Optimizing Water–Carbon Coupling Through a Trait-Based Framework Integrating WCCI and Dual-Filter CATS Model

Abstract

Ecological restoration in degraded landscapes requires understanding the factors driving ecosystem function. We ask the central question: Do microtopography and plant functional traits control water-carbon coupling efficiency (WCCI) in mining-affected grasslands? We developed a novel, decoupled WCCI metric balancing water-use efficiency (1/SLA) and carbon-stock potential (Height + Foliage Cover). We hypothesized that (1) microhabitats with severe environmental filters (e.g., drought, erosion) would exhibit the lowest WCCI, and (2) this function could be optimized by assembling species that balance these two distinct trait strategies. Our objectives were to: (i) quantify the new WCCI across five microhabitat zones (A–E); (ii) assess how soil filters shape existing community functions; and (iii) identify optimized, zone-specific species assemblages using a dual-filter CATS model that maximizes WCCI. Results show significant variability in WCCI. The most degraded zones, A (arid) and B (high erosion), exhibited the lowest functional performance (mean WCCI = 0.029 and 0.078), supporting our first hypothesis. The dual-filter CATS model, constrained by abiotic targets (Tolerance = 10%) and a diversity cap (Max Abundance = 30%), successfully generated distinct, functionally tailored species assemblages for each zone. For instance, the optimized community for arid Zone A included the drought-adapted grass Stipa capillata (15.9%), while the resource-rich Zone D was recommended Medicago lupulina (12.7%). Conclusion: These findings confirm that a “one-size-fits-all” approach is insufficient. We demonstrate the necessity of a trait-based, microhabitat-specific framework to move beyond taxonomic mimicry and truly optimize biogeochemical functions in restoration.

1. Introduction

Global terrestrial ecosystems are undergoing unprecedented degradation, driven by a combination of climate change, anthropogenic activities, and land-use change [1,2,3]. According to the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services [4], on a global scale, over 23% of agricultural lands have experienced a decline in productivity, and approximately 15% of terrestrial carbon sinks have been lost due to ecosystem degradation [5,6]. This alarming trend is further compounded by the ongoing loss of biodiversity, exacerbating the challenges for ecosystem-based services such as carbon sequestration, water regulation, and soil stabilization [7,8,9]. These developments highlight the urgent need for innovative and scalable ecological restoration strategies that can reinstate ecosystem functionality while also mitigating climate change impacts.

Ecological restoration, defined as the process of reestablishing ecological functions and services in degraded ecosystems, has become a critical area of scientific research [10,11]. While conventional restoration efforts have focused primarily on the reassembly of native species [12,13,14], these approaches often struggle to restore ecosystems’ full functionality under the pressures of climate change and other anthropogenic disturbances [15]. Thus, a shift towards more dynamic, trait-based restoration approaches that account for the complex interactions between plants, soil, and their environment has gained significant traction in recent years [16].

The integration of plant functional traits (PFTs) into restoration practices has opened new avenues for designing more resilient ecosystems [17,18,19,20,21]. Functional traits—such as leaf area, specific leaf area (SLA), plant height, and root depth—affect key ecological processes, including nutrient cycling, water regulation, and biomass accumulation [22,23,24]. Trait-based ecology shows that species’ functional strategies shape community structure. These strategies reflect how species interact with and adapt to their environment [25,26,27,28,29,30]. The Continuum Assembly of Trait-based Species (CATS) model has emerged as a particularly promising tool for restoration, enabling the optimization of species assemblages by balancing ecological constraints with functional outcomes [27,29,30]. In traditional restoration approaches, the focus is often on species identity and taxonomic diversity, which does not necessarily translate into functional recovery or resilience under changing environmental conditions [31,32,33,34]. In contrast, trait-based approaches use PFTs to predict how species will respond to various environmental filters, making it possible to optimize species selection for desired ecosystem functions [35,36]. This predictive power is crucial, particularly in disturbed environments such as degraded mining landscapes, where the species pool is often limited, and recovery trajectories can be slow and uncertain [37,38].

