Tanja Denager

Understanding local impacts of extreme weather, from e.g. precipitation or temperature, requires high-resolution downscaling of those atmospheric variables. Deterministic and classical statistical methods fail to adequately represent variability and extremes, motivating the use of probabilistic generative methods. Recent advances in generative deep learning show promise for spatial realism and ensemble generation, but their physical fidelity and limitations remain insufficiently understood. We present CEDDAR v1.0.2 (Controllable Ensemble Diffusion Downscaling for Atmospheric Rainfall), a diffusion-based generative model for daily precipitation downscaling, conditioned on large-scale (∼30 × 30 km) ERA5 precipitation fields, to produce kilometre-scale (2.5 × 2.5 km) precipitation over Denmark, using the DANish ReAnalysis product (DANRA) as reference. The modelling framework is based on an Elucidated Diffusion Model (EDM) backbone and incorporates soft physics-guidance through seasonal (Day-of-Year) and geographic conditioning with a specific land-focused design.We further introduce regularisation terms through a Signed-Distance Function weighted loss, and an auxiliary RainGate head for shaping wet-dry occurrence statistics and improving ensemble calibration. By design, the model is probabilistic and capable of producing ensembles. We further introduce an inference-time control parameter, σ^*, that allows for interpretable adjustment of stochastic scale sensitivity without retraining. For model-behaviour assessment, we introduce a multi-perspective evaluation protocol with emphasis on trade-offs across probabilistic, spatial, climatological, and scale-dependent diagnostics rather than single-metric optimisation. Our results show strong spatial and spectral realism with generally well-calibrated ensembles, but also demonstrate that visually realistic fine-scale structure does not guarantee climatological fidelity. We also find that explicitly introducing σ^* reveals systematic, though non-linear, trade-offs between large-scale coherence, fine-scale variability, and ensemble spread. This highlights the potential and the limitations of controllable diffusion-based downscaling. In conclusion, this work presents a diffusion-based downscaling framework and evaluation suite that works as a diagnostic testbed for understanding how generative models behave in physically constrained settings, and offers guidance on their appropriate use and common failure modes relevant to future operational solutions and climate-impact applications.

Understanding local impacts of extreme weather, from e.g. precipitation or temperature, requires high-resolution downscaling of those atmospheric variables. Deterministic and classical statistical methods fail to adequately represent variability and extremes, motivating the use of probabilistic generative methods. Recent advances in generative deep learning show promise for spatial realism and ensemble generation, but their physical fidelity and limitations remain insufficiently understood. We present CEDDAR v1.0.2 (Controllable Ensemble Diffusion Downscaling for Atmospheric Rainfall), a diffusion-based generative model for daily precipitation downscaling, conditioned on large-scale (∼30 × 30 km) ERA5 precipitation fields, to produce kilometre-scale (2.5 × 2.5 km) precipitation over Denmark, using the DANish ReAnalysis product (DANRA) as reference. The modelling framework is based on an Elucidated Diffusion Model (EDM) backbone and incorporates soft physics-guidance through seasonal (Day-of-Year) and geographic conditioning with a specific land-focused design.We further introduce regularisation terms through a Signed-Distance Function weighted loss, and an auxiliary RainGate head for shaping wet-dry occurrence statistics and improving ensemble calibration. By design, the model is probabilistic and capable of producing ensembles. We further introduce an inference-time control parameter, σ^*, that allows for interpretable adjustment of stochastic scale sensitivity without retraining. For model-behaviour assessment, we introduce a multi-perspective evaluation protocol with emphasis on trade-offs across probabilistic, spatial, climatological, and scale-dependent diagnostics rather than single-metric optimisation. Our results show strong spatial and spectral realism with generally well-calibrated ensembles, but also demonstrate that visually realistic fine-scale structure does not guarantee climatological fidelity. We also find that explicitly introducing σ^* reveals systematic, though non-linear, trade-offs between large-scale coherence, fine-scale variability, and ensemble spread. This highlights the potential and the limitations of controllable diffusion-based downscaling. In conclusion, this work presents a diffusion-based downscaling framework and evaluation suite that works as a diagnostic testbed for understanding how generative models behave in physically constrained settings, and offers guidance on their appropriate use and common failure modes relevant to future operational solutions and climate-impact applications.

