Buried Treasures, Hidden Thresholds: Integrating Cave and Landscape Drivers to Guide Conservation of Amazon Ferruginous Cave Biodiversity

Abstract

Iron-ore extraction plays a central role in the global economy, but several major mining areas overlap with ecologically unique ferruginous landscapes in the Brazilian Amazon, including caves that harbor endemic and highly specialized invertebrate fauna. Reconciling mineral exploitation with biodiversity conservation requires objective ecological criteria capable of supporting evidence-based decision-making. In this study, we evaluated how cave attributes and surrounding landscape features jointly structure invertebrate communities in pristine ferruginous caves of the Amazon and assessed their relative importance and environmental thresholds. Invertebrates were sampled in 69 iron-ore caves during dry and wet seasons, and 28 environmental variables related to cave morphology, microclimate, trophic resources, lithology, vegetation cover, and external climate were measured in subterranean habitats and adjacent landscapes. Our results demonstrate a clear scale-dependent pattern: cave attributes primarily regulated species richness, troglobitic richness, taxonomic distinctness, and seasonal beta diversity, whereas landscape features exerted stronger control over species composition, including troglobitic assemblages. Threshold analyses identified specific combinations of cave and landscape attributes associated with biologically pristine communities. These findings highlight that assessments of ferruginous cave biodiversity must integrate landscape-scale metrics, be conducted in unaltered environments, and prioritize networks of caves rather than isolated sites. This integrative framework provides robust ecological support for conservation planning and the sustainable management of iron-ore cave systems in the Amazon.

1. Introduction

Iron is the most widely used metal worldwide and plays a central role in the global economy [1]. In Brazil, iron ore accounts for approximately 3% of the gross domestic product and 10% of national exports [2], with extraction activities dating back to the 17th century [3]. This long-standing economic dependence directly conflicts with the ecological relevance of iron-rich landscapes. Ferruginous geosystems harbor unique ecosystems characterized by high levels of endemism and specialized biota that have been increasingly threatened by mining expansion [4,5]. In the Brazilian Amazon, one of the most biodiverse regions globally [6], ferriferous caves represent particularly important habitats, with more than 90 species described since 2005, including at least 39 obligate subterranean taxa [7,8,9]. Despite this relevance, ecological studies supporting the sustainable use and conservation of ferruginous subterranean systems remain limited [10,11,12,13,14].

Ferruginous geosystems differ markedly from other Amazonian landscapes due to their geological, climatic, and ecological characteristics. They are formed by banded iron formations (BIFs) capped by iron breccia (canga), which support shallow, nutrient-poor soils with low water retention and high thermal amplitude [15,16,17]. These conditions favor rupestrian and shrubby savannah vegetation on plateaus, while ombrophilous forests dominate surrounding hillsides [18]. Caves within these systems are typically small, shallow, and ancient [15] and are embedded within networks of centimeter- to meter-scale dissolutional and structural voids (hereafter referred to as ferruginous voids) developed within the iron duricrust and brecciated BIF matrix [13,15]. These subsurface voids promote the flow of water, nutrients, and organisms across landscape compartments, contributing to the high diversity of invertebrates, including troglobitic species, observed in ferruginous caves [10,12,13].

Mining activities intensified in the Brazilian Amazon during the 1980s, whereas biological research on ferriferous caves increased substantially only after 2005, driven mainly by environmental licensing requirements [12,19]. This temporal gap, combined with the spatial concentration of studies near mining areas, may have biased previously reported patterns of subterranean biodiversity [10,11,12]. Consequently, studies conducted in pristine ferruginous landscapes are essential to establish ecological baselines and to disentangle natural ecological drivers from mining-related disturbances.

The economic implications of cave preservation in iron-rich regions are substantial, as the protection of a single cave may imply the retention of large volumes of economically valuable ore [19]. This scenario underscores the challenge of reconciling mineral exploitation with biodiversity conservation and highlights the need for objective ecological criteria capable of supporting decision-making. In this context, identifying reliable biological metrics and environmental thresholds is fundamental to evaluate cave communities and to guide conservation strategies that balance ecological integrity and economic interests.

