Climate change is a pressing issue of today, for which data-based models and decision support techniques offer a more comprehensive understanding of its complexity. The aim of this paper is to reveal data-based techniques and their applicability in terms of climate researches. More precisely, how can Big Data, through data science answer sustainability climate issues and be applicable in scientific researches and decision sciences in an integrated manner.

The overview is guided through three closely related notions, namely, (1) data science as a novel interdisciplinary field connected to (2) machine learning that is a tool for improving automatic prediction or decision processes, and (3) Big Data which foster processing and connecting large amount of heterogeneous data. The focus point of this research is the interconnectedness of the complex climate-related systems, for which exploration Big Data provides an efficient toolbox.

Research questions formulated three aspects, which answering kept in focus through the whole paper:

The year 2015 brought about further excitement in the field of research directions concerning climate change, as the United Nations declared 17 sustainable development goals, of which SDG13 is “Take urgent action to combat climate and its impacts” (UN, 2016) and the Paris Agreement has been signed, that concerning the mitigation of greenhouse gas emissions, adaptation and finance in 2015 with the specific aim of keeping global average temperature rises well below 2°C above pre-industrial levels and then continuing efforts to keep global temperature rises below 1.5°C above pre-industrial levels, recognizing that this will significantly reduce the risks and impacts of climate change (Rogelj et al., 2016). This kind of organizing principle supports the complex analysis of the classical disciplinary sciences with a holistic, interdisciplinary approach. New types of approaches require much more complex analyses and models and, therefore, several orders of magnitude more data, which brought Big Data to life as a stand-alone scientific discipline.

Big Data-based tools are already widespread in this new complex science, for example, to monitor seasonal changes in climate change (Manogaran et al., 2018), understand climate change as a theory-guided data science paradigm (Faghmous et al., 2014), learn how to manage the risks of climate change (Ford et al., 2016), explore soft data sources, e.g., Twitter (Jang et al., 2015), or demonstrate the potential of Systems of Systems (SoS), for instance, the exploration of the structure and relationships across institutions and disciplines of a global Big Earth Data cyber-infrastructure: the Global Earth Observation System of Systems (GEOSS) (Craglia et al., 2017).

Today, it is obvious that sustainability science is intertwined with data science, however, with the support of the business model of the circular economy (Jabbour et al., 2019), the complexity of the problem repository has further increased, so there is an urgent need to include data and analysis methods in the framework, whereas research results from different fields can be used in other fields. Furthermore, trends in climate and sustainability science are driving models toward higher resolution, greater complexity, and larger ensembles, which calls for multidisciplinary approaches in climate computational sciences (Balaji, 2015). This research provides a higher-level overview of the interconnectedness of disciplines, systems, data, and tools related to climate change, exploring further focal points concerning the need a deeper level of integration, because a disconnection between important industry initiatives and scientific research is still experienced (Nobre and Tavares, 2017). We propose to solve these integration tasks and disconnections by the System of Systems thinking.

This overview seeks to address these shortcomings. Information sources (data, news, scientific databases) can be linked, drawing attention to the future importance of open linked data. The present research draws attention to System of Systems (SoS) thinking, as the drivers and effects of climate change, as well as resilience and adaptation, can only be achieved through the timely recognition and exploitation of synergies and trade-offs between the new research directions.

The research methodology outlines firstly, the identification of sustainability science problems in section 2, which revealed the connected issues and tasks as well as the requirements needed to succeed. It ensured that sustainable operation of nature and society demands the approach of systems of system along with the integration of Big Data applications into climate-related scientific, societal, and political researches. This is in line with the growing risk of uncertainty zones highlighted in the planetary boundary framework (Steffen et al., 2015). Then, the existing applications of the related data analysis in the field was explored. For a deeper and narrowed insight, literature review was based on the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) method, which contributes to the exploration and evaluation of related articles. The search has a clear and narrowed focus on the multidisciplinary nature of the issue, therefore the generic evaluation is not in purpose. Fifty-seven review articles were individually analyzed to identify focus areas and research gaps in the Big Data applications in climate change researches. Systematic meta-analysis was used to identify how data are clustering into diverse focus ares and to extract valuable structural information. The co-occurrences of keywords were examined with regard to 442 articles describing the relationship between climate change and Big Data.

