Creating a global Climate Action to leverage and speed progress on the climate agenda

A positioning paper by

July 2020 ​

1

Table of Contents

Summary 3

Objective 6

A connection gap, not an information gap 6

From knowledge management to knowledge discovery 9

The end of data silos 10 Why graph technology? 10 What is a knowledge graph? 11 How are knowledge graphs used today? 12

Benefits of an open knowledge graph for action on climate change 12 Accelerating learning 14 Creating a shared understanding through terminology 15 Expanding networks and increasing collaboration opportunities 15

The knowledge triangle 15 The power of explainable artificial intelligence through knowledge graphs 16

Developing the Climate Action Knowledge Graph 17 Steps to develop the Climate Action Knowledge Graph 18

Glossary 19

References 22

Authors 23

2

Summary

Statement of purpose

This positioning paper serves as an introduction to the often overlooked topic of how advanced data and information management practices can serve as a springboard to faster, more widespread, and more efficient climate action. It provides background on these issues for policymakers, researchers, decision makers, media and others interested in advancing climate change-related policies.

The paper focuses on current challenges and readily available solutions for data and information management to further climate change adaptation, disaster risk reduction, and sustainable development agendas. It explains how a key tool - knowledge graphs - can better connect these communities and agendas to gain more traction in practice. It outlines how this single tool also sets the stage for the use of artificial intelligence methodologies that hold greater promise for innovation and progress. The paper provides real-world examples of the use of knowledge graphs in practice.

The positioning paper proposes and argues for the development of a global Climate Action Knowledge Graph to address current barriers to the spread of needed knowledge and action. This new entity would create a comprehensive, contextualized for the climate change adaptation, disaster risk reduction and broader climate action agendas in Europe initially, and would serve as an example of good knowledge management practice for these communities globally. It would interlink relevant but currently disconnected data, information, people, organizations and businesses, so that they can learn with and from one another. It would help all these players in the global quest to implement effective policies and measures to address major climate-related issues.

The creation of a global Climate Action Knowledge Graph would enable: ● the rapid discovery of new insights and knowledge. ● the efficient exploration of important connections between domains (e.g. water, energy and agriculture) and/or stakeholders (e.g. research and industry). ● the provision of new approaches to learning. ● a foundation to set the stage to leverage powerful AI applications available now and on the technological horizon. ● a searchable, graph-based climate action knowledge management system that is easy to use, and that adds value by establishing and maintaining better links between science, policy, and climate action.

3

New advances in information technologies have a critical role to play in helping to reach global goals for climate action in line with the Paris Agreement1 and the 2030 Agenda.2 Technologies from the research frontier are also key to the success of many elements of the European Green Deal and new stimulus recovery packages proposed by the European Commission in response to the coronavirus pandemic. Specifically, investments in artificial intelligence (AI) will support aspects of European policy that are critical to achieving collective sustainable development commitments, and policy measures intended to spur economic growth, stimulate the labour market, build capacity and skills, and achieve societal resilience at all levels. As European Commission President Ursula von der Leyen has said of the recovery plan: “It turns the immense challenge we face into an opportunity…[...]...by investing in our future: the European Green Deal and digitalization will boost jobs and growth, the resilience of our societies and the health of our environment.”3

AI methods can serve as powerful tools to design and integrate responses to the world’s interlinked global missions to address climate change, reduce disaster risk, achieve sustainable development, and create healthy societies. The global aims are clear: to move “‘towards a resilient, sustainable and fair recovery path” and a “cleaner, digital and resilient future”.4

This positioning paper explores the ways in which the most cutting-edge technologies in information science can not only support but accelerate needed responses to climate change and its impacts. Such technological advancements are essential levers needed to move forward on two key issues: enhancing the use of enriched information and data in the world’s knowledge-driven economies; and creating usable, tailored, interdisciplinary and multilingual solutions for data and knowledge management systems.

For example, consider ongoing efforts to align the three global agendas of climate change, disaster risk reduction and sustainable development, and to bring together all relevant stakeholders. Such integration requires analysis and evaluation of the different goals, common objectives and potential synergies (Figure 1). This can create a foundation for practical linkages in country efforts to achieve these global goals (Dazé et al., 2018). However, making these connections is complicated by the varying conceptualization and terminology used in the different communities. Measuring ambition and tracking progress achieved within each agenda in different countries and regions through the Global Stocktake on Adaptation (UNFCCC, 2015), for example, would benefit from a full description of all knowledge domains, so that they can be seamlessly linked, compared and contrasted.

The current wave of AI technology will help knowledge managers and users to connect some of these dots in their data, and, what is more, to do so automatically. Applications that support this are generally based on knowledge graph technologies, which allow users to ​

1 https://unfccc.int/files/essential_background/convention/application/pdf/english_paris_agreement.pdf ​ 2 https://sustainabledevelopment.un.org/content/documents/21252030%20Agenda%20for%20Sustaina ble%20Development%20web.pdf 3 Europe's moment: Repair and prepare for the next generation, 27 May, 2020 press release: ​ ​ https://ec.europa.eu/commission/presscorner/detail/en/ip_20_940 4 ibid. ​ 4

map and describe complex knowledge domains and their heterogeneous data structures in an agile manner (Blumauer, 2018). This creates context and meaning across different domains and terminologies, and thereby interlinks disciplines, multiple data types, and a range of information repositories.

