Conference Agenda

Flood susceptibility mapping (FSM) is a crucial component of flood risk assessment; however, traditional statistical and machine learning methods for FSM are limited in their predictive capabilities. FSM approaches typically use static inputs, relying solely on geospatial factors, and fail to consider the spatiotemporal aspects (antecedent conditions) that trigger flood events. This study addresses this gap by developing a hybrid model that combines static geospatial features with dynamic temporal meteorological data, which is often excluded in FSM. The proposed hybrid model consists of two branches: (1) a 2D Convolutional Neural Network (CNN) to extract the features from geospatial inputs (i.e., slope and surficial geology) and (2) a Convolutional Long Short-Term Memory (ConvLSTM2D) network to learn the temporal antecedent conditions from Daymet precipitation, temperature and snow-water equivalent. This model was trained and tested in the Saint John River basin, New Brunswick, Canada — a region that has experienced significant historical flooding. Three hyperparameters were investigated: temporal sequence length (1–4-month timesteps), resampling ratio (0.1-0.7), and positive class weight (1.5 or 2.0). The optimal model was achieved with a 3-month timestep, a 0.2 resampling ratio, and a 1.5 positive class weight, resulting in an F1 score of 0.89. The model performance was highest when using a 3-month timestep, which captured the full snowmelt-to-rain spring cycle, outperforming models that used timesteps of 1, 2, or 4 months. The proposed 2D CNN-ConvLSTM2D architecture is effective in simultaneously learning the static geospatial features and temporal meteorological sequences, highlighting the importance of seasonal antecedent conditions in FSM.

Optical satellite sensors are widely used for rapid flood mapping due to their global coverage and free availability. Thresholding spectral indices, such as the Normalized Difference Water Index (NDWI), can detect water pixels rapidly and with good precision. However, small shifts in threshold values can lead to large differences in flood area and data-driven approaches for threshold selection remain a challenge. At the same time, new foundation segmentation models, such as the Segment Anything Model (SAM), can extract object boundaries from images without task-specific training, though it lacks flood-specific contextual awareness. To address these limitations, we propose a risk-guided segmentation framework that combines risk-weighted optimization of NDWI thresholding, and further refinement of the NDWI mask using SAM. The goal is to improve flood delineation by incorporating information on where a flood is more likely to occur (flood hazard maps) and how flood boundaries appear visually (SAM). We evaluate the method on the 2018 spring flood along the Saint John (Wolastoq) River in New Brunswick, Canada, across five study regions for both Sentinel-2 and Landsat-8 scenes using imagery captured on May 2, 2018 (peak flood for the study regions). We show that a higher risk score corresponds to a higher segmentation accuracy, demonstrating that flood hazard maps can help guide NDWI threshold selection. Moreover, refinement with SAM improves segmentation quality compared to the baseline NDWI masks, demonstrating that the use of risk-guided spectral thresholding with foundation models can improve flood delineation in optical satellite imagery.

Floods remain among the most damaging natural hazards worldwide. In Canada’s capital region—Ottawa and its surrounding areas—flood prediction is crucial, most especially in flood-prone zones, to mitigate recurring events such as the 2017 and 2019 Ottawa floods, which caused extensive damage to homes and infrastructure. This study integrates 18 flood conditioning factors with remote sensing indices and ensemble machine learning to improve flood susceptibility mapping in the Ottawa River watershed. A complementary HEC-RAS 2D hydraulic model simulated flow depth and velocity under a 100-year flood scenario. The ensemble model achieved strong predictive performance (Kappa, F1-score, and AUC > 0.979) and demonstrated high transferability across sub-regions (Kappa > 0.85; F1-score > 0.92; AUC > 0.99). HEC-RAS results indicated spatial variability in flood depth (up to 15 m) and velocity (up to 15 m/s). SHAP analysis identified Elevation, HAND, MNDWI, NDWI, and Aspect as the dominant flood-driving factors. The integrated framework enhances flood susceptibility assessment and supports Natural Resources Canada’s efforts to strengthen flood risk management and resilience in the Ottawa River watershed and similar regions.

