How can spatiotemporal clustering of animal movement trajectories be used to identify high-risk road crossing corridors associated with animal mortality?
This question directly builds on course topics such as trajectory mining, spatial clustering, and spatial network analysis, while extending them to an applied ecological and safety-focused domain not fully explored in the assigned readings.
Identification of High-Risk Crossing Zones
Can clustering animal trajectories near road networks reveal consistent road-crossing hotspots that are associated with increased collision risk?
This leverages spatial clustering methods discussed in the course but applies them to animal–infrastructure interactions.
Risk Patterns
How do crossing hotspots vary by time (e.g., season, time of day), and can spatiotemporal segmentation improve the detection of high-risk periods for animal–vehicle collisions?
While temporal analysis appears in the course, its use for risk-aware spatial pattern discovery remains underexplored.
Anomalous Movement Near Roads
Do anomalous deviations from typical migration corridors occur more frequently near roads, and can such deviations be automatically detected using unsupervised methods?
This connects trajectory mining with unsupervised anomaly detection, extending ideas from the reading list.
Preliminary Proposal: Detection of Unexpected Forest Fire Ignition Hotspots Using Exposure-Normalized Spatiotemporal Analysis
This project explores spatiotemporal hotspot detection and spatial pattern analysis to study forest fire ignition events. The course covers hotspot detection, spatial clustering, and spatiotemporal data mining, primarily using examples from urban systems, transportation, and human activity data. However, most papers in the reading list identify hotspots based on raw event density and implicitly assume uniform or known exposure across space and time.
In the context of forest fires, this assumption does not hold. Fire ignitions depend on multiple factors such as vegetation, fuel conditions, weather, lightning occurrence, and human activity, all of which vary across space and time and are not directly observable as a single exposure variable. As a result, regions with many reported ignitions may reflect higher ignition opportunity rather than higher underlying risk. Studying forest fire ignitions provides an opportunity to apply spatiotemporal data mining techniques from the course while addressing a key limitation that is not fully explored in the assigned readings.
How can exposure-normalized spatiotemporal hotspot detection be used to identify high-risk forest fire ignition regions when ignition opportunity varies across space and time and is not directly observable?
This question directly builds on course topics such as hotspot detection and spatial clustering, while extending them to an environmental risk application where uneven and latent exposure complicates standard hotspot analysis.
1.Identification of Exposure-Normalized Ignition Hotspots
Can normalizing observed forest fire ignition events by estimated ignition opportunity lead to different hotspot patterns compared to those obtained using raw ignition density alone? This applies classical hotspot detection methods discussed in the course but adapts them to account for uneven environmental and ignition conditions across space and time.
2.Temporal Patterns of Ignition Risk
How do ignition hotspots change over time, such as across seasons or under different weather conditions, and can spatiotemporal segmentation improve the identification of persistently high-risk ignition regions?While temporal analysis is covered in the course, its use in exposure-aware hotspot detection for environmental risk remains underexplored.
3.Unexpected Ignition Activity
Do certain regions experience more forest fire ignitions than expected given local ignition conditions, and can such regions be identified using exposure-normalized spatiotemporal analysis?This connects hotspot detection with anomaly detection by focusing on identifying areas with unexpectedly high ignition activity rather than high activity alone.