X-ray micro-tomographic images of a Luvisol (resolution 19 µm), obtained from the Soil Structure Library (https://structurelib.ufz.de/lit/, last access: 5 March 2026; Weller et al., 2022) in the form of a TIFF image stack, were used to create the state matrix.

The cellular automaton model used here is based on a set of deterministic rules applied synchronously at each simulation time step, modifying both the state matrix and the resource matrix. In each matrix, only water-filled pore grid units, resources grid units and microbial cells can change state. Within a simulation, solid phase grid units and empty pores are immutable and represent the inherent constraints of a particular matrix (Fig. 2).

The amount of resources in the system is defined at the start of the simulation in the form of resource elements, 100 of which is assigned to each resource grid unit. Resources can diffuse from their initial locations in the matrix according to a propagation model that follows Moore's neighbourhood (eight adjacent grid unit; Cellular Automata – The Moore neighbourhood, 2026). This diffusion occurs along concentration gradients, with resources spreading from resource-rich to resource-poor zones. At each iteration i, resources diffuse if the resource content of a grid unit is above a threshold value. Fixed amounts of resources are distributed from resource-rich grid units to neighbouring grid units with lower resource contents, up to a maximum diffusion rate. The change in resources in a water or resource grid element (x, y) can therefore be written as: 1Resource(resourcex,y)i+1=Resource(resourcex,y)i-∑m,n∈{-1,0,1}T(x+m,y+n)i where T(x+m,y+n)i denotes the amount of resources transferred between cell (x, y) and its neighbour (x+m, y+n) during iteration i. By convention, Ti is positive when (x, y) is more resource-rich than its neighbour (x+m, y+n), i.e., resources flow out of (x, y), and negative when the reverse holds. This transfer is constrained by a diffusion threshold (donor eligibility), such that a non-zero transfer occurs only if Resource(x,y)≠Resource(x+m,y+n), and by a fixed diffusion rate (maximum transfer per neighbour), such that the richer of the two cells loses a constant quantity to the poorer one rather than a gradient-proportional amount. This produces a stepwise and locally asymmetric redistribution process, leading to a gradual homogenization of resource distribution across the matrix without reaching instantaneous equilibrium.

Microbial cells take up all resources that are present in the grid unit that they occupy. The resources arrive in the grid unit by diffusion or were already present when a motile microbial cell came to occupy the grid unit.

At the end of each iteration, the total accumulated resource content stored in a microbial cell is used as an internal state variable (“resource points”) that determines its physiological status. This accumulated pool is then compared to the predefined thresholds that govern microbial processes such as maintenance, survival, reproduction, or mortality.

The different types of microbial cells (Table 1) transition between three states (growth, maintenance or death), depending on resource availability. The transitions are regulated by three critical resource thresholds. The maintenance threshold (Sd) triggers a state of reduced metabolism when resources become scarce. During maintenance, cells do not move or reproduce. If the resource scarcity persists and falls below the mortality threshold (Sm < Sd), the microbial cell dies and the remaining resource is transferred to the available resource pool. Conversely, the accumulation of resource above the reproduction threshold (Sr) activates cell division, in which a new cell of the same type appears in any of the eight adjacent aqueous grid units. The parent cell invests a significant proportion of its resource (25 % to 50 % depending on the microbial strategy) in this process according to the formula : 2Resource(parentx,y)i+1=Resource(parentx,y)i-Resource(descendantx,y)i+1 where Cr represents the fixed metabolic cost of reproduction (Table 1).

The location of the descendant cell is chosen randomly among the set of available neighboring cells that can be occupied (i.e., empty or otherwise permissible states depending on the model rules). If no suitable cell is available in the local neighborhood, reproduction does not occur during that iteration and is postponed until space becomes available. In practice, this situation is rare due to the spatial structure and occupancy dynamics of the cellular automaton.

The reproductive costs vary among microbial cell types, the cost being higher for K-strategists, which invest in fewer, larger, and more resource-rich offspring or cellular structures optimized for survival under stable but competitive conditions, rather than in the rapid, low-cost proliferation favored by r-strategists. This trade-off between investment per offspring and reproductive rate also extends to resource acquisition strategy, which itself carries a cost that differs between motile and immobile cell types.

