Surveys were made during low tides from August to October 2009, on days without rainfall. On each shore, two stretches of 10 m were surveyed. In each stretch, a 10-m chain was deployed along eight transects running sea parallel, haphazardly spread at different distances from low-water mark along the whole shore. In each transect, the micro-habitats present at 21 points spaced at 50-cm intervals along the chain were recorded, by assigning one of the following categories to each point: open-rock, rockpool, channel, crevice, or overhang ( Figure 1C ). The percentage frequency of occurrence was calculated for each category on each transect. The mean percentages of micro-habitat occurrence over all transects sampled on the mid-shore of the same stretch were then assessed as a measure of landscape variability. From the whole set of rockpools intersected by transects, a minimum of 25 rockpools per stretch were randomly selected and individually marked in the field for sampling (numbered removable epoxy tag on adjacent rock). We sampled 323 rockpools (Oliveirinha, 52; Queimado, 51; Almograve, 57; Cabo Sardão, 56; Alteirinhos, 53; Monte Clérigo, 54).

Each rockpool was treated as a system of three concentric micro-habitats ( Figure 1D ): i) inside-rockpool (POOL); ii) edge of rockpool (EDGE); and iii) open-rock surrounding the edge (NEAR). POOL was sampled using two equal-sized quadrats deployed on the bottom of each rockpool, encompassing inside-pool surfaces located at least 2cm away from the pool waterline. POOL quadrat-size varied with rockpool size; so that a minimum of 50% of the total surface area of rockpool-bottom was covered by the two sampling quadrats. For each rockpool, the size of the POOL quadrat was selected from eight differently sized quadrats, made from ~1.3-cm grid of plastic wire mesh, with the following areas (respective quadrat-size in number of grid units): 40 cm² (5×5), 161 cm² (10×10), 361 cm² (15×15), 640 cm² (20×20), 1,005 cm² (25×25), 1,444 cm² (30×30), 2,581 cm² (40×40), and 4,032 cm² (50×50). EDGE was defined as the thin rim comprising a 2-cm wide band surrounding the pool waterline, including a 1-cm wide marginal inside-pool surface and a 1-cm wide marginal outside-pool surface. EDGE was sampled with strings of 25×2 cm length deployed along the pool waterline, using enough strings to cover a minimum of 50% of the perimeter of each rockpool. NEAR was defined as the adjacent outside-pool area of open-rock, bounded by a halo-line positioned at approximately 2 cm away from the pool waterline and extended to a constant width of 19 cm. NEAR was sampled by two quadrats of 361 cm² (19×19 cm) randomly deployed within open-rock surfaces surrounding each rockpool. All quadrats and strings were made of flexible material to fit the topography of sampled surfaces. For each variable assessed within POOL or NEAR, a replicate consisted of the mean obtained from the two sampling quadrats of the same micro-habitat of every rockpool system. For variables assessed within EDGE, a replicate consisted of the sum obtained from all strings sampled along the edge of every rockpool. Additionally, mid-shore surfaces of open-rock located at least 25 cm away from any rockpool (FAR) were sampled by six replicate quadrats of 361 cm² per stretch. For comparison with FAR replicates, a subset of six NEAR replicates per stretch, all located around mid-shore rockpools, was selected during surveys (hereafter “MidNEAR replicates”).

Species identification, allocation to size classes (maximum shell length—MSL, measured with calipers), and counts were made for all limpets within the four micro-habitats (POOL, EDGE, NEAR, and FAR). Five size classes were considered for both Patella spp. (MSL, ≤ 1 cm; 1–2 cm, 2–3 cm, 3–4 cm, and ≥ 4 cm) and S. pectinata (MSL, ≤ 0.5 cm, 0.5–1 cm, 1–2 cm, 2–3 cm, and 3–4 cm). Every individual limpet within the sampling areas was recorded for the respective species-size class, micro-habitat, and individual rockpool system. There were two exceptions: a) juvenile limpets (MSL ≤ 1 cm for patellids and MSL ≤ 0.5 cm for S. pectinata, see Seabra et al., 2020) within POOL were sub-sampled using two quadrats of 40 cm²; and b) limpets with MSL ≤ 0.5 cm were not sampled within EDGE due to the transitional nature of this micro-habitat. All limpet counts were then standardized to densities in 50×50 cm.

