Recent years have seen a surge in interest in building deep learning-based fully data-driven models for weather prediction. Such deep learning models, if trained on observations can mitigate certain biases in current state-of-the-art weather models, some of which stem from inaccurate representation of subgrid-scale processes. However, these data-driven models, being over-parameterized, require a lot of training data which may not be available from reanalysis (observational data) products. Moreover, an accurate, noise-free, initial condition to start forecasting with a data-driven weather model is not available in realistic scenarios. Finally, deterministic data-driven forecasting models suffer from issues with long-term stability and unphysical climate drift, which makes these data-driven models unsuitable for computing climate statistics. Given these challenges, previous studies have tried to pre-train deep learning-based weather forecasting models on a large amount of imperfect long-term climate model simulations and then re-train them on available observational data. In this article, we propose a convolutional variational autoencoder (VAE)-based stochastic data-driven model that is pre-trained on an imperfect climate model simulation from a two-layer quasi-geostrophic flow and re-trained, using transfer learning, on a small number of noisy observations from a perfect simulation. This re-trained model then performs stochastic forecasting with a noisy initial condition sampled from the perfect simulation. We show that our ensemble-based stochastic data-driven model outperforms a baseline deterministic encoder–decoder-based convolutional model in terms of short-term skills, while remaining stable for long-term climate simulations yielding accurate climatology.

Accurate and reliable disaster forecasting is vital for saving lives and property. Hence, effective disaster management is necessary to reduce the impact of natural disasters and to accelerate recovery and reconstruction. Typhoons are one of the major disasters related to heavy rainfall in Korea. As a typhoon develops in the far ocean, satellite observations are the only means to monitor them. Our study uses satellite observations to propose a deep-learning-based disaster monitoring model for short-term typhoon rainfall forecasting. For this, we consider two deep learning models: a video frame prediction model, Warp and Refine Network (WR-Net), to predict future satellite observations and an image-to-image translation model, geostationary rainfall product (GeorAIn) (based on the Pix2PixCC model), to generate rainfall maps from predicted satellite images. Typhoon Hinnamnor, the worst typhoon case in 2022 in Korea, is selected as a target case for model verification. The results show that the predicted satellite images can capture the structures and patterns of the typhoon. The rainfall maps generated from the GeorAIn model using predicted satellite images show a correlation coefficient of 0.81 for 3-hr and 0.56 for 7-hr predictions. The proposed disaster monitoring model can provide us with practical implications for disaster alerting systems and can be extended to flood-monitoring systems.

Robust feature selection is vital for creating reliable and interpretable machine-learning (ML) models. When designing statistical prediction models in cases where domain knowledge is limited and underlying interactions are unknown, choosing the optimal set of features is often difficult. To mitigate this issue, we introduce a multidata (M) causal feature selection approach that simultaneously processes an ensemble of time series datasets and produces a single set of causal drivers. This approach uses the causal discovery algorithms PC$ {}_1 $ or PCMCI that are implemented in the Tigramite Python package. These algorithms utilize conditional independence tests to infer parts of the causal graph. Our causal feature selection approach filters out causally spurious links before passing the remaining causal features as inputs to ML models (multiple linear regression and random forest) that predict the targets. We apply our framework to the statistical intensity prediction of Western Pacific tropical cyclones (TCs), for which it is often difficult to accurately choose drivers and their dimensionality reduction (time lags, vertical levels, and area-averaging). Using more stringent significance thresholds in the conditional independence tests helps eliminate spurious causal relationships, thus helping the ML model generalize better to unseen TC cases. M-PC$ {}_1 $ with a reduced number of features outperforms M-PCMCI, noncausal ML, and other feature selection methods (lagged correlation and random), even slightly outperforming feature selection based on explainable artificial intelligence. The optimal causal drivers obtained from our causal feature selection help improve our understanding of underlying relationships and suggest new potential drivers of TC intensification.

Uncertainty quantification (UQ) plays a crucial role in data assimilation (DA) since it impacts both the quality of the reconstruction and near-future forecast. However, traditional UQ approaches are often limited in their ability to handle complex datasets and may have a large computational cost. In this paper, we present a new ensemble-based approach to extend the 4DVarNet framework, an end-to-end deep learning scheme backboned on variational DA used to estimate the mean of the state along a given DA window. We use conditional 4DVarNet simulations compliant with the available observations to estimate the 4DVarNet probability density function. Our approach enables to combine both the efficiency of 4DVarNet in terms of computational cost and validation performance with a fast and memory-saving Monte-Carlo based post-processing of the reconstruction, leading to the so-called En4DVarNet estimation of the state pdf. We demonstrate our approach in a case study involving the sea surface height: 4DVarNet is pretrained on an idealized Observation System Simulation Experiment (OSSE), then used on real-world dataset (OSE). The sampling of independent realizations of the state is made among the catalogue of model-based data used during training. To illustrate our approach, we use a nadir altimeter constellation in January 2017 and show how the uncertainties retrieved by combining 4DVarNet with the statistical properties of the training dataset lead to a relevant information providing in most cases a confidence interval compliant with the Cryosat-2 nadir alongtrack dataset kept for validation.

