Water management assessment with text mining

Water allocation during droughts is a challenge for policymakers, often addressed through participatory approaches. The implications of this governance mode are understudied as long-term records of the decision-making processes are often unavailable. We use natural language processing (NLP) and network analysis to extract information on water allocation decisions and climate-related issues from meeting minutes of river basin stakeholders. To test this approach, we considered the minutes of 1100 meetings held between 1997 and 2021 in the twelve basin committees of Ceará, Brazil. This region has a long history of droughts, which have strongly influenced water policies and politics. The river basin committee is currently composed of representatives of governmental and non-governmental institutions and deliberates on the water management process. To identify conflicts and relevant issues discussed during the meetings, we created a topic modeling approach consisting of: (1) sentence embedding using SBERT, (2) dimensionality reduction using UMAP, and (3) sentence clustering using K-means. Based on this, we calculated the topic frequency in each committee over time and normalized it by the number of documents registered each year. We also detected the topics mentioned in the same document to build network graphs of co-occurring topics. By using named entity recognition and dependency parsing, we identified the main actors involved during these meetings. Findings indicate that the most common topics were related to 'organic farming', 'fish mortality in reservoirs' and 'structural problems in water infrastructure'. The enhancement of water use monitoring - to identify potential water right violations - seems to be the preferred strategy to cope with droughts. During droughts, stakeholders appear to be more concerned about urban water supply than agriculture demand. We use historical data on water permit granting and water use charging to validate this finding. We also see an increase in climate-informed decisions over time, which became more frequent as new droughts affected the region. In summary, the proposed approach allows exploiting existing text data in order to identify the spatio-temporal patterns of topics related to water allocation. These data are often underexplored due to difficulties in analysing large amounts of text using conventional tools. Hence, text analysis offer exciting new opportunities for research in the field of water management.