Understanding local impacts of extreme weather, from e.g. precipitation or temperature, requires high-resolution downscaling of those atmospheric variables. Deterministic and classical statistical methods fail to adequately represent variability and extremes, motivating the use of probabilistic generative methods. Recent advances in generative deep learning show promise for spatial realism and ensemble generation, but their physical fidelity and limitations remain insufficiently understood. We present CEDDAR v1.0.2 (Controllable Ensemble Diffusion Downscaling for Atmospheric Rainfall), a diffusion-based generative model for daily precipitation downscaling, conditioned on large-scale (∼30 × 30 km) ERA5 precipitation fields, to produce kilometre-scale (2.5 × 2.5 km) precipitation over Denmark, using the DANish ReAnalysis product (DANRA) as reference. The modelling framework is based on an Elucidated Diffusion Model (EDM) backbone and incorporates soft physics-guidance through seasonal (Day-of-Year) and geographic conditioning with a specific land-focused design.We further introduce regularisation terms through a Signed-Distance Function weighted loss, and an auxiliary RainGate head for shaping wet-dry occurrence statistics and improving ensemble calibration. By design, the model is probabilistic and capable of producing ensembles. We further introduce an inference-time control parameter, σ^*, that allows for interpretable adjustment of stochastic scale sensitivity without retraining. For model-behaviour assessment, we introduce a multi-perspective evaluation protocol with emphasis on trade-offs across probabilistic, spatial, climatological, and scale-dependent diagnostics rather than single-metric optimisation. Our results show strong spatial and spectral realism with generally well-calibrated ensembles, but also demonstrate that visually realistic fine-scale structure does not guarantee climatological fidelity. We also find that explicitly introducing σ^* reveals systematic, though non-linear, trade-offs between large-scale coherence, fine-scale variability, and ensemble spread. This highlights the potential and the limitations of controllable diffusion-based downscaling. In conclusion, this work presents a diffusion-based downscaling framework and evaluation suite that works as a diagnostic testbed for understanding how generative models behave in physically constrained settings, and offers guidance on their appropriate use and common failure modes relevant to future operational solutions and climate-impact applications.

Spatially explicit knowledge of soil organic carbon and groundwater variability in peatlands is essential to support climate action, such as planning and implementing restoration projects. Geospatial machine learning is a key tool to obtain such knowledge at high spatial resolution, based on linking maps of explanatory variables with point information to train regression or classification models. Explanatory variables are usually identified through expert knowledge and derived from satellite remote sensing, digital elevation models or maps of soil types or land use. Identifying and processing relevant explanatory variables at large scale is non-trivial and cumbersome.Geospatial foundation models, such as Google’s Alpha Earth change how satellite data and other geospatial data can be utilized in downstream machine learning tasks. Such models provide analysis-ready unified layers, i.e. embeddings, that are semantically rich representations capturing the underlying input data. In the case of Alpha Earth, input data cover archives of Sentinel-1 and Sentinel-2 as well as other geospatial data sources.In the present study we introduce Alpha Earth embeddings into the modelling of soil organic carbon and groundwater across Danish peatlands at 10 m resolution. We use existing datasets and models of the two variables to benchmark the potential of foundation models for low-barrier large-scale modelling. The models trained against solely Alpha Earth embeddings are contrasted with models trained against explanatory variables selected through expert knowledge as well as with hybrid models combining basic topographical variables with Alpha Earth embeddings.The Alpha Earth model of soil organic carbon produces meaningful spatial patterns while having a 6% decrease in performance (RMSE) with respect to the expert model. The true positive rate to predict peaty and peat soils is 0.68 and 0.65 for the expert and Alpha Earth model, respectively. The hybrid model increases the performance slightly with respect to the Alpha Earth model and all models achieve very comparable result of mapping the overall peat extent. The Alpha Earth model predicting groundwater has a 3% performance decrease with respect to the expert model (RMSE). When introducing synthetic training data for drained and wet conditions to the groundwater model, the Alpha Earth model shows limited performance. However, the hybrid model can utilize the synthetic data in a more meaningful way and achieves satisfactory results with respect to performance and spatial patterns.In addition, we carry out feature importance analysis to explain the Alpha Earth embeddings, which is clear limitation in the usage of foundation models, where explainability is typically not provided.