Figure 1: Where different international policy processes connect, through direct references and through thematic linkages (Dazé et al., 2018).5

Knowledge graphs – technological tools that allow for unprecedented levels of sophistication in connecting disparate facets of knowledge – are becoming state-of-the-art in data science. What began as a initiative called Knowledge Graph in 2012 has since ​ ​ revolutionized data science, and, along with it, the ways in which people find the information ​ they seek – and the information they are able to find. Based on formal ontologies – in other ​ words, the explicit specifications of terms in given domains and the relationships among these terms (Gruber 1993) – a knowledge graph represents a collection of interlinked descriptions of entities. These are real-world objects, events, situations or abstract concepts. They contain large volumes of factual information and informal semantics, and make use of World Wide Web Consortium (W3C) principles,6 and thereby combine ​

5 http://r9f.ab1.mwp.accessdomain.com/wp-content/uploads/2018/08/napgn-en-2018-alignment-to-adva nce-climate-resilient-development-overview-brief.pdf 6 https://en.wikipedia.org/wiki/Semantic_Web ​ 5

symbolic (e.g. knowledge models) and statistical (e.g. machine learning) AI methods for high-quality, comprehensive data and information management solutions. ​ ​

Knowledge graphs can seamlessly be developed and edited over time, with collaborative input and a clearly defined governance model. The initiative proposed here would thus provide a knowledge model to describe climate change action as it grows and evolves. Such a device potentially provides a needed and valuable public service. For example, those working on National Adaptation Plan (NAP) processes and national disaster risk reduction strategies, and others working in specific sectors such as water and health, could describe their own knowledge domains and define their needs, and learn about how these domains ​ link to other knowledge domains. This paper underscores the potential benefits of knowledge graphs. It offers examples of how knowledge graphs can help those working in different areas of climate change increase their collaboration, and leverage synergies between their different agendas.

Objective

The purpose of this concept note is to highlight the transformational potential that AI methods can have for the world of climate change research policy and practice. A cutting-edge team of leading semantic technology experts (Semantic Web Company, developers of PoolParty Semantic Suite7), taxonomies (REEEP, developers of Climate Tagger8 and Secretariat of ​ the community of practice, the Climate Knowledge Brokers9), and climate change researchers (SEI, climate change researchers, and developers of weADAPT and the Connectivity Hub10) have worked together for many years, and set out below, some ways in which this can be achieved with further investment.

The work of creating a global Climate Action Knowledge Graph will itself foster an already emerging alliance of global, regional and national knowledge brokers and domain experts specializing in climate and development information, and representing diverse organizations. The building, promoting and maintaining of a Climate Action Knowledge Graph provides a shared purpose for a diverse set of information players – from governments to international organizations to research institutes, NGOs, industries and good practice networks.

A connection gap, not an information gap

The problem is clear: the world must act to mitigate and adapt to climate change – and quickly. Sharing knowledge is key to supporting individuals and communities to identify and implement the best possible actions, to reduce vulnerability and increase the resilience of people and environments to the impacts of climate change.

7 https://www.poolparty.biz ​ 8 https://www.climatetagger.net/ ​ 9 https://www.climateknowledgebrokers.net/ ​ 10 https://www.placard-network.eu/our-work/connectivity-hub/ ​ 6

Numerous innovative and effective ways to contribute to solving the climate change crisis, continue to surface and evolve. Yet, at the same time, it is extremely difficult to identify and access credible, relevant and appropriate information, available from myriad sources, in heterogeneous forms, published in different languages, and explained using different terminologies.

How can individuals seeking to address climate change impacts find the information they seek? What solutions offer to reduce both climate change vulnerability and the risk of disaster? What are others doing to address similar challenges? What have actors learned from their own efforts? What can others learn from their experiences and knowledge? What can actors teach one another?

Collective knowledge and expertise on climate action lie scattered across multiple platforms, buried in large documents, and shared in various languages that seldom speak to one another. Information languishes in silos of different but related topics and domains, each with its own specific focus and terminology. This inhibits the discovery, filtering, analysis, and application of knowledge, and knowledge transfer between complementary areas of endeavour. As a result, effective solutions (and their resulting lessons) often remain undiscovered and underutilized. Around the world, in different areas and contexts - and at great expense – people replicate efforts to solve problems that already have answers. People miss important learning opportunities. Actors make the same mistakes. Innovations fail to catch on. The cycle repeats itself.

At present, poor classifications, and the resulting limits of search potential on many platforms impinge on discoverability and uptake. For example, a World Bank study in 2014 showed that more than 31% of its own policy reports were never downloaded, and only 13% had been downloaded 250 times or more by users. And almost 87% of policy reports had never been cited. This speaks to the perpetuation of silos in which people work, and the disconnect between different disciplines, organizations, networks, and languages. Despite a recognition of the issues that underpin this, few initiatives exist to address them.

This issue worsens as websites, data catalogues and information repositories update, and as knowledge production expands. Some platforms are better than others at making effective use of keyword tags and auto-categorization for classifying content (PreventionWeb and “PANORAMA: Solutions for a healthy planet” offer two good examples). Yet, these efforts largely take place in isolation. They are topic-specific. Very few platforms use formal, open, or shared taxonomies to classify their content based on a controlled vocabulary that others ​ can adopt. Thus, these approaches fail to address the issue of knowledge fragmentation across the internet.

The development of an EU Green Taxonomy (TEG, 2019) by the European Commission and High-Level Expert Group on Sustainable Finance for the identification of green projects and green financial instruments is a step in the right direction. This incorporates climate change adaptation and disaster risk reduction indicators and metrics. The aims are to avoid inaccurate “green” labels, and to ensure that climate change adaptation and disaster risk

7

reduction are better reflected in decision-making processes and related financial flows (Leitner et al. 2020).

