The dataset of caddisfly monitoring in Luxembourg has been separated in three periods: P1 (1961–1968), P2 (1994–2002) and P3 (2007–2020).

The sampling protocol during P1 was only partially described. Hoffmann (1970a) mentioned that all the biotopes likely to shelter Trichoptera were considered. The inventories were mainly based on the capture and identification of adults, although some larvae and aquatic pupae were captured and raised in the laboratory to be identified at the adult stage. Information on the distribution of caddisflies was defined for each species as “wide” or “restricted”, but real abundances values were not available. Sampling places were projected on maps, using a 10-km UTM grid according to the work of Stroot (1984).

During the second period (i.e. P2), a total of 239 sites, distributed all over the country, were sampled twice a year (spring and summer-autumn seasons), in the different microhabitats (riffles, depositional zones, different types of vegetation). Inventories were mainly based on Trichoptera larvae, but were completed by light trap sampling of adults during one season in each site. Caddisflies were identified at a species level and abundances recorded in individuals per m². The sampling protocol was described in Dohet et al. (2002).

During the most recent period (i.e. P3), caddisflies were collected within the scope of the Water Framework Directive (Directive 2000/60/EC). During a first phase (2007–2017), 20 sites were monitored each autumn according to a standardised multi-habitat sampling protocol for benthic invertebrates (AFNOR 2010, 2015, 2016). From 2017, the survey was extended to 120 sites, using a stratified random sampling over different stream types, ecoregions and Natura 2000 delineations (Directive 92/43/EC).

A subset of spring and mainly autumn samples were monitored each year following the same protocol.

A total of 200 caddisflies species have been observed in Luxembourg since the first inventories in the Sixties. After removing species living exclusively in lentic waters (e.g. reservoirs, ponds, wetlands) and species recorded outside the defined periods, a total of 124 caddisfly species was considered in our dataset for subsequent analysis. This represents 12,045 occurrences of caddisflies: 912 in P1, 5056 in P2 and 6077 in P3. Since abundance data and precise locations of the sampling sites were not available for the first period (P1), we used presence/absence data of caddisflies collected along the three periods and the 10-km UTM grid cells as geographic units for analyses.

To classify caddisflies as specialist or generalist species, we selected a wide range of various ecological, biological and life history traits. These ecological preferences and biological characteristics were extracted from the open-access database www.freshwaterecology.info (Graf et al. 2008, 2019; Schmidt-Kloiber and Hering 2015). In this database, the occurrence of different trait modalities was coded as presence/absence, single category assignment or by fuzzy coding using a 10-point assignment system (see www.freshwaterecology.info for more information on the coding system applied). The general hypothesis governing the selection of traits was that an extensive functional space covering a wide range of ecological and biological characteristics will provide a more precise estimation of the specific requirements of the species and will enable the appropriate classification of specialist and generalist species in our dataset. For several ecological parameters, limited information is available. In order to maximise the number of species and the trait coverage across species, we only selected parameters coded for at least 50% of the species of our dataset. Finally, a total of nine parameters including 51 trait modalities was considered: stream zonation preference; hydrologic preference; microhabitat/substrate preference; feeding type; habitat specialist; temperature range preference; climate change vulnerability score; and duration of emergence period/reproductive life cycles per year. Detailed information on the assignment system and the method used to calculate single scores for each ecological/biological parameter are given in Table 1.

To address the potential impact of changing climate, several parameters enabling us to detect and evaluate temperature increases as induced by climate change, were selected. The parameter “temperature range preference” allows us to separate, amongst specialist taxa, the cold-adapted caddisfly species (i.e. cold-stenotherms) from one hand and the warm-adapted species (i.e. warm-stenotherms) from the other hand. In our proposed method, a score of “10” is given to species coded as cold-stenotherm or warm-stenotherm in the www.freshwaterecology.info database. Moreover, the parameter “climate change vulnerability score”, calculated from six autecological traits that are known to be associated with vulnerability to climate change (i.e. endemism, micro-endemism, temperature preference, altitudinal preference, stream zonation preference and life history (Hershkovitz et al. 2015), is added to calculate the score for each caddisfly species (Table 1).