However, despite their promise, current trait-based models, including CATS, face several limitations [39]. First, they often neglect the dynamic feedbacks between plants and their environment, such as soil feedbacks and interspecies interactions [40]. Additionally, these models are typically limited by the availability of trait data, which can constrain their predictive power and generalizability [41,42,43,44]. Recent developments in restoration ecology have begun to integrate biogeochemical cycles, particularly the coupling of water and carbon processes, into trait-based models [45]. Water and carbon are two of the most critical resources for plant growth, and their interactions are fundamental to ecosystem functioning, particularly in arid and semi-arid regions where water stress limits both plant productivity and carbon sequestration potential [46,47,48,49,50]. The dynamics of water–carbon coupling involve complex feedback mechanisms, where water availability influences plant carbon uptake and storage, while plant biomass and soil organic matter contribute to water retention and nutrient cycling [51,52,53]. But despite the growing recognition of the importance of water–carbon interactions, the integration of these processes into ecosystem restoration frameworks remains underexplored [54]. While traditional ecological models tend to treat water and carbon processes separately, the reality of ecosystem functioning requires a more holistic approach that accounts for their interdependencies [55,56]. For instance, plants that are efficient in water use (e.g., drought-resistant species) may also be effective in carbon storage, but the mechanisms behind this coupling are still poorly understood [57,58].

Recent studies have shown that trait-based models, particularly those focusing on water-use efficiency and biomass production, can provide valuable insights into how species can be selected for optimal water–carbon synergy [59]. By optimizing species combinations based on both water and carbon traits, restoration projects can enhance ecosystem resilience and provide greater potential for climate change mitigation. Despite these advances, integrating water–carbon coupling into trait-based models such as CATS remains a significant gap in restoration science [60].

This study confronts this gap by developing a novel, decoupled Water-Carbon Coupling Index (WCCI) and integrating it into a dual-filter CATS optimization framework. We move beyond conventional trait-matching to explicitly optimize for a core biogeochemical process. We therefore ask the central scientific question: Do microtopography and plant functional traits control water-carbon coupling efficiency (WCCI) in mining-affected grasslands? We hypothesize that: (i) Microhabitats with severe abiotic filters (e.g., high aridity in Zone A, high erosion in Zone B) will fail to recover functional synergy and exhibit significantly lower WCCI than more stable zones. (ii) Optimal WCCI is achieved not by maximizing a single trait, but by balancing the functional trade-off between community-level water-use efficiency (driven by 1/SLA) and carbon-stock potential (driven by Height and Foliage Cover). To test these hypotheses, we pursued four specific objectives:

(1)

To quantify and compare the existing WCCI across the five stratified microhabitat zones (A-E).

(2)

To evaluate how key environmental filters (soil moisture, nutrients, and erosion) shape existing community trait distributions and functional performance (WCCI).

(3)

To develop and apply a dual-filter CATS optimization model to identify zone-specific species assemblages that maximize the CWM-WCCI.

(4)

To compare the functional composition of these optimized assemblages, thereby providing practical, tailored restoration prescriptions.

2. Materials and Methods

2.1. Study Area

This study was conducted in a degraded mining ecosystem located within the Hulun Buir Grassland, Inner Mongolia, China (Figure 1). The mining site, which covers approximately 0.9 km², was abandoned in 2019 and has undergone around five years of unassisted natural recovery. This region belongs to the semi-arid continental steppe climate zone, with an annual average precipitation of 350–370 mm. January is the coldest month, with an average low temperature of −30.83 °C, while July is the hottest month, with an average high temperature of 25.84 °C. The dominant soil types include chestnut soils and chernozems, which are typical for northern China’s grasslands. Adjacent to the mining site, a pristine reference grassland was selected as a baseline ecosystem. This reference site, located approximately 100 km from the mining area, serves as a natural comparison for ecological conditions and community structure.