From this perspective, this study evaluated how cave attributes and surrounding landscape features influence invertebrate communities in pristine residual ferruginous plateaus of the Amazon. Specifically, we aimed to (1) identify which cave attributes (physical, climatic, and trophic) and landscape features (lithology, location, vegetation, and climate) influence cave community structure; (2) assess the relative importance of these predictors for maintaining subterranean biodiversity; and (3) determine environmental thresholds favoring key biological parameters of cave communities. We hypothesized that species richness, troglobitic richness, and seasonal beta diversity would be primarily influenced by cave attributes, reflecting short-term habitat conditions, whereas taxonomic distinctness and species composition would be more strongly shaped by landscape features, reflecting long-term ecological and evolutionary processes.

2. Materials and Methods

2.1. Study Site

This study was conducted in 69 iron-ore caves inserted in the Amazon biome in Serra do Tarzan, in the southeastern region of the state of Pará, Brazil (Figure 1). This region is recognized worldwide as Carajás Mineral Province due to the wide deposits of iron, nickel, copper, and gold and responsible for 42.93% of Brazilian mineral production [2]. Despite the huge amount of mineral exploitation occurring in the region, Serra do Tarzan is protected by two federal conservation units: Campos Ferruginosos National Park and Carajás National Forest. Such national reserves prevent mineral extraction, as well as the agricultural advance that surrounds these conservation units (Figure 1). Therefore, the studied area comprises the closest unit (within the regional geomorphological context) to the pristine condition of the Amazon ferruginous geosystem. The climate is characterized as continental equatorial (at altitudes lower than 350 m asl) or equatorial mesothermal (at altitudes higher than 700 m asl), with average temperature between 23 and 25 °C [15]. The rainfall presents strong seasonal variations, with dry periods between June and September (~10 to 90 mm per month) and an extremely rainy period between October and April (~160 to 340 mm per month) [20]. The average annual precipitation reaches 2400 mm [15].

2.2. Sampling of Environmental Variables and Cave Fauna

Each cave was sampled twice: the first field expedition occurred between January and February (rainy season), and the second between July and August (dry season), both in 2016. The sampling year coincided with the strong 2015–2016 El Niño event, which has been associated with extreme drought conditions across the eastern Amazon [21,22], potentially influencing invertebrate activity patterns. Overall, 28 variables divided into seven groups were used to characterize the caves (physical features, lithology, trophic characteristics, climatic factors) and surrounding landscape (location and cave insertion, climate, vegetation) (Table S1) [23,24,25,26]. Additionally, the invertebrates were collected through active capture all along the cave extension (sampling effort: 6 min per square meter per collector), with special focus on potential microhabitats for the occurrence of fauna (such as deposits of organic matter and under the rocks). Permission was given by Instituto Chico Mendes de Conservação da Biodiversidade—ICMBio, Nº 83/2016. The spatial restriction and scarcity of resources in subterranean habitats lead to a less abundant fauna, therefore making it necessary to search the entire cavity area. Collected specimens were fixed in ethanol (100%) and identified to the lowest possible taxonomic level by experts in different orders (see the Acknowledgments section for their names). Taxa identified at supra-specific levels were separated into morphospecies—an approach considered sufficient for ecological studies and conservation purposes [27,28]. The presence of troglomorphic traits was used to consider a species as troglobitic [29]. Such traits are specific for each taxonomic group (although there are some convergences such as anophthalmia, depigmentation, and appendage elongation), which were distinguished and validated by the experts responsible for the species identification. Vouchers of each taxon were deposited into reference collections (institutions and voucher numbers are presented in Table S2).