Within both POOL and NEAR of every rockpool system, we assessed the following habitat-composition variables (details in Supplementary Table S1 ): i) the percentage cover of 15 space-occupying categories (rock, sand, Lichinaceae, Verrucariaceae, cyanophytes, crustose non-coralline algae, CCA, articulated coralline algae, seaweed, Porifera, sea anemones, barnacles, mussels, sea urchins, and other sessile invertebrates), i.e., substratum types and functional groups of sessile organisms and ii) counts of two groups of non-limpet mobile grazer gastropods, i.e., trochids (including Steromphala umbilicalis, S. pennanti, Phorcus sauciatus, and P. lineatus) and littorinids (Melarhaphe neritoides), both standardized to densities in 50×50 cm. All taxa were visually identified in the field to the lowest possible taxonomic resolution and then lumped into functional groups of sessile or mobile organisms.

A set of 19 physical variables was assessed for each of the 323 sampled rockpools, in the field or through image analysis (details in Supplementary Table S1 ). Each variable corresponds to a quantitative or qualitative metric associated with size (perimeter, surface area, and volume), shape (circularity, roundness, aspect ratio, and curved or straight edges), topography (maximum and mean depth, POOL and NEAR slope, POOL and NEAR topographic complexity, “confinement”), or position (straight and contoured distance to the nearest rockpool, shore height, distance to low-water mark, and “barriers”) of rockpools and/or surrounding rock. “Confinement” was the height of the rocks surrounding the pool (1, pool flushes with surrounding rock; 2, low walls, <3 cm; 3, medium-height walls, 3–6 cm; 4, high walls, 6–10 cm; 5, very-high walls, >10 cm and often with obtuse angles). “Barriers” was presence or absence of outcrops of rock to seaward at <1 m from the pool. Although a few variables applied only to a single micro-habitat (volume, maximum, and mean depth: POOL-only; slope and topographic complexity: separately assessed inside pools and on surrounding rock), most were physical descriptors of each rockpool system across POOL, EDGE, and NEAR (shape and position variables applying to the three micro-habitats; perimeter and surface area of a rockpool are proportional to the ones of its surrounds).

Two-way non-parametric multivariate analyses of variance (PERMANOVA, Anderson, 2001) were made separately for each of the three species of limpets (P. ulyssiponensis, P. depressa, and S. pectinata), within each of the three micro-habitats (POOL, EDGE, and NEAR), to test for differences in limpet populations among the six shores and among the two stretches within each shore (shore—fixed factor with six levels; stretch—random factor with two levels, nested in shore; sample size varied between 25 and 30 rockpools per stretch).

To assess intra-specific connectivity, correlations of the total density of the same species between pairs of micro-habitats (POOL and EDGE; EDGE and NEAR; and POOL and NEAR) were made separately for the three species (P. ulyssiponensis, P. depressa, and S. pectinata) using Spearman's coefficient (n = 323 rockpool systems).

Three-way PERMANOVAs were performed separately for each of two species (P. depressa and S. pectinata) to test for differences in limpet populations living on mid-shore open-rock between areas close to and further away from pools, and among shores and stretches (proximity to rockpools—fixed factor with two levels: MidNEAR and FAR; Shore—fixed factor with six levels; Stretch—random factor with two levels, nested in Shore; n = 6). In the case of P. depressa, for which a significant interaction was found between Proximity and Shore, non-metric multidimensional scaling (MDS) was then done separately for each level of factor Proximity (MidNEAR, FAR), to visualize the distance among stretches of all shores and correlation vectors of the response variables (size-class densities of P. depressa). In duplicate MDS plots, vectors of predictors of landscape variability were overlaid (i.e., mean percentages of occurrence of five micro-habitat categories within stretches of every shore).

PERMANOVA tests employed permutation of residuals under a reduced model using 999 permutations and type III (partial) sum of squares. Where differences were detected by PERMANOVA, pair-wise tests determined which levels of each factor differed. The similarity percentages breakdown (SIMPER) procedure (Clarke, 1993) was applied to identify which response variables were the major contributors to the differences between groups detected by pair-wise tests (size classes most responsible for multivariate distances between significantly different factor levels).

