the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 12 Feb 2026
ForEdgeClim v1.0: a 3D process-based microclimate model incorporating vertical and lateral energy fluxes to simulate forest edge-to-core transitions
Abstract. Forest microclimates play a fundamental role in regulating biodiversity, ecosystem functioning, and forest resilience to climate change. However, most existing microclimate models focus on vertical processes and neglect lateral energy exchanges, limiting their ability to represent forest edge effects. This is important because of forest fragmentation dynamics and because up to 20 % of the global forest cover is less than 100 m from an edge.
Here, we introduce ForEdgeClim, a new process-based microclimate model implemented as a publicly available open-source R package that is able to simulate air and surface temperature at high spatial resolution along the forest edge-to-core continuum (here demonstrated at 1 m resolution). By explicitly leveraging high-resolution 3D forest structural data (e.g., derived from terrestrial laser scanning), the model represents a substantial advance over existing approaches that rely on simplified or spatially aggregated canopy descriptions. Building on this detailed structural representation, ForEdgeClim couples meteorological forcing with a physically based energy balance framework – including shortwave and longwave radiation, sensible and latent heat fluxes, and soil heat exchange – to resolve microclimate dynamics in three dimensions. Radiative transfer is represented using a two-stream approximation in both vertical and lateral directions, whereas the full energy balance is iteratively solved within a 3D voxel grid to account for coupled radiative and heat flux exchanges.
A Sobol sensitivity analysis indicates that heat-transfer processes dominate local air temperature dynamics (≥ 67 % of the total model output variance), whereas radiative transport plays a stronger role in controlling surface temperature and spatial temperature heterogeneity. These insights informed a targeted calibration of key model parameters. Model performance was evaluated using high-frequency in situ temperature measurements, with forest structural information derived from terrestrial laser scanning data, collected along a forest edge-to-core transect in a temperate forest in Belgium. Validation shows that ForEdgeClim successfully reproduces observed edge-to-core temperature gradients and fine-scale spatial variability in air temperature (R2 ≥ 0.87, RMSE ≤ 2.01 °C).
By combining high-resolution structural information with a physically grounded yet computationally efficient framework, ForEdgeClim bridges the gap between simplified empirical microclimate models and computationally intensive ray-tracing approaches, which typically lack a full energy balance formulation. The model thus provides a versatile platform for microclimate research, ranging from biodiversity and habitat modelling to studies of forest-climate interactions under a changing environment, especially where edge effects play a key role in fragmented landscapes.
Competing interests: At least one of the (co-)authors is a member of the editorial board of Geoscientific Model Development.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(40977 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 09 Apr 2026)
- RC1: 'Comment on egusphere-2026-649', Ilya Maclean, 17 Feb 2026 reply
-
RC2: 'Comment on egusphere-2026-649', Anonymous Referee #2, 03 Mar 2026
reply
General comments:
ForEdgeClim addresses a genuine gap in the microclimate modelling landscape. The 3D voxel framework, the coupling with high-resolution TLS data, and the transparent sensitivity and calibration analyses represent real contributions. The authors deserve credit for tackling a difficult problem and for making their code and data openly available.However, in its current form, the model is at a rather early stage. The radiative transfer component is reasonably developed, but the absence of wind-driven processes, the lack of leaf-wood separation despite having TLS data ideally suited for it, the highly simplified soil heat flux, the temperature-only output, and the evaluation at a single site all limit the model's physical realism and demonstrated generalisability. Several of the "semi-empirical" parameters (im, is, gf) are effectively doing the work of unresolved physical processes, and the future development discussion remains too vague to assess the model's trajectory.I want to emphasise that none of these issues are fatal. ForEdgeClim v1.0 provides a working framework that can serve as a foundation for meaningful advances. But the manuscript, as currently written, does not sufficiently distinguish between what the model achieves now and what it aspires to achieve. A more honest framing of the current capabilities, combined with a concrete and technically specific development roadmap, would substantially strengthen the paper and provide much greater value to the community.I encourage the authors to continue this important work. The problem they are tackling: fine-scale microclimate modelling in fragmented forests, is one of the key challenges in forest ecology and biodiversity science. With improvements, ForEdgeClim has the potential to become a genuinely useful tool for the community.Major comments:
1. Absence of wind and turbulent transportMy most fundamental concern is that wind speed and turbulent transport are entirely absent from the model. There is no representation of advection, momentum transfer, or wind-driven mixing anywhere in ForEdgeClim. The sensible heat exchange (Eq. 15) relies on a static convection coefficient gf that does not depend on wind speed, atmospheric stability, or any aerodynamic quantity. Air-to-air heat transfer between voxels is modelled as molecular conduction (Eqs. 13-14), which is orders of magnitude weaker than turbulent heat transport in real forest environments.This omission is particularly consequential for a model that specifically targets forest edges. The reason edges are microclimatically distinct is the combined effect of lateral radiation penetration wind exposure. Warm or cold air is advected horizontally from adjacent open landscapes into the forest, and the penetration depth of this thermal signal is largely governed by canopy drag on the airflow. Without wind, the model must rely on the static "distance of influence" parameters (im, is) to prescribe how far the macroenvironmental signal propagates -- but this penetration depth is physically set by wind-driven processes. The consistent convergence of im towards its upper bound (60 m) in calibration may be a symptom of the model stretching a static parameter to compensate for a missing dynamic process.Furthermore, the Priestley-Taylor formulation for latent heat (Eq. 16) is justified as appropriate when advection is negligible -- but at forest edges, advection is precisely not negligible.Given the above, I feel the current model is more accurately described as a 3D radiative-thermal equilibrium model rather than a comprehensive microclimate model. The term "microclimate" in the biophysical literature carries the expectation of capturing the dominant atmospheric processes that shape local climate, and wind is unquestionably one of them.I would suggest reframe the model's scope more precisely. The title or subtitle could be adjusted to reflect the radiative-thermal nature of the current framework (e.g., "...incorporating vertical and lateral radiative and thermal fluxes..."). Or at minimum, state clearly in the abstract and introduction, not just in Section 6.5, that wind processes are excluded. Discuss quantitatively when the no-wind assumption is expected to break down. The authors have ~2 years of meteorological station data. Checking whether model residuals correlate with observed wind speed is a straightforward analysis that would either support the no-wind assumption or clearly delineate the model's domain of validity. Also provide a concrete technical plan for incorporating wind (see Major Comment 5 below) would be another way to imrpove.2. Lack of leaf-wood separationThe model assigns a single bulk density (ρ) to each voxel and applies uniform optical properties (ω, β, β0) across all structural elements, without distinguishing leaves from woody material. This is acknowledged in Table A5 but receives almost no discussion in the main text.This simplification has cascading physical consequences: Shortwave scattering properties of leaves and wood differ substantially, particularly in the near-infrared band. A single scattering coefficient ω cannot accurately represent both. Only leaves transpire, yet the Priestley-Taylor LE is applied to the entire "forest surface." In winter for this deciduous forest (oak, beech, ash), the canopy is essentially bare wood, but the model continues to compute LE as if transpiring surfaces are present. Voxel densities are scaled monthly using PAI ratios, but optical properties are held constant year-round (Table A5). In a deciduous forest, the transition from leaf-on to leaf-off fundamentally changes the scattering regime -- this is not captured. I could list more, but i just stop here. Seperate leaf and wood component is critical, even in CLM, which is at far coarser spatial resolution, it maintains separate optical properties for leaves and stems (rhol/rhos, taul/taus, etc.) and weights them by their respective area fractions (LAI and SAI) before entering the two-stream calculation.What makes this particularly notable is that the TLS data used in this study are among the most suitable data sources currently available for leaf-wood separation