2.3. Data Analyses

2.3.1. Community Structure Parameters

All analyses were conducted in the software R (version 4.0.2) [30]. The species richness corresponds to the sum of distinct taxa found on each cave in the dry season (S_DRY), rainy season (S_WET), and both (S_TOT). The Spearman correlation was applied to select independent richness indices for the posterior elaboration of models (considered independent when rho < 0.700). The richness of troglobites represents the total records for each cavity in both seasons (S_TRO). The general species composition and troglobitic species composition were compared among cavities through a principal component analysis (PCoA), applying the Bray–Curtis dissimilarity index. Such analyses were conducted with the function capscale from the vegan package, from a matrix with standardized and log-transformed (log(x + 1)) abundance data, with “x” as the number of individuals observed for each species. The two first PCoA axes (MDS1 and MDS2) were considered proxies of the composition of species and troglobites for the subsequent models, which were applied as response variables.

The taxonomic distinctness index (Δ⁺) was calculated for 411 invertebrate species (SΔ⁺), which belong to 37 orders widely distributed in the study area (Table S3). Such groups were chosen due to the taxonomic refinement, so only orders in which all the morphotypes were identified at least to the genera level were included in the calculation of Δ⁺. The Spearman correlation was used to evaluate if the taxa selected to calculate the index satisfactorily represented all the species found in the caves, correlating S_TOT and SΔ⁺ (considered adequate if rho > 0.700 and p < 0.050). The Δ⁺ index of each cave was obtained through the function taxa2dist from the vegan package, with values plotted on funnel graphs with confidence intervals in relation to the expected average (95% and 99.99%).

The beta diversity partitioning proposed by Cardoso et al. [31,32] was applied to determine the processes responsible, using the differences in species composition between seasons. This approach was applied considering the Jaccard index calculated based on a composition matrix, what allowed for determining two components: species replacement and species-richness differences. The beta diversity partitioning was calculated through the function beta from the BAT package. The components were determined considering all the cavities to indicate which process was predominant in the study area. Then, values related to the selected component were extracted for each cavity and used as proxies for the beta diversity in the models.

The Mantel test was applied to detect if the geographical distance was responsible for the similarity in the composition of species and troglobites across caves (function mantel from vegan package). Additionally, Moran’s I index was used to evaluate if species richness, richness of troglobites, and taxonomic distinctness were influenced by geographical distance (function moran.test from the spdep package).

2.3.2. Modeling Community Structure Parameters

The 28 variables used to describe the cavities and landscape were used as model predictors (Table S1). First, the continuous predictors of each variable group were subjected to principal correlation analysis (PCA), from which the most representative ones were selected (the loading values of each axis were selected until the cumulative proportion attained at least 90% variation). Spearman correlation was used for continuous variables to check for the presence of autocorrelated parameters in the models (when rho > 0.700, p < 0.050). Similarly, Cramer’s V test was applied to the categorical variables (within and among groups) to select independent predictors (when φc < 0.700). The autocorrelation among the previously selected continuous and categorical variables was tested using the multiple correlation coefficient, prevailing in the choice for continuous variables when considered similar (if R > 0.700 and p < 0.050).

Generalized linear models (GLMs) and linear models (for species composition and the composition of troglobites) were used to assess the factors guiding the structure of communities (through the lme4 package). The species richness, richness of troglobites, taxonomic distinctness, seasonal beta diversity (presented as the most representative component), species composition, and composition of troglobites (represented by MDS1 and MDS2) were used as response variables for such analyses. On the other hand, the previously selected physical and environmental variables of caves and landscapes were used as predictors. When spatial autocorrelation was detected (indicated by the Mantel and Moran’s I tests), an additional term (the autocovariate) was included, which represents the spatial autocorrelation in the residuals of the best model [33]. In this approach, named the residual autocovariate model (RAC), the spatial autocorrelation represents the estimated strength of the relationship between the best model residuals for one given cave and the residual values for the neighboring ones [33]. The spatial autocorrelation in residuals of the RAC model was assessed by the Moran’s I test.

