Interdisciplinary Transformation University Austria

While cities are attractive places, brimming with opportunities and possibilities for their inhabitants, they have also been found to have negative consequences, especially on physical and mental health. In a world of ever-growing urban populations, it is important to understand how to make cities healthier and more pleasant places to live. In the present study, we investigated the impact of art as an urban intervention and compared it to the well-known effects of greenery (i.e., plants and vegetation) in an identically framed intervention. Specifically, we looked at how people engage with a Graetzloase (a type of parklet) and its embedding urban environment in terms of visual and spatial attraction as well as wellbeing. The Graetzloase displayed either abstract art or greenery and was placed on two distinct streets that, among other elements, also contained art and greenery. Our field study captured the ongoing experiences during people’s exploration of the urban environment by employing mobile eye-trackers and physiological devices. While our findings demonstrated a certain level of visual and spatial attraction towards the Graetzloases, it was not as pronounced as initially anticipated. Nevertheless, our analyses still inform on What decorating element should be placed in a Graetzloase, as well as Where to implement the Graetzloase. Our results suggest that artistic elements are more visually attractive (i.e., they were looked at for longer times) than the greenery, and that both visual and spatial attraction towards the Graetzloases are greatly impacted by the street context. We found that the Art Graetzloase when displayed in a wide street containing greenery elements, is visually more present in the participant’s visual field than all the other experimental combinations. The more precise analyses of the participant viewing behavior confirm this trend. Regarding wellbeing, we found no evidence for the impact of street context or the types of decorations in the Graetzloases. Our results establish an initial empirical foundation for the design and placement of not only future parklets but also urban art interventions in general.

The widespread use of mobile phones and social media platforms provides valuable information about users’ behavior and activities. Mobile phone data are rich on positional information, but lack semantic context. Conversely, geo-social media data reveal users’ opinions and activities, but are rather sparse in space and time. In the context of emergency management, both data types have been considered separately. To exploit their complementary nature and potential for emergency management, this paper introduces a novel methodology for improving situational awareness with the focus on urban events. For crowd detection, a spatial hot spot analysis of mobile phone data is used. The analysis of geo-social media data involves building spatio-temporal topic-sentiment clusters of posts. The results of the spatio-temporal contextual enrichment include unusual crowds associated with topics and sentiments derived from the analyzed geo-social media data. This methodology is demonstrated using the case study of the Vienna Pride. The results show how crowds change over time in terms of their location, size, topics discussed, and sentiments.

Detecting anomalies in general ledger data is of utmost importance to ensure the trustworthiness of financial records. Financial audits increasingly rely on machine learning (ML) algorithms to identify irregular or potentially fraudulent journal entries, each characterized by a varying number of transactions. In machine learning, heterogeneity in feature dimensions adds significant complexity to data analysis. In this paper, we introduce a novel approach to anomaly detection in financial data using Large Language Model (LLM) embeddings. To encode non-semantic categorical data (i.e., attributes lacking inherent linguistic meaning) from real-world financial records, we tested 3 pretrained general-purpose sentence-transformer models. For the downstream classification task, we implemented and evaluated 5 optimized ML models, including Logistic Regression, Random Forest, Gradient Boosting Machines, Support Vector Machines, and Neural Networks. Our experiments demonstrate that LLMs contribute valuable information to anomaly detection as our models outperform the baselines, in selected settings by a large margin. The findings further underscore the effectiveness of LLMs in enhancing anomaly detection in financial journal entries, particularly by tackling feature sparsity. We discuss a promising perspective on using SBERT embeddings for non-semantic data in the financial context and beyond.

During a disaster, the timely provision of customised and relevant data is of utmost importance. In the case of floods, data from remote sensing (satellite-based or airborne) is often used, but in recent years data from social media platforms has also been increasingly utilised. Focusing on these data sources, this study provides an in-depth assessment of requirements by emergency responders. Furthermore, the paper sheds light on the legal and ethical considerations that need to be taken into account during data collection and processing. A particular focus lies on the use of artificial intelligence (AI) for data analysis in disaster response. Topics such as privacy preservation and AI-informed decision making are highlighted throughout the paper. The investigation was carried out based on expert interviews with scientists, an extensive literature review, and workshops with emergency responders.