at fine scales. Established methods exist for classifying TLS point clouds into leaf and wood components using local geometric features (planarity, linearity), return intensity, or machine learning approaches. Moreover, the co-author team includes leading expertise in TLS-based forest structural analysis, and the model's RTM is based on ED2.2, which explicitly separates leaf and wood area indices with distinct optical properties.3. Temperature-only output for a model framed as a "microclimate model"ForEdgeClim predicts only temperature (air, surface, soil), yet in the microclimate literature, humidity (vapour pressure deficit, relative humidity) is widely recognised as a co-driver of ecological processes, species distributions, and evaporative demand. The Priestley-Taylor formulation implicitly involves humidity through the psychrometric constant, but no humidity state variable is tracked or output. For a model positioned for ecological applications (Section 6.7), this is a meaningful gap. I would like to see more discussions about whether a diagnostic humidity estimate could be derived from the existing framework. If humidity prediction is planned for a future version, state this clearly and outline how it would be implemented.4. Calibration only against air temperature, leaving surface temperature unconstrainedCalibration is performed exclusively against air temperature from TOMST sensors at 15 cm height, yet forest surface temperature is a key prognostic variable (Figs. 2c, 4b). The Sobol analysis reveals that gf dominates surface temperature variance (8.7-32%) but is excluded from calibration because air temperature is insensitive to it. This means surface temperature predictions are essentially unconstrained by observations. I would suggest adding an uncertainty envelope for surface temperature predictions based on the parameter ensemble spread from the Sobol analysis; also could discuss what observational data (e.g., thermal cameras, radiometers) would be needed for proper surface temperature validation.5. The future development roadmap needs to be concrete and specificSection 6.5 identifies several missing processes: wind, improved soil heat flux, advanced radiative transfer, but presents them as vague aspirations ("future work could...") without any technical specificity. Given how fundamental some of these missing components are, this is insufficient for a v1.0 model description paper. Readers and potential users need to assess whether ForEdgeClim has a viable path toward becoming the comprehensive tool it aspires to be, or whether its current architecture would require fundamental restructuring.Minor Comments:400 Latin hypercube samples for 25 parameters may yield unreliable higher-order Sobol indices. Convergence of total-order indices typically requires substantially more samples. I would suggest report confidence bounds on the Sobol indices, or run a convergence check (e.g., 200 vs. 400 samples) to demonstrate stability.The convergence of ks and im to their upper bounds in multiple calibrations is acknowledged but attributed only to overly restrictive priors. However, a parameter hitting its bound may also indicate structural model misspecification: a missing process being compensated by inflated parameter values (e.g., im compensating for absent wind-driven advection). Both interpretations should be discussed.The model is run at 1 m resolution but resolution is stated to be configurable. In the introdcution of the model, the author claimed that the resolution can be adjusted based on need but no test at alternative resolutions is provided.Holding ω, β, β0 constant year-round in a deciduous forest is a strong assumption. Even without leaf-wood separation, allowing these parameters to take seasonally varying values (e.g., two sets: leaf-on and leaf-off) would better represent reality. See also Major Comment 2.Runtime of 20 s to 2 min per hourly timestep is reported, but how this scales with transect length or voxel count is not discussed. From a user side, I am curious to know whether this is feasible on a standard workstation.Citation: https://doi.org/10.5194/egusphere-2026-649-RC2 -
RC3: 'Comment on egusphere-2026-649', Run Zhong, 04 Mar 2026
reply
General Comments：
The manuscript introduces ForEdgeClim v1.0, a timely and computationally innovative 3D process-based model designed to simulate the complex microclimates of fragmented forest landscapes. The authors should be commended for their ambitious integration of high-resolution Terrestrial Laser Scanning (TLS) data into a voxel-based energy balance framework, which addresses a long-standing limitation in traditional 1D microclimate modeling. The open-source nature of the code and the successful demonstration of lateral thermal gradients provide a valuable foundation for the forest ecology community.