Predictors not contributing significantly (p > 0.050) to the models were removed based on the results indicated by the function summary after conducting the Wald test. The GLMs and LMs were applied using the following error distributions: Gaussian (species composition, taxonomic distinctness, seasonal beta diversity), Poisson (richness of troglobites), or negative binomial (species richness), with a function of logistic link for the response variable. Models with ΔAICc < 7 plus the first model above this value were considered to select all the predictors acting on the response variables, since they present some support for the main hypothesis [34]. The average coefficient of such models was calculated, which allowed for detecting the predictor variables with high levels of uncertainty due to the standard error amplitude [35]. Models containing all the indicated variables were subjected to hierarchical partitioning, which calculates the independent contribution of each variable to obtain their relative importance [36,37]. Variance inflation factor (VIF) analyses (vif function from car package) were applied to the final models to identify hidden multicollinearity issues. We then removed the predictors in which VIF > 10 [38,39].

2.3.3. Delimiting Thresholds and Levels for Community Structure Parameters

The segmented regression was conducted with the significant continuous variables to identify the behavioral changes (breakpoints) in relation to the response variables through the segmented package. The values from which the response variables were favored were selected based on the breakpoints. AICc values were compared to identify which type of regression (linear or segmented) represented the best model, as indicated by previous studies [13,14,40,41]. Results were expressed considering the amplitude of variation (Δ) of the segmented model in relation to the linear one. Additionally, a pairwise test of orthogonal contrasts was used for all the categorical predictors with more than two levels to individualize the significant ones (glht function from multcomp package).

Based on our analysis, caves presenting a set of predictors values that favored each response variable were selected. These caves were highlighted on a map of geographic location, considering the quantity and identity of the response variables they presented. This map was created using the QGIS 3.10 software with an orbital image [42].

3. Results

3.1. Environmental Variables and Cave Fauna

Most surveyed caves occur along the plateau margins, predominantly in BIF and canga formations, with marked variation in size, morphology, microclimatic conditions, trophic resources, and surrounding vegetation (Figure 2, Table S4). Regarding the cave invertebrate fauna, a total of 163,092 individuals were found, which belong to 939 species, 245 families, and 51 orders (Figure 3, Table S2). Nine species presented troglomorphic traits and were therefore considered troglobitic (Figure 4).

3.2. Community Structure Parameters

The caves presented an average of 36 species (± 22) in the dry season, 41 (± 22) in the rainy season, and 60 (± 30) in total (Table S5). Due to the similar behavior of these metrics of species richness (S_DRY vs. S_WET: rho = 0.97; S_WET vs. S_TOT: rho = 0.91; S_WET vs. S_TOT: rho = 0.97; p < 0.001), only the metric related to both seasons was adopted as response variable, which was considered a proxy of the others. Furthermore, the taxa used in the analysis of taxonomic distinctness properly represented the subterranean communities of the study area (SΔ⁺ vs. S_TOT: rho = 0.97 and p < 0.001). The average value of this index was 73.87, excluding those caves out of the expected confidence interval (Figure 5A). Regarding the community composition, caves contained an average of 26.34% of species in general and 75.31% of troglobitic species in common (Figure S1). The seasonal beta diversity presented a dissimilarity of 76.939% in species composition, which was mainly determined by species-richness differences (Figure 5B). The species richness, richness of troglobites, and composition presented significant spatial autocorrelation (S_TOT Moran’s I = 0.163, p = 0.003; S_TRO Moran’s I = 0.276, p < 0.001; composition of troglobites: R_MANTEL = 0.112, p = 0.027). There was no correlation between the geographical distance and taxonomic distinctness, seasonal beta diversity, or species composition for the caves in Serra do Tarzan (Δ⁺ Moran’s I = 0.102, p = 0.054; β_rich Moran’s I = 0.067, p = 0.183; Species composition: R_MANTEL = 0.052, p = 0.069).

3.3. Community Structure Models

The principal component analysis (PCA) within groups retained 11 continuous predictors (Table S6). No autocorrelation was detected among these variables, including guano cover within caves and forest area within a 250 m buffer, both originating from groups with only one continuous predictor (Table S7). Likewise, the six categorical predictors represented independent measures (Table S8). However, the multiple correlation coefficient indicated that solar radiation and the surrounding forest area were associated with lithology and geomorphological compartments, respectively (Table S9). Consequently, 17 of the 28 initially identified predictors were retained for the models addressing community structure and threshold definition.