Machine learning for flood susceptibility mapping (FSM) has traditionally relied on narrowly scoped events and temporally constrained datasets, limiting the generalizability and long-term utility of predictive models. We present a multi-event, multi-temporal modelling framework that leverages discrete flood occurrences from 2005 to 2023 to train a unified model capable of inference across an extended temporal horizon. Each flood event was treated as a spatio-temporal marker, enabling the model to learn evolving driver–event relationships and underlying temporal trends. Dynamic inputs (e.g., climate data, land use/land cover) are integrated with static geophysical features (e.g., digital terrain model and derivatives) to capture both transient and persistent influences on flood susceptibility. An XGBoost model was trained, tested, and validated using a 70/15/15 split, achieving an overall accuracy of 0.945, with true positive and true negative rates of 0.95 and 0.94, respectively. Precision scores for wet (flood-prone) and dry (non-flood-prone) classes are 0.94 and 0.95. Generated yearly national FSM maps from 2000 to 2023 were evaluated against published flood event datasets. Validation using national flood records, climate variability bulletins, and spatio-temporal analyses of year-to-year raster correlations confirms that years with elevated predicted susceptibility correspond to observed flood events. In addition, a weighted wetness score identified the years with both widespread and extreme flood-prone conditions, highlighting the model’s ability to capture multi-scale temporal dynamics. These results demonstrate that multi-event, multi-temporal modelling enhances the temporal reach and robustness of geospatial flood prediction, providing a foundation for long-term monitoring, trend analysis, and policy-relevant scenario planning.

This study presents a high-resolution geomorphological analysis of river terraces along the urban corridor of the La Silla River (Monterrey Metropolitan Area, Mexico) using UAV-based LiDAR and photogrammetry, with a DJI Matrice 350 RTK equipped with a Zenmuse L2 sensor, generating dense point clouds, DEMs, and orthomosaics. These products allowed for the precise identification of three terrace levels (T1-T3), their geomorphometric attributes, and their lithological composition. The results reveal contrasting degrees of anthropogenic modification: while terrace 1 retains its natural morphology, terraces 2 and 3 show substantial alterations due to residential expansion, public infrastructure, and road construction, which alter the original geomorphological surfaces. Temporal satellite images also show the sensitivity of terrace geomorphology to extreme hydrometeorological phenomena, with cyclones such as Hanna (2020) and Alberto (2024) causing vegetation loss, surface restructuring, and local modification of terraces. Overall, UAV-LiDAR proved to be very effective for mapping terraces in restricted urban environments, providing essential details for monitoring, risk assessment, and sustainable management of urban rivers.

The accumulation of C in peatlands generally depends on hydrologic conditions that maintain saturated soils and impede rates of decomposition. Boreal Peatlands have provided rich reservoirs of stored C for millennia. However, with climate change, warming and drying patterns across the boreal and arctic are resulting in dramatic changes in ecosystems and putting these systems at risk. As long as peatlands are functioning hydrologically, they will continue to sequester and store carbon. The ability to retrieve and monitor soil moisture from peatlands is of interest for a wide range of applications from hydrological modeling to understanding ecosystem vulnerabilities to increased drought, decomposition and wildfire to monitoring methane flux and peatland restoration. To develop soil moisture retrieval algorithms, we studied a range of boreal peatland sites (bogs and fens) stratified across geographic regions of North America from 2010 to 2024. We developed soil moisture retrieval algorithms from polarimetric C-band (5.7 cm wavelength) and L-band (24 cm wavelength) synthetic aperture radar (SAR) data. Both multi-linear regressions and gradient boosters (XGBoost, CatBoost and Explainable Boosting Machines) were developed. We found that integrating polarimetric SAR parameters that are sensitive to vegetation structure and parameters most sensitive to surface soil moisture in the models provided the best results. Data were withheld for model testing and coefficient of determination, RMSE, unbiased RMSE are reported.

The Peace–Athabasca Delta (PAD) is undergoing long-term ecological change driven by climate warming, hydro-regulation, and fluctuating flood–dry cycles. This study uses a harmonised 40-year Landsat composite time series (1984–2024) to assess vegetation, surface-water extent, and thermal conditions across the delta. An 11-year moving-window Mann–Kendall trend analysis was applied to NDVI, EVI, MNDWI, and LST, retaining only significant Theil–Sen slopes. Significant vegetation–water trends were combined into a 10-class framework that maps greening, browning, wetting, and drying across all landscape types, including ecotones. Parallel LST trends reveal reinforcing or contrasting thermal feedbacks. It provides a coherent basis for interpreting whether vegetation and hydrologic changes reflect ecotone expansion or contraction under thermal variability.