In the following sections, the aforementioned research questions are being unfolded and answered through revealing the increasing importance of the System of Systems theorem. Synergies between new research directions and disciplines must be explored to determine the drivers and effects of climate issues as well as provide an efficient strategic adaptation and mitigation plan that also consider socio-environmental factors. Our proposed SoS framework is a response to this integrated knowledge management, as a first step toward climate computing.

In section 2, the sustainability science theorem questions are answered considering the essential need of data science applications. In section 3, heterogeneous data management as well as Big Data tools and techniques are emphasized.

The systematic review of climate change analyses can be found in section 4, which includes the connections between Big Data and climate in section 4.1 as well as a critical summary of different methods in section 4.2. The social aspects are highlighted in section 4.3. Based on the overview, from the new climate-related research findings, a specific SoS framework is presented in section 4.4 and the intertwining of the SoS and SDGs are discussed in section 5, where the suggestions for future research directions and applications are summarized.

The complexity of climate issues requires adaptive strategies for public policy (Di Gregorio et al., 2019), actions to incite social behavior (Xie B. et al., 2019), and the development of regulatory and market-simulating responses to economic life (Wright and Nyberg, 2017). To meet this complex societal need, research has focused on understanding the causes of climate change (Hegerl et al., 2019), the development of predictive models (Du et al., 2019), and mitigation solutions (Gomez-Zavaglia et al., 2020), as well as the exploration of opportunities to shape social attitudes (Iturriza et al., 2020).

An interdisciplinary approach is essential in terms of the identification of almost every climate-related problem and development of their solutions. This interdisciplinary perspective has formed sustainability science theorem to gain a comprehensive understanding of the interrelationship between environment and society (Kates et al., 2001). This theory focuses on transdisciplinary questions, which can only be answered by applying of data science tools.

To follow the aforementioned path toward sustainable dynamics of nature and society, the data science toolbox and models must be integrated into climate change-related scientific and societal research as well as political agenda. In the following, the Big Data tools and management are interpreted with a specific focus on their role in climate change and we build a System of Systems (climate computing) framework from the various applications.

The term Big Data has spread due to new technologies and innovations that have emerged over the past decade (Chen and Chiang, 2012) given the demand for the analysis of large amounts of and rapidly generated diverse data, therefore, collection and processing takes place at a high speed, which is difficult to implement with calcareous analytical tools (Constantiou and Kallinikos, 2015). The explosive leap in the amount of data has also infiltrated health, finance, and education (Benjelloun et al., 2015). With regard to the global economy, Big Data is key to understanding and increasing performance (Maria et al., 2015). Big Data is also gaining ground in the field of sustainability, so it can be used to improve social and environmental sustainability in supply chains (Dubey et al., 2019), augment the informational landscape of smart sustainable cities (Bibri, 2018), and improve the allocation and utilization of natural resources (Song et al., 2017) as well as supply chain sustainability (Hazen et al., 2016).

Big and open data from “smart” government to transformational government can facilitate collaboration. It is possible to introduce real-time solutions into agriculture, health, transport, and other challenges (Bertot et al., 2014). The Big Data approach can be the most effective tool to improve mutual governmental and civic understanding, thus embodying the principles of digital governance as the most viable public management model (Clarke and Margetts, 2014). There is a need to collect large amounts of data that can be used to model and test different scenarios to sustainably transform energy production and consumption, improve food and water security, as well as eradicate poverty. Initiatives such as the Intergovernmental Panel on Climate Change and the Global Ocean Observing System can fill gaps in scientific, technical and socio-economic data (Gijzen, 2013). The analysis of sustainable business performance forecasts through the analysis of Big Data in the context of developing countries shows that “Management and leadership style” and “Government policy” are the most significant factors at present (Raut et al., 2019).

The process of data mining is shown in Figure 1.