However, such classifications typically aim to support specific frameworks for analysis and planning, not better online knowledge management per se - even though both of these ​ ​ objectives are clearly desirable. Individual knowledge managers here and there undertake most of this curation independently. The results are not consistent between platforms, and they may even be inconsistent within platforms due to natural evolution, the work of multiple managers over time, and user error. The results? Missed opportunities, replication of efforts, and redundancy in impacts.

To break free of this trap, actors addressing climate change-related issues need better ways to connect with other people undertaking similar efforts on similar agendas. New, “smart” ways of managing the expanse of existing and continuously emerging knowledge, organizations and networks working on climate action to make them easier to find, access, use and analyse. This requires bringing greater and more sophisticated connectivity to all of this knowledge and generating greater understanding of the differing language often used to describe the same concepts. There is a need to overcome time-consuming challenges such as integrating structured (e.g. spreadsheet) and unstructured (textual) data. The urgency of these needs presents an opportunity to leverage AI techniques to enable automated connections to be made between content – thus, moving towards a “Semantic Web of Data” that seamlessly brings relevant knowledge to the user, without the need for complex queries.

More specifically, we propose that the development and uptake of a publicly available Climate Action Knowledge Graph can become a powerful solution, enabling not only the ​ identification of information that is difficult to find, but also the identification and analysis of patterns and connections within the data. This allows the discovery and combination of information to create new knowledge.

Key aspects of the development of the Climate Action Knowledge Graph include:

1. Understanding stakeholder needs ○ Initiating joint projects to connect different organisations, testing innovations, and generating new thinking on how to best meet user needs (e.g. in the form of design sprint workshops bringing together relevant stakeholders / alpha users) ○ Reaching out to other relevant actors, bringing in more and new stakeholders in the field.

2. Developing the common Climate Action Knowledge Graph ○ Developing ontologies and taxonomies (on the basis of existing ones) as a skeleton of the knowledge graph. ○ Interlinking more and more relevant data and information sources to this skeleton to enrich and enlarge the Climate Action Knowledge Graph continuously. This will (i) allow users to search and share climate-relevant

8

data efficiently; and (ii) help people identify quickly which resources are most relevant for them. ○ Iteratively expanding the input to the knowledge graph (along a newly developed knowledge graph governance model), and, in turn, expanding the body of knowledge it covers, and the sectors it spans.

3. Peer-to-peer learning ○ Offering workshops to provide opportunities for members of different communities to meet each other online or face-to-face. ○ Sharing challenges and comparing the experiences of the success and failure of certain approaches ○ Coming together to define a way forward as a community of practice. ○ Building capacity by offering hands-on advice, and insights from the wealth of experience across organizational silos.

From knowledge management to knowledge discovery

Addressing these issues demands a transformation in knowledge management that uses not only a standardized common language - a taxonomy (see Climate Tagger) - but, where ​ ​ ​ possible, knowledge graphs that translate hierarchical knowledge models into ​ ​ interconnected networks of nodes and links. Knowledge graphs can be used to map entire knowledge domains, defining highly specific semantic relationships between objects. This enables the linking of content in meaningful ways to yield new, dynamic, more integrated, and more holistic knowledge. Powerful applications for research, policy, industry and practise can be created on top of a knowledge graph to allow for the quick and easy search and discovery of relevant information from huge and disparate information systems in place. Relations become visible although information and data is used from different sources and from different stakeholders, Knowledge graphs enable deeper insights as well as a relevant holistic view on the topic. Complex and dynamic queries can be answered based on data from various sources, and more content can be analysed, digested and processed. This means reducing time, cost and effort.

The widespread uptake and application of a topic-specific climate action knowledge graph (covering, for example, climate change adaptation and disaster risk reduction, see Connectivity Hub) can initiate a cultural shift towards , wherein online content ​ connects to all other related and relevant content. This machine-readable linking of content would enable “pooling” of online content into “semantic data lakes” (Bauer and Kaltenböck, ​ ​ 2016). This facilitates increased knowledge discovery, and provides the foundation for more sophisticated analyses and accelerated learning. As well as linking related data, knowledge graphs have the power to support learning more directly. For example, the integration of glossaries and other metadata can be used to promote understanding of technical or domain-specific language and its use (see Connectivity Hub). ​ 9

To be successful, the roll-out of such technology would need to be accompanied by relevant standards in the form of a specified and agreed knowledge graph governance model, including: ● the adoption and criteria for the use of a shared terminology (a taxonomy) for describing content, ● the provision of term definitions and other metadata to promote understanding of their meaning(s) and handle ambiguity, and ● the adoption of an appropriate tool to semantically tag (and classify) content in a way that enables relevant content across different websites, data catalogues and information repositories to be interconnected to comply with Linked Open Data 11 ​ (LOD) standards.

C limate Tagger A good example of a knowledge graph-based system is the Climate Tagger12, https://www.climatetagger.net/, a web-based, free-to-use tagging tool based on a ​ ​ taxonomy of approximately 5,000 concepts in the field of climate change development and renewable energy and energy efficiency. Available in five languages, the Climate Tagger allows harmonization of tagging and powerful information retrieval from many climate change fields, including climate mitigation and adaptation, economy and green growth, and ​ ​ ​ ​ ​ ​ specific areas such as REDD+ (Reducing Emissions from Deforestation and Forest ​ ​ Degradation). ​ ​

The end of data silos

The words of 16th Century poet John Donne’s – who famously said, “No man is an island,” have never been more relevant.13 It has never been clearer that organizations and individuals and research/practice/policy communities working in isolation will struggle in many ways. The data from these disconnected departments and teams will also struggle to find their way into meaningful use in the wider world. An increasing number of organizations are looking to knowledge graphs as a way of connecting these disparate pieces of information, experiences and ideas.