Cirrus clouds are key modulators of Earth’s climate. Their dependencies on meteorological and aerosol conditions are among the largest uncertainties in global climate models. This work uses 3 years of satellite and reanalysis data to study the link between cirrus drivers and cloud properties. We use a gradient-boosted machine learning model and a long short-term memory network with an attention layer to predict the ice water content and ice crystal number concentration. The models show that meteorological and aerosol conditions can predict cirrus properties with R2 = 0.49. Feature attributions are calculated with SHapley Additive exPlanations to quantify the link between meteorological and aerosol conditions and cirrus properties. For instance, the minimum concentration of supermicron-sized dust particles required to cause a decrease in ice crystal number concentration predictions is 2 × 10−4 mg/m3. The last 15 hr before the observation predict all cirrus properties.

Satellite imagery can detect temporary cloud trails or ship tracks formed from aerosols emitted from large ships traversing our oceans, a phenomenon that global climate models cannot directly reproduce. Ship tracks are observable examples of marine cloud brightening, a potential solar climate intervention that shows promise in helping combat climate change. In this paper, we demonstrate a simulation-based approach in learning the behavior of ship tracks based upon a novel stochastic emulation mechanism. Our method uses wind fields to determine the movement of aerosol–cloud tracks and uses a stochastic partial differential equation (SPDE) to model their persistence behavior. This SPDE incorporates both a drift and diffusion term which describes the movement of aerosol particles via wind and their diffusivity through the atmosphere, respectively. We first present our proposed approach with examples using simulated wind fields and ship paths. We then successfully demonstrate our tool by applying the approximate Bayesian computation method-sequential Monte Carlo for data assimilation.

Simulating abundances of stable water isotopologues, that is, molecules differing in their isotopic composition, within climate models allows for comparisons with proxy data and, thus, for testing hypotheses about past climate and validating climate models under varying climatic conditions. However, many models are run without explicitly simulating water isotopologues. We investigate the possibility of replacing the explicit physics-based simulation of oxygen isotopic composition in precipitation using machine learning methods. These methods estimate isotopic composition at each time step for given fields of surface temperature and precipitation amount. We implement convolutional neural networks (CNNs) based on the successful UNet architecture and test whether a spherical network architecture outperforms the naive approach of treating Earth’s latitude-longitude grid as a flat image. Conducting a case study on a last millennium run with the iHadCM3 climate model, we find that roughly 40% of the temporal variance in the isotopic composition is explained by the emulations on interannual and monthly timescale, with spatially varying emulation quality. The tested CNNs outperform simple baseline models such as random forest and pixel-wise linear regression substantially. A modified version of the standard UNet architecture for flat images yields results that are as good as the predictions by the spherical CNN. Variations in the implementation of isotopes between climate models likely contribute to an observed deterioration of emulation results when testing on data obtained from different climate models than the one used for training. Future work toward stable water-isotope emulation might focus on achieving robust climate–oxygen isotope relationships or exploring the set of possible predictor variables.

Large-scale atmospheric circulation patterns, so-called weather regimes, modulate the occurrence of extreme events such as heatwaves or extreme precipitation. In their role as mediators between long-range teleconnections and local impacts, weather regimes have demonstrated potential in improving long-term climate projections as well as sub-seasonal to seasonal forecasts. However, existing methods for identifying weather regimes are not specifically designed to capture the relevant physical processes responsible for variations in the impact variable in question. This paper introduces a novel probabilistic machine learning method, RMM-VAE, for identifying weather regimes targeted to a local-scale impact variable. Based on a variational autoencoder architecture, the method combines non-linear dimensionality reduction with a prediction task and probabilistic clustering in one coherent architecture. The new method is applied to identify circulation patterns over the Mediterranean region targeted to precipitation over Morocco and compared to three existing approaches: two established linear methods and another machine-learning approach. The RMM-VAE method identifies regimes that are more predictive of the target variable compared to the two linear methods, both in terms of terciles and extremes in precipitation, while also improving the reconstruction of the input space. Further, the regimes identified by the RMM-VAE method are also more robust and persistent compared to the alternative machine learning method. The results demonstrate the potential benefit of the new method for use in various climate applications such as sub-seasonal forecasting, and illustrate the trade-offs involved in targeted clustering.

