The survey was conducted in 12 mountainous rivers assigned to three typological groups according to the Polish Water National Authority and the Water Framework Directive (Jusik et al., 2014): Tatra mountain rivers (Biały Dunajec, Dunajec, and Białka – Group 1), mountain flysch rivers (Raba, Brynica, Toszecki Potok, and Nysa Kłodzka – Group 2), and upland carbonate and silicate rivers (Sołokija, Warta, Ropa, Biała, and Odra – Group 3) (Fig. 1), varying in bed modification (incision intensity or redeposition).

The first group comprises rivers located in an alpine granitoid region, characterized by calcareous and silicate bedrock. The second group consists of rivers flowing through much lower mountain ranges (up to the timber zone), where the bedrock contains sandstone rock formations. The third group represents rivers of upland landforms with various carbonate and silicate sediments and rocks. The typology of river channel modification was obtained from field observation and channel measurements (cross-sections, longitudinal profile and cover, height of the floodplain). Narrow channels with downcutting to the floodplain and simplified channel morphology were defined as incised.

All rivers are routinely monitored by the nearest monitoring station of the Environmental Agency (Environmental Agency Data, 2018), and all 12 rivers have consistently been assigned a similar average chemical status in recent years. Analysis of variance (ANOVA) showed no variation between the river groups in incision bed modification (F=1.56, p=0.26) as well as in the following physicochemical properties: dissolved oxygen, conductivity, hardness, pHmax, NH3, NO3-, NO2-, total N, and PO43-. Only water temperature and pHmin significantly depended on the river group. All habitat variables (flow, depth, and substrate type) were significantly dependent on river group (Table 1); meanwhile the incision was not influenced by the parameters variation.

Benthic invertebrate samples were collected in two seasons: autumn (October 2017) and spring (April 2018). No flood waves occurred between these surveys, and the channel morphology remained the same throughout the sampling period. We collected 20 subsamples (1 m2 each subsample) from each low-flow channel along a representative 100 m section of each river according to the sampling procedure for the BMWP_PL index (Bis and Mikulec, 2013). A total of 480 subsamples were taken from a wide range of water depths and flow velocity. Following Jowett et al. (1991) and Muñoz-Mas et al. (2016), the substrate types were converted to a single index by summing the weighted percentages of each type.

Macroinvertebrate samples were collected with a D-frame net according to the Environmental Agency's sampling protocol for biomonitoring assessment using a kicking motion for 3 min across all habitats (Bis and Mikulec, 2013). All collected material was preserved in the field with 4 % formaldehyde. Aquatic macroinvertebrates were separated from the rest of the material in the laboratory using a stereoscopic microscope, and then they were identified to the family level (Tachet et al., 2000), except Oligochaeta, Porifera, and Hydrozoa, which were recorded as such. Due to the varied preferences of macroinvertebrates to habitat conditions, the BMWP_PL index was adopted as the best qualitative index. The Biological Monitoring Working Party (BMWP) is one of the most commonly used biotic indices in various rivers and streams around the world (Roche et al., 2010; Wyżga et al., 2013). It has been adopted in many countries, including Poland (Dz.U. 2019 poz. 2149, 2019). The BMWP index was originally developed to represent water quality, but subsequent studies showed that it reflects ecological quality of the waterbodies and can also be related to hydromorphological impoverishment such as incision or straightening (Mutz et al., 2013; Wyżga et al., 2013; Mikuś et al., 2021). This index best considers the sensitivity of invertebrates to environmental variables because families with similar stress tolerances are grouped together (Armitage et al., 1983).

ANOVA was used to verify the statistical significance of the differences in environmental data between the three river groups (Statsoft, 2013). Non-metric multidimensional scaling (NMDS) was used to test the relationship between the macroinvertebrate taxonomic composition of the assemblages of the 12 rivers assigned to three groups (Group 1, Group 2, and Group 3) and hydromorphological variables (water velocity and depth) during the spring and autumn. Descriptive physical properties (water depth and velocity) were classified into two or three categories: low, medium, and high. We used minimum and maximum values of depth and velocity range in each river group and divided them into 33 percentile ranges of the total value variability. In the case when the ranges were less than 0.5 m depth, we have chosen two groups of 50 percentiles of the depth ranges. The significance of differences between depth and velocity classes was tested by analysis of similarities (ANOSIM) (p values of pairwise comparison with Bonferroni correction) on the Bray–Curtis dissimilarity matrix with 499 permutations of the data. PAST software (version 3.13) was used to analyse NMDS and ANOSIM (Hammer et al., 2001).

