Description: Our research focuses on the detection of ocean carbon uptake regimes that are critical in the context of comprehending climate change. One observation among geoscientific data in Earth System Sciences is that the datasets often contain local and distinct statistical distributions posing a major challenge in applying clustering algorithms for data analysis. The use of global parameters in many clustering algorithms is often inadequate to capture such local distributions. In this study, we propose a novel tool to detect and visualize oceanic carbon uptake clusters. We implement a distance-variance selection method (augmented by BIC scores) on agglomerative hierarchical clustering constructed upon a regional multivariate linear regression model set. Instead of relying on a global distance, users can select the local distance and variance thresholds on our tool to detect the connections on the dendrograms that stand as potential clusters by considering both compactness and similarity.

Description: Our research focuses on detecting and tracking ocean carbon regimes, which are crucial indicators for understanding the impacts of climate change on ocean carbon uptake. Geoscientific datasets in Earth System Sciences often contain local and distinct statistical distributions at a regional scale. This poses a significant challenge in applying conventional clustering algorithms for data analysis. Based on the observed limitations of prominent methods, in our study, we propose a framework that enhances well-established unsupervised machine-learning methods tailored to applications on geoscientific datasets. We define a carbon uptake regime as a region characterized by common relationships between the carbon uptake and its drivers, as simulated by a multi-annual hydrodynamic model simulation. As a first step, we compute multivariate linear regressions capturing local spatial relations between carbon dioxide uptake and its drivers to discover such regimes. This is followed by an agglomerative hierarchical clustering constructed upon the collection of regional multivariate linear regression models. To overcome the emerging limitations of a global cut for partitioning, which is inadequate to capture the local statistical distributions, we present a novel, straightforward and adaptive approach to detect and visualize ocean carbon uptake regimes in this work. This method relies on the distance-variance selection technique and detects multiple local cuts on the dendrogram by considering both the compactness and similarity of the clusters. Detecting meaningful and well-defined carbon uptake regimes is vital for their tracking over time. The tracking is performed through a simple yet effective approach where summary structures derived from the clusters are traced over time. Applied over longer time scales, this novel method will enable marine scientists to effectively monitor the impacts of climate change on the ocean carbon cycle more

Description: Working with observational data in the context of geophysics can be challenging, since we often have to deal with missing data. This requires imputation techniques in pre-processing to obtain data-mining-ready samples. Here, we present a convolutional neural network approach from the domain of deep learning to reconstruct complete information from sparse inputs. As data, we use various two-dimensional geospatial fields. To have consistent data over a sufficiently long time span, we favor to work with output from control simulations of two Earth System Models, namely the Flexible Ocean and Climate Infrastructure and the Community Earth System Model. Our networks can restore complete information from incomplete input samples with varying rates of missing data. Moreover, we apply a bottom-up sampling strategy to identify the most relevant grid points for each input feature. Choosing the optimal subset of grid points allows us to successfully reconstruct current fields and to predict future fields from ultra sparse inputs. As a proof of concept, we predict El Niño Southern Oscillation and rainfall in the African Sahel region from sea surface temperature and precipitation data, respectively. To quantify uncertainty, we compare corresponding climate indices derived from reconstructed versus complete fields.

Description: Machine learning (ML) and in particular deep learning (DL) methods push state-of-the-art solutions for many hard problems, for example, image classification, speech recognition, or time series forecasting. In the domain of climate science, ML and DL are known to be effective for identifying causally linked modes of climate variability as key to understand the climate system and to improve the predictive skills of forecast systems. To attribute climate events in a data-driven way, we need sufficient training data, which is often limited for real-world measurements. The data science community provides standard data sets for many applications. As a new data set, we introduce a consistent and comprehensive collection of climate indices typically used to describe Earth System dynamics. Therefore, we use 1000-year control simulations from Earth System Models. The data set is provided as an open-source framework that can be extended and customized to individual needs. It allows users to develop new ML methodologies and to compare results to existing methods and models as benchmark. For example, we use the data set to predict rainfall in the African Sahel region and El Niño Southern Oscillation with various ML models. Our aim is to build a bridge between the data science community and researchers and practitioners from the domain of climate science to jointly improve our understanding of the climate system.

Description: This paper discusses the challenges of applying a data analytics pipeline for a large volume of data as can be found in natural and life sciences. To address this challenge, we attempt to elaborate an approach for an improved detection of outliers. We discuss an approach for outlier quantification for bathymetric data. As a use case, we selected ocean science (multibeam) data to calculate the outlierness for each data point. The benefit of outlier quantification is a more accurate estimation of which outliers should be removed or further analyzed. To shed light on the subject, this paper is structured as follows: first, a summary of related works on outlier detection is provided. The usefulness for a structured approach of outlier quantification is then discussed using multibeam data. This is followed by a presentation of the challenges for a suitable solution, and the paper concludes with a summary.

Description: Artificial neural networks (ANNs) are known to be powerful methods for many hard problems (e.g. image classification, speech recognition or time series prediction). However, these models tend to produce black-box results and are often difficult to interpret. Layer-wise relevance propagation (LRP) is a widely used technique to understand how ANN models come to their conclusion and to understand what a model has learned. Here, we focus on Echo State Networks (ESNs) as a certain type of recurrent neural networks, also known as reservoir computing. ESNs are easy to train and only require a small number of trainable parameters, but are still black-box models. We show how LRP can be applied to ESNs in order to open the black-box. We also show how ESNs can be used not only for time series prediction but also for image classification: Our ESN model serves as a detector for El Nino Southern Oscillation (ENSO) from sea surface temperature anomalies. ENSO is actually a well-known problem and has been extensively discussed before. But here we use this simple problem to demonstrate how LRP can significantly enhance the explainablility of ESNs.

Description: In the framework of a changing climate, it is useful to devise methods capable of effectively assessing and monitoring the changing landscape of air-sea CO2 fluxes. In this study, we developed an integrated machine learning tool to objectively classify and track marine carbon biomes under seasonally and interannually changing environmental conditions. The tool was applied to the monthly output of a global ocean biogeochemistry model at 0.25° resolution run under atmospheric forcing for the period 1958–2018. Carbon biomes are defined as regions having consistent relations between surface CO2 fugacity (fCO2) and its main drivers (temperature, dissolved inorganic carbon, alkalinity). We detected carbon biomes by using an agglomerative hierarchical clustering (HC) methodology applied to spatial target-driver relationships, whereby a novel adaptive approach to cut the HC dendrogram based on the compactness and similarity of the clusters was employed. Based only on the spatial variability of the target-driver relationships and with no prior knowledge on the cluster location, we were able to detect well-defined and geographically meaningful carbon biomes. A deep learning model was constructed to track the seasonal and interannual evolution of the carbon biomes, wherein a feed-forward neural network was trained to assign labels to detected biomes. We find that the area covered by the carbon biomes responds robustly to seasonal variations in environmental conditions. A seasonal alternation between different biomes is observed over the North Atlantic and Southern Ocean. Long-term trends in biome coverage over the 1958–2018 period, namely a 10 % expansion of the subtropical biome in the North Atlantic and a 10 % expansion of the subpolar biome in the Southern Ocean, are suggestive of long-term climate shifts. Our approach thus provides a framework that can facilitate the monitoring of the impacts of climate change on the ocean carbon cycle and the evaluation of carbon cycle projections across Earth System Models.