Data assimilation is a core component of numerical weather prediction systems. The large quantity of data processed during assimilation requires the computation to be distributed across increasingly many compute nodes; yet, existing approaches suffer from synchronization overhead in this setting. In this article, we exploit the formulation of data assimilation as a Bayesian inference problem and apply a message-passing algorithm to solve the spatial inference problem. Since message passing is inherently based on local computations, this approach lends itself to parallel and distributed computation. In combination with a GPU-accelerated implementation, we can scale the algorithm to very large grid sizes while retaining good accuracy and compute and memory requirements.

Wind speed at the sea surface is a key quantity for a variety of scientific applications and human activities. For its importance, many observation techniques exist, ranging from in situ to satellite observations. However, none of such techniques can capture the spatiotemporal variability of the phenomenon at the same time. Reanalysis products, obtained from data assimilation methods, represent the state-of-the-art for sea-surface wind speed monitoring but may be biased by model errors and their spatial resolution is not competitive with satellite products. In this work, we propose a scheme based on both data assimilation and deep learning concepts to process spatiotemporally heterogeneous input sources to reconstruct high-resolution time series of spatial wind speed fields. This method allows to us make the most of the complementary information conveyed by the different sea-surface information typically available in operational settings. We use synthetic wind speed data to emulate satellite images, in situ time series and reanalyzed wind fields. Starting from these pseudo-observations, we run extensive numerical simulations to assess the impact of each input source on the model reconstruction performance. We show that our proposed framework outperforms a deep learning–based inversion scheme and can successfully exploit the spatiotemporal complementary information of the different input sources. We also show that the model can learn the possible bias in reanalysis products and attenuate it in the output reconstructions.

In this paper, we explore the crucial role and challenges of computational reproducibility in geosciences, drawing insights from the Climate Informatics Reproducibility Challenge (CICR) in 2023. The competition aimed at (1) identifying common hurdles to reproduce computational climate science; and (2) creating interactive reproducible publications for selected papers of the Environmental Data Science journal. Based on lessons learned from the challenge, we emphasize the significance of open research practices, mentorship, transparency guidelines, as well as the use of technologies such as executable research objects for the reproduction of geoscientific published research. We propose a supportive framework of tools and infrastructure for evaluating reproducibility in geoscientific publications, with a case study for the climate informatics community. While the recommendations focus on future CIRCs, we expect they would be beneficial for wider umbrella of reproducibility initiatives in geosciences.

Simulated data from the Coupled Model Intercomparison Project Phase 6 (CMIP6) has been very important for climate science research, as they can provide wide spatio-temporal coverage to address data deficiencies in both present and future scenarios. However, these physics-based models require a huge amount of high-performance computing (HPC) resources. As an alternative approach, researchers are exploring if such simulated data can be generated by Generative Machine Learning models. In this work, we develop a model based on Pix2Pix conditional Generative Adversarial Network (cGAN), which can generate high-resolution spatial maps of global sea surface temperature (SST) using comparatively less computing power and time. We have shown that the maps generated by these models have similar statistical characteristics as the CMIP6 model simulations. Notably, we trained and validated our cGAN model on completely distinct time periods across all ensemble members of the EC-Earth3-CC and CMCC-CM2-SR5 CMIP6 models, demonstrating satisfactory results and confirming the generalizability of our proposed model.