Finally, a mean score is calculated for each species from the nine parameters selected (Tables 2, 3 and 4). The distribution of the mean-scores’ values are calculated separately for cold-adapted and warm-adapted species. Only the species showing a score higher than the 75 percentile of each distribution values are selected as specialist species, either cold-adapted or warm-adapted. The distribution of the mean-scores’ values calculated for all species of our dataset enables to select generalist species having mean scores lower than the 25 percentile of the “all species” distribution values.

Changes in the composition of caddisfly communities amongst the three periods (P1, P2 and P3) were assessed using the temporal beta‐diversity index (TBI) approach developed by Legendre (2019). The TBI is a dissimilarity index measuring the change in community composition amongst several geographic units within the same region, i.e. measuring the change in beta-diversity, between two separate surveys over time. We compared the three periods two by two, resulting in three analyses. As data from P1 were at the resolution of the 10-km UTM grid and consisted of presence-absence only, we used the UTM grid cells as geographic units (Fig. 1) and the Jaccard index to compute the TBI. The Jaccard index is an occurrence-based (i.e. binary) coefficient using only the double-presences in the calculation of dissimilarity, i.e. its value does not change when double-zeroes are added to the data (Legendre and De Cáceres 2013). This index, therefore, considers only the species present at least in one of the two periods.

We measured the dissimilarity in species composition using the TBI, based on the Jaccard index in each grid cell, between the first (i.e. P1 or P2) and the second (i.e. P2 or P3) periods. We then decomposed each TBI index into two finer indices: one for species losses (B) and one for species gains (C). B > C or B < C means that the grid cell has lost or gained species between the two periods, respectively. Grid cells with equal amounts of losses and gains would have B equal to C. We used the B and C indices to produce B-C plots, with B (losses per grid cell) in the abscissa and C (gains per grid cell) in the ordinate, as described in Legendre (2019). These plots display the relative importance of species losses and gains across the grid cells, where the green line indicates the theoretical positions of grid cells with equal amounts of losses and gains and the parallel red line indicates the centroid of all grid cells. Therefore, a red line below the green line indicates a dominance of species losses across the grid cells, while a red line above the green line indicates a dominance of species gains. Distinctive symbols are used for the grid cells dominated by gains (squares) and by losses (circles) and the symbol diameter represents the TBI value. To have an overall indication of the direction of change over all grid cells, we tested the difference between species gains and losses using a permutation paired t-test with 9999 random permutations. All calculations and plots were made using the “TBI” and “plot.TBI” functions available in the R package adespatial (Dray et al. 2019).

These analyses were performed for the three caddisfly ecological groups (i.e. the cold-adapted species, the warm-adapted species and the generalist species).

The proposed method to identify specialist and generalist species and to distinguish, amongst specialists, the species which are cold-adapted from those that are warm-adapted, resulted in a classification of 32 species (36% of the species in this dataset). Split up in distinct categories, 9.5% of species were classified as cold-adapted species, 7.5% as warm-adapted and finally 19% as generalist species (Tables 2, 3 and 4).

Values assigned to the nine selected parameters are summarised in Table 1 for cold-adapted species, in Table 2 for warm-adapted species and in Table 3 for generalist species. For cold (Table 2) and warm-adapted species (Table 3), only species with respectively “Therm – = 10” and “Therm + = 10” are classified. Only those with “TM-Mean” higher than the 75^th percentile of their value’s distributions are indicated (5.71 for cold-adapted species and 5.29 for warm-adapted species, respectively). For generalist species (Table 4), all species having a “TM-Mean” value lower than 3.50 of the all-species value’s distribution (25^th percentile) are indicated.

The results from the TBI analyses comparing the three periods two by two are summarised in Table 5 and are illustrated in the B-C plots in Figs 2, 3 and 4 for cold-adapted species, warm-adapted species and generalist species, respectively.

For each period comparison, changes in communities of cold-adapted species tend to be dominated by species losses, although none of the paired t-tests shows a significant change (p > 0.05; Table 5). Nevertheless, this tendency to species loss is higher when comparing P1 with P2 (Fig. 2a) and P1 with P3 (Fig. 2b) than when comparing the two last periods (Fig. 2c), as indicated by the red line being further below the green line on the B-C plots.