Binomial tests (Sokal and Rohlf, 1995) assessed if upward extension occurred in three species of limpets (P. ulyssiponensis, P. depressa, and S. pectinata) inside rockpools (POOL), compared to outside rockpools (NEAR). Data were assembled separately for each species, through the following steps: i) every replicate where the presence of a given species was found was selected from the 323 replicates of both POOL and NEAR; ii) for each of the two micro-habitats, the shore-height records of all selected replicates were considered; iii) from these records, the median, mean, maximum, and minimum values of shore height within each micro-habitat (four pairs of variables per species) were calculated for each sampled stretch of coast (12 stretches, corresponding to two stretches within each of six shores). The general null hypothesis that the median, mean, highest, or lowest shore height of a limpet species occurrence is similar between the two micro-habitats was analyzed by individual binomial tests. In each binomial test, 12 POOL and 12 NEAR values were compared.

Distance-based linear models (DistLM) were used to examine the relationship between the small-scale variability of individual limpet species inside pools, at the edge of pools or in the surrounding open-rock (size-class densities of a target species within a micro-habitat) and several predictors assessed for each rockpool system (physical, habitat-composition, and connectivity variables). Predictive variables were initially assembled for each target species within each micro-habitat (full list of predictive variables used to build the model for each response matrix in Supplementary Table S2 ). The total density of every other co-occurring species of limpets were included as follows: i) habitat-composition variables if co-existing with the target species in the same micro-habitat or ii) connectivity variables if estimated in an adjacent micro-habitat. Connectivity variables were coded (e.g., PU_adjacent_Pool) by the abbreviation of species name (P. ulyssiponensis—PU; P. depressa—PD; S. pectinata—SP) followed by “_adjacent_” and the micro-habitat code (Pool; Edge; Near). In the two cases where significant differences were detected among shores by the previously described two-way PERMANOVA tests (Section 2.4.1), specifically for P. depressa within NEAR and for S. pectinata within EDGE, the factor shore was included as an additional predictive variable for DistLM.

After preliminary procedures of elimination and transformation of predictive variables, their selection was made with BEST procedure and based on AIC criterion (methodological details given in Supplementary Text ). After obtaining the “BEST final model” for each response matrix, we further selected the “top predictors” as those that contributed most to explain total variation and for which the inclusion in this model added a minimum of 1% in R². Finally, distance-based redundancy analysis (db-RDA) plots were produced to identify the response variables (size classes) that best described the variability of each target species within each micro-habitat, and their association with the “top predictors.” Correlation vectors of response variables and “top predictors” were overlaid in duplicate db-RDA plots for comparison.

A second DistLM analysis was ran for each species within each micro-habitat, with the “top-predictors” of the “BEST final model” (obtained by the first DistLM analyses) classified in three indicator groups of predictors: physical, habitat-composition, and connectivity variables. We assessed the contribution of each of these groups to the overall explanation of final models.

On every shore, there was a consistent pattern of decreasing abundance of this species from pools to further away: the total mean density of P. ulyssiponensis was 10 times higher in POOL compared to EDGE, four times higher in EDGE compared to NEAR, and null within FAR ( Figure 2 ). Consistently on all shores, nearly 90% of the total mean density within pools corresponded to PU ≤ 1 cm ( Figure 2 , POOL). Within the surrounding micro-habitats, the modal size class was PU 1–2 cm on most shores ( Figure 2 , EDGE and NEAR). This species was present on mid-shore open-rock areas near pools on five shores in very low abundances (the highest total mean density of 3 per 50 × 50 cm was recorded in Queimado) with variable size structure among shores ( Figure 2 , MidNEAR).

Population size structure and density within pools did not differ among shores or stretches within each shore ( Table 2 , PU within POOL). Within the surrounding micro-habitats, differences were only found between stretches ( Table 2 , PU within EDGE and NEAR). Correlations of the total density of P. ulyssiponensis between adjacent micro-habitats were all positively significant between POOL and EDGE (r_s = 0.495, p<0.001), between EDGE and NEAR (r_s = 0.423, p<0.001), and between POOL and NEAR (r_s = 0.136, p=0.014).

On every shore, the abundance of this species was consistently higher inside rockpools compared to the surrounding micro-habitats: the total mean density of P. depressa in POOL was three times higher than in EDGE and five times higher than in NEAR ( Figure 3 ). Inside pools and consistently on all shores, more than 90% of the total mean density corresponded to juveniles (PD ≤ 1cm) ( Figure 3 , POOL). The modal size class along pool edges was PD 2–3 cm on all shores except Cabo Sardão, where the size structure was dominated by PD 1–2 cm ( Figure 3 , EDGE). On the open-rock surfaces, the modal size class was either PD 1–2 cm or PD 2–3 cm, the former consistently on the two steeper shores (Almograve and Cabo Sardão) and the latter consistently on the three flatter shores ( Figure 3 , NEAR, MidNEAR, and FAR).