2.2. Sampling Design and Field Data Collection

A stratified sampling design was implemented to capture the maximum environmental heterogeneity of the degraded landscape. We selected five general sampling locations across the mining slope (representing variations in aspect and elevation) and established three 1 × 1 m sampling plots within each location, for a total of 15 plots within the mining site. This initial sampling design allowed us to collect a comprehensive dataset for subsequent empirical stratification. In the reference ecosystem, 20 sampling locations were selected in two zones: (i) Transition Zone (8 plots within a 0.5–2 km radius from the mining site) and (ii) Regional Background Zone (12 plots across a 100 km radius). These reference plots represented undisturbed ecosystems and were used to characterize natural community dynamics.

In each quadrat, plant community structure was quantified using the Braun-Blanquet cover-abundance scale [61]. For each species, we recorded the total vegetation cover percentage, species composition, and mean plant height. Additionally, functional traits including leaf length (LL), leaf width (LW), leaf area (LA), specific leaf area (SLA), and leaf circumference (C) were measured for all 107 species identified across the mining and reference sites. For each species, three healthy individuals were selected, and leaf samples (n = 5 per individual) were pressed, scanned, and analyzed using a CI-202 scanner (±0.1 cm² accuracy).

Soil samples were collected to a depth of 30 cm using a stainless-steel auger. These samples were homogenized and stored in light-proof containers for laboratory analysis. The following soil properties were measured: (i) Physical properties: Bulk density [62], moisture content (SMC) [63]; (ii) Chemical properties: Total nitrogen (TN) [64], phosphorus (TP) [64], and organic matter (SOM) [64]; (iii) Erosion estimation [65]: Soil erosion rates were determined using ¹³⁷Cs activity measured by gamma-ray spectrometry (E).

2.3. Objective 1 and 2: Quantifying Environmental Filters and WCCI

2.3.1. Microhabitat Stratification

To convert the 15 sampled plots into distinct, functionally relevant microhabitat zones, we performed a Ward’s hierarchical clustering analysis based on the measured soil properties (SMC, SOM, TN, TP, E). This approach was necessary because the initial aspect-based sampling did not sufficiently capture the primary environmental gradients. The clustering successfully grouped the 15 plots into five statistically distinct zones (A–E), which then served as the final basis for all subsequent functional analyses. Analysis of Variancee (ANOVA) with Tukey’s HSD post hoc tests was used to confirm significant differences in these variables between the resulting zones.

2.3.2. Calculation of the Water-Carbon Coupling Index (WCCI)

To quantify functional synergy, we developed a novel, decoupled WCCI based on distinct functional trait proxies representing “efficiency” (water) and “stock” (carbon).

(1)

Water Conservation Score (${\mathrm{W}}_{\mathrm{s}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}}$): This represents the community’s strategy for efficient resource use. We selected the inverse of Specific Leaf Area (1/SLA) as the sole proxy. 1/SLA (equivalent to Leaf Mass per Area, LMA) is a cornerstone trait of the Leaf Economics Spectrum (LES). Higher 1/SLA values strongly indicate a “conservative” resource strategy, characterized by denser leaf tissues, enhanced drought tolerance, and higher intrinsic water-use efficiency (WUE) under water-limited conditions.

(2)

Carbon Accumulation Score ($C_{score}$): This represents the community’s potential for biomass storage (“carbon stock”). It is calculated as the aggregate of Plant Height (H) and Foliage Cover (FC). Taller stature (H) indicates greater vertical biomass potential, while higher foliage cover (FC) represents horizontal community-level carbon stocks. Together, they provide a robust proxy for the standing carbon capacity of the vegetation.

