Machine learning for stochastic parametrisations
Environmental Data Science Cambridge University Press 3 (2025) e38
Abstract:
Atmospheric models used for weather and climate prediction are traditionally formulated in a deterministic manner. In other words, given a particular state of the resolved scale variables, the most likely forcing from the subgrid scale processes is estimated and used to predict the evolution of the large-scale flow. However, the lack of scale separation in the atmosphere means that this approach is a large source of error in forecasts. Over recent years, an alternative paradigm has developed: the use of stochastic techniques to characterize uncertainty in small-scale processes. These techniques are now widely used across weather, subseasonal, seasonal, and climate timescales. In parallel, recent years have also seen significant progress in replacing parametrization schemes using machine learning (ML). This has the potential to both speed up and improve our numerical models. However, the focus to date has largely been on deterministic approaches. In this position paper, we bring together these two key developments and discuss the potential for data-driven approaches for stochastic parametrization. We highlight early studies in this area and draw attention to the novel challenges that remain.
Source of rainfall above Mediterranean caves (Chauvet and Orgnac) and long-term trend of cave dripping oxygen isotopes based on 20 years monitoring records: Importance for speleothem-based climate reconstructions
Quaternary Science Reviews 349 (2025) 109145
Abstract:
Understanding the factors that shape climate and influence the isotopic composition of precipitation is crucial for paleoclimate reconstructions, especially in regions with Mediterranean climates where rainfall is influenced by both Atlantic and Mediterranean moisture sources. This study examines the relationship between moisture origins, climatic variables, and the stable isotopic composition of precipitation and cave drip water in the Orgnac and Chauvet caves, located in southern France, over a 20-year period. The research reveals notable seasonal variations in rainfall 未18O values, driven by temperature and Rayleigh distillation processes. As shown in our previous work in Villars Cave (SW-France), temperature changes alone cannot fully explain the observed isotopic variability. We observed that winter precipitation tends to have lower 未18O values due to longer transport distances from distant oceanic sources, while summer precipitation displays higher 未18O values due to shorter transport paths. Additionally, the study highlights the influence of sea surface wind speeds and evaporation rates on water vapor isotopes, further shaping the seasonal 未18O patterns. As rainwater infiltrates the soil and percolates into the karst system, the seasonal 未18O signal in drip water is often dampened due to mixing in the reservoirs above the caves, which typically reduces seasonality. The key findings include: (1) a multi-year increasing trend in drip water 未18O, likely associated with reduced local water excess and the effects of global warming, with significant implications for speleothem isotope records, and (2) moisture from the Mediterranean Sea contributes to 10% of the total precipitation source, despite the region's proximity to the sea, especially during intense storm events. This study provides new insights into the complex interactions between moisture sources, temperature, and isotopic signatures in Mediterranean climate regions, with implications for improving speleothem-based paleoclimate reconstructions.
3D Cloud reconstruction through geospatially-aware Masked Autoencoders
Workshop paper at 鈥淢achine Learning and the Physical Sciences鈥, NeurIPS (2024)
Abstract:
Clouds play a key role in Earth's radiation balance with complex effects that introduce large uncertainties into climate models. Real-time 3D cloud data is essential for improving climate predictions. This study leverages geostationary imagery from MSG/SEVIRI and radar reflectivity measurements of cloud profiles from CloudSat/CPR to reconstruct 3D cloud structures. We first apply self-supervised learning (SSL) methods-Masked Autoencoders (MAE) and geospatially-aware SatMAE on unlabelled MSG images, and then fine-tune our models on matched image-profile pairs. Our approach outperforms state-of-the-art methods like U-Nets, and our geospatial encoding further improves prediction results, demonstrating the potential of SSL for cloud reconstruction.
Updraft Width Modulates Ambient Atmospheric Controls on Convective Cloud Depth
Journal of Geophysical Research: Atmospheres American Geophysical Union 129:23 (2024) e2024JD041769
Abstract:
The depth of convective clouds affects vertical transport of atmospheric constituents, influencing downstream weather and climate. Atmospheric controls on the maximum depth reached by moist convection are investigated with radar鈥恡racked convective cells tagged with sounding鈥恉erived atmospheric parameters from a field campaign in central Argentina. Regression analyses show that narrow (<12鈥恔m diameter) and wide (>16鈥恔m diameter) cell depths respond to disparate factors, where cell areas are defined using composite reflectivity signatures. Undiluted lifted parcel indices including convective available potential energy (CAPE) and level of neutral buoyancy (LNB) are top predictors of wide cell maximum depth while mid鈥恡ropospheric relative humidity is the top predictor of narrow cell maximum depth. Because narrow cells are more numerous than wide cells, the overall outcome of the full cell population does not strongly correlate with CAPE and LNB conditions. Tracked cells and atmospheric conditions in a simulation with 3鈥恔m grid spacing covering the field campaign produce similar results to those observed. Narrow cells that are relatively deep have a cooler and moister mid鈥恡roposphere with weaker free tropospheric subsidence, while relatively deep wide cells have much warmer and moister lower tropospheric conditions. These atmospheric differences are present 1 hr before cell initiation at both a fixed observing site and variable cell initiation locations. Simulated narrow cell maximum equivalent potential temperature decreases with height at a rate similar to the ambient vertical gradient, causing these cells to fall short of their LNB and 91探花ing the view that entrainment鈥恉riven dilution is a dominant control on their depth.Dependencies of Simulated Convective Cell and System Growth Biases on Atmospheric Instability and Model Resolution
Journal of Geophysical Research: Atmospheres American Geophysical Union 129:22 (2024) e2024JD041090