Population size structure and density inside and around the edge of pools did not differ among shores but were significantly different among stretches within shores ( Table 2 , PD within POOL and EDGE). On the open-rock near pools, there were significant differences in population size structure and density among shores and among stretches; pair-wise tests detected differences between Cabo Sardão and three other shores (Oliveirinha, Queimado, and Alteirinhos), and between the two stretches of Alteirinhos and Monte Clérigo ( Table 2 , PD within NEAR). Differences among shores were mostly explained by higher densities of the two smallest size classes (mainly PD 1–2 but also PD ≤ 1) in Cabo Sardão than in the other shores (SIMPER, Figure 3 ). Cabo Sardão was the shore where the highest abundance was recorded for this species within NEAR (total mean density of 52 limpets per 50×50 cm) ( Figure 3 ). Correlations of the total density of P. depressa between adjacent micro-habitats were positively significant between POOL and EDGE (r_s = 0.610, p<0.001) and between EDGE and NEAR (r_s = 0.120, p=0.031), but non-significant between POOL and NEAR (r_s = 0.076, p=0.173).

Regarding population size structure and density of P. depressa on mid-shore open-rock surfaces at different proximity to rockpools, a significant interaction was found between factors Proximity and Shore, and significant differences were found among stretches at Oliveirinha and Alteirinhos ( Table 3 ). Pair-wise tests on Proximity detected no differences between mid-shores surfaces close to and far from pools on any shore ( Table 3 , MidNEAR = FAR). Pair-wise tests on Shore revealed a different among-shore pattern for the two categories of Proximity. For mid-shores surfaces near to pools, the most evident pattern of variation was between Cabo Sardão and all shores other than Almograve ( Table 3 , MidNEAR); this was due to a higher density of PD 1–2 cm and PD ≤ 1 cm in Cabo Sardão (SIMPER, Figures 3 , 5A ); Cabo Sardão was the shore with the lowest relative proportion of open-rock ( Figure 5B ) (20% for the average of the two stretches, Supplementary Figure S3 ). For mid-shores surfaces far from pools, significant differences were found between the three flatter shores and the group of two steeper and one intermediate shores ( Table 3 , FAR); this reflected a higher density of PD 1–2 cm and PD ≤ 1 cm in the group formed by steeper and intermediate shores (Almograve, Cabo Sardão, and Alteirinhos) (SIMPER, Figures 3 , 5C ); these shores had more crevices and less open-rock ( Figure 5D ), their relative proportions of crevices (33%) and open-rock (23%) being four times higher and two times lower on average compared to the ones recorded in flatter shores (8% and 49%, respectively) ( Supplementary Figure S3 ).

This species ( Supplementary Figure S1 ) did not occur at Queimado, and only one individual was recorded at Oliveirinha. It was consistently found in very low densities across the other four shores—the highest total mean density of three limpets per 50×50 cm was recorded inside pools in Alteirinhos. The presence of P. vulgata was most consistently found: (i) along pool edges, compared to the other micro-habitats and (ii) in Cabo Sardão, compared to the other shores. Inside pools (POOL), P. vulgata individuals were either juveniles (PV ≤ 1cm) or small-sized (PV 1–2 cm). Within the surrounding micro-habitats (EDGE and NEAR), the commonest size classes were PV 1–2 cm and PV 2–3 cm. Within open-rock far from pools (FAR), only PV ≤ 1cm or PV 2–3 cm were recorded.

This species ( Supplementary Figure S2 ) was absent from POOL and EDGE and did not occur at Oliveirinha or Queimado, being consistently present on four shores within NEAR in very low densities—the highest total mean density of four limpets per 50×50 cm was recorded in Almograve—and variable size-structure—the modal size class was either PR 1–2 cm or PR 2–3 cm, depending on the shore. Only a few sampled individuals occurred on the mid-shore open-rock of the two steeper shores: Almograve (both within MidNEAR and FAR) and Cabo Sardão (only within FAR).