Species richness was higher in larger caves and those with compound planimetric patterns, low humidity amplitude (i.e., high minimum humidity), and animal-originated trophic resources (bat guano, feces and carcasses of non-flying vertebrates), as well as those at low altitudes (

Table 1. Summary of the best models explaining species richness, troglobitic richness, taxonomic distinctness, species composition, troglobitic composition, and seasonal beta diversity (richness-difference component). Model type is indicated for each response variable (LM = Linear Model; GLM = Generalized Linear Model), followed by the selected predictors, their respective predictor group, variance inflation factor (VIF), test statistics (t or z values), and p-values. Estimates and standard error (SE) correspond to averaged model coefficients. Relative importance indicates the independent contribution of each predictor based on the hierarchical partitioning of full models. Predictors with an SE exceeding estimate values were considered non-significant. Response variables showing spatial autocorrelation were evaluated using the residual autocovariate (RAC) approach.

Response Variable	Model	Predictor Variables	Group	VIF	t/z	p	Estimate	SE	Variable Relative Importance (%)
Species richness	GLM (with RAC approach)	Area	4	4.964	10.220	<0.001	0.002	0.000	100
		Bat guano	6	3.851	−7.261	<0.001	−0.011	0.003	100
		Vertebrate traces	6	1.660	7.157	<0.001	0.495	0.138	100
		RAC term	-	1.098	6.851	<0.001	1.788	0.522	100
		Planimetric pattern	4	1.371	−4.679	<0.001	−0.234	0.100	100
		Altitude	1	1.783	4.355	<0.001	0.002	0.001	100
		Minimum humidity	7	1.455	4.538	<0.001	0.012	0.005	99.872
Troglobites	GLM (with RAC approach)	Minimum humidity	7	1.328	3.048	0.002	0.002	0.001	99.178
		Area	4	3.654	2.696	0.007	5.765	5.344	93.690
		RAC term	-	1.157	1.939	0.053	0.044	0.029	76.533
		Bat guano	6	3.066	−1.173	0.241	−0.005	0.011	43.645
Taxonomic distinctness	GLM	Hydric features	7	1.001	2.212	0.031	2.344	2.094	79.986
		Mean temperature	2	1.001	1.095	0.277	1.266	2.290	37.959
Species composition (MDS1)	LM	Forest cover	3	1.379	3.116	0.003	0.091	0.048	100
		Scarp height	1	1.054	−3.607	0.001	−0.073	0.044	99.686
		Altitude	1	2.006	−1.966	0.053	−0.002	0.004	47.336
		Depth	4	1.883	2.123	0.038	0.039	0.058	46.761
		Area	4	6.266	−2.029	0.047	−0.001	0.002	31.340
		Bat guano	6	3.796	1.720	0.090	0.004	0.012	29.045
Species composition (MDS2)	LM	Mean temperature	2	1.002	−1.984	0.052	−0.556	0.581	67.412
		Hydric features	7	1.505	2.224	0.030	0.598	0.704	62.165
		Bat guano	6	1.503	−1.984	0.052	−0.008	0.010	55.033
Composition of troglobites (MDS1)	LM (with RAC approach)	Scarp height	1	1.006	−7.894	<0.001	−0.077	0.020	100
		RAC term	-	1.004	7.914	<0.001	8.103	2.058	100
		Longitude	1	1.035	−8.926	<0.001	0.000	0.000	100
		Maximum humidity	7	1.031	−6.726	<0.001	−0.066	0.020	99.999
Composition of troglobites (MDS2)	LM (with RAC approach)	RAC term	-	1.020	5.798	<0.001	8.708	3.019	100
		Mean temperature	2	1.020	−3.831	<0.001	−0.684	0.359	99.653
β diversity (Species-richness differences component)	GLM	Forest cover	3	1.000	−3.166	0.002	−0.027	0.017	98.269
		Maximum humidity	7	1.000	−2.094	0.040	0.017	0.016	75.295

). All these predictors contributed significantly (i.e., the confidence intervals do not include zero) and presented similar relative importance for species richness.

