Maarten van Reeuwijk
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Maarten van Reeuwijk has a total of 166 co-authors. Here are the top co-authors by number of shared publications:
- Christopher Edward Wilson
Department of Civil and Environmental Engineering, Imperial College London, London SW7 2AZ, United Kingdom
19 shared publications - Dipanjan Majumdar
Department of Civil and Environmental Engineering, Imperial College London, London SW7 2AZ, United Kingdom
18 shared publications - Sam Oliver Owens
Department of Civil and Environmental Engineering, Imperial College London, London SW7 2AZ, United Kingdom
18 shared publications Paul Bartholomew
16 shared publications- Claire E. Heaney
Department of Earth Science and Engineering, Imperial College London, SW7 2AZ London, UK
10 shared publications Ruiyue Huang
Department of Civil and Environmental Engineering, Imperial College London, SW7 2AZ London, UK
9 shared publications
Latest Publications
- Open Access
- 21 September 2026
- Published by Copernicus GmbH in Geoscientific Model Development
- Vol. 19 (18) , 8801-8814
Parameter estimation for land-surface models using Neural Physics
We propose a novel inverse-modelling approach that estimates the parameters of a simple land-surface model (LSM) by assimilating data into a differentiable, physics-based forward model formulated using convolutional operations. The governing equations are expressed within the Neural Physics framework, allowing direct gradient-based optimisation of time-dependent parameters without the need to derive and maintain adjoint formulations. The model parameters are estimated by minimising the mismatch between model predictions and synthetic or observational data. Although differentiability is enabled through machine-learning libraries, the forward model itself remains entirely physics-based and neither the forward model nor the parameter estimation procedure involve training. To evaluate the approach, we first generate synthetic observations of soil temperature by running the forward model with known parameter values and subsequently treat these parameters as unknown in an inverse problem. We show that observations of soil temperature at a single depth are insufficient to reliably constrain the model parameters. Using observations at two depths, however, does yield reliable parameter estimates, although the individual contributions of latent and sensible heat fluxes cannot be distinguished. We also apply the approach to urban flux tower data from Phoenix, United States, and show that the thermal conductivity, volumetric heat capacity and the combined sensible-latent heat transfer coefficient can be reliably estimated whilst using an observed value for the effective surface albedo. The resulting model accurately predicts the outgoing longwave radiation, conductive soil fluxes and the combined sensible-latent heat fluxes, demonstrating that the Neural Physics framework can be used to accurately determine the parameters of the particular LSM used here. This model is intentionally simple and does not include, for example, a subsurface moisture model. This simplicity facilitates an exploration of parameter identifiability, confounding and equifinality.
- 21 September 2026
- Published by Copernicus GmbH in Geoscientific Model Development
- Vol. 19 (18) , 8801-8814
Open AccessParameter estimation for land-surface models using Neural Physics
We propose a novel inverse-modelling approach that estimates the parameters of a simple land-surface model (LSM) by assimilating data into a differentiable, physics-based forward model formulated using convolutional operations. The governing equations are expressed within the Neural Physics framework, allowing direct gradient-based optimisation of time-dependent parameters without the need to derive and maintain adjoint formulations. The model parameters are estimated by minimising the mismatch between model predictions and synthetic or observational data. Although differentiability is enabled through machine-learning libraries, the forward model itself remains entirely physics-based and neither the forward model nor the parameter estimation procedure involve training. To evaluate the approach, we first generate synthetic observations of soil temperature by running the forward model with known parameter values and subsequently treat these parameters as unknown in an inverse problem. We show that observations of soil temperature at a single depth are insufficient to reliably constrain the model parameters. Using observations at two depths, however, does yield reliable parameter estimates, although the individual contributions of latent and sensible heat fluxes cannot be distinguished. We also apply the approach to urban flux tower data from Phoenix, United States, and show that the thermal conductivity, volumetric heat capacity and the combined sensible-latent heat transfer coefficient can be reliably estimated whilst using an observed value for the effective surface albedo. The resulting model accurately predicts the outgoing longwave radiation, conductive soil fluxes and the combined sensible-latent heat fluxes, demonstrating that the Neural Physics framework can be used to accurately determine the parameters of the particular LSM used here. This model is intentionally simple and does not include, for example, a subsurface moisture model. This simplicity facilitates an exploration of parameter identifiability, confounding and equifinality.