The calculation followed a multi-step normalization process. (i) Trait Normalization: Each of the three raw traits (H, FC, and 1/SLA) was independently normalized to a 0–1 scale across all species using min-max scaling; (ii) Component Score Calculation:

$$W_{score}={\left(1/SLA\right)}_{norm}$$

(1)

$$C_{score}=H_{norm}+{FC}_{norm}$$

(2)

This process yields a single WCCI value for each species, ranging from 0 (very poor coupling) to 1 (optimal coupling), which was then used to calculate community-weighted means (CWM-WCCI) for each plot and simulated community.

$${WCCI}_i=\alpha ·W_{{score}_i}+\left(1-\alpha \right)·C_{{score}_i}$$

(3)

where W_score_i and C_score_i represent the normalized scores for water-use efficiency and carbon sequestration, respectively. The coefficient α was set to 0.5 to balance the weight of water and carbon in the index, though alternative weighting schemes were also considered in sensitivity analyses. This coupling index was used to assess the water–carbon synergy across different species assemblages optimized for each microhabitat.

This process yields a single WCCI value per species (0 = poor coupling, 1 = optimal coupling), which serves as the core functional target for community-level analyses and optimization.

2.3.3. Trait-Environment Analysis [66]

For each species in the study, we calculated the community-weighted mean (CWM) of seven key functional traits (LL, LW, LA, SLA, H, C, FC). CWM values were computed using the formula:

$${CWM}_i=\sum _{i=1}^S{p}_i{t}_i$$

(4)

where p_i is the relative abundance of species ii in the community, and t_i is the mean trait value of species i. These CWM values were used to quantify the functional composition of plant communities across both the mining and reference ecosystems.

To understand how environmental factors shape plant trait distributions, we calculated Pearson correlation coefficients between key soil variables (SMC, SOM, TN, TP, E) and CWM trait values [67]. This analysis aimed to identify the environmental factors most strongly influencing the distribution of functional traits in the study areas. The results provided essential data for model construction and species optimization, highlighting which environmental gradients drive specific trait expressions.

2.4. Objective 3 and 4: The Dual-Filter CATS Optimization Framework

To identify microhabitat-specific species assemblages that maximize WCCI and provide tailored restoration prescriptions, we developed and applied a three-stage, dual-filter CATS optimization model. The overall conceptual logic of this optimization framework is illustrated in Figure 2.

2.4.1. Environmental Filtering and Target Trait Prediction

To ensure the optimized communities were adapted to the local environment, we first established “abiotic trait targets” for each microhabitat:

(1)

Filter 1 (Training): We used the randomForest package to train a Random Forest model on the reference ecosystem data (N = 20 plots). This model learned the relationship between environmental predictors (SMC, SOM, TN, TP, E) and the observed CWM traits (CWM-H, CWM-FC, CWM-SLA).

(2)

Filter 2 (Prediction): This trained model was then used to predict the “target” CWM trait values for each of the five mining microhabitats (A–E) based on their specific soil conditions.

2.4.2. Optimization: Maximizing WCCI via Linear Programming

Finally, we used linear programming (LP) to find the optimal relative abundance $\left(p_i\right)$ for each species that would maximize the CWM-WCCI, subject to a series of strict constraints:

(1)

Objective: $Maximize={\sum }_{i=1}^{107}{p}_i\times {WCCI}_i$

(2)

Subject to Constraints:

• Sum Constraint: ${\sum }_{i=1}^{107}{p}_i=1$
• Abiotic Filter Constraint: The CWM traits of the final community must fall within a strict 10% tolerance window of the “target” values from Filter 2. This ensures microhabitat-specificity.
• Diversity Constraint: The relative abundance of any single species was capped at 30% to prevent mono-dominance.

This LP model was run independently for each of the five microhabitat zones (A–E) to generate five distinct optimal assemblages.

2.5. Statistical Analysis

To quantitatively stratify the microhabitats, we performed a Ward’s hierarchical clustering analysis [68] based on the key soil properties (SMC, SOM, TN, TP, E). All statistical analyses were conducted using R (version 4.2.2). Pearson correlation coefficients [69] were used to analyze the relationship between functional traits and environmental factors. Random forest models [70] were used for trait prediction based on environmental variables, with model performance validated through cross-validation (R²). Monte Carlo simulations [71] (1000 iterations) were performed to propagate uncertainty in the trait imputation process and assess its impact on the final species optimization.