Snow is a crucial element of the sea ice system, affecting the sea ice growth and decay due to its low thermal conductivity and high albedo. Despite its importance, present-day climate models have a very idealized representation of snow, often including just one-layer thermodynamics, omitting several processes that shape its properties. Even though sophisticated snow process models exist, they tend to be excluded in climate modeling due to their prohibitive computational costs. For example, SnowModel is a numerical snow process model developed to simulate the evolution of snow depth and density, blowing snow redistribution and sublimation, snow grain size, and thermal conductivity in a spatially distributed, multilayer snowpack framework. SnowModel can simulate snow distributions on sea ice floes in high spatial (1-m horizontal grid) and temporal (1-hour time step) resolution. However, for simulations spanning over large regions, such as the Arctic Ocean, high-resolution runs face challenges of slow processing speeds and the need for large computational resources. To address these common issues in high-resolution numerical modeling, data-driven emulators are often used. However, these emulators have their caveats, primarily a lack of generalizability and inconsistency with physical laws. In our study, we address these challenges by using a physics-guided approach in developing our emulator. By integrating physical laws that govern changes in snow density due to compaction, we aim to create an emulator that is efficient while also adhering to essential physical principles. We evaluated this approach by comparing three machine learning models: long short-term memory (LSTM), physics-guided LSTM, and Random Forest, across five distinct Arctic regions. Our evaluations indicate that all models achieved high accuracy, with the physics-guided LSTM model demonstrating the most promising results in terms of accuracy and generalizability. Our approach offers a computationally faster way to emulate the SnowModel with high fidelity and a speedup of over 9000 times.

Comprehensive housing stock information is crucial for informing the development of climate resilience strategies aiming to reduce the adverse impacts of extreme climate hazards in high-risk regions like the Caribbean. In this study, we propose an end-to-end workflow for rapidly generating critical baseline exposure data using very high-resolution drone imagery and deep learning techniques. Specifically, our work leverages the segment anything model (SAM) and convolutional neural networks (CNNs) to automate the generation of building footprints and roof classification maps. We evaluate the cross-country generalizability of the CNN models to determine how well models trained in one geographical context can be adapted to another. Finally, we discuss our initiatives for training and upskilling government staff, community mappers, and disaster responders in the use of geospatial technologies. Our work emphasizes the importance of local capacity building in the adoption of AI and Earth Observation for climate resilience in the Caribbean.

Outdoor air pollution is estimated to cause a huge number of premature deaths worldwide. It catalyzes many diseases on a variety of time scales, and it has a detrimental effect on the environment. In light of these impacts, it is necessary to obtain a better understanding of the dynamics and statistics of measured air pollution concentrations, including temporal fluctuations of observed concentrations and spatial heterogeneities. Here, we present an extensive analysis for measured data from Europe. The observed probability density functions (PDFs) of air pollution concentrations depend very much on the spatial location and the pollutant substance. We analyze a large number of time series data from 3544 different European monitoring sites and show that the PDFs of nitric oxide ($ NO $), nitrogen dioxide ($ {NO}_2 $), and particulate matter ($ {PM}_{10} $ and $ {PM}_{2.5} $) concentrations generically exhibit heavy tails. These are asymptotically well approximated by $ q $-exponential distributions with a given entropic index $ q $ and width parameter $ \lambda $. We observe that the power-law parameter $ q $ and the width parameter $ \lambda $ vary widely for the different spatial locations. We present the results of our data analysis in the form of a map that shows which parameters $ q $ and $ \lambda $ are most relevant in a given region. A variety of interesting spatial patterns is observed that correlate to the properties of the geographical region. We also present results on typical time scales associated with the dynamical behavior.

The growing demand for global wind power production, driven by the critical need for sustainable energy sources, requires reliable estimation of wind speed vertical profiles for accurate wind power prediction and comprehensive wind turbine performance assessment. Traditional methods relying on empirical equations or similarity theory face challenges due to their restricted applicability beyond the surface layer. Although recent studies have utilized various machine learning techniques to vertically extrapolate wind speeds, they often focus on single levels and lack a holistic approach to predicting entire wind profiles. As an alternative, this study introduces a proof-of-concept methodology utilizing TabNet, an attention-based sequential deep learning model, to estimate wind speed vertical profiles from coarse-resolution meteorological features extracted from a reanalysis dataset. To ensure that the methodology is applicable across diverse datasets, Chebyshev polynomial approximation is employed to model the wind profiles. Trained on the meteorological features as inputs and the Chebyshev coefficients as targets, the TabNet more-or-less accurately predicts unseen wind profiles for different wind conditions, such as high shear, low shear/well-mixed, low-level jet, and high wind. Additionally, this methodology quantifies the correlation of wind profiles with prevailing atmospheric conditions through a systematic feature importance assessment.