Why graph technology? Experts in machine learning and AI technologies have raised hopes that the problem of ​ disconnected and low-quality data can be fixed automatically. However, machines, as yet, are not able to learn from any kind of data, especially unstructured information. ​ ​

11 The Climate Tagger (https://www.climatetagger.net/) (see Climate Tagger) is one such tool that is ​ ​ ​ ​ being optimized for climate change adaptation (CCA) and disaster risk reduction (DRR) through the PLAtform for Climate Adaptation And Risk reduction (PLACARD) project. The Climate Tagger will be further optimized to support automatic keyword tagging of content. 12 The tool is operated by The Renewable Energy and Energy Efficiency Partnership (REEEP) 13 https://de.wikipedia.org/wiki/John_Donne ​ 10

Algorithms such as “deep learning” only work well when a lot of data (more than most ​ ​ 14 organizations even have in place) of the same kind is available, and even then, only rather simple cognitive processes such as “classification” can be automated.

As Portland State University computer science professor Melanie Mitchell15 pointedly stated in The New York Times, “Today’s A.I. systems sorely lack the essence of human intelligence: 16 understanding the experienced situations, being able to grasp their meaning.” This is where graph technologies come into play: they let machines (and ultimately people) better understand how things relate to each other. Putting things in context and simultaneously relating them to each other are at the very core of any knowledge-oriented data model.

What is a knowledge graph? The words “knowledge” and “graph” were carefully selected. In comparison to a knowledge base, the knowledge domain is organized as a graph. As such, it connects different types of objects in a systematic way; and as a knowledge graph, it encodes knowledge arranged in a network of flexible nodes and links – rather than in of rows and columns. Using knowledge graph technology allows machines and people to scale their data and knowledge dynamically by growing a of facts about things. This is useful for data integration, knowledge discovery and in-depth analyses. In short, users of knowledge graph technology benefit from a holistic view of their knowledge domains. Users can take advantage of storing information in a form that is easy to reuse by way of a knowledge graph (Figure 2).

Figure 2: An example knowledge graph (Blumauer and Nagy, 2020).17

14 Google (2018): Relational inductive biases, deep learning, and graph networks - https://ai.google/research/pubs/pub47094 15 https://en.wikipedia.org/wiki/Melanie_Mitchell ​ 16 Melanie Mitchell: Artificial Intelligence Hits the Barrier of Meaning - https://www.nytimes.com/2018/11/05/opinion/artificial-intelligence-machine-learning.html 17 https://www.poolparty.biz/the-knowledge-graph-cookbook/ ​ 11

How are knowledge graphs used today? LinkedIn, Google, Amazon and Microsoft are a few of the global organizations that operate their own knowledge graphs as part of their infrastructures – and these operations affect how people around the world use technology today. Many people use graph technology without even realizing it. Knowledge-graph technology is at work when people use Amazon’s Alexa, , and Google Home to answer questions. Day-to-day searches, and the resulting recommendations rely on knowledge graphs. Widespread usage also helps companies such as Google and Amazon refine, improve and scale up their knowledge graphs, through machine learning. ​ ​

18 19 Google’s Knowledge Graph was originally derived from , t​ he CIA World 20 21 Factbook, , and various other data sources. Since then, Google has continuously extended its Knowledge Graph using a combination of machine learning22 and manual curation.

The Google project is only one of more than 1,200 publicly visible knowledge graphs that are not only highly structured and interconnected, but are also most frequently available for download and reuse as standards-based knowledge bases. This “graph of graphs” is the ​ ​ publicly visible part of the “Semantic Web”. However, large parts of it are built out as ​ ​ “enterprise knowledge graphs” with access only possible from behind corporate firewalls. ​ ​

Benefits of an open knowledge graph for action on climate change Knowledge graphs provide rich and interlinked information and result in powerful knowledge discovery and semantic search. Climate change data, information and knowledge management systems can benefit from this.

With so much digital information available, most people still rely on search boxes to find what they want. This search only reveals what is directly tied to a term, leaving out all the available context. Using knowledge graphs to map the meaningful context allows users to browse vast knowledge repositories easily and more meaningfully. This tool - the Climate Action Knowledge Graph - connects isolated information sources, and provides fresh and timely context. This can be especially useful for users who are unfamiliar with the knowledge domain and do not use the most optimal search queries. That is, they will come across information they would not otherwise find with a simple search.

18 https://developers.google.com/knowledge-graph/ ​ 19 https://www.wikidata.org/wiki/Wikidata:Main_Page ​ 20 https://www.cia.gov/library/publications/the-world-factbook/ ​ 21 Data Dumps | Freebase API (Deprecated) ​ 22 [Google (2014): Knowledge Vault: A Web-Scale Approach to Probabilistic Knowledge Fusion ​ 12

For example, Laniakea23, a visualization of English Wikipedia, offers an interesting example of connecting data and information to allow more sophisticated exploration of a given topic. Laniakea lets users easily browse the complex landscape of information, and allows them to dive into a wealth of knowledge in a previously unknown way, in similarity to the Connectivity Hub (see Connectivity Hub). However, for the field of climate change research, it is also ​ ​ important to know which organizations are working on which topics, and how language is being used to describe their work.