Abstract. Long short-term memory (LSTM) networks have demonstrated state-of-the-art performance for rainfall-runoff hydrological modelling. However, most studies focus on predictions at a daily scale, limiting the benefits of sub-daily (e.g. hourly) predictions in applications like flood forecasting. Moreover, training an LSTM network exclusively on sub-daily data is computationally expensive and may lead to model learning difficulties due to the extended sequence lengths. In this study, we introduce a new architecture, multi-frequency LSTM (MF-LSTM), designed to use input of various temporal frequencies to produce sub-daily (e.g. hourly) predictions at a moderate computational cost. Building on two existing methods previously proposed by the co-authors of this study, MF-LSTM processes older inputs at coarser temporal resolutions than more recent ones. MF-LSTM gives the possibility of handling different temporal frequencies, with different numbers of input dimensions, in a single LSTM cell, enhancing the generality and simplicity of use. Our experiments, conducted on 516 basins from the CAMELS-US dataset, demonstrate that MF-LSTM retains state-of-the-art performance while offering a simpler design. Moreover, the MF-LSTM architecture reported a 5 times reduction in processing time compared to models trained exclusively on hourly data.

Platforms have a problem with harmful or illegal content online.Flagging, which is an empowering tool for users to report violating content.A new European Union law, the Digital Services Act (DSA), seeks to harmonize the regulation of the flagging process.This paper examines how these flagging mechanisms support user action through semi-structured interviews (N=12) with regulatory authorities and professional reporting experts, using a walkthrough approach (with case studies based on flagging systems on Facebook and TikTok).We found tensions between the empowerment of users with additional reporting options and how it burdens users within service interfaces and processes; users need to understand the law, participate in a legal process, and differentiate between legal options and terms of service.Design choices, like the length of necessary reporting steps, also impacted expectations on the transparency of the reporting process.We close with design insights on support for users and stakeholders in the reporting process.

Abstract. Long Short-Term Memory (LSTM) networks have demonstrated state-of-the-art performance for rainfall-runoff hydrological modeling. However, most studies focus on daily-scale predictions, limiting the benefits of sub-daily (e.g. hourly) predictions in applications like flood forecasting. Moreover, training an LSTM exclusively on sub-daily data is computationally expensive, and may lead to model-learning difficulties due to the extended sequence lengths. In this study, we introduce a new architecture, multi-frequency LSTM (MF-LSTM), designed to use input of various temporal frequencies to produce sub-daily (e.g. hourly) predictions at a moderate computational cost. Building on two existing methods previously proposed by coauthors of this study, the MF-LSTM processes older inputs at coarser temporal resolutions than more recent ones. The MF-LSTM gives the possibility to handle different temporal frequencies, with different number of input dimensions, in a single LSTM cell, enhancing generality and simplicity of use. Our experiments, conducted on 516 basins from the CAMELS-US dataset, demonstrate that MF-LSTM retains state-of-the-art performance while offering a simpler design. Moreover, the MF-LSTM architecture reported a 5x reduction in processing time, compared to models trained exclusively on hourly data.

Abstract. Deterministic model predictions can struggle to adequately capture extreme events such as floods and droughts, which are of particular relevance in hydrology. This limitation arises because deterministic models collapse the conditional runoff distribution to a single point estimate. Probabilistic modeling provides a way to address this issue by explicitly representing uncertainty and assigning non-zero probabilities to a range of possible outcomes, including rare and extreme events, thereby capturing the full range of plausible hydrological responses. Motivated by this perspective, we examine whether probabilistic Long Short-Term Memory (LSTM) models improve the representation of extreme events in rainfall–-runoff simulations across Switzerland. Overall, the probabilistic models show good calibration, although some miscalibration remains for the extremes. Differences between models mainly manifest in how uncertainty is distributed: some approaches produce narrower and lighter-tailed distributions, while others yield broader distributions with heavier tails. These trade-offs highlight that probabilistic models differ not only in sharpness but also in how their calibration for rare events. We observe this tradeoff also in models' accuracy metrics. When evaluating the mean of the probabilistic predictions using the Nash–Sutcliffe efficiency (NSE), none of the probabilistic approaches outperform the deterministic LSTM in terms of average predictive accuracy. However, a clear advantage over the determinsitc models emerges when focusing on the tail of the discharge distribution. For the most extreme events (top 0.1 % of the discharge distribution), the deterministic LSTM underestimates more than 90 % of observed values (since it provides estimates of an expectation), whereas probabilistic predictions can capture a substantially larger fraction (67 %) of these extremes within their upper predictive bounds. Building on the additional information provided by probabilistic runoff predictions, we further show how they can be translated into actionable flood warnings using reinforcement learning. To this end, we introduce a Flood Risk Communication Agent (FRiCA) that operates on probabilistic runoff predictions and learns decision rules for issuing warnings of varying intensity. The FRiCA is implemented as an LSTM-based policy network and is trained by rewarding correct warning levels while penalizing the underestimation of flood severity. Results indicate that the FRiCA outperforms simple fixed heuristics, such as issuing warnings based on the predictive mean or a fixed high quantile (e.g., the 99th percentile). While this behavior already demonstrates the potential of reinforcement learning for improved flood risk communication, it also motivates further exploration of better reward design and policy network definition for context-dependent decision policies that adapt to varying hydrological and societal contexts.