On every shore, this species was consistently more abundant inside rockpools compared to the surrounding micro-habitats: on average, the total mean density of S. pectinata in POOL was eight times higher than in EDGE and 19 times higher than in NEAR ( Figure 4 ). Inside pools, the size structure of S. pectinata was dominated by juveniles (SP ≤ 0.5cm) on all shores (74% on average), although a comparatively lower proportion of juveniles (54%) and exceptionally higher densities of the three larger size classes were found in Monte Clérigo ( Figure 4 ). Within pool surrounding areas, the modal size class depended on the shore, being either SP 0.5–1 cm, SP 1–2 cm, or equal proportions of these two size classes ( Figure 4 , EDGE, NEAR, and MidNEAR). Away from pools, the modal size class was SP 0.5–1 cm on all shores ( Figure 4 , FAR). Monte Clérigo had the highest abundance of S. pectinata within all micro-habitats (POOL, EDGE, and NEAR) and mid-shore categories (MidNEAR and FAR) ( Figure 4 ). However, no significant differences were found among shores or stretches within each shore in population size structure and density within pools ( Table 2 , SP within POOL). Within pool edges, differences in population size structure and density were significant among shores and not among stretches, with Monte Clérigo differing significantly from Almograve, Cabo Sardão, and Alteirinhos ( Table 2 , SP within EDGE). Differences between Monte Clérigo and the other shores were mostly explained by higher densities of SP 1–2 cm and SP 2–3 cm within rockpool edges of Monte Clérigo (SIMPER, Figure 4 ). Moreover, the highest total mean density within EDGE (15 individuals per 50×50 cm) was at Monte Clérigo ( Figure 4 ). On open-rock near pools, only differences among stretches were significant ( Table 2 , SP within NEAR). Correlations of the total density of S. pectinata between adjacent micro-habitats were positively significant between POOL and EDGE (r_s = 0.455, p<0.001) and between EDGE and NEAR (r_s = 0.214, p<0.001) but non-significant between POOL and NEAR (r_s = 0.066, p=0.236). Population size structure and density of S. pectinata on mid-shore open-rock did not differ with respect to proximity to rockpools, or among shores or stretches ( Table 3 ).

The highest abundance inside rockpools was consistently recorded at the second lowest interval of shore height (1.5–2 m above C.D.), decreasing progressively at higher shore levels. This was observed on all shores except Monte Clérigo, where a slightly higher abundance was found on the upper mid shore (2–2.5 m) ( Figure 6 , POOL). The highest abundance of this species in the open-rock surrounding rockpools was always recorded at the lowest shore level available on each of the six shores, declining upshore to complete absence (in Oliveirinha and Queimado), or sharply on the other four shores ( Figure 6 , NEAR). A similar pattern was found for rockpool edges, but with a less-abrupt decrease in abundance with increasing shore height ( Figure 6 , EDGE).

Within every micro-habitat, this species was generally most abundant at mid-shore levels; the exception was Monte Clérigo, where abundances of P. depressa within all the three micro-habitats were similar across all three shore levels ( Figure 6 ). On the other five shores within pools and their edges, very low densities or complete absence were found at the lowest shore level, with most individuals occurring from 1.5 to 3 m above C.D. at Oliveirinha and Queimado and higher than 2 m above C.D. at Almograve, Cabo Sardão, and Alteirinhos ( Figure 6 , POOL and EDGE). Outside rockpools, P. depressa was almost absent for the two higher shore levels at Oliveirinha and Queimado, while it was present across all shore levels at Almograve, Cabo Sardão, and Alteirinhos; the highest abundances were found at the 1.5–2 m level on three shores (Oliveirinha, Queimado, and Alteirinhos) and from 2 to 3 m above C.D. at Almograve and Cabo Sardão ( Figure 6 , NEAR).

There were no clear vertical patterns in this low-density range-edge species; where present, it occurred mostly on pool edges at various heights on different shores ( Figure 6 ).

This species was most abundant at the highest shore levels, being usually found on the steeper shores (Almograve and Cabo Sardão) ( Figure 6 ).