C onnectivity Hub The PLACARD Connectivity Hub24is a cutting-edge search and discovery tool that brings together resources (articles, topics, organisations) relevant to professionals working in the areas of climate change adaptation (CCA) and disaster risk reduction (DRR), in a fast and efficient way. The Hub dynamically links content from several major European and international knowledge management platforms (e.g. Climate-ADAPT, weADAPT, PreventionWeb, Eldis) by leveraging their tagging systems.

In doing so, the Hub brings content to the user that they may not have known was relevant to their search e.g. if they are new to a topic area. In this way, the Hub is designed to accelerate learning. “Search and discovery” will be further enhanced by the PLACARD CCA and DRR taxonomy, developed from the tagging systems to provide context and ​ ​ background on how terminology is used and how it has evolved over time. This is important for helping non-experts understand and navigate terms used by multiple communities of practice but with differing interpretations. Furthermore, the Connectivity Hub can be a test bed for the use of artificial intelligence (AI) and machine learning if the taxonomy is enhanced with relational and/or semantic data. Resulting new and unexpected combinations of information can produce powerful, policy-relevant insights e.g. supporting learning from relevant successful climate actions elsewhere that are otherwise difficult to find (Barrott et al., 2020). Bringing together the CCA and DRR sectors in ways that foster better collaboration and unified views across their two communities and missions offers an excellent showcase for the utilitarian potential and power of knowledge graph technology.

A knowledge graph supports decision and process augmentation based on linked data, explicitly for the particular organisation for which it has been developed. Usually, an organization’s knowledge graph cannot be accessed, viewed or reused by other parties. Why

23 http://laniakea.fathom.info/ ​ 24 https://www.placard-network.eu/our-work/connectivity-hub/ ​ 13

are organizations exploring a more graph-based approach to managing their data? What are the benefits of building an organization’s knowledge graph?

Five typical reasons for an organization to use a knowledge graph are:

1. To efficiently handle hierarchical or strongly linked datasets. 2. To gain new insights based on entity-centric views (e.g. thematic topics related to climate change), in contrast to document-centric views. 3. To understand and calculate the relationships and causalities in a knowledge domain. 4. To integrate heterogeneous data sources (structured and unstructured) based on a “schema-late” approach in an agile way. ​ ​ 5. To create unified views across multiple data silos within an organisation.

As a multidimensional index that is able to house all types of organizational information, knowledge graphs should be an integral part of any organization’s data governance framework and AI strategy. Different types of organisations (enterprises, NGOs, research bodies, governments, etc.) can use knowledge graphs to address these five key data management issues. The same can be said for situations that arise within or across different sectors within climate change. This is where a publicly available graph, such as the proposed Climate Action Knowledge Graph, can become a powerful solution and resource, enabling both a specialised view of a particular sector, as well as a linked integrated, and more holistic view of how this knowledge fits within other climate change knowledge.

From a climate change perspective, there are several ways in which an enterprise knowledge graph could support climate change research, policy and practise.

1. Accelerating learning A graph connecting resources and knowledge across organizational silos and different communities will offer brand new opportunities for learning and sharing about past and ongoing work. Connecting resources, and people across different areas of expertise will encourage exchange and collaboration on a non-virtual level as well. Rather than relying on a traditional text search, a climate action-focused knowledge graph can be used to create a network that allows users to dive into thousands of documents from different areas and publishers. Applications can be built onto the graph, allowing different ways to discover knowledge, and enabling new research.

Knowledge graphs enable users who are unfamiliar with a knowledge domain to find information they would not otherwise locate through a simple search. In addition, in AI-powered learning management systems a set of concepts can be modelled to achieve proficiency in a particular field. Recommendation engines can then be used to suggest content associated with the edges of these concepts in a student's knowledge domain as they progress in their learning.

14

2. Creating a shared understanding through terminology Eliciting all the diverse terms used in the climate change field, their disparate interpretations, and the different ways in which they link with each other presents an enormous challenge. The constant evolution of terminology adds complexity to the task. Success will depend on consultation with experts, and the creation of working groups to help validate and extend new and existing taxonomies into knowledge graphs, and to keep pace with ongoing evolution.

Thus, developing knowledge graphs starts with collaboration across domains. Creating the Climate Action Knowledge Graph will bring together experts from different fields to provide input to the knowledge model with new connections and relevant context – essential components for enabling more cross-sectoral research and collaboration.

3. Expanding networks and increasing collaboration opportunities Using semantic technologies and linked data aims to improve access to timely and relevant information by coordinating and orchestrating efforts across groups. Bringing communities closer together promises to increase efficiency, effectiveness and collaboration in response to global challenges. The ultimate aim is to find ways to help people act to address climate change-related issues as effectively and swiftly as possible.

The knowledge triangle

Figure 3 shows a knowledge triangle,25 which offers a way to visualize how a knowledge graph works, and its development steps from raw data to information and knowledge. In the case of such a diverse area as climate change – which incorporates numerous sectors, affects a wide range of stakeholders on different scales, and which produces an ever-growing amount of new data and information – graph technology is a revolutionary approach. It can make an important contribution towards increased efficiency and knowledge from i) the potential to gain insights more quickly; ii) creating greater understanding of relationships and casualties; and, iii) fostering the integration of more varied data sources.

25 https://medium.com/@dmccreary/the-knowledge-triangle-c5124637d54c ​ 15

Figure 3: A knowledge triangle (Source: Don McCreary).