<strong class="journal-contentHeaderColor">Abstract.</strong> Deterministic model predictions can struggle to adequately capture extreme events such as floods and droughts, which are of particular relevance in hydrology. This limitation arises because deterministic models collapse the conditional runoff distribution to a single point estimate. Probabilistic modeling provides a way to address this issue by explicitly representing uncertainty and assigning non-zero probabilities to a range of possible outcomes, including rare and extreme events, thereby capturing the full range of plausible hydrological responses. Motivated by this perspective, we examine whether probabilistic Long Short-Term Memory (LSTM) models improve the representation of extreme events in rainfall&ndash;-runoff simulations across Switzerland. Overall, the probabilistic models show good calibration, although some miscalibration remains for the extremes. Differences between models mainly manifest in how uncertainty is distributed: some approaches produce narrower and lighter-tailed distributions, while others yield broader distributions with heavier tails. These trade-offs highlight that probabilistic models differ not only in sharpness but also in how their calibration for rare events. We observe this tradeoff also in models' accuracy metrics. When evaluating the mean of the probabilistic predictions using the Nash&ndash;Sutcliffe efficiency (NSE), none of the probabilistic approaches outperform the deterministic LSTM in terms of average predictive accuracy. However, a clear advantage over the determinsitc models emerges when focusing on the tail of the discharge distribution. For the most extreme events (top 0.1 % of the discharge distribution), the deterministic LSTM underestimates more than 90 % of observed values (since it provides estimates of an expectation), whereas probabilistic predictions can capture a substantially larger fraction (67 %) of these extremes within their upper predictive bounds. Building on the additional information provided by probabilistic runoff predictions, we further show how they can be translated into actionable flood warnings using reinforcement learning. To this end, we introduce a Flood Risk Communication Agent (<em>FRiCA</em>) that operates on probabilistic runoff predictions and learns decision rules for issuing warnings of varying intensity. The <em>FRiCA </em>is implemented as an LSTM-based policy network and is trained by rewarding correct warning levels while penalizing the underestimation of flood severity. Results indicate that the <em>FRiCA </em>outperforms simple fixed heuristics, such as issuing warnings based on the predictive mean or a fixed high quantile (e.g., the 99th percentile). While this behavior already demonstrates the potential of reinforcement learning for improved flood risk communication, it also motivates further exploration of better reward design and policy network definition for context-dependent decision policies that adapt to varying hydrological and societal contexts.

<strong class="journal-contentHeaderColor">Abstract.</strong> Long Short-Term Memory (LSTM) networks have demonstrated state-of-the-art performance for rainfall-runoff hydrological modeling. However, most studies focus on daily-scale predictions, limiting the benefits of sub-daily (e.g. hourly) predictions in applications like flood forecasting. Moreover, training an LSTM exclusively on sub-daily data is computationally expensive, and may lead to model-learning difficulties due to the extended sequence lengths. In this study, we introduce a new architecture, multi-frequency LSTM (MF-LSTM), designed to use input of various temporal frequencies to produce sub-daily (e.g. hourly) predictions at a moderate computational cost. Building on two existing methods previously proposed by coauthors of this study, the MF-LSTM processes older inputs at coarser temporal resolutions than more recent ones. The MF-LSTM gives the possibility to handle different temporal frequencies, with different number of input dimensions, in a single LSTM cell, enhancing generality and simplicity of use. Our experiments, conducted on 516 basins from the CAMELS-US dataset, demonstrate that MF-LSTM retains state-of-the-art performance while offering a simpler design. Moreover, the MF-LSTM architecture reported a 5x reduction in processing time, compared to models trained exclusively on hourly data.

ReVerie is an Interactive AI installation that collects and visualizes textual dream data from the artist and the audience and creates a collective dream-reliving experience. Within Dream science, ‘Dreamwork’ encompasses techniques such as dream analysis, interpretation, and exploration aimed at uncovering insights into the subconscious mind. Central to ‘Dreamwork’ is the concept of re-experiencing dreams—immersing oneself in the recollection of dream memories, and emotions. Through a 3D generative diffusion model, the ‘ReVerie’ system translates the whispering of dream objects into 3D immersive visualization in real time to facilitate dream re-experiencing and a collective dream fly-through experience.