To develop habitat suitability functions of macroinvertebrates, reflecting the optimal conditions in the river, generalized additive model (GAM) procedures were chosen. The advantage of the method described by Jowett and Davey (2007) is that it calculates the probability of relations between dependent biotic variables and independent flow parameters. To choose the best-fitting model, we have ranked the available models according to the Akaike information criterion procedure and Δ AICc values, which reflect the difference of AIC between a given model and the lowest AIC. The best-fitting model, describing the relationship between independent variables (depth and velocity and two-way interaction between them) and the macroinvertebrate BMWP_PL index, was the generalized additive model with Poisson error distribution and log link function. We have also measured the accuracy of the GAM procedures (Shearer et al., 2015). The total deviance explained calculated as the relative difference between the residual and the null deviances of the model ([null deviance-residual deviance] / null deviance) was adopted. The course of the regression line of the BMWP-PL and depth and velocity for each group of the bed material rivers was obtained using CurveExpert software, where the best-fitted line for the set of non-linear curves was applied and ranked. The BMWP_PL curve maximum values were regarded as the most optimal for invertebrates and the most preferred. We were interested in calculation of optimal condition for depth and velocity separately to obtain the optimal conditions, allowing us to calculate the discharge which is needed for hydraulic and CCHE2D modelling. The preferred depths and velocities for each season and riverbed material groups were used to calculate the hydraulic discharges which are the most optimal for BMWP_PL variables and recognized as environmental flow.

We used the hydraulic method for the assessment of the environmental flow of each river because of the relationship between the hydraulic parameters of watercourses (depth and velocity) and the quality of the aquatic environment (BMWP_PL–GAM relations). We used rating curves for each river describing the water depth–flow relations to obtain environmental flow for a given optimal depth. Detailed description of the applied hydraulic method of environmental flow calculation is given in Książek et al. (2019). To compare the environmental flow in relation to hydromorphological parameters (incision and redeposition), we used the proportion of environmental flow (Qenv) to mean hydraulic parameters of the minimum discharge: low low flow (LLF – the lowest low flow), mean low flow (MLF – average of the minimum annual flows), and mean annual flow (MAF – average of the annual flows). These metrics show the position of the calculated environmental flow in relation to available water volume (flow characteristics from hydrological year-to-year 1961 to 2017 observations).

We provided the detailed CCHE2D modelling of randomly chosen incised and redeposited rivers using simple randomization procedure based on the single sequence of throwing a dice. The model is a depth-averaged two-dimensional numerical model for simulating unsteady, turbulent, free-surface flow in open channels with a moveable bed. The CCHE2D model solves depth-integrated shallow water equations for all hydraulic calculations (Wu et al., 2000; Duan et al., 2001). The CCHE2D package consists of two modules: a mesh generator (MG) and a graphical user interface (GUI). The main function of the MG is designing a complex mesh system. The mesh is generated based on the surveyed topography and/or a digital terrain model (DTM). The model was applied in two representative rivers, varying in riverbed morphology – from incised bed rock channels to a channel with natural sediment structures (with redeposition). The mesh for each sector of the river was generated by interpolating cross-sections. A total of 5112 observations were used, Raba 3033 (incision) and Ropa 2079 (redeposition). The shape of the channels was fairly regular along the reach under study, and its pattern presented little complexity (i.e. a single channel with no islands), but riffle–pool sequences were observed. The 153–200 m long meshes were composed of cells and nodes (length and number of modes, respectively, for Ropa 153 m and 49 715 and for Raba 200 m and 99 200). Data used for the initial conditions were extracted from field measurements. Special attention was devoted to bed roughness due to its importance for water surface level. Roughness values ranged from 0.01 in hydraulic smooth bed zones to 0.07 in rough areas. Finally, the model time step was defined at 0.1 or 0.25 s, depending on the model structure. The model was calibrated by comparing the measured and computed water surface levels for measured discharges in all cells and nodes (Fig. 2).

In the case of the Raba River, for 70 % of the calculated nodes, the difference between the calculated and measured water surface level (WSL) was in the range ±0.02 m. A total of 84 % of Ropa River nodes were in the range of -0.02 to 0.06 m. In all described models, Δh in the main channel does not cross ±0.02 m, but the visible differences are related to the horizontal layout of WSL in the cross-section. Evaluation of the compatibility measures of the numerical model showed very good agreement (Książek et al., 2010), and the prepared models did not need recalibration.