Concerning the changes in communities of warm-adapted species (Fig. 3), although not significant (p > 0.05), the tendency is opposite when comparing the historical period (P1) to the recent periods P2 (Fig. 3a) and P3 (Fig. 3b), with grid cells generally dominated by species gains. On the contrary, the most recent changes (i.e. between P2 and P3) were more dominated by species losses as for cold-adapted communities (Fig. 3c)), although this change is not significant (p > 0.05).

Changes in the community composition of generalist species follow the same tendencies as changes in warm-adapted communities, but with significant trends (Table 5). Between P1 and P2, changes are significantly dominated by species gains (Fig. 4a), whereas they are significantly dominated by species losses between P2 and P3 (Fig. 4c).

The maps of the Fig. 5 illustrate the geographical distribution over the three periods of an example of a cold-adapted caddisfly species, Lithax niger (Fig. 5a), a warm-adapted species, Leptocerus interruptus (Fig. 5b) and a generalist species, Allogamus auricollis (Fig. 5c).

The relatively large number of species listed as specialists (17%, summing cold and warm-adapted species) confirms the high potential of caddisflies as a species-rich and ecologically diverse insect order to reflect the intensity of different stressors on aquatic ecosystems (Hering et al. 2009; Conti et al. 2014).

In each category, several species are nowadays considered as very or extremely rare. We can particularly mention Ernodes articularis and Adicella filicornis (only observed during the Sixties and now probably extinct in the country), Lithax niger, Plectrocnemia brevis and Rhyacophila philopotamoides amongst the cold-adapted species; Agapetus laniger (only observed during the Sixties and probably now extinct in the country), Hydroptila angulata, Hydroptila lotensis, Orthotrichia costalis, Oxyethira flavicornis, Ceraclea fulva and Leptocerus interruptus amongst the warm-adapted species. Even amongst generalists, numerous species show moderate to strong decrease during the recent years, like Brachycentrus montanus, Hydropsyche botosaneanui, Hydropsyche dinarica, Annitella obscurata or Tinodes assimilis. Some other generalist taxa have always been very rare in the area, for example, Hydropsyche bulgaromanorum, Hydropsyche exocellata, Orthotrichia angustella, Limnephilus ignavus, Limnephilus sparsus, Stenophylax mitis or Stenophylax mucronatus. Parachiona picicornis is probably extinct. We should, however, mention that several species enumerated here as extremely rare or extinct species nowadays are also characteristic of very specific habitats (e.g. springs, swampy springs) or, can hardly be identified at larval stages (e.g. Hydroptilidae). Therefore, they are likely to be absent from usual monitoring lists if these specific habitats are not sufficiently monitored or if adults or aquatic pupae are not found in benthic samples.

In this study, we analysed the long-term temporal changes of caddisfly communities in Luxembourg. We compared the changes that occurred in specific groups of species (i.e. specialist and generalist species) that are known to react differently to environmental stressors (Hering et al. 2009; Conti et al. 2014).

As expected, the overall change in species composition between periods for cold-adapted species was dominated by a loss of species (Table 5, Fig. 2). This general trend was more preeminent when comparing the historical and contemporary data than when comparing the two most recent periods. Numerous studies reported clear benthic invertebrate’s declines, especially when specialist or cold-adapted species were considered (Durance and Ormerod 2007; Domisch et al. 2011, 2013; Conti et al. 2014; Dohet et al. 2015; Haase et al. 2019). A large-scale loss of the potential distribution ranges of the entire cold-stenothermic aquatic insect assemblages has been predicted by the end of this century (Sauer et al. 2011).

Even though, long-term datasets covering more recent periods have also related stabilisation of cold-adapted species richness in montane and subalpine/alpine streams (Gebert et al. 2022). These results align with those indicating that freshwater insects, contrary to terrestrial insects, have frequently been reported increasing in abundance within these later years (van Klink et al. 2020). This current state of increasing taxonomic richness has been interpreted as a transient condition that could eventually result in the local extinction of cold-adapted species and the prevalence of more generalist warm-adapted and eurytherm species, which are better adapted to the new climatic conditions (Steinbauer et al. 2018; Haase et al. 2019; Gebert et al. 2022).