- Open Access
- 4 September 2026
- Published by Wiley in Quarterly Journal of the Royal Meteorological Society
Multi‐scale flow analysis for scale‐aware urban canopy models
As numerical weather prediction (NWP) models approach hectometric resolution, they operate increasingly in a regime where urban heterogeneity, specifically the morphological heterogeneity and resulting flow heterogeneity, is only partially resolved and the assumptions underlying conventional urban canopy models (UCMs) become questionable. To address this scale gap, we apply the multi‐scale coarse‐graining framework to building‐resolving large‐eddy simulations (LES) of the University of Bristol campus, a realistic heterogeneous urban environment. Two related morphologies are considered: an original layout containing large open‐space contrasts and a modified configuration with these regions infilled. By filtering the LES fields systematically, we quantify how flow heterogeneity is partitioned between resolved and unresolved contributions as the averaging length changes and identify a characteristic urban length‐scale of flow heterogeneity at which resolved and unresolved variability are comparable. This scale is strongly morphology‐dependent, with m for the original layout and m for the modified case, demonstrating that neighbourhood‐scale organisation can remain dynamically important at resolutions relevant to next‐generation NWP. Using this framework, we perform an a priori assessment of distributed drag and turbulent‐stress parameterisations. The results show that parameterisations derived from idealised geometries perform reasonably well only at sufficiently large averaging length‐scales (, corresponding to coarse resolutions), where horizontal transport is negligible and the flow appears approximately homogeneous. At smaller averaging lengths, their fidelity degrades rapidly due to increasing heterogeneity and filter‐to‐filter variability in morphology. These limitations are more pronounced in realistic layouts than in idealised cuboid arrays. Overall, the results highlight that the applicability of urban parameterisations depends critically on the relationship between model resolution and the characteristic urban length‐scale of morphology‐dependent flow heterogeneity. The framework provides a systematic route to diagnose this scale and to guide the development of scale‐aware urban canopy models for high‐resolution NWP.
- 4 September 2026
- Published by Wiley in Quarterly Journal of the Royal Meteorological Society
Open AccessMulti‐scale flow analysis for scale‐aware urban canopy models
As numerical weather prediction (NWP) models approach hectometric resolution, they operate increasingly in a regime where urban heterogeneity, specifically the morphological heterogeneity and resulting flow heterogeneity, is only partially resolved and the assumptions underlying conventional urban canopy models (UCMs) become questionable. To address this scale gap, we apply the multi‐scale coarse‐graining framework to building‐resolving large‐eddy simulations (LES) of the University of Bristol campus, a realistic heterogeneous urban environment. Two related morphologies are considered: an original layout containing large open‐space contrasts and a modified configuration with these regions infilled. By filtering the LES fields systematically, we quantify how flow heterogeneity is partitioned between resolved and unresolved contributions as the averaging length changes and identify a characteristic urban length‐scale of flow heterogeneity at which resolved and unresolved variability are comparable. This scale is strongly morphology‐dependent, with m for the original layout and m for the modified case, demonstrating that neighbourhood‐scale organisation can remain dynamically important at resolutions relevant to next‐generation NWP. Using this framework, we perform an a priori assessment of distributed drag and turbulent‐stress parameterisations. The results show that parameterisations derived from idealised geometries perform reasonably well only at sufficiently large averaging length‐scales (, corresponding to coarse resolutions), where horizontal transport is negligible and the flow appears approximately homogeneous. At smaller averaging lengths, their fidelity degrades rapidly due to increasing heterogeneity and filter‐to‐filter variability in morphology. These limitations are more pronounced in realistic layouts than in idealised cuboid arrays. Overall, the results highlight that the applicability of urban parameterisations depends critically on the relationship between model resolution and the characteristic urban length‐scale of morphology‐dependent flow heterogeneity. The framework provides a systematic route to diagnose this scale and to guide the development of scale‐aware urban canopy models for high‐resolution NWP.