Motile cells can search for and exploit available resources within a configurable perception radius R. Motile cells move at a speed of one grid unit per iteration towards the nearest and most resource-concentrated resource grid unit within their perception radius. Due to the cost of motility, motile cells focus only on resources with a resource content above a defined threshold equal to 5 resource elements (Table 1). The cost of motility (Cm) is deducted for each move according to: Resource(t+1)=Resource(t)-Cm. The inability of mobile cells to anticipate obstacles, such as pore walls, when moving toward the nearest resources introduces a realistic constraint that limits their effectiveness in complex environments. Immobile cells rely entirely on the diffusion of resources to acquire the resource and matter necessary for maintenance and growth. This resource acquisition strategy is energetically more economical, due to the absence of motility costs.

For each of the 10 pore-space images, four distinct initial conditions were generated by combining two moisture levels (water-saturated and low-water conditions) with two resource spatial distributions (resources located either in large pores or in small pores). This results in a total of 40 distinct initial matrices (= 10 images × 2 moisture levels × 2 resource distributions). For each initial matrix, simulations were then performed with different microbial community configurations, combining two initial microbial densities (low: 23 cells, and high: 56 cells) and three microbial diversity treatments (r-strategists only, K-strategists only, and mixed r + K communities).

Each unique combination of initial matrix and microbial configuration was replicated 20 times and simulated over 200 iterations. This full factorial design resulted in a total of 4800 simulations runs.

In addition to the simulations presented above, analyses were performed to establish the sensitivity of the model to (1) matrix moisture, (2) initial resource concentration, and (3) the parameters governing the r- and K-strategies. Details of the methods and results obtained for these additional simulations are presented in the supplementary data.

At the beginning of the simulations, microbial biomass represented between 9.0 % (low microbial density) and 16.5 % (high microbial density) of total soil organic carbon (SOC). These values were chosen as they are representative of the rhizosphere, a highly active soil compartment characterized by elevated microbial densities relative to total organic carbon (Zhao et al., 2022). Although microbial biomass is often reported to represent a smaller fraction of SOC in bulk soils, typically reaching up to ∼6 % in field measurements (He et al., 2021), estimates based on fumigation–incubation methods indicate microbial biomass carbon contents ranging from 300 to 1200 mg C kg⁻¹ in natural silt–clay soils (Franzluebbers et al., 1994; Jenkinson and Powlson, 1976; Sparling and West, 1988), corresponding to approximately 1.5 %–12 % SOC in soils containing 1 %–2 % total organic carbon. The values used in the present model therefore fall within, or slightly above, the upper range reported in the literature, consistent with a focus on the rhizosphere rather than bulk soil. Within the model, the use of relatively high initial microbial densities aimed to ensure the establishment and persistence of microbial populations over the simulation period, while allowing a representative distribution of different functional types across heterogeneous microenvironments (pores of varying size and resource availability). Although the model assumes an identical total amount of carbon between pore types, the differential accessibility generated by diffusion constraints and spatial geometry conceptually reproduces empirical gradients. Cells were initially distributed randomly in space to avoid imposing predefined spatial advantages among microbial types (mobile vs. immobile) and life-history strategies (r vs. K), thereby allowing differences in colonization dynamics to emerge from intrinsic traits and interactions.

In the model, the recycling rate of microbial biomass (cellular resource) was set to 100 %, meaning that the entire resource content of dead cells was instantaneously transferred to the pool of assimilable resources. This assumption implies an immediate decomposition of cellular structures and the absence of a non-assimilable organic matter compartment. Empirical studies indicate that microbial biomass recycling efficiency in soils is much lower. Although microbes are able to feed on microbial necromass, recycling efficiencies rarely exceed 12 % (Kästner et al., 2021), and microbial necromass can constitute more than half of total soil organic carbon stocks (Liang et al., 2019). The use of a 100 % recycling rate therefore does not aim to represent realistic microbial necromass processing, but rather reflects a deliberate structural simplification of the model. All cellular resource is assumed to be potentially accessible to other microbes, corresponding to an idealized and instantaneous transformation of biomass into assimilable resources. As a consequence, the model accelerates carbon transformation processes compared to real soil systems and likely overestimates the efficiency and speed of microbial biomass recycling. This limitation should be kept in mind when interpreting model outcomes.