Big Data is a rapidly generated amount of information from a variety of sources and in a different format. Data analysis is the examination and transformation of raw data into interpretable information, while data science is a multidisciplinary field of various analyses, programming tools, and algorithms, forecasting analysis statistics as well as machine learning that aims to recognize and extract patterns in raw data. Thus, Big Data primarily looks at ways to analyse, systematically extract or otherwise handle data from datasets that are too large or complex to handle with traditional data processing application software that requires significant scaling (multiple nodes) to process efficiently. In other words, Big Data can be defined by the 5V key characteristics, i.e., volume, velocity, variety, veracity, and value (Laney, 2001).

The storage, sustainability, and analysis of massive content is a challenge that the current state of algorithms and systems cannot handle (Trifu and Ivan, 2014) in an integrated manner, therefore the synergies of the different sources are not sufficiently exploited. The purpose of using Big Data is to provide data management and analysis tools for the ever-increasing amount of data (Anuradha et al., 2015). As is shown in Figure 2, data analysis can be divided into four general categories (Erl et al., 2016). In the environments of Big Data analytics, data analytics involves the use of highly scalable distributed frameworks and technologies to extract meaningful information from large amounts of raw data that requires the use of different data analysis methods (Rajaraman, 2016).

Big Data is usually associated with two technologies, cloud computing and the Internet of Things (IoT) (Honti and Abonyi, 2019). Cloud computing accelerates unlimited data storage, parallel data processing, and analysis (Inukollu et al., 2014). The key benefits of cloud computing are improved analysis, simplified infrastructure, and cost reduction. IoT offers the ability to connect computing devices, mechanical and digital machines as well as objects and people (Lavin et al., 2015). With the advent of the IoT, huge amounts of data can be collected using smart devices connected via the Internet (Suchetha et al., 2015).

The applicability of Big Data techniques is also significantly enhanced by the novel tools that support data collection and integration. The interoperability of the systems can be improved by data warehouses and the related ETL (extract, transform, load) functionalities that can also be used to gather information from multiple models and data sources. The benefit of these structure are demonstrated in the EC4MACS (European Consortium for Modeling of Air Pollution and Climate Strategies) data warehouse that establishes a suite of modeling tools for a comprehensive integrated assessment of the effectiveness of emission control strategies for air pollutants and greenhouse gases. In this system the integrated data are loaded into the GAINS (Greenhouse gas-Air pollution Interactions and Synergies) Data Warehouse. This assessment brought together expert knowledge in the fields of energy, transport, agriculture, forestry, land use, atmospheric dispersion, health and vegetation impacts, and it developed a coherent outlook into the future options to reduce atmospheric pollution in Europe (Nguyen et al., 2012).

The integration of different information can also be supported by ontology-based linked data. Ontology Web Language (OWL) models enables the semantic characterization of the different events that can describe the climate change story from multiple perspectives, including scientific, social, political, and technological ones (Pileggi et al., 2020).

Artificial intelligence (AI) and machine learning (ML) are also the key enabler technologies of big data analysis. This paper focuses on the applicability of ML-based models. AI is mainly used to support decision-making, but it also can skilfully fill observational gaps when combined with numerical climate model data. An example of this application can be found in the extension of historical temperature measurements used in global climate datasets like HadCRUT4 (Kadow et al., 2020).

Analysis of Big Data combines traditional methods of statistical analysis with computational approaches. Based on the complexity between the variables and the type of results required, data analysis can be a simple data set query or a combination of sophisticated analysis techniques (Al-Shiakhli, 2019). The analysis of Big Data is a synthesis of quantitative and qualitative analyses. Climate computing combines multidisciplinary researches in regard to climatic, data and system sciences to efficiently capture and analyse climate-related Big Data as well as to support socio-environmental efforts. Underlying this aspect, a complex model of the earth system is continuously developed by DKRZ using supercomputers relying on Big Data, numerical computations, and simulation models to enable scientists to integrate chemical and biological processes, as well as investigate the interaction of the climate and the socio-economic system (Klimarechenzentrum, 2021).