In the present dataset, the temporal trends of changes within warm-adapted species mirrored those of the generalist species (Table 5, Figs 3 and 4). Between historic period and contemporary periods, changes in warm-adapted and generalist communities were dominated by species gains, whereas they were dominated by species losses between the two contemporary periods. The concomitant decline of cold-adapted species and increase in species adapted to higher temperatures and generalist species suggests the replacement of local species by other, more widespread species (Clavel et al. 2011; Haase et al. 2019). This process can potentially lead to novel assemblages which could affect community dynamics, through, for example, biotic and functional homogenisation of communities and ultimately biodiversity maintenance and ecosystem functioning (Clavel et al. 2011; Gebert et al. 2022). However, during the last decades (comparison of P2 and P3: Figs 3c and 4c), changes in the community composition of generalist species and warm-adapted species were dominated by species losses. This result suggests that affected insect groups not only include specialists that occupy particular ecological niches, but also common and generalist species (Sánchez-Bayo and Wickhuys 2019). In contrast, on a comparable timespan and in the same region (central Europe), Haase et al. (2019) reported profound changes in stream invertebrate communities. A warming of 0.5 °C over the past 25 years was associated with significant increases in total abundance and richness of warm-adapted and eurytherm species (Haase et al. 2019). This pattern was interpreted as an increase in primary productivity induced by slight increases in temperature in these streams, which provides additional food resources and niche space for invaders, enabling a higher abundance and richness of consumer species (Haase et al. 2019). The decline of cold-adapted species and increase in species preferring higher temperatures suggests a similar underlying process as the one found by Steinbauer et al. (2018) with repeated plant surveys from mountain summits across Europe. Shifts in relative abundance were considered as early warning signals of upcoming local extinctions because competitive replacement of resident species required that colonisers build up sufficiently large populations. Local extinctions should, hence, follow colonisation with a time-lag (Steinbauer et al. 2018).

Long time-series datasets, like the one available in Luxembourg for caddisflies, are paramount to predict future trends and to better understand trajectories and the degree of change of biodiversity (Larsen et al. 2018). Though, our dataset was subject to several constraints. In particular, the lack of abundance data for the historical period limits our interpretation of the comparison of species losses and gains within the three periods analysed in this study. Species’ abundance is potentially more sensitive to global change than a binary presence/absence change and understanding the causes of changes in species’ abundances is crucial to assess consequences for ecosystem functioning and extinction risk (Bowler et al. 2017). Furthermore, the sampling methods and study designs changed between the three study periods, which could have influenced our results. Particularly, the lack of homogeneity in sampling seasonality and spatial coverage between periods does not allow us to draw definitive conclusions on the reliable absence of either species in each grid cell and period. Indeed, the absence of a species could result from an insufficient sampling effort in a particular region or stream type and/or because the survey effort was restricted to a particular season which did not enable the specific identification of the larval stages of this species. Finally, if specialist and generalist benthic invertebrates can be regarded as reliable integrative indicators of water temperature fluctuations, overriding effects of different other stressors (e.g. land use, eutrophication, siltation, toxic substances, invasive species) should not be neglected when interpreting long-time series dataset.

To remediate to these weaknesses, our caddisfly dataset will be completed with a new three-year sampling cycle (2021–2023) in the near future. These complementary data will partly solve the discrepancies in sampling coverage and seasons between the two more recent study periods. When accounted for the differences in sampling procedures and design, the abundance of caddisflies will be used for a better understanding of the temporal trends of changes within communities between the more contemporary periods. Finally, temporal beta-diversity analysis will be tested to the subgroup of sites corresponding to the different stream types present in Luxembourg (Ferréol et al. 2005). We postulate that the evaluation of species losses and gains amongst stream types along a sixty years’ timespan will enable us to investigate the influence of the upstream–downstream gradient in structuring communities. Elevational range shifts may be especially important for stream invertebrates because streams and their resident communities follow a natural elevation gradient from source to mouth and stream organisms like caddisflies, have higher dispersal capacities within compared to between river networks (Haase et al. 2019).