- Open Access
- 30 June 2026
- Published by Elsevier in SSRN Electronic Journal
Inpainting U-Net for seamless pedestrian-level wind prediction across urban morphologies
Pedestrian-level wind prediction is essential for urban wind-comfort assessment, but LES remains computationally expensive, preventing its use for rapid evaluation. This study develops a two-stage U-Net framework for efficient prediction of time-averaged wind speed over urban morphologies. The model is trained and evaluated using the UrbanTALES dataset, which contains city configurations under different wind directions. In the first stage, a baseline U-Net model (M1) predicts the wind field patch-by-patch from building height and fetch information. This patch-wise formulation allows the framework to be applied to arbitrary-size urban domains, but it can introduce discontinuities at patch boundaries because each patch is inferred independently, while limited attention has been paid to assembling patch-wise predictions into seamless full-field predictions. To address this, a second U-Net model (M2) is introduced as an inpainting-based refinement model. M2 takes a larger contextual window containing the M1 prediction and morphology, and reduces discontinuities in the M1 prediction using neighbouring flow information. During full-field inference, M2 is iteratively applied until convergence. The results show that M1 captures the spatial distribution of pedestrian-level wind speed and performs well for low- and moderate-velocity regions, although high-velocity peaks are less accurate. M2 substantially reduces patch-boundary artefacts and improves the spatial coherence of the full-field prediction. Across unseen cases, the proposed framework reproduces the mean velocity and spatial variability well, while maximum velocities remain underestimated. Overall, the proposed framework provides an efficient and flexible surrogate model for high-resolution pedestrian-level wind prediction across realistic urban morphologies.
- 30 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessInpainting U-Net for seamless pedestrian-level wind prediction across urban morphologies
Pedestrian-level wind prediction is essential for urban wind-comfort assessment, but LES remains computationally expensive, preventing its use for rapid evaluation. This study develops a two-stage U-Net framework for efficient prediction of time-averaged wind speed over urban morphologies. The model is trained and evaluated using the UrbanTALES dataset, which contains city configurations under different wind directions. In the first stage, a baseline U-Net model (M1) predicts the wind field patch-by-patch from building height and fetch information. This patch-wise formulation allows the framework to be applied to arbitrary-size urban domains, but it can introduce discontinuities at patch boundaries because each patch is inferred independently, while limited attention has been paid to assembling patch-wise predictions into seamless full-field predictions. To address this, a second U-Net model (M2) is introduced as an inpainting-based refinement model. M2 takes a larger contextual window containing the M1 prediction and morphology, and reduces discontinuities in the M1 prediction using neighbouring flow information. During full-field inference, M2 is iteratively applied until convergence. The results show that M1 captures the spatial distribution of pedestrian-level wind speed and performs well for low- and moderate-velocity regions, although high-velocity peaks are less accurate. M2 substantially reduces patch-boundary artefacts and improves the spatial coherence of the full-field prediction. Across unseen cases, the proposed framework reproduces the mean velocity and spatial variability well, while maximum velocities remain underestimated. Overall, the proposed framework provides an efficient and flexible surrogate model for high-resolution pedestrian-level wind prediction across realistic urban morphologies.
- Open Access
- 22 June 2026
- Published by Copernicus GmbH
Across-Scale boundary layer processeS in complex URban Environments: the ASSURE/urbisphere/UrbanAIR observation network in Bristol, UK.
Climate change already affects most of the world’s urban population. Developing resilient urban environments requires improving both weather and climate modelling. Heterogeneity exists from street (100 m) to neighbourhood (1 km) to city (10 km) scales due to urban form and function. Hectometric-scale numerical weather prediction (NWP) may be starting to resolve neighbourhood-scale heterogeneity but how can observation networks be designed to capture spatial variation in urban climates (horizontally and vertically) to evaluate these model predictions robustly?To address these challenges, we combine city-scale field observations, resident interviews, high-resolution numerical (Large Eddy Simulation, NWP) and wind-tunnel modelling. Three research projects (ASSURE, urbisphere, and UrbanAIR) have collaborated to design and maintain an urban observation network in the UK. The focus is on Bristol as it is compact, has representative land-use, is close to the coast, and lies in relatively low-lying complex terrain. Bristol City authorities had previously used Met Office climate simulations at 2.2 km resolution to plan for urban heat vulnerability. Both partners were engaged in our design of a network of over 40 lamp-post mounted automatic weather stations that has run since spring 2024. A network of ground-based remote sensing (Doppler wind lidars, automatic lidar ceilometers) was deployed at six sites across the city to observe boundary layer development in response to the urban surface and orography. Other observations included indoor climate, radiation, vegetation, and tracer gas dispersion experiments. The core Bristol field campaign ran from spring 2024-2025 but a sub-set of measurements is ongoing. This follows other year-long urbisphere campaigns in Berlin, Paris, Freiburg, and Heraklion.This talk provides an overview of the observation network in the Bristol project and the data-set obtained. Key findings will be presented where field and modelling approaches were combined to design the deployment, e.g., a Virtual Doppler Lidar approach using LES, lessons learned from a testbed of a high-resolution NWP ensemble.