Exploratory Data Analysis (EDA) techniques are approaches for analysing large data sets. These techniques make the main features clearer by hiding other aspects. Most EDA techniques are graphical in nature, with some non-graphical additions. Some basic EDA tools are histograms, quantile quantile plots (Q-Q-plots), scatter plots, box plots, stratification, log transformation, and other summary statistics (Komorowski et al., 2016). Qualitative models can be classified into qualitative causal models and abstraction hierarchies. The causal models can be classified into Digraphs, Fault Trees, and Qualitative Physics. Abstraction hierarchies consist of two important components: structural and functional (Venkatasubramanian et al., 2003).

Data mining is a set of methods that extracts certain information from large and complex databases. Data discovery uses automated, software-based techniques to eliminate randomness and uncover hidden patterns and trends (Fayyad and Simoudis, 1997). The classification of data mining techniques is summarized in Table 1 (Zaki and Ho, 2000), including a straightforward description of the method, common analytical techniques, the definition of relevant application areas and examples related to climate studies.

Classification is fundamental in terms of data mining techniques (Zaki and Ho, 2000). Classification models define the similarity structure of the variables and are partitioned into groups (classes) (Aggarwal, 2015). In Big Data-based climate studies, classification models and techniques are greatly utilized. Two streams with different hydroclimatologies were studied in the United States using an artificial neural network (ANN). The analysis identified a large effect on a variety of factors such as average runoff, flow variability, flood frequency and baseline flow stability (Poff et al., 1996). To overcome the great uncertainties inherent in climate models, an alternative neural network-based climate model has been developed that increases the efficiency of large climate model sets by at least one order of magnitude. Based on this, it can be concluded that heating exceeds the surface heating range estimated by the IPCC for almost half of the members of the ensemble (Knutti et al., 2003). This neural network is an effective tool for dealing with such difficult and challenging problems, moreover, has been widely used to explore the mechanisms of climate change and predict trends is climate change that take full advantage of the unknown information hidden in climate data, however, it cannot decipher it.

General Circulation Models (GCMs)—the most advanced tools for estimating future climate change scenarios- operate on a coarse scale, which can be downscaled by support vector machine (SVM) approaches, training meteorological subdivisions (MSDs) and developing a downscaling model (DM) that has been shown to be better than conventional downscaling using multilayered regenerative artificial neural networks (Tripathi et al., 2006). The utilization of solar energy is evolving dynamically in connection with SDG 7, but power plant performance may fluctuate due to the diversity of meteorological conditions, which can be compensated by satellite imagery and SVM learning scheme to predict the motion vector of clouds (Jang et al., 2016). Object-based image analysis (OBIA) and support vector machine (SVM) combined with a decision-tree classification are suitable for mapping mangrove areas that was impossible by traditional remote sensing methods other than rough spatial resolution (Heumann, 2011). Decision tree algorithms consistently outperform maximum likelihood and linear discriminant function classifiers in terms of land cover mapping problems classification accuracy (Friedl and Brodley, 1997). Using a weather-generating model,which allows the nearest neighbor to be re-sampled by disturbing historical data, it is possible to create a set of climatic scenarios based on probable climatic scenarios to produce meteorological data that can be used to assess the vulnerability of the river basin to extreme events (Sharif and Burn, 2006). The ability of the Bayesian Network (BN) to predict long-term changes in the shoreline associated with rises in sea level and quantitatively estimate forecast uncertainty renders it suitable for research into the effects of climate change (Gutierrez et al., 2011). It has been used successfully to assess the effects of climate change disturbances on the structure of coral reefs (Franco et al., 2016) and in terms of belief updating concerning the reality of climate change in response to presenting information concerning the scientific consensus on anthropogenic global warming (AGW) (Cook and Lewandowsky, 2016). Using genetic algorithm and occurrence data from museum specimens, ecological niche models were developed for 1,870 species occurring in Mexico and projected onto two climatic surfaces modeled for 2055 (Peterson et al., 2002). A multi-objective genetic algorithm for optimizing water distribution systems (WDS) was used as a discovery tool to examine trade-offs between traditional economic goals and minimize greenhouse gas emissions (Wu et al., 2010). The European territory was subdivided into similar regions of predicted climate change based on simulations of total daily precipitation as well as recent (1986–2005) and long-term future (2081–2100) temperatures using K-mean cluster analysis (Carvalho et al., 2016). An automated procedure based on a cluster initialization algorithm is proposed and applied to changes in the 27 climatic extremes. The proposed method requires, on average, 40% fewer scenarios to meet the 90% threshold than k-means clustering (Cannon, 2015).

