- A Tipping Point – VIDI Project
In this repository, you will find a summary of all the projects where I was involved as a post-doc at TU Delft, Netherlands.
Abstract Climate change is one of the most pressing topics of the 21st century. As the time to mitigate its possible impacts has passed, now it is important that individuals and societies start to adapt. This makes the study of individuals’ and societies’ drivers and constraints to climate change adaptation of paramount importance, but so far there is a lack of standardized information to analyze such topics worldwide. To fill this gap, we designed a pipeline to derive databases using peer-reviewed articles’ text. We use Machine Learning and Natural Language Processing to extract the articles’ findings. Then, we derived a database that, later, we used to perform descriptive and statistical analyses of climate change adaptation to floods and sea-level rise, as well as, to study farmers’ incremental and transformational adaption worldwide. We argue that the use of transparent methodologies to derive data from text is increasingly necessary as, in general, the amount of literature grows exponentially.
The main three outcomes for this project are:
- An algorithm to visually explore the factors associated with climate change adaptation.
- A database of articles about individuals’ and societies’ factors associated with climate change adaptation.
- Usage of the algorithm and the database to study two different climate change topics: (a) adaptation to floods and sea-level rise, and (b) farmers’ incremental and transformational adaptation to climate change.
In the following sections I further elaborate on them.
Acknowledgments This work was supported by the Netherlands Organization for Scientific Research NWO VIDI grant number 191015. This is the main project to which I was part of. For it, I worked together with the SC3 members: Prof. Tatiana Filatova (Principal Investigator) and the PhD candidates Thorid Wagenblast and Joos Akkerman.
Abstract The fast-growing number of research articles makes it problematic for scholars to keep track of the new findings related to their areas of expertise. Furthermore, linking knowledge across disciplines in rapidly developing fields becomes challenging for complex topics like climate change that demand interdisciplinary solutions. The rise of machine learning-supported text analysis has been instrumental in processing thousands of articles. Yet, how text relationships are built remains a Black-Box for domain experts, making it difficult to relate connected concepts to existing theories conceptualizing cause-effect relationships and permitting hypothesis building and testing. This paper presents an approach to sensibly use Natural Language Processing by extracting variable relations and synthesizing their findings using networks while relating to key concepts dominant in relevant disciplines. As an example, we apply our methodology to analyze farmers’ adaptation to climate change and compare the results with mainstream text summarization methods. Furthermore, we validate our methodology by asking experts to score and compare the outcomes. Results show that using Natural Language Processing and networks descriptively offer an interpretable way to synthesize literature review findings that outperform mainstream text summarization methods. This methodology gives not only the direction and the frequency of the association between the words but also the frequency the word appears in the articles, which is instrumental when performing literature reviews.
This work is currently under revisions in a journal. The working paper can be accessed here and the codes to reproduce and replicate the work are here.
2. Database of articles about individuals’ and societies’ factors associated with climate change adaptation
The data for our two applications comes from Scopus (ELSEVIER, 2023). The data was retrieved in two batches: during the last week of August 2022, and during the last week of January 2024. We derived the data from articles that were about human adaptation to climate change and that belonged to the categories: Multidisciplinary, Social Sciences, Arts and Humanities, and Environmental Science. For both batches, we searched for articles related to adaptation to climate change using search terms related with climate change (e.g. “climat[a-z]* change”,“…”).
The first data batch was retrieved during the last week of August 2022, resulting in 30,000 articles. As not all of the articles were about the studied topic (factors associated with undertaken CCA measures), we needed to categorize the articles as “relevant” or “irrelevant.” For this, we organized a labeling session where, besides us, eight PhD candidates experts in CCA participated. To simplify labelers work, we performed a Non-Negative Matrix Factorization on the articles’ titles and abstracts and then give the labelers the articles corresponding to one cluster.
Using the articles clustered in that way, we categorized 5% of the articles as relevant or irrelevant to the study of CCA using ASReview (ASReview LAB developers, 2022). To train the participants, we explained the problem to them and we showed them several examples of the type of articles’ abstract that were useful to our study (Appendix B). In terms of the active learning classifier, we used the following settings: feature extraction technique – TF-IDF; classifier – Naïve Bayes; query strategy – Mixed (95% Maximum and 5% Random); and balance strategy – Dynamic resampling (Double). The labeling session lasted 4 hours and, at the end, we had 1600 labeled articles where 33% were relevant.
To classify the rest of the articles, we trained a simple perceptron classifier with a Binary Cross Entropy and Sigmoid loss function (BCEWithLogitsLoss function from the python package PyTorch (PyTorch, n.d.)). We split the labeled database into 64% training, 16% validation, and 20% test sets. The model had 79% accuracy and resulted in 2438 articles classified as related to human adaptation to climate change.
To validate that the 2438 articles were about human adaptation to climate change and that they talked about the possible drivers and constraints, we (together with a student assistant and two PhD candidates) read the abstracts from the articles and when the article was relevant we downloaded its pdf. This final step took between December 2022 and February 2023, and resulted in the retrieval of around 842 articles.