In the model, carbon use by microbial communities is described through their carbon use efficiency (CUE), a widely used metric in soil microbial ecology that reflects how efficiently microorganisms convert the carbon they take up into new biomass, as opposed to losing it through respiration. A high CUE indicates that a large proportion of assimilated carbon is invested in growth, whereas a low CUE indicates that most of the carbon taken up is respired to meet maintenance and metabolic costs, with proportionally less contributing to biomass production. CUE is computed at each iteration based on carbon fluxes at the level of the microbial community. The total carbon uptake by microbes includes both the portion converted into biomass (growth) and the portion lost through respiratory processes (maintenance and metabolic costs). CUE is therefore defined as the ratio of carbon allocated to biomass to total carbon uptake: CUE=(carbonallocatedtobiomass)/(totalcarbonuptake). This ratio made it possible to assess the influence of multiple factors on the metabolic efficiency of microbial communities: resource location, water regime, adaptive strategies, and microbial motility.

Analysis of microtomographic images to extract pore-space images and initialize the simulation matrices was carried out in Python while the cellular automaton model itself was developed in R. Python was only used during the pre-processing stage to extract pore-space images and initialize the simulation matrices from the image data. The spatial simulations and visualizations were implemented using custom R scripts rather than relying on an existing cellular automaton framework or dedicated spatial modeling library. The code used for this work can be found at 10.5281/zenodo.21336136 (Maelmstr., 2026) and model current development is available at https://github.com/Maelmstr/soil-cellular-automaton (last access: 10 October 2026).

All statistical analyses were performed using R (R, 2026). The demography of microbial populations, their use of resources for maintenance, growth and movement, and the temporal dynamics of resources were monitored for each simulation. The first 25 iterations of each simulation were treated as a warm-up phase and excluded from CUE calculations, in order to limit the influence of the random initial spatial placement of microbial cells on the resulting efficiency estimates. During this early period, the dominant process is the spatial redistribution of mobile cells toward accessible resource-rich areas, with population density curves showing that microbial populations already increase rapidly within the first iterations. As a result, the resource distribution at iteration 25 is no longer strictly equivalent to the initial condition, but this shift was expected and intentional: our aim was to assess microbial efficiency once the initial, largely arbitrary spatial configuration had reorganized into a more ecologically meaningful pattern shaped by microbial movement and local resource accessibility, rather than to capture the transient dynamics from initial conditions. Since this warm-up period was applied uniformly across all simulations, any resulting differences in resource availability at iteration 25 affect all treatments comparably, so that relative comparisons among treatments remain valid. Furthermore, because the warm-up phase represents only a small fraction of the total simulation length, its exclusion is expected to have a limited effect on the overall CUE estimates.

Analysis of the calculated CUE was done using analysis of variance (ANOVA), with location of resources in pore size, moisture condition, and mobility as fixed factors. Tukey Honest Significant Differences post-hoc pairwise comparisons were then performed to account for multiple comparisons.

The initial microbial cell density had no effect on the population dynamics profiles of r- or K-strategists and therefore the results for the more abundant initial cell numbers are presented in the main text (Fig. 3) and those the less abundant initial populations are shown in the Supplement (Fig. S1). The mortality of immobile K-strategists was very high during the first 50 iterations (Fig. 3a) and was directly linked to their initial position in the matrix: if they were positioned too far from resources, the diffusion rates of resources were not sufficiently high to satisfy their energetic requirements for survival. Compared to immobile K-strategists, immobile r-strategists (Fig. 3b) reproduced more easily, due to their lower reproduction threshold. As a result, a small population peak was observed between iterations 50 and 80 (Fig. 3b). Thereafter, resource limitation caused a decrease in population size as cell death rates surpassed reproduction rates. It is interesting to note that the largest immobile populations were found in the large pores. However, immobile populations persisted longer in small pores, especially at low moisture levels. The most unfavourable initial configuration for the persistence of immobile populations over time was when resources were in large, water-saturated pores.

The population dynamics of the motile populations showed an initial increase in size during the first 50 iterations (Fig. 3), which was followed by a decrease, as the populations became resource limited. The results observed for motile populations differed according to the life strategy: the population of r-strategists increased to over 100 cells/matrix while the populations of K-strategists remained more modest at fewer than 40 cells/matrix (Fig. 3c). The populations of K-strategists reached a maximum when the resources were located in small pores, whereas r-strategist populations thrived when resources were in large, water-saturated pores. Neither r- nor K-strategists thrived in the other's preferred environments. The least favourable situation for motile r-strategists was when resources were in small pores with low water content. In this scenario, the environment impeded motility, thus reducing both the rate at which resources were acquired and the reproduction rate.