Clustering-based analyses are widely accepted data mining techniques, however, improvements in terms of time and cost savings are constantly required due to the management of an increasing amount of data (Shirkhorshidi et al., 2014). Regarding its usage in climatic analyses, a clustering-based spatio-temporal analysis framework of atmospheric data was developed to support both governmental and industrial decision-making processes (Cuzzocrea et al., 2019). To assess erosivity risk, clustering and classification analyses were applied on the national level in Turkey, moreover, an artificial neural network-based prediction was also made. The results identified an increasing risk of soil erosion in the southern and western regions of Turkey, which demands erosion control practices (Aslan et al., 2019). Research has been conducted to regionalize Europe according to similar surface temperatures based on data between 1986 and 2005. The differences between long-term predictive data (CMIP5) and historical data were analyzed with k-means clustering analyses to determine grid points (Carvalho et al., 2016). A fuzzy c-means approach regionalization was determined in western India for the analysis of meteorological drought homogeneous regions to provide effective support for water resources planning and management during droughts (Goyal and Sharma, 2016). Clustering techniques can support simulation and predict models by grouping large-scale data. “Wind energy production is expected to be affected by shifts in wind patterns that will accompany climate change.” In California, wind patterns have been clustered using model simulations from the variable-resolution Community Earth System Model (VR-CESM) and analyzed according to the change in the frequency of clusters and changes in winds within clusters. The changes in capacity factor have significant influence with regard to energy generation (Wang M. et al., 2020).

Regression analysis sought to reveal functional relationships between variables that can further support predictive and forecasting models. Urbanization tends to have a significant impact on climate change, as underlined by an Australian study which determined that changes in land use and vegetation as a result of shifts in urbanization that affect the local climate and water cycle as well as its impacts are considered to be local specific (Maheshwari et al., 2020). Multiple regression-based analysis has been used to determine flood risk in urban catchments by combining multiple linear regression, multiple nonlinear regression and multiple binary logistics regression. This framework sought to support action plans concerning drainage management and maximize the impacts of flood susceptibility strategic implementations (Jato-Espino et al., 2018). Regarding water management, the influence of climate change on the hydrological cycle in the Yangtze River Basin has been analyzed using a regression analysis model and geographic information system (Keliang, 2019). Soil plays a significant role in carbon sequestration, therefore, moderate undesired climatic effects. A model has been designed regarding the top 25 cm of topsoil of the Sierra Morena (Red Natura 2000) area to determine the relationship between independent variables and soil organic carbon (SOC), moreover, by the use of multiple linear regression analysis examined the effects of these variables on SOC content. The results indicated that “SOC in a future scenario of climate change depends on average temperature of coldest quarter (41.9%), average temperature of warmest quarter (34.5%), annual precipitation (22.2%), and annual average temperature (1.3%).” The comparison between the current (2016) and future situations reflects a reduction of 35.4% SOC content and a trend in northward migration (Olaya-Abril et al., 2017).