During manuscript preparation, we used the generative AI tool ChatGPT (OpenAI, GPT-4, 2023) to assist in language polishing and in drafting preliminary conceptual descriptions. In addition, AI was employed to generate an initial version of one conceptual diagram, which was subsequently reviewed and substantially revised by the authors. No AI tools were used for data collection, analysis, or interpretation. The authors carefully checked and edited all AI-assisted content to ensure accuracy and originality, and take full responsibility for the final version of the manuscript.

3. Results

3.1. Objective1: WCCI Is Lowest in Severely Degraded Zones

The integrated functional performance, quantified by the decoupled WCCI, revealed stark functional disparities among microhabitat zones (Figure 3). Communities in the severe environments of Zone A and Zone B exhibited the lowest functional synergy, with mean WCCI values of only 0.029 (±0.001) and 0.078 (±0.025), respectively. In contrast, zones with more favorable conditions (Zones C, D, and E) achieved significantly higher functional coupling, with mean WCCI values converging around 0.10 (Zone C: 0.102; Zone D: 0.100; Zone E: 0.103). Notably, Zone E displayed the highest internal variability (CV = 67.7%), ranging from functionally poor plots (WCCI = 0.034) to the highest performing plot in the entire study (WCCI = 0.219), reflecting extreme spatial heterogeneity. These results quantitatively confirm that natural recovery has failed to restore functional water-carbon synergy in the most degraded microhabitats (A and B), identifying them as priority targets for active restoration.

3.2. Objective2: Environmental Filters Drive Trait Distributions

To better understand how local abiotic conditions shape community assembly and functional processes, it was first necessary to identify the degree of environmental heterogeneity within the study site. Through Ward’s hierarchical clustering analysis of key soil physicochemical properties, the study site was quantitatively stratified into five statistically distinct microhabitat zones (Figure 4). The environmental conditions varied significantly across these zones (ANOVA, p < 0.05 for all tested variables).

Zone B exhibited the highest mean Soil Moisture Content (SMC) at 13.71 ± 5.2% (mean ± SD), making it the most mesic microhabitat. In contrast, Zone A was the most arid, with a mean SMC of only 7.14 ± 1.5%. Soil fertility gradients were equally pronounced. Zone D was the most fertile, possessing the highest mean Soil Organic Matter (SOM) (16.31 ± 2.1 g/kg) and Total Nitrogen (TN) (1.17 ± 0.3 g/kg). Conversely, Zone B, despite its high moisture, was nutrient-poor, showing the lowest mean SOM (7.23 ± 1.8 g/kg) and TN (0.49 ± 0.2 g/kg). Furthermore, the soil erosion index (E) revealed different geomorphological dynamics: Zone B experienced the most severe erosion (E = −144.4 ± 35.1 t/ha), while Zone D was a depositional area (E = 41.0 ± 15.2 t/ha). These results confirm that the microtopography creates a complex mosaic of distinct environmental filters, each posing unique challenges and opportunities for plant establishment and growth.

3.3. Objective2: Trait-Environment Filtering in Plant Communities

The functional composition of the extant plant communities strongly reflected the underlying environmental gradients, providing clear evidence of non-random, trait-based assembly processes (Figure 5). Across all the plots, community-weighted mean (CWM) traits varied widely, with CWM Specific Leaf Area (SLA) ranging from 45.2 to 315.6 mm²/mg and CWM Leaf Area (LA) from 0.09 to 19.06 cm².