Mountain headwaters of the Eastern Slopes of Alberta (ES) are the primary source of freshwater of major easterly flowing basins in western Canada, supplying a significant volume of water to about four million people. However, increasing temperatures is altering mountain aquatic (open water areas, lakes, reservoirs, rivers, ponds, wetlands) and riparian vegetation (herbaceous and woody/shrub) ecosystems. The ES, Canada, has demonstrated landcover and process changes associated with climate warming, e.g., increases in the air temperature [1] have led to earlier snowmelt, and increased glacier wastage, resulting in higher river flows over a shorter period, which can result in expansion of open water areas during and following peak flow periods [2]. The impacts on wetlands are less visible or well known, and there is a need to evaluate spatial and temporal changes and trends in wetland loss, growth, or genesis across this mountainous ecosystem. Here, we provide a framework for quantifying and assessing multi-decadal wetland extents over the large spatial scale of the ES from 1984 to 2023. We used the historical Landsat archive to produce a remote sensing-based time series landcover classification over the last 40 years in the ES.

Functional Data Analysis (FDA) provides a powerful framework for representing temporal dynamics in remote-sensing data. Building on this concept, this study develops a transfer learning framework using a minimally trained Functional Principal Component Analysis (FPCA)-based feature extraction engine (“FPC engine”) to map dynamic wetlands at large scale. A small set of training locations from Ontario was used to train the FPC engine, which captures dominant seasonal backscatter patterns of open water, shallow water, and marsh-like vegetation. The trained engine was then transferred to the Prairie Pothole Region (PPR) to delineate dynamic wetland classes without extensive local calibration. This label-efficient design—supervised in selecting training locations but unsupervised in feature extraction—reduces field data needs while maintaining strong generalization. Validated results show that the transferred FPC engine effectively separates dynamic wetland classes across contrasting climatic and geomorphic conditions, supporting scalable and cost-efficient monitoring with Sentinel-1 SAR data.

Wetland mapping in boreal environments remains challenging due to complex vegetation structure, subtle and variable terrain gradients, diverse wetland types, and the proportion of treed wetlands. This study develops and evaluates a framework to remotely identify wetland types in Pukaskwa National Park (Ontario, Canada) by integrating multi-scale terrain metrics with multi-temporal Sentinel-2 spectral derivatives. Five years (2017–2021) of Sentinel-2 data were used to derive harmonic NDVI metrics, including linear trend, amplitude, and phase of the first Fourier component, capturing seasonal vegetation and hydrologic dynamics. These spectral predictors effectively delineated open water and non-treed peatlands but struggled in densely forested wetlands where canopy obscures surface moisture signals. To address this limitation, Gaussian scale-space analysis was applied to the Copernicus GLO-30 DSM, informed by FFT-based evaluation of terrain wavelengths (100 m–10 km), to generate multi-scale Local Relief Models and curvature metrics representing depressional and convex landforms. A hierarchical workflow masked open water using Sentinel-1, removed upland convex terrain using LRM-curvature rules, then applied Random Forest classification using field training data and combined spectral-terrain predictors. Accuracy assessment stratified by terrain context showed strong performance in low-lying depressional areas and suppression of false wetland detections in high terrain with local depressions. Reduced accuracy in relatively flat areas was attributed to DSM vertical uncertainty limiting detection of shallow depressions beneath dense canopy, resulting in reliance on optical separability that weakens under closed canopy but improves where tree cover is sparse. Overall, results demonstrate the value of combining Fourier-based temporal descriptors with multi-scale terrain analysis for boreal wetland mapping.

This work proposes a semantic-enhanced dynamic spatiotemporal model that integrates temporal attention, dynamic graph learning, and semantic module to better capture the complexity of human mobility. By combining dynamic adjacency learning with geographic and semantic structures, the model identifies both physical and functional relationships between zones. Results on TELUS mobility data demonstrate that semantic-enhanced graph construction improves prediction accuracy and robustness, offering a more meaningful representation of urban mobility dynamics and providing a strong foundation for future mobility forecasting and city-scale analytics.