Simulations in which both r- and K-strategists were present together showed that competition affected both populations, whether they were motile or immobile, from approximately the 40th iteration onwards, resulting in decreases in population size (Fig. 4). This was due to resources becoming more limited relative to the needs of the populations. The competitive interactions resulted in r-strategists dominating the matrix. As stated above, the same general dynamics were observed with a larger initial cell supply (56 r-strategists and 56 K-strategists) (Fig. S2).

The CUE differed markedly among microbial strategies and initial community sizes (Fig. 5). Overall, communities dominated by K-strategists exhibited higher CUE than either those dominated by r-strategists or mixed communities. The highest CUE values were observed in K-strategist communities with low initial cell densities. In contrast, r-strategist communities consistently showed a lower CUE under comparable conditions (Tukey: P<0.001).

Increasing the initial number of cells led to a decrease in CUE for both r- and K-strategist communities, indicating a negative effect of initial density on carbon use efficiency. Mixed communities composed of both r- and K-strategists exhibited the lowest CUE values across all scenarios. Statistical support for these effects was provided in Fig. S3.

In addition to the effects of metabolic strategy and the initial cell numbers, the location of resources in specific pores also had significant effects (F=101.53; P<0.001), as did microbial cell motility (F=138.89; P<0.001) and their interaction (F=328.03; P<0.001).

Firstly, higher CUE values (0.40±0.04 for K cells and 0.39±0.04 for r cells) were always obtained for immobile cells under low water conditions. The next highest, but significantly lower (Tukey: P<0.001), CUE values were obtained when resources were in water saturated small pores (0.36±0.07 for K cells and 0.35±0.04 for r cells). When the resources were in large pores, immobile cells showed the lowest CUE, in particular under water-saturated conditions (0.24±0.04 for K cells and 0.23±0.02 for r cells).

The motile cells with the largest CUE were observed in the large pores (Fig. 5). In the case of motile K-strategists, these were large, water-saturated pores (CUE=0.49±0.05), while in the case of motile r-strategists, large pores, whether water-saturated or not, were indifferently associated with the highest CUE (Tukey: P=0.50). In contrast to what was described for immobile cells, motile cells in small pores had reduced CUE, with the lowest CUE found in small pores with low water content (CUE=0.28±0.04 for K-strategists and CUE=0.31±0.04 for r cells).

The simulations with both r and K strategists enabled us to determine the effect of competition between the different cell types on CUE. Increasing the initial cell density significantly reduced CUE, the strategy of the cells also had a significant effect, with K-strategists having higher CUE values. The K-strategist cells were more sensitive to the presence of r-strategists. In contrast, the CUE of r-strategists was less sensitive to the presence of K-strategists in the matrices. This competitive effect was largely modulated by the initial conditions of the matrices, the size of the resource distribution pores, but also humidity (Fig. 6).

The results showed that the initial configurations of the matrix had different effects on the cells; the two most distinct configurations were those in which the pore network was saturated with water, with resources distributed only in the large pores (Fig. 6a), and in which water was limited and resources were found only in the small pores. When the pore network was water-saturated and resources were only in large pores (Fig. 6a), motile r-strategists were not affected by increasing cell density in the initial state (Tukey: P=0.32), nor by competition with motile K-strategies in the mixed situation (Tukey: P=0.99). However, this was not the case for motile K-strategists, whose CUE was reduced when there were motile r-strategists in the matrix (Tukey: P<0.001). In the case of immobile cells, r-strategists were affected at greater initial cell densities (Tukey: P<0.001), regardless of whether these were r- or K-strategists (Tukey: P=0.36). The CUE of immobile K-strategists was reduced when the initial number of K-strategists was high (Tukey: P<0.001), and reduced further when in the presence of r-strategists (Tukey: P<0.001).