Random Forest (RF) regression analysis, which captured the multi-factor, non-linear relationships, revealed significant trait-environment couplings. We observed a strong negative coupling between CWM-SLA and Soil Moisture Content (SMC), which was strongly driven by Total Nitrogen (TN) (R² = 0.544). As soil moisture increased, the dominant community-level leaf strategy shifted demonstrably from a resource-acquisitive strategy in drier zones to a more conservative one in wetter zones (Figure 5a). Concurrently, CWM-LA was strongly and positively coupled with Soil Organic Matter (SOM) and Total Nitrogen (TN) (R² = 0.456), with SWM being the top driver (Figure 5b). This indicates that communities in more fertile soils were dominated by species with larger leaf surfaces, maximizing light interception and photosynthetic capacity. Additionally, CWM plant height (H) showed a strong coupling with Soil Water Content (SMC) (R² = 0.531). This suggests that adequate moisture availability is the primary structural filter, allowing for the dominance of taller, more competitive species (Figure 5c).

3.4. Objective 3 and 4: Optimized, Zone-Specific Community Assemblages

Applying the dual-filter optimization framework, which constrained by both abiotic trait targets and a species dominance cap to ensure diversity. We generated distinct, microhabitat-specific assemblages designed to maximize the CWM-WCCI (Figure 6). The model’s recommendations demonstrate clear functional differentiation across the five zones, successfully tailoring configurations to local filters.

Zone A, the assemblage for the most arid and nutrient-poor zone is dominated by Oxytropis myriophylla (30.0%) and Descurainia sophia (30.0%). Crucially, the model selected the drought-adapted grass Stipa capillata (15.9%) as a key component, complemented by Kalimeris indica (13.1%) and the nitrogen-fixing Astragalus melilotoides (5.7%) to cope with the harsh conditions. Zone B, characterized by severe erosion, required a functionally similar but distinct configuration. While sharing the same dominant species (O. myriophylla, 30.0%; D. sophia, 30.0%), the model replaced the grass Stipa with the legume Medicago lupulina (12.7%) and Kalimeris indica (12.2%), likely prioritizing species with soil-binding or pioneering capabilities. In stark contrast, the assemblage for the resource-rich Zone D also shows a different combination. While still anchored by O. myriophylla (30.0%) and D. sophia (30.0%), it features Medicago lupulina (12.7%) and Elymus repens (12.5%), species known for higher resource acquisition and competitive ability. About Zone C and E, these zones demonstrated further specificity. Zone C’s configuration included Salsola collina (13.1%), while Zone E introduced Artemisia sieversiana (10.9%) and Lepidium apetalum (10.2%), highlighting the model’s sensitivity to subtle environmental shifts.

This demonstrates the framework’s ability to move beyond a ‘one-size-fits-all’ solution, tailoring restoration prescriptions to the unique abiotic filters of each microhabitat while optimizing for a core biogeochemical function (WCCI).

4. Discussion

This study advances a trait-based restoration framework by moving beyond the conventional goals of species reassembly to the targeted optimization of a critical ecosystem process: water-carbon coupling. Our findings demonstrate that by integrating a novel functional index (WCCI) into a spatially explicit modeling framework, it is possible to design functionally superior plant communities tailored to the heterogeneous conditions of degraded landscapes.

4.1. An Innovative Framework for Optimizing Biogeochemical Function

The central innovation of this study is the development of a decoupled Water-Carbon Coupling Index (WCCI) and its integration into a dual-filter CATS optimization framework. By defining water function based on the “conservative” trait (1/SLA) and carbon function based on “stock” traits (H + FC). This new metric provided clearer and more ecologically sound insights. Observed communities exhibited large functional disparities across microhabitats (Figure 5), with the most severely degraded zones, Zone A (arid) and Zone B (high erosion), exhibiting the lowest functional performance (mean WCCI = 0.029 and 0.078, respectively). This contrasts with the more stable zones (C, D, E), which achieved higher mean WCCI scores.

This result strongly supports our first hypothesis: severe abiotic filters suppress functional recovery. Furthermore, by applying this validated WCCI as the optimization criterion, our dual-filter CATS model produced assemblages with significantly higher functional potential. For instance, the optimized CWM-WCCI for Zone A reached 0.46, representing a more than 15-fold increase from its observed baseline (0.029). This confirms the framework is not simply descriptive but prescriptive, capable of identifying configurations that dramatically outperform natural recovery.