The power of explainable artificial intelligence through knowledge graphs Knowledge graphs as a form of artificial intelligence can help organisations and stakeholders improve decision making by providing better discoverability through holistic integration of knowledge across many domains. However, “black box” AI can be limited by its inability to ​ ​ explain its reasoning to human users. This is where “explainable AI” has a role to play. ​ ​

Explainable AI can provide useful insights regarding the i) data used; ii) use of existing / trained algorithms; and, most importantly, iii) the results and thereby the value of an AI engine. As the name suggests, explainable AI provides transparency - the reasons for given results. Thus, it allows a user to understand such results. This, in turn, supports better decision making. Furthermore, the reasons and benefits for explainable AI increase with the cost/risk involved in the decision (see Figure 4).

“...[this is]...the next stage of human augmentation by machines, when AI will empower humans to take corrective actions according to the explanations given. Within three years, we believe...[explainable AI]... will have come to dominate the AI landscape for businesses—because it will enable people to understand and act responsibly, as well as creating effective teaming between human and machines.” Source: Accenture Labs, Understanding Machines, Explainable AI, 201826 ​

The “right to explanation” Explainability is a practical imperative. It will also be required because of ethical and/or legal requirements, such as the introduction under the EU’s General Data Protection Regulation (GDPR) of the “right to explanation” about algorithm-derived decisions. Moreover, explainability is vital because it puts people in control. This means that AI augments human ​ skills, rather than trying to replace them. For all these reasons, AI needs to go beyond machine learning to the next stage: Explainable AI. Source: https://en.wikipedia.org/wiki/Right_to_explanation ​ ​

26 https://www.accenture.com/_acnmedia/pdf-85/accenture-understanding-machines-explainable-ai.pdf ​ 16

The newly established European AI Alliance27 also has a strong focus on explainable AI as well as ethical issues in AI - both areas that can be supported by knowledge graphs.

Figure 4: The need for explainable AI rises with the potential cost of poor decisions.

If AI lacks the ability to explain itself in the above listed areas, then the risk of it supporting a wrong decision may outweigh the benefits it could bring in terms of the speed, accuracy and efficiency of decision-making. The effect would be to severely limit its usage and confidence in the approach. This is particularly relevant to an area such as climate change where the impacts of poor decisions can have a high cost. Transparency in how decisions are made is key to building trust and buy-in. Knowledge graphs provide meaning and context about an application, its data and important background information or common knowledge to understand the topic. Making this information visible to the end user fosters trust and more evidence-based decision making.

Developing the Climate Action Knowledge Graph

Following the building and publishing of the initial graph, services such as the Connectivity Hub can offer powerful gateways into previously unconnected silos of information. For example, using graph technology in the Connectivity Hub, a policymaker could conduct a simple search on flood management, and be suggested further (directly and indirectly related) resources to rapidly identify available solutions, tried-and-tested responses, suitable experts, and countries and organizations pioneering in the field. That same policymaker could almost immediately reach out directly to relevant people and organizations. All this information would come from a single, efficient gateway, connecting many disparate sources of data.

27 https://ec.europa.eu/futurium/en/european-ai-alliance ​ 17

Assessing use cases, developing controlled vocabularies and building links to create a graph connecting the fields of CCA and DRR are tasks that can be managed by a consortium. However, the involvement and feedback of the relevant communities will need to be integrated from the start. This not only ensures the quality and relevance of the knowledge graph, but also increases interest and commitment to the underlying technology by the stakeholders. The process to create the graph will involve workshops with members of different climate change communities, interviews with individual experts, and feedback and testing rounds of agreed use-cases.

A first version of the Climate Action Knowledge Graph should incorporate information from relevant knowledge sources, including, but not limited to: sectors, technologies, projects, policies, research papers, organizations, and initiatives. Knowledge graph development is a combination of manual work (involving relevant stakeholders and experts) and continuous machine-supported automated enrichment activities. Furthermore, a governance model and communication strategy need to be developed to ensure the sustainability and use of the Climate Action Knowledge Graph in different communities working on climate change.

Steps to develop the Climate Action Knowledge Graph

1. Identify all key, relevant stakeholders in the field. 2. Make key decisions about basic requirements of the knowledge graph, including its objectives, usage, licensing and publishing scenarios. 3. Develop a governance model that specifies how future changes and enrichments of the Climate Action Knowledge Graph will take place. This step clarifies who makes decisions and how, and defines the envisaged focus and scope. 4. Conduct interviews and hold workshops with stakeholders to understand the sector and to elicit the requirements of all stakeholders involved. 5. Evaluate existing knowledge models, taxonomies, data, and information in the field that should be taken into account in the graph. Build on relevant existing models 6. Specify a set of (prioritized) core questions that the Climate Action Knowledge Graph should be able to answer. This step helps to narrow the scale and objectives. It also provides an evaluation test that can be used continuously throughout the development phase and beyond. 7. Develop the core structure taking into account the results from stakeholder interviews and workshops. 8. Enrich the Climate Action Knowledge Graph with more documents and data that are interlinked by making use of the core structure (by means of tagging, linking, and expansion of this structure). 9. Build more advanced artificial intelligence applications on top of the knowledge model to further demonstrate its added value and potential. 10. Promote and communicate the knowledge graph to stakeholders and others in the field. This step ensures awareness about and use of the knowledge graph. Invite stakeholders to provide feedback (continuously) to refine the knowledge graph.

18

For a more detailed version of these steps, please see the advanced road map in Barrott et ​ ​ al. (2020).28

Shared ontologies and open knowledge graphs are key to accelerating climate action. If undertaken on a platform-by-platform basis, these approaches are resource heavy to implement in terms of time, finance and expertise. Such costs put these technologies out of reach for platforms run by smaller teams and with limited budgets. The situation thus biases the knowledge domain, and potentially misses important knowledge. This can contribute to an inequity of representation of knowledge from the Global South or least-developed countries, for example, or from specific sectors and domains.