This article addresses the challenges of assessing pedestrian-level wind conditions in urban environments using a deep learning approach. The influence of large buildings on urban wind patterns has significant implications for thermal comfort, pollutant transport, pedestrian safety, and energy usage. Traditional methods, such as wind tunnel testing, are time-consuming and costly, leading to a growing interest in computational methods like computational fluid dynamics (CFD) simulations. However, CFD still requires a significant time investment for such studies, limiting the available time for design modification prior to lockdown. This study proposes a deep learning surrogate model based on a MLP-mixer architecture to predict mean flow conditions for complex arrays of buildings. The model is trained on a diverse dataset of synthetic geometries and corresponding CFD simulations, demonstrating its effectiveness in capturing intricate wind dynamics. The article discusses the model architecture and data preparation and evaluates its performance qualitatively and quantitatively. Results show promising capabilities in replicating key wind features with a mean error of 0.3 m/s and rarely exceeding 0.75 m/s, making the proposed model a valuable tool for early-stage urban wind modelling.

Climate models are biased with respect to real-world observations. They usually need to be adjusted before being used in impact studies. The suite of statistical methods that enable such adjustments is called bias correction (BC). However, BC methods currently struggle to adjust temporal biases. Because they mostly disregard the dependence between consecutive time points. As a result, climate statistics with long-range temporal properties, such as the number of heatwaves and their frequency, cannot be corrected accurately. This makes it more difficult to produce reliable impact studies on such climate statistics. This article offers a novel BC methodology to correct temporal biases. This is made possible by rethinking the philosophy behind BC. We will introduce BC as a time-indexed regression task with stochastic outputs. Rethinking BC enables us to adapt state-of-the-art machine learning (ML) attention models and thereby learn different types of biases, including temporal asynchronicities. With a case study of number of heatwaves in Abuja, Nigeria and Tokyo, Japan, we show more accurate results than current climate model outputs and alternative BC methods.

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.

Bias correction is a critical aspect of data-centric climate studies, as it aims to improve the consistency between observational data and simulations by climate models or estimates by remote sensing. Satellite-based estimates of climatic variables like precipitation often exhibit systematic bias when compared to ground observations. To address this issue, the application of bias correction techniques becomes necessary. This research work examines the use of deep learning to reduce the systematic bias of satellite estimations at each grid location while maintaining the spatial dependency across grid points. More specifically, we try to calibrate daily precipitation values of tropical rainfall measuring mission based TRMM_3B42_Daily precipitation data over Indian landmass with ground observations recorded by India Meteorological Department (IMD). We have focused on the precipitation estimates of the Indian Summer Monsoon Rainfall (ISMR) period (June–September) since India gets more than 75% of its annual rainfall in this period. We have benchmarked these deep learning methods against standard statistical methods like quantile mapping and quantile delta mapping on the above datasets. The comparative analysis shows the effectiveness of the deep learning architecture in bias correction.

Sea Surface Height Anomaly (SLA) is a signature of the mesoscale dynamics of the upper ocean. Sea surface temperature (SST) is driven by these dynamics and can be used to improve the spatial interpolation of SLA fields. In this study, we focused on the temporal evolution of SLA fields. We explored the capacity of deep learning (DL) methods to predict short-term SLA fields using SST fields. We used simulated daily SLA and SST data from the Mercator Global Analysis and Forecasting System, with a resolution of (1/12)° in the North Atlantic Ocean (26.5–44.42°N, −64.25–41.83°E), covering the period from 1993 to 2019. Using a slightly modified image-to-image convolutional DL architecture, we demonstrated that SST is a relevant variable for controlling the SLA prediction. With a learning process inspired by the teaching-forcing method, we managed to improve the SLA forecast at 5 days by using the SST fields as additional information. We obtained predictions of 12 cm (20 cm) error of SLA evolution for scales smaller than mesoscales and at time scales of 5 days (20 days) respectively. Moreover, the information provided by the SST allows us to limit the SLA error to 16 cm at 20 days when learning the trajectory.