The vertical distribution patterns of this species were the most variable among shores; the highest abundances within pools were found on the mid-shore and/or upper shore ( Figure 6 , POOL). On all shores except Monte Clérigo, a consistent absence or low abundance both inside and in the edge of pools was found at the two lower shore levels ( Figure 6 , POOL and EDGE). Inside pools, it was found at high shore levels, being most abundant at 2.5–3 m at Oliveirinha, Almograve, Cabo Sardão, and Alteirinhos; the distribution inside pools at Queimado was vertically discontinuous as mostly restricted to two separate levels (2–2.5 m and >3 m), with maximum abundance at the highest level ( Figure 6 , POOL). On pool edges, it was most abundant at mid-tidal heights in Oliveirinha and Queimado (1.5–3 m above C.D.), while only being found at higher levels (>2.5 m above C.D.) in Almograve, Cabo Sardão, and Alteirinhos ( Figure 6 , EDGE). Outside rockpools, it was mostly found on the lower shore levels in Oliveirinha and Queimado, and at mid-tidal levels (from 2 to 3 m above C.D.) at Almograve, Cabo Sardão, and Alteirinhos ( Figure 6 , NEAR). In Monte Clérigo, the highest abundances within both pools and pool edges were found at the highest level, whereas the highest abundance on open-rock was at the lowest level ( Figure 6 ).

Our hypothesis, that inside rockpools (POOL) each of the most common species (P. ulyssiponensis, P. depressa, and S. pectinata) would occur higher up the shore than on open-rock (NEAR), was generally supported ( Figure 7 ; Table 4 ). This pattern was consistent across most of the 12 sampled coastal stretches, both when considering extreme (minimum and maximum) or central-location (median and mean) values of shore height ( Table 4 ; Figures 7A ). There was an exception: the lowest shore height where the presence of P. ulyssiponensis was recorded in each stretch did not differ between pools and open-rock ( Table 4 ). Compared to open-rock, the median shore height inside rockpools was extended upwards in 0.6 m for P. ulyssiponensis, 0.2 m for P. depressa, and 0.7 m for S. pectinata ( Figures 7B ).

The models retrieved by DistLM analyses for P. ulyssiponensis inside rockpools (PU POOL) and on the open-rock surrounding rockpools (PU NEAR) explained identical proportions of the variability among replicates (44%), higher than the model regarding this species in the edge of rockpools (PU EDGE, 34%) ( Table 5 ).

In both PU POOL and PU NEAR models, habitat-composition variables were the most important group of “top-predictors” for structuring P. ulyssiponensis size-class densities (40% and 39%, respectively, Table 5 ). Two habitat-composition variables estimated as percentage cover were selected in common for these models ( Table 5 ): articulated coralline algae (the first selected variable in both models, contributed alone to explain 13% and 30% of the variation inside and outside rockpools, respectively) and CCA (explained 5% and 2% of the variation inside and outside rockpools, respectively). Both were positively associated with overall abundance of this species within both micro-habitats: articulated coralline algae was mostly related with density of P. ulyssiponensis juveniles (PU ≤ 1cm) inside rockpools and with densities of P. ulyssiponensis adults (all size classes except PU ≤ 1 cm) outside rockpools; whereas CCA was mostly related with P. ulyssiponensis adults inside and outside rockpools ( Figures 8A, C ). Other habitat-composition predictors were selected exclusively for the PU POOL model: the cover of sand (also including detritus and cobbles in minor proportions, Supplementary Table S2 ) was negatively related with densities of P. ulyssiponensis adults, whereas both the cover of mussels and sea urchins were positively related with densities of P. ulyssiponensis, particularly juveniles ( Figure 8A ). Both the cover of sea anemones and barnacles were selected exclusively for the PU NEAR model ( Table 5 ), both positively related with density of P. ulyssiponensis juveniles outside rockpools and responsible for the segregation of a few samples (squares in the bottom part of the db-RDA plot, Figure 8C ).

The group of physical variables was also important for PU POOL and PU NEAR (adding another 5% of explanation to the models), while connectivity variables were not selected for these models ( Table 5 ). Shore height was selected as a “top predictor” for both models ( Table 5 ), with a general negative association found between this predictor and P. ulyssiponensis abundance inside and outside rockpools, particularly negatively related with density of juveniles inside rockpools and with densities of adults outside rockpools ( Figures 8A, C ). Outside rockpools, distance to low-water mark was also negatively related with densities of P. ulyssiponensis adults ( Figure 8C ).