For each research section, we chose 20 points at each subsampled area differing in water velocity and water depth as the main environmental variables creating habitat heterogeneity for macroinvertebrates. Then, according to the relationship between hydromorphological habitat attributes (water depth and velocity) and the BMWP_PL index values (describing the ecological quality of the river), we constructed a GAM model as the best-fitted method to mark out the range of hydromorphological attributes (where the BMWP_PL suitability index obtained from the GAM model curve is the highest). Based on the optimal depth values, environmental flow was established using rating curves.

Two rivers (located in the same Carpathian region) representing opposite bed modifications (incision and redeposition) were chosen for the model as a case study. The modelled sectors of the river had channels with a pool–riffle sequence and fluvial deposits but varied in terms of degradation of the bed structure. The hydrological characteristics of the modelled river are presented in Fig. 3.

The Raba was selected to represent incised channel rivers (bottom material mainly gravel and small stones, substrate index 14.9). The Dobczyce retention reservoir, which influences the hydrology and morphology of the river, is located upstream of the examined sector of the river (12 km). Construction of the retention reservoir in 1986 led to a significant decline in average annual flow values (MAF values varied from 12.22 m3s-1 in 1951–1985 to 10.57 m3s-1 in 1986–2015, F=49.90, p<0.0001) and broke the continuity of the sediment transport. The reduction in flow, blockade of sediment supply, and longitudinal training work of the Raba led to incision of the riverbed and permanent compactness of the bed material. The Ropa River, chosen to represent the redeposition processes, was located among upland, carbonate, and silicate rivers, with the lowest human impact, i.e. agricultural land. The bottom material consists mainly of gravel and sand (substrate index 7.2), where bedload transport remains undisturbed.

We also wanted to estimate minimum flow values for two rivers which were modelled using CCHE2D. The values of depth and velocity corresponding to the highest BMWP_PL, obtained from the GAM model for each group of river and season, were plotted against the number of pixels having optimal values. Given those calculations, we were able to obtain the weighted usable area of macroinvertebrate communities (WUA) showing the most optimal habitat parameters (GAM depth and GAM velocity). WUA is often defined as an index to various ecological parameters at different organization levels: population (such as biomass, microhabitat area, size classes) (Muñoz-Mas et al., 2016) or other community level (diversity indices or ecological metrics) (Jowett, 1997, 2003; Theodoropoulos et al., 2015; Pander et al., 2019). Each pixel covered 0.25 m2 of total river area, so the numbers of counted cells of the given values of velocity and depth were summarized and multiplied by the surface area. Based on these calculations using the CCHE2D model, we were able to find the relationship between usable area and flow values. To calculate the optimal environmental flow values, the curve between flow and optimal area was created. The low border of optimum of environmental flow was estimated as 50 % of WUA values (Jowett et al., 2008) for CCHE2D modelled rivers.

A hydraulic habitat 2D model of each river section was used for spring and autumn as an example to estimate habitat prediction in terms of calculated environmental flow during the season. Environmental flow that did not meet the conditions of 100 % habitat suitability for macroinvertebrates was expressed as the critical instream environmental flow value (Qenv critical), below which the parameters of aquatic macroinvertebrate communities dramatically declined.

A total of 151 466 individuals belonging to 92 benthic invertebrate families from 480 macroinvertebrate assemblages were identified. High variation was shown in the taxonomic composition of aquatic invertebrates, depending on the hydromorphological parameters (water depth and velocity) and the season (Fig. 4). In the case of rivers classified as Group 1, water velocity was found to significantly affect the taxonomic composition of the macroinvertebrates in both spring and autumn (Table 2).

In spring, there were significant differences between velocity classes (low and high and medium and high), while in autumn, before overwintering, significant differences were only noted for medium and high classes. In neither season were the differences noted in taxonomic composition depending on the range of depth statistically significant in the case of rivers of the second abiotic group (Group 2); more significant differences were observed between velocity and depth classes (three depth classes were adopted due to the greater amplitude of these parameters). In the spring, significant differences were visible in all velocity classes, while in the case of depth they were noted only in the comparison of the low and middle depth classes. In autumn, differences were found for all classes in the case of variation in both velocity and depth. In the case of Group 3 rivers (carbonate and silicate fine sediments and rocks), the velocity parameter only taxonomically differentiated macroinvertebrate communities in the spring between the high and medium velocity classes. In the case of depth, differences were observed in both seasons – in spring between the deepest and shallowest environments and those with medium depth and in autumn only between the deepest and the shallowest zones (Table 2).