The second data batch was retrieved during the last week of January 2024. For this, we constraint the search to articles published between August 2022 and January 2024. This resulted in 10279 articles. To categorize the data as relevant and irrelevant, we used the 842 articles obtained for the first batch to update the simple perceptron classifier with a Binary Cross Entropy and Sigmoid loss function. This gave us 548 articles, which we manually validated and downloaded their pdfs. The second batch of data consists of 253 articles.
To clean the text, we followed the steps 1 to 4 showed and explain in our algorithm article. In this step, we also add the authors’ affiliation and the articles’ researched countries. The authors’ affiliation was extracted from Scopus metadata. The articles researched countries were extracted coding an algorithm that uses sentences’ Name-Entity-Recognition (NER) together with the “en_core_web_trf” spaCy model (spaCy, n.d.-a) to extract the articles’ Geopolitical entity (GPE), Locations (LOC), and Nationalities (NORP). The algorithm tries to identify the country based on the GPE, LOC, and NORP. It follows the next sequence. It starts in the title, if it does not find the country based on the GPE, LOC, and NORP, then it moves to the abstract. Finally, if it does not find the country neither in the title nor in the abstract, then it looks for the region (such as, North America, Sub-Saharan Africa) using regex. By doing this, we identified all the authors’ affiliation and 83% of the countries and regions studied. We validated the results by manually checking 22% of the articles.
The codes to reproduce and replicate the work are here.
(a) Patterns in Reported Adaptation Constraints: Insights from Peer-Reviewed Literature on Flood and Sea-Level Rise
Abstract Understanding which climate change adaptation constraints manifest for different actors – governments, communities, individuals and households – is essential, as adaptation is turning into a matter of survival. Though rich qualitative research reveals constraints for diverse cases, methods to consolidate knowledge and elicit patterns in adaptation constraints for various actors and hazards are scarce. We fill this gap by analyzing associations between different adaptations and actors’ constraints in adaptation to climate-induced floods and sealevel rise. Our novel approach derives textual data from peer-reviewed articles (published before February 2024) by using natural language processing, supervised learning, thematic coding books, and network analysis. Results show that social capital, economic factors, and government support are constraints shared among all actors. With respect to adaptation types, communities are frequently associated with maladaptation, while individuals and households are frequently associated with transformational adaptation.
This work is currently in the second round of revisions in a journal. The working paper can be accessed here. The codes to reproduce and replicate the work are here. The final database that results from the analysis is stored in DANS under:
- Gil-Clavel, Sofia; Filatova, Tatiana, 2024, “Interrelated Climate Change Adaptation Measures and Factors”, https://doi.org/10.17026/SS/PYZCXK, DANS Data Station Social Sciences and Humanities, DRAFT VERSION; FloodsSLRDataFrame_Phase1.csv [fileName].
(b) Incremental and Transformational Climate Change Adaptation Factors in Agriculture Worldwide: A Natural Language Processing Comparative Analysis
Abstract Climate change is expected to adversely affect agriculture worldwide. This is especially true if farmers fail to at least incrementally adapt early in the twenty-first century, and fail to pursue the transformational adaptation necessary to withstand future climate changes. Many publications discuss the underlying mechanisms of autonomous adaptation to climate change using quantitative, qualitative, and mixed methods. However, the review of empirical evidence on adaptation is usually performed on articles with quantitative data using metanalysis, omitting much of the vast body of literature evidence based on qualitative work. We address this gap by performing a comparative analysis of factors associated with farmers’ climate change adaptation in both quantitative and qualitative literature using Natural Language Processing and generalized linear models. By retrieving publications from Scopus, we derive a database with metadata and associations from both quantitative and qualitative findings. We use the derived data as input for generalized linear models to analyze whether farmers’ climate change adaptation factors differ by type of adaptation (incremental vs. transformational) and across different regions of the world. Results show that factors related to informal institutions’ adaptive capacity are more likely to be associated with transformational adaptation than with incremental adaptation. Regarding world regions, results highlight uneven access to income and infrastructure, with farmers in high-income countries having an advantage, whereas farmers in low- and middle-income countries require these the most for effective adaptation to climate change.
This work is currently in the second round of revisions in a journal. The working paper can be accessed here and the codes to reproduce and replicate the work are here. The final databases that result from the analysis is stored in DANS under:
-
Gil-Clavel, Sofia; Filatova, Tatiana, 2024, “Interrelated Climate Change Adaptation Measures and Factors”, https://doi.org/10.17026/SS/PYZCXK, DANS Data Station Social Sciences and Humanities, DRAFT VERSION; FarmersDataFrame_Phase1.csv [fileName].
-
Gil-Clavel, Sofia; Filatova, Tatiana, 2024, “Interrelated Climate Change Adaptation Measures and Factors”, https://doi.org/10.17026/SS/PYZCXK, DANS Data Station Social Sciences and Humanities, DRAFT VERSION; FarmersDataFrame_Phase2_AllConnections.csv [fileName].