Urban quality of life (QoL) assessments often rely on objective spatial indicators such as infrastructure access, land use, and environmental conditions. However, these metrics may overlook how residents subjectively perceive their surroundings. This disconnect reflects a methodological gap in urban studies: the lack of frameworks that integrate both objective and perceptual aspects of urban quality. In response, this study introduces a Perception-Based Urban Quality of Life Index (PUQLI) derived from street view imagery and deep learning and compares it with a composite objective indicator built from 13 spatially measured indicators across seven QoL domains. Rather than replacing conventional QoL assessments, PUQLI is intended to capture the visual-perceptual or experiential dimension of urban quality as inferred from street-level imagery. Each indicator was normalized and spatially joined to a hexagonal grid system. Pearson correlation revealed only modest associations between PUQLI and the objective indicators, indicating that subjective and objective urban quality are related but not equivalent. A mismatch index was then computed to quantify perception–provision gaps, revealing statistically significant and spatially patterned divergences (t = –10.535, p < 0.0001). Positive mismatch clustered in mixed-use urban centers, whereas negative mismatch aligned with documented environmental and infrastructural stressors; together with the significantly negative mean mismatch, this indicates a structural perception–provision gap in which measurable provision does not always translate into favorable lived experience. These findings highlight the need to integrate subjective perception into urban quality assessment and position the mismatch index as a practical diagnostic tool for perception-informed urban planning.

This research develops a pedestrian behavior model that incorporates the existence and dynamics of pedestrian groups. Using high-precision laser sensor data collected in the atrium of a hospital, the research first defines spatiotemporal parameters representing interpersonal distance, relative speed, and walking direction between pedestrians. Based on these parameters, machine learning techniques, including Support Vector Machine (SVM) and Random Forest algorithms, were employed. The SVM demonstrated superior accuracy and stability, successfully identifying groups even under complex walking conditions. Building on these results, the pedestrian behavior model described by psychological stress factors, such as stress from other pedestrians, obstacles, and group dispersion, is improved to account for the behavior of pedestrian groups. Model parameters were calibrated using laser sensor trajectory data with individual attributes (sex, staff, mobility aid usage). The proposed model accurately reproduced observed walking trajectories, with errors within 80 cm for approximately 80% of pedestrians. Finally, the model was applied to evaluate pedestrian spaces by mapping spatial distributions of psychological stress. Pedestrian stress was highest around reception areas, while group dispersion stress was greater in low-density zones where groups tend to spread out. These findings demonstrate that incorporating group behavior enhances the realism and applicability of pedestrian models for evaluating and designing public spaces. Future work will focus on applying the model to diverse facilities and pedestrian environments.

This study presents a workflow that links roadside LiDAR sensing with the modelling of pedestrian crossing behavior. Using self-collected LiDAR data from an informal mid-block crossing in Munich, the workflow includes object detection, tracking, trajectory reconstruction, event extraction, and contextual feature engineering. Behaviour-based yielding and stepping-out moments are used to identify pedestrian decision moments, which are subsequently labelled as gap-accepted or gap-rejected according to gap-acceptance theory. For each decision moment, features describing pedestrian state, social context, and vehicle context are extracted from the reconstructed trajectories. A logistic regression classifier is applied as an interpretable baseline to estimate gap-acceptance decisions under varying traffic conditions. The preliminary results indicate satisfactory predictive performance and show intuitive coefficient patterns, highlighting the influence of vehicle time gaps, pedestrian standing position, and peer presence. Overall, the study demonstrates the effectiveness of LiDAR-derived trajectories as a behavioral sensing foundation for modelling pedestrian crossing decisions.

This study proposes a ring-based SpatialTransformer to learn how building uses at different distances from a railway station interact to generate pedestrian flow. Concentric ring buffers at 100-meter intervals up to 800 meters were defined around 100 randomly selected stations in Tokyo, treating each ring as a spatial token. Self-Attention was applied to learn inter-zone interactions directly from data, without prior structural assumptions. GPS-derived walking trip counts served as the target variable and Geographically Weighted Regression as the baseline. Across 30 independent trials, the SpatialTransformer consistently outperformed GWR in predictive accuracy. SHAP analysis revealed that mid-to-outer distance zone features dominate pedestrian flow prediction, while features from the 0-100m zone contributed little. The attention matrix showed that each distance zone attends most strongly to spatially distant zones, demonstrating that pedestrian flow is regulated by structural interactions across the entire catchment area rather than by any single zone in isolation. These findings challenge the compact city assumption that station-proximate development maximizes pedestrian flow, and suggest that land use distribution across the full walkable catchment area deserves greater consideration in urban planning practice.