The presence of troglobites depended mainly on high minimum humidity values (relative importance close to 100%), cave area (93.690%), and proximity between caves (76.533%). Only the presence of hydric features favored the occurrence of phylogenetically distant groups (relative importance close to 80%) (Table 1). Regarding community composition, the MDS1 axis indicated that caves mainly surrounded by forest areas and situated on small scarps presented fauna distinct from the other cavities. The standard error amplitude of the other model predictors of MDS1 and MDS2 demonstrated that they were not deemed significant enough to explain species composition. The composition of troglobitic species changed according to the scarp height, location (proximity between the sites and longitudinal component), and maximum humidity measured in the cave, as well as the average temperature of the external landscape. Seasonal differences in species richness were greatest in caves bordered by canga vegetation (i.e., lower areas of forest; relative importance of 98.269%) and characterized by high humidity (75.295%). No spatial autocorrelation was observed between the residuals of species richness and troglobite models subjected to the RAC approach (S_TOT residuals: Moran’s I = 0.092, p = 0.081; S_TRO residuals: Moran’s I = −0.070, p = 0.374; Troglobites’ composition residuals: MDS1 Moran’s I = −0.045, p = 0.718 and MDS2 Moran’s I = −0.138, p = 0.081).

3.4. Thresholds and Levels for Community Structure Parameters

Species-rich communities were found in caves larger than 123.102 m², those presenting compound (spongework) planimetric patterns, those with guano piles occupying an area up to 0.890 m², those with low humidity amplitude (minimum superior to 81.000% UR), and those located in the plateau or in the basis of Serra do Tarzan (Figure 6A–E). Caves with an area and minimum humidity superior to 56.145 m² and 81.335% UR, respectively, shelter more troglobitic species (Figure 6F,G). Caves with more than 16 ha of surrounding forest and situated on scarps inferior to 3.956 m favor the representativeness of species (dissimilar communities) (Figure 6H,I). Additionally, caves situated at the easternmost portion of this geomorphological unit on scarps superior to 7.232 m, under temperatures superior to 23.183 °C and humidity from 96.468% UR onwards, favor the occurrence of distinct troglobitic species (Figure 6J–M). Finally, caves with limited forest cover (<11.153 ha) and extreme humidity (>98.543% UR) are more susceptible to seasonal shifts in species composition (Figure 6N,O).

Based on the indicated thresholds, 19 caves present characteristics that favor a parameter responsible for the structuring of cave communities (Figure 7). Among them, cave ST_0056 stands out, with five of the six parameters evaluated. Two caves (ST_0003 and ST_0042) presented three parameters important for the subterranean fauna, while five caves included two of them, and eleven presented only one. Considering the response variables, only two caves include characteristics that favor rich (ST_0042 and ST_0056) or dissimilar (ST_0050 and ST_0055) communities. On the other hand, 15 cavities are prone to the presence of a high number of troglobitic species.

4. Discussion

Our study confirms that cave attributes and landscape features jointly structure ferruginous cave communities, with their relative influence varying among biological parameters, as hypothesized. Species richness, troglobitic richness, and seasonal beta diversity were primarily driven by cave attributes, including area, humidity, depth, and hydric features, whereas landscape features exerted stronger control over taxonomic distinctness and species composition. This partitioning reflects the scale-dependent nature of subterranean community assembly, in which landscape-scale filters define the species pool and local cave conditions modulate community structure.

Landscape attributes were consistently associated with species composition, reinforcing the role of epigean–hypogean connectivity in structuring cave assemblages [28]. Vegetation cover, geographic position, and cave insertion height emerged as key predictors, indicating that cave communities partially reflect surrounding surface ecosystems. Nestedness patterns between forest and canga environments support this interpretation, with forested areas providing higher productivity, humidity, and thermal stability than canga plateaus [43]. Altitudinal gradients further influenced richness through changes in productivity and microclimate [44], while external temperatures around 23 °C appeared optimal for troglobitic persistence, highlighting the sensitivity of these taxa to subtle climatic variation [45].

