The variants of BULC : Bayesian Updating of Land Cover
Figure 8 from Lee et al., 2020.
Land-cover maps through time of the town of Querência, located in sector C2 of the study area (see Figure 4 and Figure 5). Top left: BULC-U 1986. Top right: Landsat 5 1986. Middle left: 1995 BULC-U. Middle left: Landsat 5 1995. Bottom left: BULC-U 2000. Bottom right: Landsat 5 2000.
|
Monitoring land cover change is challenging because satellite data come from many sensors with different resolutions, noise levels (i.e. cloudy regions), and revisit times (especially early years), making consistent, time-series analysis difficult. We developed the BULC framework, a Bayesian data fusion approach that integrates classifications from any sensor into a single, low-noise time series.
Its variants (BULC-D and BULC-U) enable rapid disturbance detection and large-scale mapping even with limited labeled data. This approach improves the accuracy, timeliness, and scalability of land-cover monitoring for environmental science, hazard response, and land-use policy. Click 'Learn More' for more information about a series of projects relating to BULC and its variants. |
Synthetic aperture radar (SAR) detection
Figure 4 from Flores-Anderson et al., 2025. Before and After of an Early deforestation event as seen by HLS Sentinel-2 images (Masek et al., 2021) in false color combination (Short-Wave Infrared (SWIR)/Near Infrared (NIR)/Green) and the corresponding detection results including date of change for maps.
|
Early tropical deforestation and agricultural expansion are difficult to detect because clouds obscure optical imagery and short-wavelength radar struggles to sense subtle structural change beneath forest canopies. We showed that integrating L-band SAR into a Bayesian, multi-sensor framework enables reliable detection of early logging within weeks, while combining L- and C-band SAR with optical data supports consistent national-scale cropland tracking in data-sparse regions.
This work enables earlier intervention against illegal deforestation and improves annual land-use monitoring to support conservation and land-management policy. Click 'Learn More' below for additional information on a series of projects relating to SAR detection in a variety of ways. |
Other projects
Deep learning for early fire mapping
Figure 2 from Boothman & Cardille, 2022. Results Overview
|
New techniques for old fires: Using deep learning to augment fire maps from the early satellite era
In this paper, we aimed to address a major gap in land-cover science: the lack of reliable land-cover and fire maps for the 1970s due to sparse, low-information satellite imagery. Large forested regions of Canada, including Quebec, remain poorly documented prior to the modern satellite era. We developed deep learning and Bayesian methods that recover fire scars and land-cover patterns from early Landsat MSS imagery, extending consistent mapping across Canada back more than a decade.
This work provides critical historical baselines for wildfire regimes, carbon accounting, and long-term ecosystem change. |
Multi-sensor change detection for forest disturbance
Figure 1 from Cardille et al., 2022. False Colour Infrared from 2017 of the study area encompassing the Elephant Hill fire and forest harvest in British Columbia, Canada. The highlighted areas indicate the four disturbance outcomes relevant for change capture. 1) Winter Harvest 2) Summer Harvest 3) Fire (early season) and 4) Fire (late season).
|
Multi-sensor change detection for within-year capture and labelling of forest disturbance
In this paper, our research addresses a critical gap in wildfire monitoring: the trade-off between timely fire information and the spatial detail needed for effective emergency response and planning. Existing products are either fast but coarse or accurate but delayed. We developed a multi-sensor system that fuses Landsat and Sentinel-2 data to map fire growth at 30 m resolution in near-real time.
This approach delivers accurate, low-latency fire maps that directly support hazard response, management decisions, and national burned-area reporting. |