To mitigate agricultural greenhouse gas emissions, Danish ministerial agreements have initiated a land-use transformation of historical dimensions, focusing on restoring and rewetting extensive peatland areas currently used for agriculture. In addition, a CO2-equivalent tax on emissions from organic peatlands is scheduled to be implemented from 2028. This study informs discussions on requirements and best practices for rewetting and peatland restoration and highlights the importance of including changing climate conditions and rewetting management scenarios in future peatland management strategies.The study integrates process-based hydrological modeling and empirical CO2 flux modeling at a daily temporal resolution to evaluate how peatland hydrology influences CO2 emissions under scenarios of rewetting and climate change.Following the calibration of a three-dimensional transient distributed hydrological model for a peat-dominated catchment, daily groundwater table dynamics were simulated to represent hydrological conditions in drained peat soils. These simulations were coupled with an empirical CO2 flux model, developed from a comprehensive daily dataset of groundwater table depth, temperature, and soil CO2 flux measurements. The novel empirical CO2 flux model captures a clear temperature-dependent response of soil CO2 emissions to variations in groundwater table depth.By applying this coupled modeling framework, we quantified CO2 emissions at daily timescales. The results demonstrate that incorporating both temperature sensitivity and high-resolution temporal variability in water level significantly influence projections of CO2 fluxes. In particular, high CO2 emissions are expected in cases of co-occurrence of elevated air temperature and low groundwater tables. Using 17 different climate projections from the Euro-CORDEX regional climate modeling project, we simulated future groundwater table depth and temperature-dependent CO2 emissions. We find increased emissions due to increased temperatures, which, however, can be counter-balanced (in the Danish case) or amplified depending on the future trend in groundwater table depth.Our results further demonstrate that rewetting strategies that achieve near-surface groundwater tables mainly during winter result in only marginal emission reductions compared to drained conditions. Conversely, near-surface groundwater tables in summer offer more effective reductions (up to 50%).The study illustrates the value of combining detailed hydrological simulations with emission models.

Understanding local impacts of extreme weather, from e.g. precipitation or temperature, requires high-resolution downscaling of those atmospheric variables. Deterministic and classical statistical methods fail to adequately represent variability and extremes, motivating the use of probabilistic generative methods. Recent advances in generative deep learning show promise for spatial realism and ensemble generation, but their physical fidelity and limitations remain insufficiently understood. We present CEDDAR v1.0.2 (Controllable Ensemble Diffusion Downscaling for Atmospheric Rainfall), a diffusion-based generative model for daily precipitation downscaling, conditioned on large-scale (∼30 × 30 km) ERA5 precipitation fields, to produce kilometre-scale (2.5 × 2.5 km) precipitation over Denmark, using the DANish ReAnalysis product (DANRA) as reference. The modelling framework is based on an Elucidated Diffusion Model (EDM) backbone and incorporates soft physics-guidance through seasonal (Day-of-Year) and geographic conditioning with a specific land-focused design.We further introduce regularisation terms through a Signed-Distance Function weighted loss, and an auxiliary RainGate head for shaping wet-dry occurrence statistics and improving ensemble calibration. By design, the model is probabilistic and capable of producing ensembles. We further introduce an inference-time control parameter, σ^*, that allows for interpretable adjustment of stochastic scale sensitivity without retraining. For model-behaviour assessment, we introduce a multi-perspective evaluation protocol with emphasis on trade-offs across probabilistic, spatial, climatological, and scale-dependent diagnostics rather than single-metric optimisation. Our results show strong spatial and spectral realism with generally well-calibrated ensembles, but also demonstrate that visually realistic fine-scale structure does not guarantee climatological fidelity. We also find that explicitly introducing σ^* reveals systematic, though non-linear, trade-offs between large-scale coherence, fine-scale variability, and ensemble spread. This highlights the potential and the limitations of controllable diffusion-based downscaling. In conclusion, this work presents a diffusion-based downscaling framework and evaluation suite that works as a diagnostic testbed for understanding how generative models behave in physically constrained settings, and offers guidance on their appropriate use and common failure modes relevant to future operational solutions and climate-impact applications.

