Seasonal Arctic sea ice forecasting with probabilistic deep learning

Increased methane emissions from oil and gas following the Soviet Union's collapse

Global atmospheric methane concentrations rose by 10 to 15 ppb/y in the 1980s before abruptly slowing to 2 to 8 ppb/y in the early 1990s. This period in the 1990s is known as the "methane slowdown" and has been attributed in part to the collapse of the former Soviet Union (USSR) in December 1991, which may have decreased the methane emissions from oil and gas operations. Here, we develop a methane plume detection system based on probabilistic deep learning and human-labeled training data. We use this method to detect methane plumes from Landsat 5 satellite observations over Turkmenistan from 1986 to 2011. We focus on Turkmenistan because economic data suggest it could account for half of the decline in oil and gas emissions from the former USSR. We find an increase in both the frequency of methane plume detections and the magnitude of methane emissions following the collapse of the USSR. We estimate a national loss rate from oil and gas infrastructure in Turkmenistan of more than 10% at times, which suggests the socioeconomic turmoil led to a lack of oversight and widespread infrastructure failure in the oil and gas sector. Our finding of increased oil and gas methane emissions from Turkmenistan following the USSR's collapse casts doubt on the long-standing hypothesis regarding the methane slowdown, begging the question: "what drove the 1992 methane slowdown?"

A deep learning model for reconstructing centenary water storage changes in the Yangtze River Basin

Since 2002, the Gravity Recovery and Climate Experiment (GRACE) and its Follow-On mission (GRACE-FO) have facilitated highly accurate observations of changes in total water storage anomalies (TWSA). However, limited observations of TWSA derived from GRACE in the Yangtze River Basin (YRB) have hindered our understanding of its long-term variability. In this paper, we present a deep learning model called RecNet to reconstruct the climate-driven TWSA in the YRB from 1923 to 2022. The RecNet model is trained on precipitation, temperature, and GRACE observations with a weighted mean square error (WMSE) loss function. The performance of the RecNet model is validated and compared against GRACE data, water budget estimates, hydrological models, drought indices, and existing reconstruction datasets. The results indicate that the RecNet model can successfully reconstruct historical water storage changes, surpassing the performance of previous studies. In addition, the reconstructed datasets are utilized to assess the frequency of extreme hydrological conditions and their teleconnections with major climate patterns, including the El Niño-Southern Oscillation, Indian Ocean Dipole, Pacific Decadal Oscillation, and North Atlantic Oscillation. Independent component analysis is employed to investigate individual climate patterns' unique or combined influence on TWSA. We show that the YRB exhibits a notable vulnerability to extreme events, characterized by a recurrent occurrence of diverse extreme dry/wet conditions throughout the past century. Wavelet coherence analysis reveals significant coherence between the climate patterns and TWSA across the entire basin. The reconstructed datasets provide valuable information for studying long-term climate variability and projecting future droughts and floods in the YRB, which can inform effective water resource management and climate change adaptation strategies.

Unsupervised feature extraction of aerial images for clustering and understanding hazardous road segments

Aerial image data are becoming more widely available, and analysis techniques based on supervised learning are advancing their use in a wide variety of remote sensing contexts. However, supervised learning requires training datasets which are not always available or easy to construct with aerial imagery. In this respect, unsupervised machine learning techniques present important advantages. This work presents a novel pipeline to demonstrate how available aerial imagery can be used to better the provision of services related to the built environment, using the case study of road traffic collisions (RTCs) across three cities in the UK. In this paper, we show how aerial imagery can be leveraged to extract latent features of the built environment from the purely visual representation of top-down images. With these latent image features in hand to represent the urban structure, this work then demonstrates how hazardous road segments can be clustered to provide a data-augmented aid for road safety experts to enhance their nuanced understanding of how and where different types of RTCs occur.

Newly reconstructed Arctic surface air temperatures for 1979-2021 with deep learning method

A precise Arctic surface air temperature (SAT) dataset, that is regularly updated, has more complete spatial and temporal coverage, and is based on instrumental observations, is critically important for timely monitoring and improving understanding of the rapid change in the Arctic climate. In this study, a new monthly gridded Arctic SAT dataset dated back to 1979 was reconstructed with a deep learning method by combining surface air temperatures from multiple data sources. The source data include the observations from land station of GHCN (Global Historical Climatology Network), ICOADS (International Comprehensive Ocean-Atmosphere Data Set) over the oceans, drifting ice station of Russian NP (North Pole), and buoys of IABP (International Arctic Buoy Programme). The last two are crucial for improving the representation of the in-situ observed temperatures within the Arctic. The newly reconstructed dataset includes monthly Arctic SAT beginning in 1979 and daily Arctic SAT beginning in 2011. This dataset would represent a new improvement in developing observational temperature datasets and can be used for a variety of applications.

Future climate imposes pressure on vulnerable ecological regions in China

Ecological regions of medium fragility account for 55 % of China's land. Large-scale afforestation and land reclamation have been carried out in these areas over the past few decades. However, how future climate change poses risks and challenges to them remains unclear. By establishing a multi-algorithm framework combining machine learning algorithms with multi-source dataset, our work predicts Normalized Difference Vegetation Index (NDVI, a proxy for vegetation greenness) and its variations in the 21st century under different climate scenarios. We find that vegetation greening (i.e., NDVI increase) in northern and southwestern China is unstable over four 20-year periods from 2020 to 2100. However, a strikingly prominent greening is expected to occur on the Qinghai-Tibet Plateau until the end of this century. Future warming can not only exacerbate the difficulties of vegetation conservation and restoration in vulnerable ecological regions, also threaten these new croplands, stymieing ambitions to increase crop production in China. Our results underscore the crucible that a warming climate presents to current restoration projects. We highlight the urgency of adapting to climate change to achieve ambitious goals of carbon sequestration and food security in China.

