggesting the region should be investigated further.

Very high phosphate concentrations were much rarer than those for nitrate. From the 428 samples across the two sampling events, only five recorded the highest available interval, > 1 mg/l. One of these was in the Gander, where additionally two samples in the 0.5–1 mg/l interval were observed, strengthening the suggestion the stream should be investigated further. The Mullerthal region with high nitrate concentrations did not provide any extreme phosphate results, though 2 of the 15 measurements put it in the 0.5–1 mg/l range. Nonetheless, the nitrate concentrations are likely to be a more serious issue in the area than phosphates. Another high phosphate reading came from the Syre river at Munsbach in the east, though this reading represents an outlier for the river. In fact, another user recorded a reading in the 0.02–0.05 mg/l at the same location two days earlier, suggesting that the anomalously high reading could be an error. Two high phosphate reading were taken in the small stream Tretterbaach in the north by one user. Other readings in the stream are not as high, though one put phosphate concentration in the 0.2–0–5 mg/l interval, where it could be classed as unsatisfactory. While these results are not conclusive, they do warrant investigation.

Scientific community

Citizen science continues to face scepticism is some quarters, with its value sometimes poorly communicated. While not suggesting that the role of citizen science should be downplayed, we follow Buytaert et al. (2014) to suggest that limitations need to be clearly stated, and realistic goals need to be set. These limitations, and a clear representation of uncertainties, along with recommendations about which of them can be addressed and how, are critical when presenting results to policy- and decision-makers. Competing interests and apparent contradictions make environmental decision-making very complex, and different aspects of uncertainty will affect the decisions made. Furthermore, results of projects incorporating citizen science should, possibly in addition to scientific literature, be presented in ways that do not require expert technical knowledge, and data should be freely available (as is the case with Earthwatch and FWW initiatives).

Project coordination

While Earthwatch and FWW projects have successfully engaged volunteers in several countries, it is important to recognise the idiosyncrasy of individual water systems and regions, and accordingly adapt the sampling protocol. To this end, the University of Luxembourg is engaged in identifying relevant parameters to be collected through a citizen science app. When organising projects, it would also be beneficial to liaise with agencies responsible for water monitoring. This would allow determination of key parameters to monitor, identification of potential areas of interest, and decisions on the appropriate timing of the measurements. The parameters and areas of interest would be particularly relevant near possible pollution sources, as this would allow for flagging, or even identifying, potential events with more certainty than was possible for the Beggen plant in 2019. Appropriate timing, to stage the sampling events at times when other, independent datasets are collected, would allow for mutual validation, increasing the Pedigree score of all datasets, and improving the quality of the conducted research. More frequent, regular sampling events would furthermore allow for a study of trends in nutrient concentration. With the AGE data being collected at least quarterly, this would allow a glimpse at each season. While the responses to the participants’ survey were not sufficient for a full analysis, respondents indicated an overwhelming interest to participate in further citizen science events, should these be organised more frequently. The onus on more frequent citizen science campaigns lies with the organisers, not the participants.

Citizen scientists

It really is stating the obvious that there can be no citizen science without the citizens. By their active involvement, the public do, or should, assume a certain level of responsibility, whatever their level of involvement. The organisers have the right to expect that every participant collects data to the best of their ability. From the participation survey, it is clear that public are willing to invest extra time to achieve that.

Official data

Public access to environmental information has been granted by the Aarhus Convention of the United Nations Economic Commission for Europe, applied in the EU in its legislation with the WFD. In Luxembourg, publicly accessible geospatial information shows the results of evaluations of water quality in a map of rivers subject to monitoring. However, a study interviewing officials found some of them voicing fears concerning the publication of these data (Pickar 2021). The interpretation of datasets is often complex even for specialists, and non-specialists may reach wrong conclusions if not informed accordingly—for example, high nitrate concentrations might lead to panic if the consequences are exaggerated. It is therefore critical that published data are accompanied by narrative reports to put data in the relevant context. These reports should also incorporate a discussion of uncertainty. While all AGE numbers were treated as exact measure, with the highest possible score in the “Empirical” column of the pedigree matrix (Table 4), there must be a level of uncertainty in each measurement. These become especially relevant given the quality regimes are separated by thresholds (Table 3). If a nitrate concentration of 11.2 is “good” while 11.3 only “moderate,”’, it needs to be clearly stated how certain it is that a reading falls in the particular regime. The other pertinent column of the pedigree matrix is Validation. Even if a reading scores highly on being an “exact measure” in a “large sample set” collected with “best available practice,” it would benefit from comparisons with independent measurements, which citizen science initiatives provide.

Governance

Governance of natural resources is inherently poly-centric, with citizen science projects clearly demonstrating the shortcomings of the technocratic model of knowledge production. De-centering of the scientific knowledge process is required within the social process of co-creation of actionable knowledge that should form the basis for all actions associated with the adaptive and transformative governance of natural resources, such as fresh water.