We have developed probabilistic models to estimate the likelihood of harmful algae presence and outbreaks along the Norwegian coast, which can help optimization of the national monitoring program and the planning of mitigation actions. We employ support vector machines to calibrate probabilistic models for estimating the presence and harmful abundance (HA) of eight toxic algae found along the Norwegian coast, including Alexandrium spp., Alexandrium tamarense, Dinophysis acuta, Dinophysis acuminata, Dinophysis norvegica, Pseudo-nitzschia spp., Protoceratium reticulatum, and Azadinium spinosum. The inputs are sea surface temperature, photosynthetically active radiation, mixed layer depth, and sea surface salinity. The probabilistic models are trained with data from 2006 to 2013 and tested with data from 2014 to 2019. The presence models demonstrate good statistical performance across all taxa, with R (observed presence frequency vs. predicted probability) ranging from 0.69 to 0.98 and root mean squared error ranging from 0.84% to 7.84%. Predicting the probability of HA is more challenging, and the HA models only reach skill with four taxa (Alexandrium spp., A. tamarense, D. acuta, and A. spinosum). There are large differences in seasonal and geographical variability and sensitivity to the model input of different taxa, which are presented and discussed. The models estimate geographical regions and periods with relatively higher risk of toxic species presence and HA, and might optimize the harmful algae monitoring. The method can be extended to other regions as it relies only on remote sensing and model data as input and running national programs of toxic algae monitoring.

Twelve climate models and observations are used to attribute the global mean surface temperature (GMST) changes from 1900 to 2014 to external climate forcings. The external forcings are decomposed into the effects of the well-mixed greenhouse gas concentration variation, the effects of anthropogenic aerosol concentration changes, and the effects of natural forcings. First, a convolutional neural network (CNN) is trained to estimate the simulated historical GMST from single-forcing experiments using outputs from the multi-model ensemble. We then use this CNN to solve the attribution problem using an original variational inversion approach. The variational inversion is first validated using historical climate simulations as pseudo-observations. Then we perform an inversion from observations. This provides a distribution of the GMST resulting from the three forcings. For 2014, inversions estimate that the greenhouse gases changes are responsible for a GMST anomaly within [0.8 $ {}^{\circ } $ C,1.9 $ {}^{\circ } $ C], while anthropogenic aerosols and natural forcings anomalies are within [−0.7 $ {}^{\circ } $ C,−0.1 $ {}^{\circ } $ C] and [−0.1 $ {}^{\circ } $ C,0.3 $ {}^{\circ } $ C], respectively. The method designed here can be adapted and extended to attribute the changes of other variables or to focus on the regional scale.

Canopy habitats challenge researchers with their intrinsically difficult access. The current scarcity of climatic data from forest canopies limits our understanding of the conditions and environmental variability of these diverse and dynamic habitats. We present 307 days of climate records collected between 2019 and 2020 in the tropical rainforest canopy of the Yasuní National Park, Ecuador. We monitored climate with a 10-min temporal resolution in the middle crowns of eight canopy trees. The distance between canopy climate stations ranged from 700 m to 10 km. Apart from air temperature, relative humidity, leaf wetness, and photosynthetically active radiation (PAR), measured in each canopy climate station, global radiation, rainfall, and wind speed were measured in different subsets of them. We processed the eight data series to omit erroneous records resulting from sensor failures or lack of the solar-based power supply. In addition to the eight original data series, we present three derived data series, two aggregating canopy climate for valleys or for ridges (from four stations each), and one overall average (from the eight stations). This last derived data series contains 306 days, while the shortest of the original data series covers 22 days and the longest 296 days. In addition to the data, two open-source tools, developed in RStudio, are presented that facilitate data visualization (a dashboard) and data exploration (a filtering app) of the original and aggregated records.

Transparent, understandable, and persuasive recommendations support the electricity consumers’ behavioral change to tackle the energy efficiency problem. This paper proposes an explainable multi-agent recommendation system for load shifting for household appliances. First, we extend a novel multi-agent approach by designing and implementing an Explainability Agent that provides explainable recommendations for optimal appliance scheduling in a textual and visual manner. Second, we enhance the predictive capacity of other agents by including weather data and applying state-of-the-art models (i.e., k-nearest-neighbors, extreme gradient boosting, adaptive boosting, Random Forest, logistic regression, and explainable boosting machines). Since we want to help the user understand a single recommendation, we focus on local explainability approaches. In particular, we apply post-model approaches local, interpretable, model-agnostic explanation and SHapley Additive exPlanations as model-agnostic tools that can explain the predictions of the chosen classifiers. We further provide an overview of the predictive and explainability performance. Our results show a substantial improvement in the performance of the multi-agent system while at the same time opening up the “black box” of recommendations.