4.2. The Mechanisms of Functional Optimization: Filters, Keystones, and Soil Feedbacks

The observed improvement in WCCI through optimization invites the question of why certain communities can achieve higher water–carbon coupling. Our results provide three complementary mechanistic explanations.

First, environmental filtering at the microhabitat scale emerged as a decisive constraint on functional potential [72]. For example, the observed mean WCCI in fertile Zone D (0.100) was more than three times higher than in arid Zone A (0.029). This functional gap underscores that soil conditions set an upper limit on natural recovery. However, optimization successfully closed this gap: our optimized communities in both Zone A (optimized CWM-WCCI ≈ 0.462) and Zone D (optimized CWM-WCCI ≈ 0.462) reached a nearly identical high functional potential. This demonstrates a critical finding: the limitation in Zone A is one of community assembly, not intrinsic site potential, which strongly justifies active restoration [73].

Second, functional performance was driven by a new, optimized set of keystone species assemblages, which were selected by our dual-filter mechanism. In the natural communities, performance was weak. Our optimized assemblages (Figure 6), in contrast, were not driven by the original dominant species, but by a new, functionally optimal assembly. This new assembly was built on two distinct functional roles: (i) A “Functional Core”: Species like Oxytropis myriophylla and Descurainia sophia were selected in nearly all zones up to their 30% abundance cap because their traits provided the highest intrinsic WCCI (the objective). (ii) “Environmental Tuners”: Species like the grass Stipa capillata (in arid Zone A, 15.9%) or the legume Medicago lupulina (in fertile Zone D, 12.7%) were selected not for their WCCI score, but because their specific traits were required to tune the community’s average traits to fit within the strict 10% abiotic tolerance window of that specific microhabitat. Therefore, the keystone was not a single species, but this combination created by our dual-filter constraints.

Third, links between functional optimization and soil processes suggest potential feedback mechanisms. As noted in our original analysis, communities with higher CWM-WCCI showed positive correlations with soil moisture and organic matter, even though these relationships were not statistically significant given sample size. This pattern, still relevant today, implies that establishing these new, functionally efficient communities could initiate positive soil feedbacks, contributing to soil stabilization and nutrient retention, and reinforcing the long-term sustainability of the restoration.

4.3. Practical Implications: From Precision Restoration to Sustainable Agronomy

From an applied perspective, the WCCI–CATS framework provides a critical, two-step approach for precision restoration. First, the WCCI serves as a robust diagnostic tool for prioritizing intervention. Our analysis (Figure 5) quantitatively identifies Zones A (mean WCCI = 0.029) and B (mean WCCI = 0.078) as functionally collapsed areas that require immediate, active restoration. By contrast, Zone D (mean WCCI ≈ 0.100) is functionally more stable and may only require minimal, passive management.

Second, the CATS optimization model generates actionable prescriptions (Figure 6). For each microhabitat, the framework provides zone-specific solutions tailored to the environmental constraints: (i) In the arid Zone A, the model prioritized the stress-tolerant grass Stipa capillata (15.9%) to meet the low water target (W-score). (ii) In the resource-rich Zone D, the optimized assemblage included the nitrogen-fixing legume Medicago lupulina (12.7%) to leverage the higher nutrient availability (C-score).

This function-first approach is highly transferable beyond mine restoration, directly addressing the core concerns of sustainable agronomy. The WCCI can inform the design of multifunctional systems: (i) Cover Crop Design: The framework can be used to select cover crop mixtures optimized for both high biomass accumulation (C-score) and enhanced drought resilience (W-score), a crucial balance in water-limited agriculture. (ii) Pasture Revitalization: The model can identify specialized grass-forb combinations that maximize forage productivity while simultaneously improving the soil’s water retention capacity. (iii) Buffer Zone Management: Deploying high-WCCI plant assemblages along riparian or field edges can maximize erosion control and nutrient retention in agro-ecosystems.