Long Short-Term Memory (LSTM) networks have shown strong performance in rainfall&amp;#8211;runoff modelling, often surpassing conventional hydrological models in benchmark studies. However, recent studies raise questions about their ability to extrapolate, particularly under extreme conditions that exceed the range of their training data. This study examines the performance of a stand-alone LSTM trained on 196 catchments in Switzerland when subjected to synthetic design precipitation events of increasing intensity and varying duration. The model&amp;#8217;s response is compared to that of a hybrid model and evaluated against hydrological process understanding. Our study reiterates that the stand-alone LSTM is characterised by a theoretical prediction limit, and we show that this limit is below the range of the data the model was trained on. We show that saturation of the LSTM cell states alone does not fully account for this characteristic behaviour, as the LSTM does not reach full saturation, particularly for the 1-day events. Instead, its gating mechanisms prevent new information about the current extreme precipitation from being incorporated into the cell states. Adjusting the LSTM architecture, for instance, by increasing the number of hidden states, and/or using a larger, more diverse training dataset can help mitigate the problem. However, these adjustments do not guarantee improved extrapolation performance, and the LSTM continues to predict values below the range of the training data or show hydrologically unfeasible runoff responses during the 1-day design experiments. Despite these shortcomings, our findings highlight the inherent potential of stand-alone LSTMs to capture complex hydro-meteorological relationships. We argue that more robust training strategies and model configurations could address the observed limitations, ensuring the promise of stand-alone LSTMs for rainfall&amp;#8211;runoff modelling.

Studies across various task types suggest that collaboration between humans and AI leads to improved and more satisfying outcomes. However, the effects of prolonged use of AI on users’ skills and perceptions remain unclear. This study involved 12 participants using an AI-based assistant in a multitasking balancing game over five days. Our findings indicate that AI assistance improved participants’ performance, even for the condition that was not supported by AI, showing no deskilling effect. Additionally, participants experienced significantly lower cognitive load (measured via ocular pupil diameter) in the AI-supported condition. This suggests that AI-assisted training can enhance multitasking motor skills in a less stressful and cognitively demanding manner. We also found a positive correlation between users’ understanding of the AI assistant and its acceptance, highlighting the importance of transparency and effectively communicating the capabilities of AI applications.

<strong class="journal-contentHeaderColor">Abstract.</strong> Long Short-Term Memory (LSTM) networks have demonstrated state-of-the-art performance for rainfall-runoff hydrological modeling. However, most studies focus on daily-scale predictions, limiting the benefits of sub-daily (e.g. hourly) predictions in applications like flood forecasting. Moreover, training an LSTM exclusively on sub-daily data is computationally expensive, and may lead to model-learning difficulties due to the extended sequence lengths. In this study, we introduce a new architecture, multi-frequency LSTM (MF-LSTM), designed to use input of various temporal frequencies to produce sub-daily (e.g. hourly) predictions at a moderate computational cost. Building on two existing methods previously proposed by coauthors of this study, the MF-LSTM processes older inputs at coarser temporal resolutions than more recent ones. The MF-LSTM gives the possibility to handle different temporal frequencies, with different number of input dimensions, in a single LSTM cell, enhancing generality and simplicity of use. Our experiments, conducted on 516 basins from the CAMELS-US dataset, demonstrate that MF-LSTM retains state-of-the-art performance while offering a simpler design. Moreover, the MF-LSTM architecture reported a 5x reduction in processing time, compared to models trained exclusively on hourly data.