However, while the structural framework is impressive, the physical representation of certain dynamic processes requires further refinement. A primary concern is the current model’s reliance on several static assumptions—most notably regarding vegetation phenology, radiative transfer anisotropy, and the absence of advective heat transport. For a model aiming at "high-resolution" 3D simulation, the gap between structural complexity (via TLS) and process simplification remains significant. Major revisions are suggested to either incorporate these dynamic factors or, at a minimum, provide a rigorous discussion of how these limitations constrain the model’s applicability across different seasons and environments.
Specific Comments：
Point 1: The Paradox of Structural Simplification
The decision to omit the Clumping Index and LAD as independent parameters is a critical weakness. While the authors may frame this as a "trade-off" between computational efficiency and data availability, this logic is inherently flawed for a model utilizing TLS data. TLS is precisely the tool meant to resolve fine-scale clustering and structural orientation. By treating voxels as homogeneous turbid media, the model misses the "gap fraction" dynamics essential for edge effects, where lateral light penetration is governed by canopy gaps rather than bulk density. This simplification leads to a systematic underestimation of radiation transmission to the sub-canopy and biases the resulting thermal equilibrium.
Point 2: Incorporating Seasonal Phenology into Optical Properties
For a "process-based" model applied to deciduous forests, the assumption of static leaf optical properties (e.g., constant reflectance in the 400–700 nm range) may limit accuracy during transitional periods. In reality, leaf spectral signatures and Leaf Area Density (LAD) shift significantly from summer greenness to winter senescence. While I recognize that on-site spectral measurements may not always be available, a high-resolution 3D model should ideally account for this temporal resolution. At a minimum, the authors should explicitly discuss this limitation and its potential impact on energy balance during spring and autumn. If empirical leaf-off/leaf-on scaling is not feasible in v1.0, this should be identified as a critical area for future development.
Point 3: Physical Consistency of the 3D Radiative Transfer Scheme
The application of the two-stream approximation in a 3D voxelized framework at forest edges introduces a fundamental physical inconsistency. The classical two-stream theory (as presented in Eqs. 2-5) was originally derived for 1D horizontally homogeneous media, assuming an isotropic diffuse radiation field. However, at a forest edge, the radiative environment is characterized by extreme anisotropy due to the lateral influx of direct and diffuse solar radiation.
In Section 2.1.1 and Figure 1, the model handles lateral fluxes (Φx and Φy) by extending the vertical two-stream equations. While this approach simplifies the computation, it may not fully capture the directional nature of photon interception at the edge, which is governed by the specific geometry of canopy gaps and the solar position. Although the authors describe this as a "3D process-based" model, the current use of two-stream approximations for lateral fluxes effectively treats the complex 3D light field as a series of coupled 1D problems. This simplification could potentially limit the model's ability to accurately resolve the extinction coefficient and the resulting thermal gradients in highly heterogeneous transition zones.
Given that the model already operates on a 3D voxel grid, more advanced schemes such as the Discrete Ordinates (DO) method or Spherical Harmonics expansion would be more robust for discretizing the angular domain and capturing the non-isotropic nature of the edge radiation field. At a minimum, I recommend that the authors provide a more detailed justification or a sensitivity analysis to demonstrate the validity of the isotropic assumption, particularly at the immediate forest edge where the thermal gradient is most pronounced.
Summary Recommendation:
The model shows great promise, but the transition from a "static structural model" to a "dynamic process model" requires a more rigorous treatment of vegetation phenology and fluid dynamics.
Citation: https://doi.org/10.5194/egusphere-2026-649-RC3 -
RC4: 'Comment on egusphere-2026-649', Vivienne Groner, 11 Mar 2026
reply
General comments
This manuscript presents a new model for a 3D process-based microclimate model that incorporates both vertical and lateral energy fluxes to simulate forest microclimates. The authors provide a detailed description of the model structure, governing equations, and parameterization, together with a calibration procedure and a demonstration case study along a core-to-edge transect in a temperate forest in Belgium. The aim of the model is to provide a tool to better study microhabitats in forest edge-to-core transition zones.
