In October 2021, the Swedish Meteorological and Hydrological Institute (SMHI) launched a novel national system for impact-based weather warnings, moving from the traditional format for meteorological, hydrological, and oceanographic warnings towards an assessment process that includes collaboration and consultation with regional stakeholders. For certain types of warnings, joint assessments of the potential impacts of weather events for a specific geographic area and time frame are made in collaboration with local and regional actors. As part of this new system, local and regional administrative efforts are made to create assessment-support documentation which are collated by practitioners at the municipal or organizational level, drawing on local knowledge, and subsequently compiled by the County Administrative Board. This process aims to support the collaborative decision-making processes ahead of the publication and in the evaluation of issued weather warnings. This paper explores the potential of integrating long- and short-term perspectives in societal response to climate change impacts with focus on extreme weather events. We present a case of AI-based technology to support processes linked to the national system for impact-based weather warnings and its integration with local and regional climate adaptation processes. We explore opportunities to integrate an AI-based pipeline, employing AI-based image and text analysis of crowdsourced data, in the processes of the warning system, and analyse barriers and enablers identified by local, regional, and national stakeholders. We further discuss to what extent data and knowledge of historical extreme weather events can be integrated with local and regional climate adaptation efforts, and whether these efforts could bridge the divide between long-term adaptation strategies and short-term response measures related to extreme weather events. Thus, this study unfolds the existing and perceived barriers to this integration and discusses possible synergies and ways forward in risk management and climate adaptation practice.

Extreme weather events, such as flooding, are expected to increase in frequency and intensity. Therefore, the prediction of extreme weather events, assessment of their local impacts in urban environments, and implementation of adaptation measures are becoming high-priority challenges for local, regional, and national agencies and authorities. To manage these challenges, access to accurate weather warnings and information about the occurrence, extent, and impacts of extreme weather events are crucial. As a result, in addition to official sources of information for prediction and monitoring, citizen volunteered geographic information (VGI) has emerged as a complementary source of valuable information. In this work, we propose the formulation of an approach to complement the impact-based weather warning system that has been introduced in Sweden in 2021 by making use of such alternative sources of data. We present and discuss design considerations and opportunities towards the creation of a visual analytics (VA) pipeline for the identification and exploration of extreme weather events and their impacts from VGI texts and images retrieved from social media. The envisioned VA pipeline incorporates three main steps: (1) data collection, (2) image/text classification and analysis, and (3) visualization and exploration through an interactive visual interface. We envision that our work has the potential to support three processes that involve multiple stakeholders of the weather warning system: (1) the validation of previously issued warnings, (2) local and regional assessment-support documentation, and (3) the monitoring of ongoing events. The results of this work could thus generate information that is relevant to climate adaptation decision making and provide potential support for the future development of national weather warning systems.

Extreme weather events are expected to increase in frequency and intensity due to global warming. During disaster events, up-to-date relevant information is crucial for early detection and response. Recently, Twitter emerged as a potentially important source of volunteered geographic information of key value for global monitoring systems and increasing situational awareness. While research on the use of machine learning approaches to automatically detect disaster events from social media is increasing, the visualization and exploration of the identified events and their contextual data are often neglected. In this paper, we address this gap by proposing a visual analytics pipeline for the identification and flexible exploration of extreme weather events, in particular floods, from Twitter data. The proposed pipeline consists of three main steps: (1) text classification, (2) location extraction, and (3) interactive visualization. We tested and assessed the performances of four classification algorithms for classifying relevant tweets as flood-related, applied an algorithm to assign location information, and introduced a visual interface for exploring their spatial, temporal, and attribute characteristics. To demonstrate our work, we present an example use case where two independent flooding events were identified and explored. The proposed approach has the potential to support real-time monitoring of events by providing data on local impacts collected from citizens and to facilitate the evaluation of extreme weather events to increase adaptive capacity.

This paper presents an evaluation of how data augmentation and inter-class transformations can be used to synthesize training data in low-data scenarios for single-image weather classification. In such scenarios, augmentations is a critical component, but there is a limit to how much improvements can be gained using classical augmentation strategies. Generative adversarial networks (GAN) have been demonstrated to generate impressive results, and have also been successful as a tool for data augmentation, but mostly for images of limited diversity, such as in medical applications. We investigate the possibilities in using generative augmentations for balancing a small weather classification dataset, where one class has a reduced number of images. We compare intra-class augmentations by means of classical transformations as well as noise-to-image GANs, to interclass augmentations where images from another class are transformed to the underrepresented class. The results show that it is possible to take advantage of GANs for inter-class augmentations to balance a small dataset for weather classification. This opens up for future work on GAN-based augmentations in scenarios where data is both diverse and scarce.