When resource was in small pores in low water conditions (Fig. 6b), the CUE of r-strategists was unaffected by the increase in initial cell density (Tukey HSD: P=0.001), and did not differ from the CUE of K-strategists (Tukey: P=0.98). The effects were different for motile K-strategists, whose CUE was not affected by high initial densities of K-strategists (Tukey: P=0.99), but were affected when in the presence of motile r-cells (Tukey: P<0.001). As far as immobile cells are concerned, the effects were identical for both cell strategies, with the CUE lowered by higher initial cell densities, regardless of their identity.

Additional simulations were performed to test the sensitivity of the model to moisture level, initial resource concentration, and the individual parameters governing r- and K-strategists life histories.

Cell sensitivity to moisture was only detectable at the lowest humidity levels tested; above a water film thickness of 6 pixels, most pores were nearly saturated, reducing the differences observed between conditions. When resources were located in large pores, moisture had no significant effect on either motile or immobile cell populations or their CUE. In contrast, when resources were distributed in small pores, moisture had a marked effect: at the lowest moisture level, immobile cells reached higher mean population sizes (26.3±12.3) than motile cells (12.8±8.6, P<0.001), and immobile CUE (0.41±0.06) exceeded motile CUE (0.28±0.12, P<0.001). This pattern was reversed at higher moisture levels, where motile cells reached higher mean populations (52.4±23.6) and CUE (0.41±0.10, P<0.001), while immobile populations (16.3±15.3) and CUE (0.20±0.15) decreased (P<0.001) (Fig. S7).

Varying the initial resource concentration (10–300 resource units per resource cell) showed that at low concentrations (10–25 units), immobile cells reached higher mean populations than motile cells (P<0.0001). As resource concentration increased, motile cells increasingly outperformed immobile cells, particularly when resources were located in large pores (P<0.0001). CUE tended to be stable above a threshold concentration: motile cell CUE plateaued rapidly at 0.38±0.09 in large pores, while immobile cell CUE continued to increase with resource concentration, from 0.02±0.04 to 0.32±0.07. In small pores, CUE dynamics were similar between motile and immobile cells (Fig. S8).

Systematic variation of the four parameters governing r- and K-strategist life histories (motility cost, maintenance cost, reproduction cost, reproduction threshold) showed that pore size had a stronger effect on CUE (F=1149.13, P<0.001) than any individual life-history parameter, and also altered which parameter combinations maximised CUE. In large pores, maintenance cost was the dominant factor, explaining more than 70 % of the variance in CUE for both motile and immobile cells. In small pores, reproduction threshold was the dominant factor, explaining more than 45 % of the variance. Correlation analysis showed that the canonical r (0000) and K (1111) parameter combinations were never the most divergent in their effect on CUE, indicating non-additive interactions among parameters; these interactions accounted for a larger share of the variance in small pores than in large pores (Figs. S9–S14).

Full methodological details and statistical outputs for all sensitivity analyses are provided in the Supplementary Material.

As outlined in the introduction, spatially explicit, mechanistic models have considerably advanced our understanding of soil physical structure (Ray et al., 2017; Ritschel and Totsche, 2019; Rötzer et al., 2026; Zech et al., 2022). However, these approaches have largely developed along two separate lines: models that prioritize the physical heterogeneity of the soil matrix, generally without resolving microbial trait diversity, and trait-based microbial models that capture life-history strategies but are typically developed under homogeneous or implicit spatial conditions, with few frameworks operating at the interface of the two. A notable exception is the DEMENT model (Allison, 2012), which combines microbial trait diversity with spatially explicit dynamics; however, DEMENT is typically implemented on simplified or generic spatial grids rather than on pore-scale structures derived from empirical imaging data. The framework presented here contributes to bridging this gap by explicitly coupling a pore-scale spatial structure, derived from X-ray tomographic imaging, with microbial functional traits (r- and K-strategists, distinct motility types) and local, rule-based interactions. Compared with models that prioritize physical heterogeneity but omit microbial trait diversity (e.g., Ray et al., 2017; Ritschel and Totsche, 2019; Zech et al., 2022), our approach makes it possible to examine how microbial life-history strategies interact with, and are constrained by, realistic pore-scale structure. Conversely, compared with trait-based microbial models developed under homogeneous conditions (Panikov, 1999), our framework allows resource accessibility and microbial performance to emerge from explicit spatial constraints rather than being assumed uniform. At the same time, this focus on spatial structure and microbial traits comes at the cost of the biochemical and physiological detail found in more mechanistic models suggesting that future work coupling these complementary modeling traditions could yield a more comprehensive representation of microbial carbon cycling in structured soils.