Domain adaptation is important in agriculture because agricultural systems have their own individual characteristics. Applying the same treatment practices (e.g., fertilization) to different systems may not have the desired effect due to those characteristics. Domain adaptation is also an inherent aspect of digital twins. In this work, we examine the potential of transfer learning for domain adaptation in pasture digital twins. We use a synthetic dataset of grassland pasture simulations to pretrain and fine-tune machine learning metamodels for nitrogen response rate prediction. We investigate the outcome in locations with diverse climates, and examine the effect on the results of including more weather and agricultural management practices data during the pretraining phase. We find that transfer learning seems promising to make the models adapt to new conditions. Moreover, our experiments show that adding more weather data on the pretraining phase has a small effect on fine-tuned model performance compared to adding more management practices. This is an interesting finding that is worth further investigation in future studies.

This paper proposes an approach to diagnosing the skill of a machine-learning prediction model based on finding combinations of variables that minimize the normalized mean square error of the predictions. This technique is attractive because it compresses the positive skill of a forecast model into the smallest number of components. The resulting components can then be analyzed much like principal components, including the construction of regression maps for investigating sources of skill. The technique is illustrated with a machine-learning model of week 3–4 predictions of western US wintertime surface temperatures. The technique reveals at least two patterns of large-scale temperature variations that are skillfully predicted. The predictability of these patterns is generally consistent between climate model simulations and observations. The predictability is determined largely by sea surface temperature variations in the Pacific, particularly the region associated with the El Nino-Southern Oscillation. This result is not surprising, but the fact that it emerges naturally from the technique demonstrates that the technique can be helpful in “explaining” the source of predictability in machine-learning models.

We present a data-driven emulator, a stochastic weather generator (SWG), suitable for estimating probabilities of prolonged heat waves in France and Scandinavia. This emulator is based on the method of analogs of circulation to which we add temperature and soil moisture as predictor fields. We train the emulator on an intermediate complexity climate model run and show that it is capable of predicting conditional probabilities (forecasting) of heat waves out of sample. Special attention is payed that this prediction is evaluated using a proper score appropriate for rare events. To accelerate the computation of analogs, dimensionality reduction techniques are applied and the performance is evaluated. The probabilistic prediction achieved with SWG is compared with the one achieved with a convolutional neural network (CNN). With the availability of hundreds of years of training data, CNNs perform better at the task of probabilistic prediction. In addition, we show that the SWG emulator trained on 80 years of data is capable of estimating extreme return times of order of thousands of years for heat waves longer than several days more precisely than the fit based on generalized extreme value distribution. Finally, the quality of its synthetic extreme teleconnection patterns obtained with SWG is studied. We showcase two examples of such synthetic teleconnection patterns for heat waves in France and Scandinavia that compare favorably to the very long climate model control run.

Residential energy efficiency programs play an important role in combating climate change. More precise quantification of the magnitude and timing of energy savings would bring large system benefits, allowing closer integration of energy efficiency into resource adequacy planning and balancing variable renewable electricity. However, it is often difficult to quantify the efficacy of an energy efficiency intervention, because doing so requires consideration of a hypothetical counterfactual case in which there was no intervention, and randomized control trials are often implausible. Although quasi-experimental econometric evaluation sometimes works well, we find that for a set of energy efficiency rebate programs in Northern California, a naïve interpretation of econometric measurement finds that rebate participation is associated with an average increase in electricity consumption of 7.2% [4.5%, 10.1%], varying in magnitude and sign depending on the type of appliance or service covered by the rebate. A subsequent household survey on appliance purchasing behavior and analysis of utility customer outreach data suggest that this regression approach is likely measuring the gross impact of buying a new appliance but fails to adequately capture a counterfactual comparison. Indeed, it is unclear whether it is even possible to construct a suitable counterfactual for econometric analyses of these rebate programs using data generally available to electric utilities. We view these results as an illustration of a limitation of econometric methods of program evaluation and the importance of weighing engineering modeling and other imperfect methods against one another when attempting to provide useful evaluations of real-world policy interventions.

