General Events

View Full Calendar

PhD Defense | Evaluating the Influence of Biocontrol Program on the Colorado River Biodiversity with Multi-Source Time Series Imagery

Event Type
Seminar/Symposium
Sponsor
Department of Geography & GIS
Location
2049 Natural History Building and on Zoom
Date
Jun 9, 2025   11:00 am  
Speaker
Yilun Zhao, Geography PhD candidate
Cost
This dissertation defense is free and open to the public.
Registration
Zoom link
Contact
UIUC Geography & GIS
E-Mail
geography@illinois.edu
Originating Calendar
Geography and Geographic Information Science

Biological control is a widely adopted strategy for managing invasive plant species through the release of natural enemies. While effective in reducing the abundance of target species, the broader ecological consequences, particularly for plant biodiversity, remain poorly understood. This dissertation presents a remote sensing framework to evaluate changes in plant biodiversity following biocontrol interventions, using the case of saltcedar (Tamarix spp.) management in the southwestern United States.

High-resolution PlanetScope imagery was calibrated using fused MODIS and Harmonized Landsat Sentinel-2 (HLS) data to extract spring leaf phenological metrics at the individual tree level. The calibrated time series enabled the detection of budburst and leaf expansion, which proved effective for distinguishing species based on their unique seasonal growth patterns. Phenological traits derived from this approach provided valuable insights into species differentiation and functional diversity estimation.

To identify biocontrol-induced changes in vegetation, a Bidirectional Self-Attention Long Short-Term Memory (BiSA-LSTM) based change detection system was applied to Landsat time series. This model captured subtle and gradual changes in land cover associated with defoliation and vegetation regrowth, highlighting the complex and often prolonged ecological responses to biocontrol agents.

Spectral diversity was analyzed using a contrastive autoencoder-based model that clustered latent representations of phenological spectral signatures. This unsupervised method reduced the need for extensive field data and revealed increases in spectral heterogeneity following vegetation recovery, potentially indicating improved biodiversity conditions after biocontrol.

By combining phenological analysis, change detection, and spectral diversity estimation, this research advances remote sensing applications for ecological monitoring. The proposed framework offers scalable methods for detecting biodiversity responses across space and time and provides practical tools to support conservation efforts, guide biocontrol management, and inform restoration strategies in dynamic and data-limited environments.

link for robots only