The model results suggest that the CUE of microorganisms is largely determined by the pore environment. It has already been shown that organic matter turnover varies across different soil pore sizes (Chenu et al., 2025; Li et al., 2024; Maestrali et al., 2026; Pfeiffer et al., 2001; Xia et al., 2022) because the microbes that dominate in these different pore sizes have strategies that are adapted to their microenvironment such as fluctuations in humidity (Ceriotti et al., 2022; Kravchenko et al., 2020), available space (Chenu et al., 2001; Gupta and Germida, 1988), nutrient availability (Bouckaert et al., 2013) and predation pressure (Erktan et al., 2020; Wright et al., 1995). This study suggests that the geometry of the pores themselves selects the microbes capable of colonising the environment. Indeed, when confined in micropores, organic matter is physically protected not only through exclusion by pore size, but also due to pore tortuosity, which modulates both substrate diffusion and microbial access. In our simulations, we observed that even when motile cells were close to small pores, their effective access to resources was limited by the tortuosity of the pore network. Although motile bacteria are capable of detecting chemical gradients, the establishment and continuity of these gradients depend on the connectivity of the pores and the continuity of the water film. In highly tortuous or poorly connected pore networks, chemical gradients can be weak, discontinuous, or spatially misleading, limiting the ability of motile cells to effectively reach nearby resource areas. For example, a flagellated bacterium may not effectively reach an organic matter particle located just a few micro-meters away if the accessible pathways are long or disconnected, whereas a non-motile bacterium already positioned nearby can exploit the resource via diffusion without incurring the energy cost of motility. These results are consistent with previous studies showing that tortuosity significantly influences bacterial mobility by increasing the complexity of the paths that bacteria must follow in porous media, thereby reducing their effective diffusion and transport rates. Using magnetic resonance imaging, it has been shown that the swimming behavior of bacteria such as Pseudomonas putida is strongly impeded by the pore structure (Olson et al., 2005). Bacterial motility patterns, particularly path lengths and turning angles, interact with surface adhesion in porous media, further increasing apparent tortuosity and affecting bacterial movement (Narayanaswamy et al., 2009). Experimental and modeling studies have demonstrated that smaller pores and more tortuous paths reduce the average free-swimming distances of bacteria, hindering their self-propulsion (Dehkharghani et al., 2023; Duffy et al., 1995).

This work suggests that the water regime (low water or water-saturated) modulates the influence of pore size on microbial communities. Under conditions with low water and organic matter in the macropores, motile cells achieve a higher CUE, suggesting that water creates connected pathways without completely restricting motility. This configuration allows motile cells to access resources while experiencing fewer diffusional constraints: resources diffuse only in the available water film and are therefore less “diluted” compared to a situation where pores are completely saturated with water. Our simulations suggest that r-strategist populations, characterized by their rapid growth, exploit this situation particularly well by efficiently colonizing accessible surfaces.

Water connectivity, rather than the nominal spatial distribution of resources, appears to be a key factor shaping resource accessibility at the pore scale, and this may help explain the effects of pore size reported above. Indeed, while the spatial distribution of organic resources remains identical across moisture treatments (Fig. 1), the volume of water accessible to a given cell, and therefore its access to dissolved substrates, differs substantially depending on the extent of water connectivity. Under low-water conditions, microbial cells and dissolved substrates are largely restricted to the remaining water films, so that resource diffusion, and hence the resupply of dissolved carbon, is limited to the connected volume of water film surrounding a given cell. Isolated, disconnected water pockets can still sustain local microbial activity as long as locally available resources persist, but they cannot receive further resource input from the rest of the pore network. This constraint affects motile and immobile strategists differently. Because active movement is itself limited to the same connected water film as diffusion, motile cells cannot bypass disconnected zones; however, within a connected volume, they can actively explore and relocate toward locally resource-rich areas rather than depending passively on diffusion to bring resources to them. Immobile cells, in contrast, remain confined to their initial location and depend entirely on the volume of water connected to that specific position for access to dissolved carbon. Consequently, immobile cells located in small or poorly connected water volumes under low-moisture conditions are more likely to become resource-limited, independently of the amount of resource nominally present in their vicinity. Under water-saturated conditions, by contrast, connected water volumes are maximized throughout the matrix, and the distinction between accessible and inaccessible resources becomes largely irrelevant, allowing both motile and immobile strategists more uniform access to substrates. Taken together, these considerations suggest that the differences in microbial dynamics and CUE observed between moisture treatments arise primarily from changes in the volume of connected water mediating resource accessibility, rather than from differences in the underlying spatial distribution of resources, and that this effect is expected to be particularly pronounced for immobile microbial types.