Some knowledge platforms are testing methods to leverage AI approaches to enhance knowledge discovery, support data integration, enable data analytics, and reduce human error in categorizing content. Currently, approaches utilizing AI, knowledge graphs and semantic technologies are most prevalent in large industry sectors where information and data has become a key commodity. In the climate and development sector these technologies are still largely untapped. The AI technologies and steps described here and the road map in Barrott et al. (2020) present an opportunity to do things differently and make the changes necessary to meet global climate and development challenges, before it is too late.

Glossary

Artificial intelligence (AI) - a wide-ranging branch of computer science concerned with ​ building smart machines capable of performing tasks that typically require human intelligence. AI is an interdisciplinary science with multiple approaches, but advancements in machine learning and deep learning are creating a paradigm shift in virtually every sector of the technology industry.29 For details on the different types of AI, such as statistical AI and symbolic AI, as well as AI history, please see Wikipedia.30 Machine learning is a branch of artificial intelligence.

Black box system - In science, computing, and engineering, a black box is a device, ​ ​ system or object which can be viewed in terms of its inputs and outputs (or transfer characteristics), without any knowledge of its internal workings. Its implementation is "opaque" (black). Almost anything might be referred to as a black box: a transistor, an engine, an algorithm, the human brain, an institution or government.31

Climate action – stepped-up efforts to reduce greenhouse gas emissions and strengthen ​ resilience and adaptive capacity to climate-induced impacts, including: climate-related hazards in all countries; integrating climate change measures into national policies, strategies and planning; and improving education, awareness raising and human and

28 https://www.placard-network.eu/transforming-knowledge-management-for-climate-action/ ​ 29 https://builtin.com/artificial-intelligence ​ 30 https://en.wikipedia.org/wiki/Artificial_intelligence#Statistical ​ 31 https://en.wikipedia.org/wiki/Black_box ​ 19

institutional capacity with respect to climate change mitigation, adaptation, impact reduction and early warning.32

Deep learning - a type of machine learning that trains a computer to perform humanlike ​ tasks, such as recognizing speech, identifying images or making predictions.33

Controlled vocabulary - A controlled vocabulary is a standardized common language, an ​ organized arrangement of words and phrases used to index content and/or to retrieve content through browsing or searching. It typically includes preferred and variant terms and has a defined scope or describes a specific domain.34 ​

Data - unprocessed raw information in the forms of binary codes, numeric codes, dates, ​ strings, and full-text descriptions in documents.

Data lake - a vast pool of raw data (structured or unstructured), the purpose for which has ​ not yet been defined.35

Enterprise knowledge graph - a model of a knowledge domain created by subject-matter ​ experts with the help of intelligent machine learning algorithms. It provides a structure and common interface for all of your data and enables the creation of smart multilateral relations throughout your databases. Structured as an additional virtual data layer, the Knowledge Graph lies on top of your existing databases or data sets to link all your data together at scale – be it structured or unstructured.36

Explainable AI - Explainable AI (XAI) refers to methods and techniques in the application of ​ artificial intelligence (AI) technology such that the results of the solution can be understood by human experts. It contrasts with the concept of the "black box" in machine learning where even their designers cannot explain why the AI arrived at a specific decision. XAI is an implementation of the social right to explanation.37

Information - more useful than raw data, as it is now processed (or enriched) in some way ​ (by extracting things that are meaningful to an issue), but information itself consists of islands of disconnected items, such as entities: people, places, events, and concepts. Linking information together creates knowledge.

38 Knowledge (computer science) - connected-information that is query ready. ​

Knowledge graph - a fabric of concepts, classes, properties, relationships, and entities ​ covering multiple domains, various levels of granularity, and data from multiple sources. It

32 https://sustainabledevelopment.un.org/sdg13 ​ 33 https://www.sas.com/en_us/insights/analytics/deep-learning.html ​ 34 https://www.getty.edu/research/publications/electronic_publications/intro_controlled_vocab/what.pdf ​ 35 https://en.wikipedia.org/wiki/Data_lake ​ 36 https://www.poolparty.biz/what-is-a-knowledge-graph/ ​ 37 https://en.wikipedia.org/wiki/Explainable_artificial_intelligence ​ 38 https://medium.com/@dmccreary/the-knowledge-triangle-c5124637d54c ​ 20

functions as background knowledge for various applications (e.g. , data integration and machine learning).39

Linked Open Data - The concept of Linked Open Data defines a vision of globally accessible ​ and linked data on the internet. LOD are often thought of as a virtual data cloud where anyone can access any data they are authorized to see, and may also add to any data without disturbing the original data source. This provides an open environment where data can be created, connected and consumed on an internet scale. A basic theory of LOD is that data have more value if they can be connected to other data.40 ​

Machine learning - a method of data analysis that automates analytical model building. It is ​ a branch of artificial intelligence based on the idea that systems can learn from data, identify patterns, and make decisions with minimal human intervention.41

Metadata - Metadata are data that provide information about other data - in other words, ​ "data about data".42

Ontology (information science) - a representation, formal naming, and definition of the ​ categories, properties, and relations between the concepts, data, and entities that substantiate one, many, or all domains of discourse. More simply, an ontology is a way of showing the properties of a subject area, and how they are related, by defining a set of concepts and categories that represent the subject.43

Schema - refers to the organization of data as a blueprint of how the database is constructed ​ (divided into database tables in the case of relational databases). The formal definition of a database schema is a set of formulas (sentences) called integrity constraints imposed on a database.44

Schema-late - allows the interpretation of data to be changed at any time without having to ​ change the physical format of the data, but rather reshaping data in the best possible manner as a solution is developed. This becomes increasingly important as the size of the dataset increases.