Urban accessibility is a key factor in understanding spatial inequality, as it conditions residents’ ability to reach essential services and urban opportunities. This study analyses accessibility in the intermediate city of Los Ángeles, Chile, characterized by a centralized concentration of services and expanding peripheral residential areas. Accessibility to educational, healthcare, and commercial facilities was evaluated using approximate travel times generated through the TravelTime API, considering walking, public transport, and private vehicle modes. Travel times were calculated from the centroids of residential census blocks, and opportunity-based accessibility was assessed using travel-time thresholds to identify the range of accessible commercial establishments.The results reveal marked spatial disparities. Central areas exhibit the highest levels of accessibility due to the density and diversity of amenities, with walking emerging as the most efficient mode for short distances. In contrast, peripheral neighbourhoods show limited access to healthcare and educational facilities and depend largely on private vehicles to reach central services, despite having higher population densities. Commercial accessibility in these areas is primarily restricted to small-scale neighbourhood establishments. These findings indicate that accessibility is influenced not only by travel time and transport networks but also by the spatial distribution and variety of urban functions. The study highlights the usefulness of routing APIs as an alternative methodological tool for accessibility analysis in contexts where official mobility data are outdated or incomplete, offering valuable insights for urban planning and policies aimed at reducing spatial inequalities.

Rapid urbanization is accelerating the fragmentation of blue–green spaces and the degradation of ecosystem services, while widening inequalities in environmental exposure and access to ecological benefits. Taking the “Geo-information Town” as a case study, this paper develops an integrated 3D framework linking urban form, human behavior and spatial interactions. First, UAV oblique images are semantically segmented to identify blue–green–grey features and to jointly assess and filter image quality. Second, multi-source spatial data, including Gaode POIs, nighttime lights, urban land use, OSM road networks, vector base maps and Baidu heat maps, are used to characterize urban functions and vitality patterns related to catering, sightseeing, shopping and cultural–educational services. Third, social media check-in data from Xiaohongshu and Weibo are incorporated to capture residents’ subjective evaluations and place preferences for different spatial units. An XGBoost–SHAP modelling framework is employed to quantify the relationships between these subjective evaluations and blue–green–grey indicators, and to interpret the marginal contributions of different environmental and functional attributes. The results reveal how perceived landscape qualities and service functions jointly shape spatial attractiveness and human–landscape interactions at the neighborhood scale. Finally, we discuss future research on 3D indicator systems, fine semantic segmentation of blue–green spaces, multi-source big data fusion and perception–behavior–function coupling, providing methodological support for perception-oriented assessment of residential environmental quality and optimization of blue–green urban landscapes.

Active Mobility Accessibility Index (AMAI) quantifies the competitiveness of walking and cycling relative to driving using travel-time and distance ratios on identical sampled origin-destination pairs, reflecting network structure rather than destination choice. AMAI combines time parity and distance parity in a simple diagnostic score, using equal weights as a default specification for interpretation and policy use. Applied across the five Tyne and Wear local authorities, it demonstrates that cycling is more competitive than walking against driving. The median origin-level cycling AMAI is 0.820 and the median walking AMAI is 0.645. Parity remains limited where the share of origins at or above parity is 10.0% for cycling and 1.7% for walking. Initial API-based tests suggested that time-of-day effects are limited for the short local trips studied here, supporting development of a scalable in-house routing workflow for the main analysis. Validation against OA-level Census 2021 mode shares, with controls for terrain gradient and commute-distance composition, suggests that AMAI captures a relevant behavioural signal, while its main value lies in diagnosing local network competitiveness for policy and planning.

Understanding pedestrian behaviors is the foundation of simulation for space planning. However, conventional behavior modeling methods are insufficient for learning detailed interactions, and deep learning methods often lack interpretability. This study aims to develop a pedestrian trajectory modeling approach based on discovering causal relationships among pedestrians. The proposed method consists of two parts: analyzing causal relationships among pedestrians using statistical causal discovery methods and predicting trajectories using attention-based deep learning methods. The first part employs a semi-parametric method to identify the causal relationships underlying observed pedestrian behavior and construct a spatial-temporal graph based on these causal relationships. The second part primarily uses the graph attention network to learn interactions among pedestrians. The experimental results demonstrate that the proposed method achieves a good balance between prediction accuracy and interpretability, while also identifying limitations, including at low-density scenes and due to causal model assumptions.