Scarp height modulated dispersal pathways into caves, with increasing height reducing faunal exchange with soil and leaf-litter compartments and enhancing compositional dissimilarity among caves. Nevertheless, troglobitic assemblages remained relatively homogeneous up to a threshold associated with the superficial nature of ferruginous formations, indicating effective dispersal through ferruginous voids. This interpretation is further supported by the spatial autocorrelation detected for species richness and troglobitic composition, which indicates functional connectivity among neighboring cavities through subterranean void networks [11,13]. Thus, macro-cavities represent accessible components of a broader subterranean continuum rather than isolated ecological units.

Subterranean traits such as cave area, compound planimetry, high humidity, depth, and perennial water were central predictors of richness, community stability, and reduced seasonal beta diversity. Larger and more complex caves exhibited greater habitat heterogeneity, supporting species–area relationships until thresholds beyond which resource limitations constrained further richness gains [5]. Animal-derived trophic resources, particularly bat guano, influenced general species richness mainly through presence rather than quantity, with small deposits sufficient to attract diverse assemblages, whereas larger accumulations favored specialized taxa and potentially reduced overall richness [46]. In caves with extensive guano deposits, resource specialization may promote the dominance of guanophilic species, resulting in hyper-abundant populations that competitively exclude less-specialized taxa. This shift in community structure may ultimately reduce overall taxonomic richness despite the high availability of organic matter [47]. Hydric habitats further enhanced taxonomic distinctness by enabling the coexistence of phylogenetically distant groups, consistent with the ecological role of habitat heterogeneity [48].

Troglobitic species accounted for less than 1% of the sampled cave community. This low proportion likely results from complementary factors. Ferruginous caves in Carajás are typically small and shallow, remaining strongly influenced by epigean climatic conditions [15,49], which limits the development of extensive aphotic and environmentally buffered habitats commonly associated with obligate subterranean lineages. In addition, recent taxonomic refinement in the region, reflected in the increase from only two formally described cave species prior to 2008 to approximately 90 by 2025 [7,8,9], has brought about more conservative criteria for defining troglomorphic traits, particularly given the legal implications of this classification under Brazilian environmental legislation [19], thereby reducing the likelihood of overestimating troglobitic diversity. Finally, growing evidence of faunal similarity between Amazonian edaphic and cave environments suggests that environmental convergence may favor shared species pools rather than strict subterranean specialization [14,43].

From a conservation perspective, these results demonstrate that protecting caves in isolation is insufficient to maintain subterranean biodiversity. Even minor reductions in vegetation cover within a 250 m buffer significantly altered species composition, corroborating evidence that mining activities near caves selectively impact sensitive troglobitic taxa [11]. Threshold-based analyses revealed that only a subset of caves simultaneously favored multiple biological parameters, underscoring the need for objective, criteria-based conservation strategies. The spatial dependence among caves further supports management approaches focused on conserving networks of cavities rather than individual sites.

5. Conclusions

Our study provides an integrated ecological framework for understanding the structure of ferruginous cave communities in pristine Amazonian landscapes. By jointly evaluating cave attributes and landscape features, we demonstrate that different biological parameters respond to distinct ecological drivers operating across spatial scales. While cave attributes primarily regulate species richness, troglobitic richness, and seasonal beta diversity, landscape features play a dominant role in shaping taxonomic distinctness and species composition. The identification of environmental thresholds associated with biologically pristine communities offers an objective basis for prioritizing caves according to specific conservation goals. Moreover, the spatial connectivity among caves mediated by ferruginous voids highlights the importance of conserving networks of cavities rather than isolated sites. Our findings further indicate that conservation actions restricted to cave interiors are insufficient, as modifications in surrounding vegetation can rapidly alter community composition. By establishing ecological baselines derived from pristine ferruginous systems, this study provides essential guidance for environmental licensing, conservation planning, and sustainable management in regions under increasing mining pressure. Integrating landscape-scale processes with subterranean habitat attributes is therefore critical to reconcile mineral exploitation with the long-term conservation of ferruginous subterranean biodiversity.