Machine Learning Model-Based Ice Cover Forecasting for a Vital Waterway in Large Lakes

The St. Marys River is a key waterway that supports the navigation activities in the Laurentian Great Lakes. However, high year-to-year fluctuations in ice conditions pose a challenge to decision making with respect to safe and effective navigation, lock operations, and ice breaking operations. The capability to forecast the ice conditions for the river system can greatly aid such decision making. Small-scale features and complex physics in the river system are difficult to capture by process-based numerical models that are often used for lake-wide applications. In this study, two supervised machine learning methods, the Long Short-Term Memory (LSTM) model and the Extreme Gradient Boost (XGBoost) algorithm are applied to predict the ice coverage on the St. Marys River for short-term (7-day) and sub-seasonal (30-day) time scales. Both models are trained using 25 years of meteorological data and select climate indices. Both models outperform the baseline forecast in the short-term applications, but the models underperform the baseline forecast in the sub-seasonal applications. The model accuracies are high in the stable season, while they are lower in the freezing and melting periods when ice conditions can change rapidly. The errors of the predicted ice-on/ice-off date lie within 2–5 days.

A deep-learning estimate of the decadal trends in the Southern Ocean carbon storage

Uptake of atmospheric carbon by the ocean, especially at high latitudes, plays an important role in offsetting anthropogenic emissions. At the surface of the Southern Ocean south of 30∘S, the ocean carbon uptake, which had been weakening in 1990s, strengthened in the 2000s. However, sparseness of in-situ measurements in the ocean interior make it difficult to compute changes in carbon storage below the surface. Here we develop a machine-learning model, which can estimate concentrations of dissolved inorganic carbon (DIC) in the Southern Ocean up to 4 km depth only using data available at the ocean surface. Our model is fast and computationally inexpensive. We apply it to calculate trends in DIC concentrations over the past three decades and find that DIC decreased in the 1990s and 2000s, but has increased, in particular in the upper ocean since the 2010s. However, the particular circulation dynamics that drove these changes may have differed across zonal sectors of the Southern Ocean. While the near-surface decrease in DIC concentrations would enhance atmospheric CO2 uptake continuing the previously-found trends, weakened connectivity between surface and deep layers and build-up of DIC in deep waters could reduce the ocean's carbon storage potential.

An Evaluation of the Performance of Sea Ice Thickness Forecasts to Support Arctic Marine Transport

In response to declining sea ice cover, human activity in the Arctic is increasing, with access to the Arctic Ocean becoming more important for socio-economic reasons. Accurate knowledge of sea ice conditions is therefore becoming increasingly important for reducing the risk and operational cost of human activities in the Arctic. Satellite-based sea ice charting is routinely used for tactical ice management, but the marine sector does not yet make optimal use of sea ice thickness (SIT) or sea ice concentration (SIC) forecasts on weekly timescales. This is because forecasts have not achieved sufficient accuracy, verification and resolution to be used in situations where maritime safety is paramount, and assessing the suitability of forecasts can be difficult because they are often not available in the appropriate format. In this paper, existing SIT forecasts currently available on the Copernicus Marine Service (CMS) or elsewhere in the public domain are evaluated for the first time. These include the seven-day forecasts from the UK Met Office, MET Norway, the Nansen Environmental and Remote Sensing Center (NERSC) and the European Centre for Medium-Range Weather Forecasts (ECMWF). Their forecast skills were assessed against unique in situ data from five moorings deployed between 2016 and 2019 by the Barents Sea Metocean and Ice Network (BASMIN) and Barents Sea Exploration Collaboration (BaSEC) Joint Industry Projects. Assessing these models highlights the importance of data assimilation in short-term forecasting of SIT and suggests that improved assimilation of sea ice data could increase the utility of forecasts for navigational purposes. This study also demonstrates that forecasts can achieve similar or improved correlation with observations when compared to a persistence model at a lead time of seven days, providing evidence that, when used in conjunction with sea ice charts, SIT forecasts could provide valuable information on future sea ice conditions.

Neural Network Prediction for Ice Shapes on Airfoils Using iceFoam Simulations

In this article the procedure and method for the ice accretion prediction for different airfoils using artificial neural networks (ANNs) are discussed. A dataset for the neural network is based on the numerical experiment results—obtained through iceFoam solver—with four airfoils (NACA0012, General Aviation, Business Jet, and Commercial Transport). Input data for neural networks include airfoil and ice geometries, transformed into a set of parameters using a parabolic coordinate system and Fourier series expansion. Besides input features include physical parameters of flow (velocity, temperature, droplets diameter, liquid water content, time of ice accretion) and angle of attack. The novelty of this work is in that the neural network dataset includes various airfoils and the data augmentation technique being a combination of all time slices. Several artificial neural networks (ANNs), fully connected networks (FCNNs), and convolutional networks (CNNs) were trained to predict airfoil ice shapes. Two different loss functions were considered. In order to improve performance of models, batch normalization and dropout layers were used. The most accurate results of ice shape prediction were obtained using CNN and FCNN that applied batch normalization and dropout layers to output neurons of each layer.