Each of the hydromorphological parameters was evaluated by the GAM model, which provided the best fit to the data (Table 3). There were significant effects of depth and velocity and its combination on variation of BMWP_PL index. Generally, the percentage of the total deviance was the highest for the combination of both hydrological parameters; however the depth parameter alone described a similar level of the total deviance. Velocity explained 38.1 % and 44.5 % of the total deviance of BMWP_PL variation in the mountain rivers (Group 1) for spring and autumn respectively. In other river groups the total deviance described for velocity varied between 6 % and 29 %. Bringing into consideration that both hydrological parameters alone described more of the total deviance, we regarded them in further analyses separately. The curves of the generalized additive models for the biotic index BMWP_PL in spring and autumn are presented in Fig. 5.

These models were made for each of the three river groups: calcareous and silica bedrock alpine rivers (Group 1), sandstone mountain rivers (Group 2), and carbonate and silicate upland rivers (Group 3). In the first group, with a gravel bottom, the BMWP_PL index reached its highest values at high water velocity and in shallower zones (by the shores). In the second group of river, the BMWP_PL index was highest at medium velocities in spring and at high velocities in autumn. In both seasons, higher values for the biotic index were associated with shelf environments, as in the case of Group 1. Similar relationships with depth were noted in the Group 3 rivers, where BMWP_PL values were highest in the shallow environments at low velocity in both spring and autumn (Fig. 5).

Using the optimal depth characteristics reflecting the habitat suitability (Fig. 5), the environmental flow based on hydraulic method (rating curve) was defined. The results are shown in Table 4.

There is a high variation of the Qenv, related to its own channel properties and volume of water. To obtain the relation to hydraulic river parameters, the mean Qenv relative similarity to MAF, MLF, and LLF was measured. There was no relation to the abiotic group of river (Table 5). The only significant relation was linked to channel modification (Fig. 6). In all cases, the relative similarity of flow was significantly higher in incised channels than redeposited ones.

In each type of flow (MAF, MLF, and LLF), the relative similarity was higher in incised rivers than redeposited, showing that the incised rivers needed much more volume of water to sustain appropriate conditions for macroinvertebrates compared with the redeposited ones. More detailed analysis and visualization of spatial modelling were predicted by 2D modelling of randomly chosen rivers, presented below as a case study.

We calculated the detailed 2D modelling for two randomly chosen incised and redeposited rivers. According to the GAM macroinvertebrate habitat suitability model, WUA–flow curves were calculated for rivers with varying intensity of bed modification, Raba (incised) and Ropa (redeposited), as shown in Fig. 7.

The environmental flow was defined as the lowest flow corresponding to 50 % of the value of the usable area, which ensures minimum optimal conditions for the development and functioning of aquatic macroinvertebrates (Jowett et al., 2008). Analysis of the curves for the Raba River shows a 50 % reduction in the usable area at the flow of about 10 m3s-1 for both spring and autumn. In the case of the Ropa River, the WUA–flow curves show a 50 % reduction in the usable area at the flow of about 2 m3s-1 in spring and 3 m3s-1 in autumn (Table 6).

A spatial visualization of macroinvertebrate habitat suitability for Qenv optimal conditions is presented in Fig. 8. In the case of the strongly incised Raba River, a very small optimal habitat area was observed, covering only the shelf zone. In the case of the Ropa River, where sediment transportation occurs, the usable areas constitutes more than 20 % of the environmental flow area. The modelling was also used to determine Qenv critical, at which the most valuable areas in terms of habitat (over 80 % suitability) disappear (Fig. 8 and Table 6). Below this Qenv critical value, a dramatic decline in macroinvertebrate diversity should be expected.

A comparison of the Qenv values (optimal and critical) and means, annual flow (MAF) and low flow (MLF), for the two types of rivers is presented in Table 7. In the highly incised river (Raba River), the Qenv optimal requirement for spring was lower but for autumn was higher than MAF, and Qenv critical was always higher than MLF. In the redeposited Ropa River, in spring as well as in the autumn season, Qenv optimal requirements were much lower than MAF, and MLF was higher than Qenv critical. Both findings are congruent with the former hydraulic calculations for all rivers.