Frequent itemset/pattern mining is a commonly used technique to extract knowledge from databases. The handling of an increasing amount of heterogeneous data is becoming ever more difficult, therefore, “an efficient algorithm is required to mine the hidden patterns of the frequent itemsets within a shorter run time and with less memory consumption while the volume of data increases over the time period” (Chee et al., 2019). Association rule mining (ARM) models have been built for atmospheric environment monitoring based on the Apriori algorithm and D-S theory/ER algorithm. These techniques provide both technical and theoretical support to prevent as well as manage air pollution (Li et al., 2019). Association rule mining has also been used in terms of monitoring weather behavioral data to develop a prediction model for climate variability (Rashid et al., 2017). Furthermore, climate variability has an impact on agriculture, which demands a greater understanding with regard to the impact of the climate on crop production and food security. Therefore, the impact of seasonal rainfall on rice crop yield was determined based on ARM techniques (Gandhi and Armstrong, 2016). For the understanding of wind conditions, multidimensional sequential pattern mining is used that can define which pattern is suitable for wind energy (by taking into consideration the factors of space, time, and height). According to a study on the Netherlands, 68.97% of the country covered by a suitable wind pattern (at 128 m) and already has wind turbines installed (Yusof et al., 2017). A spatio-temporal pattern-based sequence classification framework was built to estimate the extent of deforestation. This approach was applied on a Tunisian case study that took into consideration 15 years of satellite images and historical wildfire GIS data (Toujani et al., 2020).

Visualization methods sought to explore the interconnections between data by simplifying multivariate data. Self-organizing map neural network (SOMN) method has been used to analyse anomalous atmospheric circulation patterns in China with regard to surface temperature anomalies between 1979 and 2017 (Gao et al., 2019). This method is greatly used for mapping changes, e.g., regarding urban flood hazards (Rahmati et al., 2019). A study on the city of Amol in Iran was conducted and according to the aforementioned model of urban flood hazard mapping, 23% of the land area of the city is expected to high or very high levels of flood risk, which demands efficient flood risk management. SOMN and grid cells method were applied to determine changes in spatio-temporal land cover in Inner Mongolia between 2004 and 2014 (Li et al., 2018). The Principal Component Analysis (PCA) technique has been used to assess the vulnerability of the coastal region of Bangladesh while taking into consideration the IPCC framework. The study used 31 indicators (24 socio-economic, 7 natural). PCA was applied and determined seven eigenvectors [Demographic Vulnerability (PC1), Economic Vulnerability (PC2), Agricultural Vulnerability (PC3), Water Vulnerability (PC4), Health Vulnerability (PC5), Climate Vulnerability (PC6), and Infrastructural Vulnerability (PC7)] that take into consideration climate change scenarios from 2013 to 2050 (Uddin et al., 2019). PCA has also been used to build the composite drought vulnerability index (Balaganesh et al., 2020).

This paper described the essential need for research and development objectives to realize and manage the complex issues of climate change through Big Data tools. Data-driven applications were reviewed through the co-occurrence analysis of keywords, which showed the widespread application of Big Data technologies and tools, however, comprehensively utilized and integrative analyses are less prevalent.

This research aimed to highlight the perspective of systems of systems (SoS) as the drivers and effects of climate as well as that their resilience and adaptation cannot be determined without the exploration of the synergies between new research trends and disciplines. Based on the recommended SoS viewpoint, the specific results of sustainability-related research and development projects can be integrated, enhancing knowledge accumulation and utilization. The tools of data and systems sciences can play a crucial role in recognition of climate challenges and mitigation opportunities thanks to the integration of heterogeneous data and models, and the exploration of the relationship between environmental and social factors. This integrated thinking lays the groundwork for promising future trends in climate computing.

It can be claimed that the exclusive analysis of climatic factors cannot bring about sufficient strategic adaptation by itself, rather the socio-environmental factors must be integrated the climate change models.

Mitigating the impacts of climate change and successful adaptation requires effective climate change strategic planning by countries worldwide whose decision-making requires complex models and sources of information. The Big Data toolkit enables the systematization, processing, and evaluation of heterogeneous data and information sources, which is unfeasible with traditional disciplinary analysis tools. The harmonization of the ever-expanding scientific knowledge and diversified data sources related to climate change may be one of the most urgent tasks for researchers in the future. This research presented Big Data analytics tools and their contribution toward exploring the characteristics of climate change as well as climate action-related counterparts such as sustainability and social sciences that are essential for the successful development and implementation of strategies.

VS: conceptualization, validation, investigation, writing-original draft, visualization. JA: conceptualization, validation, resources, writing-review and editing, supervision, and funding acquisition. TC: writing-original draft, investigation, visualization, and validation. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.