The simulations suggest that motility does not confer a universal competitive advantage, but instead leads to different outcomes depending on pore-scale environmental conditions. In well-connected, carbon-rich environments, motile cells benefit from increased access to spatially distributed resources, resulting in higher growth and competitive success. In contrast, in poorly connected or resource-limited environments, the energetic costs associated with motility reduce carbon allocation to growth, leading to lower CUE compared to immobile strategies. This trade-off reflects the additional biosynthesis, repair, and operating costs associated with flagellar motility (Kempes et al., 2017), and is consistent with experimental observations showing reduced growth under carbon limitation when motility is constitutively expressed (Lisevich et al., 2025). Here, we show that pore-scale physical constraints modulate this trade-off, determining when motility enhances or limits microbial efficiency.

In fully saturated macropores, bacteria can exhibit efficient “run-and-tumble” movements with long ballistic trajectories (Zhang et al., 2024). In contrast, in micropores, physical confinement forces motile cells to adopt a less efficient “swim-and-stop” or “jump-and-trap” movement, significantly eroding the advantages of motility. Because motility itself carries an intrinsic energetic cost, this loss of efficiency in confined spaces translates into a lower resource return per unit of energy invested in movement. The chemotaxis machinery further amplifies these resource costs (substrate detection, flagellar or ciliary activity, etc.), compounding the energetic burden of motility in already unfavourable micropore environments and further reducing net CUE (van Albada et al., 2009; Hoffmann et al., 2017). In this context, the relationship between motility and carbon distribution is crucial. As this study demonstrates, motile bacteria have maximum CUE when organic matter is located in macropores, where chemotactic movement towards carbon gradients is optimal. When organic matter is confined to micropores, the motility advantage decreases significantly because bacteria cannot physically access these areas. In direct contrast, in our simulations, sessile bacteria dominate in micropores, maximising CUE by minimising motility-related costs (13 %–19 % of the cellular resource budget in our models) and redirecting resources towards stress tolerance and growth efficiency. Indeed, in low-connectivity environments, Jin and Sengupta (2024) observe that passive dispersal (expansion through growth, attachment to motile vectors) can surpass active motility in resource efficiency, explaining the good performance of immobile microbes when organic matter is confined to micropores. Thus, since the prevalence of bacterial flagellar motility is strongly associated with soil carbon accessibility (Ramoneda et al., 2024), cell motility should be particularly associated with a copiotrophic lifestyle, which explains the large populations observed in the case of motile r-strategists with resource distribution in the largest pores. Ultimately, the environmental dependencies observed (pore size, moisture) suggest that microbial communities are evolving towards motility patterns optimised for their specific environments. Microbial motility is not universally advantageous, but rather represents a finely tuned adaptation to the interaction between pore architecture, carbon heterogeneity and resource trade-offs, with cascading effects on the carbon cycle at the ecosystem scale. Furthermore, under favourable moisture conditions, active cell motility plays a key role in the early colonisation process: bacterial communities with cellular motility are pioneers and often the first to colonise new surfaces, followed by other bacteria with other functional traits (Wang et al., 2025), leading to the colonisation of new microenvironments in the soil.

Our results suggest that the distinction between motile and immobile microbial strategies remains relevant even when considering larger spatial scales, as the benefits of motility are highly context-dependent rather than universal. In our simulations, the effectiveness of motility depends strongly on pore connectivity, water-film continuity, tortuosity, and the energetic costs associated with movement, indicating that microbial access to resources cannot simply be reduced to an optimization process independent of microbial traits. Instead, spatial structure determines when motility becomes advantageous relative to immobile strategies, and when it does not. At coarser spatial resolutions, however, it may not be necessary to explicitly simulate individual microbial movement: the pore-scale effects highlighted here could instead be represented indirectly through emergent parameters capturing substrate accessibility, spatial connectivity, diffusion limitation, or temporal delays in resource exploitation (i.e., a form of lag phase). Exploring how such emergent representations could substitute for an explicit motile/immobile distinction in larger-scale models represents a promising direction for future work, though it falls beyond the scope of the present study.