In the PU EDGE model, the most prominent predictor was a connectivity variable, which alone explained 22% of the model variation: the total density of P. depressa on the surrounding open-rock (PD_adjacent_Near) ( Table 5 ). This variable was positively associated with densities of P. ulyssiponensis of the smallest size classes (PU 0.5–1 cm, PU 1–2 cm) around rockpool edges ( Figure 8 , middle). The other variables responsible for the spatial pattern of P. ulyssiponensis at rockpool edges were distance to low-water mark (higher shore heights had lower overall abundance of P. ulyssiponensis), shore height, and the abundance of the congener P. depressa (PD) (both negatively associated with the abundance of larger P. ulyssiponensis) ( Table 5 ; Figure 8B ).

Both DistLM models produced for P. depressa inside rockpools (PD POOL) and on the open-rock surrounding rockpools (PD NEAR) explained 47% of variation among rockpool systems, while the model regarding the EDGE explained 26% of the variability ( Table 5 ).

The group of habitat-composition predictors was the most important for explaining P. depressa size-class densities on both PD POOL and PD NEAR models (43% and 25%, respectively); the cover of CCA and mussels contributed to more than 20% of the variation inside and outside rockpools respectively ( Table 5 ). These predictors greatly influenced spatial structuring within each micro-habitat (samples along the horizontal axis db-RDA1), with CCA positively associated with overall abundance of P. depressa inside rockpools and mussels positively associated with overall abundance of P. depressa outside rockpools ( Figures 9A, C ). The other habitat-composition variables that also explained variation inside rockpools were cover of sea urchins—negative relationship with juveniles (PD ≤ 1cm); total density of P. ulyssiponensis within the same micro-habitat (PU) and cover of articulated coralline algae—negative relationships with the largest size class (PD 4–5 cm); total density of S. pectinata within the same micro-habitat (SP)—positive relation with P. depressa adults; and cover of sand—negative relation with P. depressa adults ( Table 5 ; Figure 9A ). Outside rockpools, besides the above-mentioned positive association with mussels, Verrucariaceae cover was negatively associated with density of small-sized P. depressa (PD 1–2 cm), and the total density of P. ulyssiponensis within the same microhabitat (PU) was negatively associated with density of large-sized P. depressa (PD 2-3 cm) ( Figure 9C ).

The group of physical predictors was also important for the spatial patterns of both PD POOL and PD NEAR (adding 4% and 5% of explanation to these models, respectively). Distance to low-water mark (included on both models) and/or shore height (included on PD NEAR model) were the physical variables most important for explaining the variation in P. depressa among samples within each micro-habitat ( Table 5 ). Inside rockpools, distance to the low-water mark was positively associated with densities of mid- and large-sized P. depressa (PD 2–3 cm, and PD 3–4cm) ( Figure 9 , top). Outside rockpools, shore height was negatively associated with density of PD 3–4 cm ( Figure 9C ). An additional physical variable, the slope of the rockpool bottom, was negatively associated with PD ≤ 1cm inside rockpools ( Figure 9A ).

While no connectivity variable was selected for the PD POOL model ( Table 5 ), the total density of P. ulyssiponensis in rockpool edges (PU_adjacent_Edge) was selected for the PD NEAR model, being positively associated with density of PD 1–2 cm outside rockpools ( Figure 9 , bottom). Furthermore, shore was included as a factor in the PD NEAR model (contributing to 12% of total explanation, Table 5 ), with the following positive associations: i) Cabo Sardão and Almograve with densities of the smallest size classes (PD ≤ 1 cm and PD 1–2 cm); ii) Oliveirinha and Queimado with density of the largest size class (PD 4–5 cm); and iii) Alteirinhos and Monte Clérigo with densities of mid and large-sized P. depressa (PD 2–3 cm and PD 3–4 cm) ( Figure 9C ).

For P. depressa present around rockpool edges (PD EDGE), the structuring among samples was mainly driven by association with two connectivity variables acting in opposite directions (accounting for 12% of explanation, Table 5 ): a negative relationship with the total density of P. ulyssiponensis on the surrounding open-rock (PU_adjacent_Near) and a positive relationship with the total density of P. ulyssiponensis inside rockpools (PU_adjacent_Pool) ( Figure 9B ). The second group of predictors selected for the PD EDGE model was the one of physical variables, namely, distance to low-water mark (positive relation with overall abundance of P. depressa), shore height, and circularity (negative association with abundance of P. depressa in general) and confinement ( Figure 9B ). It is worth noting the negative relationship between very-high confinement (category 5, the most-recessed pools) and density of large-sized P. depressa (PD 3–4 cm) ( Figure 9 , middle). Finally, a habitat-composition variable was also selected for this PD EDGE model: abundance of P. ulyssiponensis within the same micro-habitat (PU)— negatively related with large-sized P. depressa ( Table 5 ; Figure 9B ).