- 22 June 2026
- Published by Copernicus GmbH
Open AccessAcross-Scale boundary layer processeS in complex URban Environments: the ASSURE/urbisphere/UrbanAIR observation network in Bristol, UK.
Climate change already affects most of the world’s urban population. Developing resilient urban environments requires improving both weather and climate modelling. Heterogeneity exists from street (100 m) to neighbourhood (1 km) to city (10 km) scales due to urban form and function. Hectometric-scale numerical weather prediction (NWP) may be starting to resolve neighbourhood-scale heterogeneity but how can observation networks be designed to capture spatial variation in urban climates (horizontally and vertically) to evaluate these model predictions robustly?To address these challenges, we combine city-scale field observations, resident interviews, high-resolution numerical (Large Eddy Simulation, NWP) and wind-tunnel modelling. Three research projects (ASSURE, urbisphere, and UrbanAIR) have collaborated to design and maintain an urban observation network in the UK. The focus is on Bristol as it is compact, has representative land-use, is close to the coast, and lies in relatively low-lying complex terrain. Bristol City authorities had previously used Met Office climate simulations at 2.2 km resolution to plan for urban heat vulnerability. Both partners were engaged in our design of a network of over 40 lamp-post mounted automatic weather stations that has run since spring 2024. A network of ground-based remote sensing (Doppler wind lidars, automatic lidar ceilometers) was deployed at six sites across the city to observe boundary layer development in response to the urban surface and orography. Other observations included indoor climate, radiation, vegetation, and tracer gas dispersion experiments. The core Bristol field campaign ran from spring 2024-2025 but a sub-set of measurements is ongoing. This follows other year-long urbisphere campaigns in Berlin, Paris, Freiburg, and Heraklion.This talk provides an overview of the observation network in the Bristol project and the data-set obtained. Key findings will be presented where field and modelling approaches were combined to design the deployment, e.g., a Virtual Doppler Lidar approach using LES, lessons learned from a testbed of a high-resolution NWP ensemble.
- Open Access
- 20 March 2026
- Published by Elsevier in SSRN Electronic Journal
Hydrodynamic coupling and retention time of inertial spheres crossing and bouncing at density interfaces
We study how solid spheres settle through a fluid made up of two layers with different densities, and specifically examine the ``bouncing’’ phenomenon, where the sphere’s downward motion is temporarily halted and reversed at the interface. Using synchronized particle tracking and time-resolved PIV/PTV, we compare two systems with similar density jumps but different viscosities: water–salt and water–glycerol solutions. As the sphere crosses the interface, it leaves a wake of lighter fluid. When this lighter wake detaches and returns upward, it slows the sphere, causing the interface to rebound. The resulting increase in viscous drag can briefly lift the sphere, producing the observed bounce. Shortly after the upward motion, as the interface relaxes, the sphere resumes sinking. We demonstrate that standard models considering only density differences fail to predict these dynamics when viscosities differ between layers. To address this, we present a new criterion based on the ratio of cross-interface Froude numbers ($Fr_2/Fr_1 \lesssim 0.14$), which incorporates both density and viscosity effects via the viscosity ratio ($\nu_1/\nu_2$). This criterion yields improved predictions of the duration that particles remain near the interface. Our results help refine models of particle transport in stratified fluids, with implications for environmental and industrial processes.
- 20 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessHydrodynamic coupling and retention time of inertial spheres crossing and bouncing at density interfaces
We study how solid spheres settle through a fluid made up of two layers with different densities, and specifically examine the ``bouncing’’ phenomenon, where the sphere’s downward motion is temporarily halted and reversed at the interface. Using synchronized particle tracking and time-resolved PIV/PTV, we compare two systems with similar density jumps but different viscosities: water–salt and water–glycerol solutions. As the sphere crosses the interface, it leaves a wake of lighter fluid. When this lighter wake detaches and returns upward, it slows the sphere, causing the interface to rebound. The resulting increase in viscous drag can briefly lift the sphere, producing the observed bounce. Shortly after the upward motion, as the interface relaxes, the sphere resumes sinking. We demonstrate that standard models considering only density differences fail to predict these dynamics when viscosities differ between layers. To address this, we present a new criterion based on the ratio of cross-interface Froude numbers ($Fr_2/Fr_1 \lesssim 0.14$), which incorporates both density and viscosity effects via the viscosity ratio ($\nu_1/\nu_2$). This criterion yields improved predictions of the duration that particles remain near the interface. Our results help refine models of particle transport in stratified fluids, with implications for environmental and industrial processes.