The simulations show that pore architecture fundamentally structures microbial ecological strategies. It is known that micropores constitute preferential niches for oligotrophic taxa (Acidobacteriales, Frankiales, Firmicutes) specialised in the processing of complex carbon compounds (Li et al., 2024). These organisms have specific morphological adaptations, reduced cell sizes and filamentous shapes, that maximise nutrient absorption under resource-limited conditions (Chen et al., 2022).

Our simulations indicate that the environment, and in particular pore architecture, can strongly modulate the influence of life cycle parameters on carbon use efficiency (CUE). While r and K strategies represent well-established ecological archetypes (Pianka, 1970; Li et al., 2024), the conditions under which they maximise CUE are not universal but depend heavily on pore size and cell motility. Using a spatially-explicit model, our study shows that in large pores, maintenance cost emerged as the dominant parameter, influencing CUE outcomes differently for motile and immobile cells (Fig. 5). In small pores, the reproduction threshold played a central role. These results indicate that the effect of each parameters cannot be dissociated from the environmental context: the same parameter can increase CUE under certain pore conditions, but cause it to be reduced under others. Similar results have been reported in microbial ecology, where the spatial heterogeneity of soil pores is recognised as a determining factor in microbial metabolic strategies (Nunan et al., 2020; Portell et al., 2018).

Carbon distribution in the soil pores network appears to be the main mechanistic link between pore size and the selection of microbial strategies. Chen et al. (2022) showed that carbon-rich rhizospheric soils favour copiotrophs, while poor environments select oligotrophs, which accentuates contrasts in resource availability between microsites. The microarchitecture of pores induces marked biogeochemical gradients over a few micrometres: Li et al. (2024) measured significant variations in oxygen, water film thickness and diffusion between pore classes, with macropores exhibiting more dynamic physicochemical conditions, while micropores maintain more stable but resource-poor environments. This trade-off between stability and availability is a fundamental axis for sorting ecological strategies, with r-strategists exploiting large, variable but rich pores, and K-strategists specialising in small, stable but limited pores. These spatio-temporal dynamics are also consistent with the phenomenon described by Zelenev et al. (2000) in the BACWAVE model, which simulated `bacterial waves' generated by the movement of a nutrient pulse and translated into spatial waves of microbial growth and mortality along roots. Similarly, our model highlights a similar phenomenon where motile cells, particularly r-strategists, move along resource gradients formed on pore walls, producing wave-like dynamics of colonisation and consumption comparable to those observed empirically in the rhizosphere.

Overall, the sensitivity analyses (supplementary data) showed that microbial responses strongly depended on interactions between pore structure, moisture conditions and microbial traits. Importantly, no single parameter combination emerged as universally optimal across all simulated conditions. Instead, the results suggest that microbial performance arises from trade-offs between physiological traits in interaction with pore-scale environmental structures. These findings support the relevance of the selected parameter ranges while also indicating that the framework could be extended beyond the classical r/K dichotomy. We chose to focus specifically on r- and K-like strategies because they represent well-established ecological archetypes that provide a robust conceptual framework for exploring microbial trade-offs. Their widespread use in ecology makes them particularly suitable for investigating how different life-history strategies interact with pore-scale environmental heterogeneity.

Our cellular automaton succeeds in reproducing several well-documented mechanisms of soil microbial ecology, such as the trade-offs between r and K strategies, the different dynamics of motile and immobile cells, and the strong selective role of pore architecture on CUE. This convergence between model outputs and empirical knowledge highlights the potential of minimalistic, mechanistic approaches to capture emergent ecosystem patterns without requiring a high degree of parametrization. By making explicit the links between cellular traits, spatial constraints and carbon cycling, our framework provides a transparent test bed for exploring hypotheses that are difficult to address empirically at the microscale. Beyond confirming the relevance of r and K archetypes, our results pave the way for extending the model towards a broader trait-based framework, such as the Y-A-S triangle (Fig. S15) (Ho et al., 2013; Malik et al., 2020), in order to represent the continuum of microbial strategies observed in soils.