The paper addresses a very timely and relevant scientific modelling question within the scope of GMD, namely the role of microclimate in governing ecological processes, and presents a novel modelling tool suitable for addressing important research questions in this area and in the broader scope of EGU. The work represents a substantial advancement in the field of microclimate modelling, as it brings in a whole new dimension/set of processes, and opens up potential for new research in ecology and climate science.
The methods and assumptions appear valid and are clearly outlined. The manuscript provides a clear and very detailed description of the governing equations, parameterizations, calibration and validation procedures, and overall model structure, illustrated by a schematic figure. In addition, the extensive supplementary material and the reference to the code repository on GitHub further support reproducibility and transparency. The code itself is well written and documented, meeting a high code quality standard. This level of detail should allow other researchers to reproduce and build upon the work. The assumptions and limitations of the approach are discussed in detail and are generally supported by previously published studies. The presented results are sufficient to support the interpretations and conclusions presented in the manuscript. The figures are generally well chosen and support the main messages of the paper.
The overall presentation is well structured and clear, and I enjoyed reading the manuscript. The title clearly reflects the contents of the paper, and the abstract provides a concise and complete summary of the study. The language is generally fluent and precise, although in a few places sentences appear a bit unclear or inconsistent , potentially due to the use of LLM for tidying the final version. Mathematical formulation, symbols, abbreviations, and units are generally defined and used appropriately. The number and quality of references are appropriate and the cited literature is recent. The supplementary material is extensive and of good quality, and appropriately complements the methods presented in the main text. I recommend minor revisions.
Specific comments
1. Time step / temporal representation
It was not entirely clear to me what temporal resolution the model operates at. From the introduction, my understanding is that the model assumes an equilibrium state; however, the manuscript also mentions a time interval of one hour. It would be helpful if the authors could clarify how time is represented in the model. Following that, I would be interested how would the model perform over multiple time steps potentially integrated with other ecological models, like a forest gap model.
2. Temperature only
Right now, the model only outputs temperature, which is common in this field. But since microclimates usually involve more than just temperature, it could be interesting to extend it to other variables too, like relative humidity or wind speed. Could you comment on the potential for the model to do so?
Citation: https://doi.org/10.5194/egusphere-2026-649-RC4
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 208 | 90 | 19 | 317 | 30 | 45 |
- HTML: 208
- PDF: 90
- XML: 19
- Total: 317
- BibTeX: 30
- EndNote: 45
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
There is undoubtedly a strong and growing need for robust microclimate modelling approaches, particularly in forested systems where fine-scale thermal heterogeneity can strongly influence ecological processes. The manuscript addresses an important problem, and the attempt to develop a tractable, process-based framework is welcome.
However, the current presentation would benefit from greater physical clarity, closer engagement with the extensive microclimate modelling literature, and more precise definition of terms and units. Throughout the manuscript, model equations are presented without consistently specifying the units of each term. For a physically based model, dimensional consistency is important, and equations would benefit from explicit unit definitions (e.g. W m⁻², m s⁻¹, J m⁻³ K⁻¹, etc.) to avoid ambiguity and ensure reproducibility.
Specific comments:
Lines 100–109 – Iterative solution of the energy balanceIt is not made clear why the surface energy balance must be solved iteratively. In many comparable contexts, the system can be solved more efficiently using a Penman–Monteith formulation or related bulk-transfer approaches. The authors should clarify why iteration is necessary here, and whether a more computationally efficient analytical solution is possible.