By shifting the focus from simply counting species to optimizing functional synergy (WCCI), this trait-based framework offers a generalizable strategy for designing resilient plant communities. This integrated approach, which links microhabitat-specific diagnosis and functional optimization, provides a direct and quantitative pathway for climate-resilient land management (Figure 7).

4.4. Limitations and Future Directions

While the WCCI–CATS framework represents a methodological advance, several limitations should be acknowledged.

First, the analysis is based primarily on above-ground functional traits (e.g., SLA, height, foliage cover). Although these traits are strong predictors of water–carbon processes, they cannot fully capture below-ground dynamics such as rooting depth, hydraulic lift, or soil–microbe interactions [74,75,76,77,78]. This simplification may underestimate the role of traits directly linked to water acquisition and carbon stabilization in soils. Second, the framework is inherently static. It identifies optimal assemblages for a given set of environmental conditions, but does not simulate successional trajectories, dispersal, or species turnover [79,80]. As a result, the predicted communities represent functional “endpoints,” and the long-term stability of these optimized assemblages under climate variability or disturbance remains uncertain. Third, the correlation between WCCI and soil health indicators (e.g., r = 0.41 for soil moisture, r = 0.36 for organic matter) was encouraging but not statistically significant, partly due to limited sample size. This highlights the need for larger-scale and longitudinal datasets to rigorously test whether communities optimized for high WCCI can indeed reinforce soil quality and ecosystem resilience through feedback processes. Finally, the species pool used in the optimization is site-specific. While the framework is transferable, the actual species lists must be recalibrated to local conditions, which requires reliable trait databases and careful validation in different ecological contexts.

Future research should therefore pursue four directions: (i) expanding trait datasets to include below-ground, hydraulic, and physiological traits [77,79]; (ii) coupling the optimization framework with dynamic, spatially explicit models that simulate recruitment, mortality, and successional change [81]; (iii) conducting field trials to compare the establishment, persistence, and soil feedbacks of WCCI-optimized communities against conventional restoration mixes; and (iv) testing the framework across diverse degraded ecosystems (e.g., semi-arid grasslands, riparian corridors, sloping agricultural lands) to evaluate its generalizability.

By addressing these limitations, the WCCI–CATS framework can be further refined into a robust, predictive, and widely applicable tool for designing resilient ecosystems under conditions of land degradation and climate stress.

5. Conclusions

This study investigated the role of microtopography and functional traits in controlling water-carbon coupling efficiency (WCCI) in a degraded mining landscape. Our results provide strong quantitative support for our primary hypotheses.

First, we confirmed our hypothesis that severe abiotic filters directly suppress functional recovery. Our novel, decoupled WCCI metric revealed that the most degraded microhabitats, Zone A (arid) and Zone B (high erosion), were functionally collapsed. Their mean WCCI scores (0.029 and 0.078, respectively) were significantly lower than the more stable microhabitats (mean WCCI ≈ 0.10). We demonstrated that this functional failure was directly linked to environmental filtering, as community traits showed significant correlations with limiting soil factors.

Second, we demonstrated our second hypothesis that a trait-based optimization framework can successfully design functionally superior, diverse, and site-specific communities. Our dual-filter CATS model, which balanced the trade-off between water-use efficiency traits (1/SLA) and carbon-stock traits (H + FC), successfully generated distinct assemblages. The model’s specificity, driven by a strict 10% trait tolerance, was evident: the prescription for arid Zone A included the drought-adapted grass Stipa capillata (15.9%), while the prescription for resource-rich Zone D was tailored to include the legume Medicago lupulina (12.7%). The 30% abundance cap constraint was also critical, ensuring all optimized communities were diverse (5–6 species per zone) rather than unrealistic monocultures.

This research moves beyond taxonomic-based strategies, offering an integrated, ecosystem-level perspective on ecological recovery. We provide not only a quantitative diagnosis of functional failure but also a transparent, reproducible framework (RF + LP) for prescribing restoration solutions that are optimized for biogeochemical function (WCCI).