Complex physical processes that are inherent to rainfall lead to the challenging task of its prediction. To contribute to the improvement of rainfall prediction, artificial neural network (ANN) models were developed using a multilayer perceptron (MLP) approach to predict monthly rainfall 2 months in advance for six geographically diverse weather stations across the Benin Republic. For this purpose, 12 lagged values of atmospheric data were used as predictors. The models were trained using data from 1959 to 2017 and tested for 4 years (2018–2021). The proposed method was compared to long short-term memory (LSTM) and climatology forecasts (CFs). The prediction performance was evaluated using five statistical measures: root mean square error, mean absolute error, mean absolute percentage error, coefficient of determination, and Nash–Sutcliffe efficiency (NSE) coefficient. Furthermore, Taylor diagrams, violin plots, box error, and Kruskal–Wallis test were used to assess the robustness of the model’s forecast. The results revealed that MLP gives better results than LSTM and CF. The NSE obtained with the MLP, LSTM, and CF models during the test period ranges from 0.373 to 0.885, 0.297 to 0.875, and 0.335 to 0.845, respectively, depending on the weather station. Rainfall predictability was more accurate, with 0.512 improvement in NSE using MLP at higher latitudes across the country, showing the effect of geographic regions on prediction model results. In summary, this research has revealed the potential of ANN techniques in predicting monthly rainfall 2 months ahead, supplying valuable insights for decision-makers in the Republic of Benin.

Many single-output regression problems require estimates of uncertainty along with the point predictions. For this purpose, there exists a class of regression algorithms that predict a conditional distribution rather than a point estimate. The off-the-shelf options are much more limited, however, when the prediction output is multivariate and a joint measure of uncertainty is required. In this paper, we predict a distribution around a multivariate random vector of dimension P, such that the joint uncertainty would quantify the probability of any vector in P-dimensional space. This is more expressive than providing separate uncertainties in each dimension. To enable joint probabilistic regression, we propose a natural gradient boosting approach based on nonparametrically modeling the conditional parameters of the multivariate predictive distribution, where we focus on the multivariate Gaussian distribution. Our method is robust, can be easily trained without extensive tuning, and performs competitively in comparison to existing approaches. The motivating application of our methodology is to predict two-dimensional oceanographic currents measured by freely floating Global Drifter Program drifters using remotely sensed data. We also demonstrate the method’s performance on simulated data. We find this method excels when strong correlation between output dimensions is present. As part of this work, we have added the model to the open source package at github.com/stanfordmlgroup/ngboost.

We compared climatic relationships to insurance loss across the inland Pacific Northwest region of the United States, using a design matrix methodology, to identify optimum temporal windows for climate variables by county in relationship to wheat insurance loss due to drought. The results of our temporal window construction for water availability variables (precipitation, temperature, evapotranspiration, and the Palmer drought severity index [PDSI]) identified spatial patterns across the study area that aligned with regional climate patterns, particularly with regards to drought-prone counties of eastern Washington. Using these optimum time-lagged correlational relationships between insurance loss and individual climate variables, along with commodity pricing, we constructed a regression-based random forest model for insurance loss prediction and evaluation of climatic feature importance. Our cross-validated model results indicated that PDSI was the most important factor in predicting total seasonal wheat/drought insurance loss, with wheat pricing and potential evapotranspiration having noted contributions. Our overall regional model had a $ {R}^2 $ of 0.49, and a RMSE of $30.8 million. Model performance typically underestimated annual losses, with moderate spatial variability in terms of performance between counties.

Precipitation is one of the most relevant weather and climate processes. Its formation rate is sensitive to perturbations such as by the interactions between aerosols, clouds, and precipitation. These interactions constitute one of the biggest uncertainties in determining the radiative forcing of climate change. High-resolution simulations such as the ICOsahedral non-hydrostatic large-eddy model (ICON-LEM) offer valuable insights into these interactions. However, due to exceptionally high computation costs, it can only be employed for a limited period and area. We address this challenge by developing new models powered by emerging machine learning approaches capable of forecasting autoconversion rates—the rate at which small droplets collide and coalesce becoming larger droplets—from satellite observations providing long-term global spatial coverage for more than two decades. In particular, our approach involves two phases: (1) we develop machine learning models which are capable of predicting autoconversion rates by leveraging high-resolution climate model data, (2) we repurpose our best machine learning model to predict autoconversion rates directly from satellite observations. We compare the performance of our machine learning models against simulation data under several different conditions, showing from both visual and statistical inspections that our approaches are able to identify key features of the reference simulation data to a high degree. Additionally, the autoconversion rates obtained from the simulation output and satellite data (predicted) demonstrate statistical concordance. By efficiently predicting this, we advance our comprehension of one of the key processes in precipitation formation, crucial for understanding cloud responses to anthropogenic aerosols and, ultimately, climate change.