This study integrates process-based hydrological modeling and empirical CO₂ flux modeling at a daily temporal resolution to evaluate how peatland hydrology influence CO₂ emissions under scenarios of rewetting and climate change. Following the calibration of a three-dimensional transient physically-based hydrological model for a peat-dominated catchment, daily groundwater table dynamics were simulated to represent hydrological conditions in drained peat soils. These simulations were coupled with an empirical CO₂ flux model, developed from a comprehensive daily dataset of groundwater table depth, temperature, and soil CO₂ flux measurements. The empirical CO₂ flux model captures a clear temperature-dependent response of soil CO₂ emissions to variations in groundwater table depth. By applying this coupled modeling framework, we quantified CO₂ emissions at daily timescales. The results demonstrate that incorporating both temperature sensitivity and high-resolution temporal variability in water level significantly influences projections of CO₂ fluxes. Especially the co-occurrence of elevated air temperature and low groundwater table significantly influence CO₂ emissions under scenarios of rewetting and climate change. These insights highlight the importance of including changing climate conditions in future peatland management strategies for emission inventories. The study illustrates the value of combining detailed hydrological simulations with emission models. It also emphasizes the need for detailed monitoring of greenhouse gas emissions across multiple sites and the development of robust empirical models that can be generalized and spatially upscaled.

This study integrates process-based hydrological modeling and empirical CO₂ flux modeling at a daily temporal resolution to evaluate how peatland hydrology influence CO₂ emissions under scenarios of rewetting and climate change. Following the calibration of a three-dimensional transient groundwater flow model for a peat-dominated catchment, daily groundwater table dynamics were simulated to represent hydrological conditions in drained peat soils. These simulations were coupled with an empirical CO₂ flux model, developed from a comprehensive daily dataset of groundwater table depth, temperature, and soil CO₂ flux measurements. The empirical CO₂ flux model captures a clear temperature-dependent response of soil CO₂ emissions to variations in groundwater table depth. By applying this coupled modeling framework, we quantified CO₂ emissions at daily timescales. The results demonstrate that incorporating both temperature sensitivity and high-resolution temporal variability in water level significantly influences projections of CO₂ fluxes. Especially the co-occurrence of elevated air temperature and low groundwater table significantly influence CO₂ emissions under scenarios of rewetting and climate change. These insights highlight the importance of including changing climate conditions in future peatland management strategies for emission inventories. The study illustrates the value of combining detailed hydrological simulations with emission models. It also emphasizes the need for detailed monitoring of greenhouse gas emissions across multiple sites and the development of robust empirical models that can be generalized and spatially upscaled.

This study integrates process-based hydrological modeling and empirical CO₂ flux modeling at a daily temporal resolution to evaluate how peatland hydrology influence CO₂ emissions under scenarios of rewetting and climate change. Following the calibration of a three-dimensional transient groundwater flow model for a peat-dominated catchment, daily groundwater table dynamics were simulated to represent hydrological conditions in drained peat soils. These simulations were coupled with an empirical CO₂ flux model, developed from a comprehensive daily dataset of groundwater table depth, temperature, and soil CO₂ flux measurements. The empirical CO₂ flux model captures a clear temperature-dependent response of soil CO₂ emissions to variations in groundwater table depth. By applying this coupled modeling framework, we quantified CO₂ emissions at daily timescales. The results demonstrate that incorporating both temperature sensitivity and high-resolution temporal variability in water level significantly influences projections of CO₂ fluxes. Especially the co-occurrence of elevated air temperature and low groundwater table significantly influence CO₂ emissions under scenarios of rewetting and climate change. These insights highlight the importance of including changing climate conditions in future peatland management strategies for emission inventories. The study illustrates the value of combining detailed hydrological simulations with emission models. It also emphasizes the need for detailed monitoring of greenhouse gas emissions across multiple sites and the development of robust empirical models that can be generalized and spatially upscaled.