Semantic Web (also called Web of Data) - an extension of the Web through standards ​ 45 developed by the World Wide Web Consortium (W3C) t​ hat promote common data formats ​ ​ and exchange protocols on the Web. The goal of the Semantic Web is to make internet data machine readable.46

39 https://semantic-web.com/glossary/knowledge-graph/?uri=http://vocabulary.semantic-web.com/ppknowledg egraph/292 40 https://www.w3.org/egov/wiki/Linked_Open_Data ​ 41 https://www.sas.com/en_us/insights/analytics/machine-learning.html ​ 42 https://en.wikipedia.org/wiki/Metadata ​ 43 https://en.wikipedia.org/wiki/Ontology_(information_science) ​ 44 https://en.wikipedia.org/wiki/Database_schema ​ 45 https://www.w3.org/ ​ 46 https://en.wikipedia.org/wiki/Semantic_Web ​ ​ 21

Tagging - In information systems, a tag is a keyword or term assigned to a piece of ​ information (such as an Internet bookmark, digital image, database record, or computer file). This kind of metadata helps describe an item and allows it to be found again by browsing or searching.47

Taxonomy - the practice and science of classification. Originally, taxonomy referred only to ​ the classification of organisms or a particular classification of organisms. In a wider, more general sense, it may refer to a classification of things or concepts, as well as to the principles underlying such a classification.48 In knowledge management a taxonomy is a hierarchical structure or order of terms of a knowledge subject49 that supports content organisation and retrieval.50

World Wide Web Consortium - The World Wide Web Consortium (W3C) is an international ​ community where Member organizations, a full-time staff, and the public work together to develop Web standards. Led by Web inventor and Director Tim Berners-Lee and CEO Jeffrey Jaffe, W3C's mission is to lead the Web to its full potential.51

References

Barrott, J., Bharwani, S. and Brandon, K. (2020). Transforming knowledge management for climate action: a road map for accelerated discovery and learning. PLACARD project, FC.ID: Lisbon. https://www.placard-network.eu/transforming-knowledge-management-for-climate-action/

Bauer, F. and Kaltenböck, M. (2016) Linked Open Data: The Essentials (2nd Edition). Climate Knowledge Brokers. https://www.climateknowledgebrokers.net/linked-open-data-essentials-climate-knowledge-br okering-edition/

Bharwani, S. Barrott, J., Lokers, R., Houtkamp, J., Costello, R. and Foddering, R. (2019) The PLACARD Connectivity Hub. https://www.placard-network.eu/our-work/connectivity-hub/ ​

Blumauer, A., (Posted August 23, 2018) Knowledge Graphs – Connecting the Dots in an Increasingly Complex World https://semantic-web.com/2018/08/23/knowledge-graphs-connecting-dots-increasingly-compl ex-world/

Blumauer, A., Nagy H., 2020. The Knowledge Graph Cookbook. edition mono/monochrom https://www.amazon.com/Knowledge-Graph-Cookbook-Andreas-Blumauer/dp/3902796707

47 https://en.wikipedia.org/wiki/Tag_(metadata) ​ 48 https://en.wikipedia.org/wiki/Taxonomy_(general) ​ 49 https://en.wikipedia.org/wiki/Taxonomy_for_search_engines ​ 50 https://semantic-web.com/glossary/taxonomy ​ 51 https://www.w3.org/Consortium/ ​ 22

Dazé, A., Terton, A., and Maass, M. (2018) Alignment to Advance Climate-Resilient Development Overview Brief 1: Introduction to Alignment. http://r9f.ab1.mwp.accessdomain.com/wp-content/uploads/2018/08/napgn-en-2018-alignmen t-to-advance-climate-resilient-development-overview-brief.pdf

T. R. Gruber. (1993) A Translation Approach to Portable Ontologies. Knowledge Acquisition, 5(2):199–220.

Leitner, M., Buschmann, D., Capela Lourenço, T., Coninx, I. and Schmidt A. 2020. Bonding CCA and DRR: recommendations for strengthening institutional collaboration and capacities. PLACARD project, FC.ID: Lisbon.

TEG (2019). Financing a Sustainable European Economy: Taxonomy Technical Report. EU Technical Expert Group on Sustainable Finance. https://ec.europa.eu/info/files/200309-sustainable-finance-teg-final-report-taxonomy_en

UNFCCC (2015) Paris Agreement. United Nations. https://unfccc.int/process-and-meetings/the-paris-agreement/the-paris-agreement

Authors

Martin Kaltenböck, Co-Founder and Managing Director at Semantic Web Company, ​ https://www.linkedin.com/in/martinkaltenboeck/ Semantic Web Company, https://www.semantic-web.com ​ PoolParty Semantic Suite, https://www.poolparty.biz ​

Denise Recheis, Knowledge and taxonomy manager at REEEP ​ REEEP www.reeep.org ​ Climate Tagger www.climatetagger.net ​

Sukaina Bharwani, Senior researcher and weADAPT Coordinator, SEI, Oxford, UK ​ https://www.sei.org/people/sukaina-bharwani/

Julia Barrott, Researcher and weADAPT Knowledge Manager, SEI, Oxford, UK ​ https://www.sei.org/people/julia-barrott/

23