Accurately estimating the risks of weather-related impacts requires comprehensively simulating weather conditions that could occur but have not occurred in the historical record. This is the aim of weather generators. Analog-based weather generators exploit the fact that the large-scale atmospheric circulation constrains regional weather and generate multivariate spatiotemporal meteorological fields by resampling historical data. During the resampling, constraints are employed to ensure that successive samples have consistent circulation patterns. Compared to other types of weather generators, resampling-based methods have the advantage that dependencies between variables and between locations are automatically correctly captured. However, the generated time series are limited to observed ranges, and even &amp;#8220;close&amp;#8221; analogs in the historical record are relatively far away from each other.We overcome these limitations by constructing a (daily) analog weather generator using ECMWF extended ensemble forecast hindcast (re-forecast) data, which provides a much larger sample size and the ability to sample values larger than the observed records. We choose this dataset because it has high spatial resolution and provides a large set of states from a relatively constant climate, while model biases remain limited because the forecasts are initialized from reanalysis data. With the ensemble hindcasts, we can also assess how &amp;#8220;close&amp;#8221; analogs are compared to typical ensemble spreads. We test our methodology by applying it to simulate weather over a European domain. Analogs are defined in terms of geopotential height at 500hPa and computed over an extended region including parts of the North Atlantic. With our approach, we can find better analogs compared to a baseline using only ERA5 data. We evaluate key properties of the simulated time series, such as their annual cycle, extremes, and lengths of wet and dry spells. The weather generator can be widely applied to estimate potential climate impacts, for instance with impact models. It is especially useful in cases where an accurate representation of dependencies between variables or across space is important for the impacts, which is the case for a number of different types of compound events.

Objectives: Due to political polarization, adherence to public health measures varied across US states during the COVID-19 pandemic. Although social media posts have been shown effective in anticipating COVID-19 surges, the impact of political leaning on the effectiveness of different topics for early warning remains mostly unexplored. Our study examines the spatio-temporal early warning potential of different geo-social media topics across republican, democrat, and swing states. Methods: Using keyword filtering, we identified eight COVID-19-related geo-social media topics. We then utilized Chatterjee's rank correlation to assess their early warning capability for COVID-19 cases 7 to 42 days in advance across six infection waves. A mixed-effect model was used to evaluate the impact of timeframe and political leaning on the early warning capabilities of these topics. Results: Many topics exhibited significant spatial clustering over time, with quarantine and vaccination-related posts occurring in opposing spatial regimes in the second timeframe. We also found significant variation in the early warning capabilities of geo-social media topics over time and across political clusters. In detail, quarantine related geo-social media post were significantly less correlated to COVID-19 cases in republican states than in democrat states. Further, preventive measure and quarantine-related posts exhibited declining correlations to COVID-19 cases over time, while the correlations of vaccine and virus-related posts with COVID-19 infections. Conclusion: Our results highlight the need for a dynamic spatially targeted approach that accounts for both how regional geosocial media topics of interest change over time and the impact of local political ideology on their epidemiological early warning capabilities.

Deterministic model predictions can struggle to adequately capture extreme events such as floods and droughts, which are of particular relevance in hydrology. This limitation arises because deterministic models collapse the conditional runoff distribution to a single point estimate. Probabilistic modeling provides a promising way to address this issue by explicitly representing uncertainty and assigning non-zero probabilities to a range of possible outcomes, including rare and extreme events, thereby capturing the full range of plausible hydrological responses. Motivated by this perspective, we investigate how long short-term memory (LSTM) based probabilistic models can be used for rainfall–runoff simulation across Switzerland. Overall, the probabilistic models show good calibration, although some miscalibration remains at the extremes. Differences between models mainly manifest in how uncertainty is distributed: some approaches produce narrower but lighter-tailed distributions, while others yield broader distributions with heavier tails. These trade-offs highlight that probabilistic models differ not only in sharpness but also in how they represent extreme outcomes. We also observe this trade-off in terms of the models’ single-point accuracy metrics. When evaluating the mean of the probabilistic predictions using the Nash–Sutcliffe efficiency (NSE), none of the probabilistic approaches outperform the deterministic LSTM in terms of average predictive accuracy. However, a clear advantage emerges when focusing on the tail of the discharge distribution. For the most extreme events (top 0.1% of the sorted discharge values), the deterministic LSTM underestimates more than 90% of observed values (since it provides estimates of an expectation), whereas probabilistic predictions can capture a substantially larger fraction of these extremes within their upper predictive bounds. Building on the additional information provided by probabilistic runoff predictions, we further show how such forecasts can be translated into discrete and actionable flood warnings using reinforcement learning. To this end, we introduce a Flood Risk Communication Agent (FRiCA) that operates on probabilistic runoff predictions and learns decision rules for issuing warnings of varying intensity. The FRiCA is implemented as an LSTM-based policy network and is trained by rewarding correct warning levels while penalizing the underestimation of flood severity. Results indicate that the FRiCA outperforms simple fixed heuristics, such as issuing warnings based on the predictive mean or a fixed high quantile (e.g., the 99th percentile). While this behavior already demonstrates the potential of reinforcement learning for improved flood risk communication, it also motivates future work toward more flexible and context-dependent decision strategies that adapt to varying hydrological and societal contexts.