- Open Access
- 18 March 2026
- Published by Wiley in Quarterly Journal of the Royal Meteorological Society
- Vol. 152 (777) , e70154
Parameter investigation for urban surface‐energy balance: A large‐eddy simulation study
- 18 March 2026
- Published by Wiley in Quarterly Journal of the Royal Meteorological Society
- Vol. 152 (777) , e70154
Open AccessParameter investigation for urban surface‐energy balance: A large‐eddy simulation study
- Open Access
- 14 March 2026
- Published by Copernicus GmbH
Across-scale boundary layer processes in complex urban environments: the urbisphere - ASSURE Bristol project
Climate change will affect most of the world’s urban population. Developing resilient urban environments requires improved weather and climate modelling. Heterogeneity exists from street (100m) to neighbourhood (1km) to city (10km) scales due to urban form and function. How should it be parameterised, given that it influences atmospheric processes acting over a similar range of scales?To address this challenge, we combine city-scale field observations, resident interviews, high-resolution numerical (LES, NWP) and wind-tunnel (WT) modelling. The focus is on Bristol, UK, as it is compact, has representative land-use, and has coastal proximity and complex terrain. It follows other year-long urbisphere project campaigns in Berlin, Paris, Freiburg, and Heraklion.This talk provides an overview of the WT, LES and NWP modelling and observations thus far in the project. A case study is described where the sub-neighbourhood scale Avon River Gorge influences boundary layer and dispersion processes.
- 14 March 2026
- Published by Copernicus GmbH
Open AccessAcross-scale boundary layer processes in complex urban environments: the urbisphere - ASSURE Bristol project
Climate change will affect most of the world’s urban population. Developing resilient urban environments requires improved weather and climate modelling. Heterogeneity exists from street (100m) to neighbourhood (1km) to city (10km) scales due to urban form and function. How should it be parameterised, given that it influences atmospheric processes acting over a similar range of scales?To address this challenge, we combine city-scale field observations, resident interviews, high-resolution numerical (LES, NWP) and wind-tunnel (WT) modelling. The focus is on Bristol, UK, as it is compact, has representative land-use, and has coastal proximity and complex terrain. It follows other year-long urbisphere project campaigns in Berlin, Paris, Freiburg, and Heraklion.This talk provides an overview of the WT, LES and NWP modelling and observations thus far in the project. A case study is described where the sub-neighbourhood scale Avon River Gorge influences boundary layer and dispersion processes.
- Open Access
- 5 January 2026
- Published by Copernicus GmbH
- Vol. 2026, 1
Reply on CEC1
The Neural Physics approach is used to determine the parameters of a simple land-surface model using PyTorch’s backpropagation engine to carry out the optimisation. In order to test the inverse model, a synthetic dataset is created by running the model in forward mode with known parameter values to create soil temperature time series that can be used as observations for the inverse model. We show that it is not possible to obtain a reliable parameter estimation using a time series of soil temperature observed at a single depth. Using measurements at two depths, reliable parameter estimates can be obtained although it is not possible to differentiate between latent and sensible heat fluxes. We apply the inverse model to urban flux tower data in Phoenix, United States, and show that the thermal conductivity, volumetric heat capacity, and the combined sensible-latent heat transfer coefficient can be reliably estimated using an observed value for the effective surface albedo. The resulting model accurately predicts the outgoing longwave radiation, conductive soil fluxes and the combined sensible-latent heat fluxes.
- 5 January 2026
- Published by Copernicus GmbH
- Vol. 2026, 1
Open AccessReply on CEC1
The Neural Physics approach is used to determine the parameters of a simple land-surface model using PyTorch’s backpropagation engine to carry out the optimisation. In order to test the inverse model, a synthetic dataset is created by running the model in forward mode with known parameter values to create soil temperature time series that can be used as observations for the inverse model. We show that it is not possible to obtain a reliable parameter estimation using a time series of soil temperature observed at a single depth. Using measurements at two depths, reliable parameter estimates can be obtained although it is not possible to differentiate between latent and sensible heat fluxes. We apply the inverse model to urban flux tower data in Phoenix, United States, and show that the thermal conductivity, volumetric heat capacity, and the combined sensible-latent heat transfer coefficient can be reliably estimated using an observed value for the effective surface albedo. The resulting model accurately predicts the outgoing longwave radiation, conductive soil fluxes and the combined sensible-latent heat fluxes.