The DistLM models for S. pectinata inside rockpools (SP POOL) and in the edge of rockpools (SP EDGE) explained similar proportions of the variability (26% and 22%, respectively), higher than the model for this species on the open-rock surrounding rockpools (SP NEAR—15%).

In SP POOL and SP NEAR models, habitat-composition predictors were most important for the spatial patterns of S. pectinata (17% and 11%, respectively) ( Table 5 ). The first selected predictor in the SP POOL model was the total density of P. depressa in the same micro-habitat (PD) ( Table 5 ), for which a positive relation with density of S. pectinata juveniles (SP ≤ 0.5cm) was suggested ( Figure 10A ). The other habitat-composition predictor included in the SP POOL model was the cover of articulated coralline algae ( Table 5 ), for which a negative relationship was suggested with the density of the middle size class (SP 1–2 cm) ( Figure 10A ). For the SP NEAR model, three different habitat-composition predictors were included: cover of crustose non-coralline algae was positively related with densities of small-sized S. pectinata (SP ≤ 0.5 cm and SP 0.5–1 cm); the variable “Other sessile invertebrates” was negatively associated with densities of larger S. pectinata (SP 1–2 cm and SP 2–3 cm); and the total density of P. depressa in the same micro-habitat (PD) was positively associated with densities of small-sized S. pectinata (SP ≤ 0.5 cm and SP 0.5–1 cm) ( Table 5 ; Figure 10C ).

The group of physical predictors also contributed to the SP POOL and SP NEAR models (adding another 9% and 4% of explanation, respectively). Common to both models was the selection of shore height, which was positively associated with S. pectinata abundance inside rockpools, namely, with juveniles within POOL (SP ≤ 0.5cm), and negatively associated with S. pectinata abundance outside rockpools, namely, with mid- and large-sized S. pectinata within NEAR (SP 1–2 cm and SP 2–3 cm) ( Table 5 ; Figures 10A, C ). The other physical predictors selected in SP POOL model were the following: confinement—particularly the negative relationship between high confinement and the largest size classes (SP 2–3 cm and SP 3–4 cm); distance to low-water mark—positive association with these largest size classes; and roundness—negative relation with juveniles (SP ≤ 0.5cm) ( Table 5 ; Figure 10A ). The other physical variable selected in SP NEAR model was Slope NEAR, showing a negative relationship with mid- and large-sized SP (SP 1–2 cm and SP 2–3 cm) ( Table 5 ; Figure 10C ).

In both the SP POOL and SP NEAR models, no connectivity variables were selected ( Table 5 ). In contrast, three connectivity variables were selected for the SP EDGE model (responsible for 6% of total explanation): P. ulyssiponensis within open-rock surfaces surrounding rockpools (PU_adjacent_Near)—negative association with mid-sized SP (SP 1–2 cm); P. depressa within open-rock surfaces surrounding rockpools (PD_adjacent_Near)—negative association with small-sized S. pectinata (SP 0.5–1 cm); and P. depressa inside rockpools (PD_adjacent_Pool)—positive association with small-sized S. pectinata ( Table 5 ; Figure 10B ). However, in the SP EDGE model, the variable that stood out was confinement, a physical predictor that explained 11% of the variation ( Table 5 ), the most evident pattern being the positive associations between low confinement (category 2) and small-sized S. pectinata (SP 0.5–1 cm) and between very-low confinement (category 1) and the largest size classes (SP 2–3 cm and SP 3–4 cm) ( Figure 10B ). Besides the connectivity variables and confinement, the factor shore was selected in this model ( Table 5 ), for which the most evident pattern was the positive association between the largest size classes (SP 2–3 cm and SP 3–4 cm) and Monte Clérigo ( Figure 10B ). This shore had the highest percentage of very-low confinement (category 1): 24% of rockpools ( Supplementary Figure S7 ).

In summary, the cover of articulated coralline algae was a mutual habitat-composition “top predictor” in models of the three species inside rockpools and of P. ulyssiponensis on the open-rock, whereas the cover of mussels and crustose non-coralline algae were the most relevant “top predictors,” respectively, for P. depressa and S. pectinata on the open-rock ( Table 5 ).