Section 2.1.2 – Below-canopy air temperatureThe physical treatment of below-canopy air temperature is fairly questionable. It is now well established that Lagrangian approaches to fluid mechanics are required to rigorously describe scalar transport within and below plant canopies (see the seminal work of Raupach and subsequent developments). Air temperature at a given point below the canopy is not determined solely by local leaf temperature or local energy exchange. Rather, it reflects the cumulative influence of upstream canopy elements through advection and turbulent mixing. In other words, the temperature field represents an integrated effect of heat exchange along air parcel trajectories as they move through and downwind of the canopy. The current formulation appears to treat temperature as a locally determined variable, which risks neglecting this trajectory-dependent behaviour. The authors need to explain why a local formulation is sufficient in this context. In addition, foliage density does not appear to influence the formulation. If fluxes are expressed in standard units (e.g. W m⁻²), canopy leaf area density or LAI should explicitly affect the magnitude of exchange. As written, it is unclear how foliage density affects the flux exchange between the canopy and the air.
Sections 2.1.3 and 2.2.3 – Soil surface temperature and ground heat fluxThe soil surface temperature equation appears dimensionally inconsistent. If fluxes are expressed in W m⁻², there should be no need to introduce an explicit cross-sectional area term. Greater clarity on units is needed here. The assumption that temperature at 8 cm depth can be treated as fixed is problematic. At this depth, temperatures exhibit clear diurnal and seasonal variability. One typically needs to approach depths of ~2 m before assuming quasi-constant temperature. Without a multi-layer soil heat transfer model, soil surface temperature cannot be derived mechanistically. As written, the approach appears to depend on continuous measurement at 8 cm depth, which limits transferability. Thermal conductivity (k) is also treated as constant, yet it varies strongly with soil moisture content. I missed discussion of this issue.
In Section 2.2.3, the ground heat flux method will not reproduce the well-established quarter-cycle phase shift (diurnal or seasonal) between surface radiation forcing and subsurface heat flux. Instead, fluxes appear to mirror radiation directly, meaning the heat storage effect is not properly represented. It would be worth exploring the de Vries / Van Wijk method as described in Campbell & Norman, which provides a more physically consistent treatment of soil heat storage consistent with the framework presented in this paper
Section 2.2.4 – Sensible heat fluxThe terminology becomes imprecise with respect to thermal energy transfer from forest surfaces to the surrounding air. You introduce a forest–air convection coefficient, g, but in conventional notation g typically denotes leaf boundary layer conductance. The discussion also refers to molecular diffusion, whereas in forest environments free and forced laminer convection dominate at the leaf surface., and turbulent heat transfer is often critical within the air. Given the method used to compute air temperature, turbulent exchange likely needs to be explicitly accounted for sperate to leaf boundary layer conductance. The authors should clearly define the physical meaning of g, specify its units, and describe how it is parameterised and scaled.
Section 2.2.5 – Latent heat flux. This section is also problematic. The discussion focuses on evaporation, but in forest systems transpiration typically represents a substantial component of latent heat flux. Transpiration is controlled by stomatal conductance, which does not appear to be included in the formulation. The manuscript argues that evaporation is primarily driven by available radiative energy and surface humidity. However, radiative energy drives evapotranspiration both through direct surface energy availability and via its effects on PAR and stomatal conductance. Only the most direct pathway appears to be represented. Ignoring stomatal control risks substantial overestimation of latent heat flux, particularly under conditions where stomata are partially closed.Some of the limitations outlined above would be more easily reconciled if the model demonstrated strong predictive performance. Figure 3 is interesting, but the discrepancies between modelled and measured values appear large relative to the effect sizes being examined. In several cases, the measured differences between forest interior and core are smaller than the deviations between modelled and observed values. This raises concerns about whether the model can reliably resolve the spatial contrasts it seeks to quantify.
Overall assessmentThe model is not without merit and addresses an important problem and does seem to perform moderately well. However, it rests on several simplifying assumptions that limit its physical realism. Greater clarity on units, stronger engagement with existing microclimate modelling literature, and more rigorous treatment of canopy turbulence, soil heat storage, and stomatal control would substantially strengthen the manuscript.