- Open Access
- 5 January 2026
- Published by Copernicus GmbH
- Vol. 2026, 1
Reply on RC1
The Neural Physics approach is used to determine the parameters of a simple land-surface model using PyTorch’s backpropagation engine to carry out the optimisation. In order to test the inverse model, a synthetic dataset is created by running the model in forward mode with known parameter values to create soil temperature time series that can be used as observations for the inverse model. We show that it is not possible to obtain a reliable parameter estimation using a time series of soil temperature observed at a single depth. Using measurements at two depths, reliable parameter estimates can be obtained although it is not possible to differentiate between latent and sensible heat fluxes. We apply the inverse model to urban flux tower data in Phoenix, United States, and show that the thermal conductivity, volumetric heat capacity, and the combined sensible-latent heat transfer coefficient can be reliably estimated using an observed value for the effective surface albedo. The resulting model accurately predicts the outgoing longwave radiation, conductive soil fluxes and the combined sensible-latent heat fluxes.
- 5 January 2026
- Published by Copernicus GmbH
- Vol. 2026, 1
Open AccessReply on RC1
The Neural Physics approach is used to determine the parameters of a simple land-surface model using PyTorch’s backpropagation engine to carry out the optimisation. In order to test the inverse model, a synthetic dataset is created by running the model in forward mode with known parameter values to create soil temperature time series that can be used as observations for the inverse model. We show that it is not possible to obtain a reliable parameter estimation using a time series of soil temperature observed at a single depth. Using measurements at two depths, reliable parameter estimates can be obtained although it is not possible to differentiate between latent and sensible heat fluxes. We apply the inverse model to urban flux tower data in Phoenix, United States, and show that the thermal conductivity, volumetric heat capacity, and the combined sensible-latent heat transfer coefficient can be reliably estimated using an observed value for the effective surface albedo. The resulting model accurately predicts the outgoing longwave radiation, conductive soil fluxes and the combined sensible-latent heat fluxes.
- Open Access
- 5 January 2026
- Published by Copernicus GmbH
- Vol. 2026, 1
Reply on RC2
The Neural Physics approach is used to determine the parameters of a simple land-surface model using PyTorch’s backpropagation engine to carry out the optimisation. In order to test the inverse model, a synthetic dataset is created by running the model in forward mode with known parameter values to create soil temperature time series that can be used as observations for the inverse model. We show that it is not possible to obtain a reliable parameter estimation using a time series of soil temperature observed at a single depth. Using measurements at two depths, reliable parameter estimates can be obtained although it is not possible to differentiate between latent and sensible heat fluxes. We apply the inverse model to urban flux tower data in Phoenix, United States, and show that the thermal conductivity, volumetric heat capacity, and the combined sensible-latent heat transfer coefficient can be reliably estimated using an observed value for the effective surface albedo. The resulting model accurately predicts the outgoing longwave radiation, conductive soil fluxes and the combined sensible-latent heat fluxes.
- 5 January 2026
- Published by Copernicus GmbH
- Vol. 2026, 1
Open AccessReply on RC2
The Neural Physics approach is used to determine the parameters of a simple land-surface model using PyTorch’s backpropagation engine to carry out the optimisation. In order to test the inverse model, a synthetic dataset is created by running the model in forward mode with known parameter values to create soil temperature time series that can be used as observations for the inverse model. We show that it is not possible to obtain a reliable parameter estimation using a time series of soil temperature observed at a single depth. Using measurements at two depths, reliable parameter estimates can be obtained although it is not possible to differentiate between latent and sensible heat fluxes. We apply the inverse model to urban flux tower data in Phoenix, United States, and show that the thermal conductivity, volumetric heat capacity, and the combined sensible-latent heat transfer coefficient can be reliably estimated using an observed value for the effective surface albedo. The resulting model accurately predicts the outgoing longwave radiation, conductive soil fluxes and the combined sensible-latent heat fluxes.