Time-series analysis of synthesized active fire imagery from multiple satellite sensors in Google Earth Engine
Abstract
The 2017 fire season was the largest on record in total area burned for British Columbia (BC), Canada. It is important to accurately map these fires to monitor within-year fire progression and to quantify resulting forest-disturbance impacts. Single-date imagery of active fires, however, often contain clouds, smoke, and haze that inevitably create inconsistencies in a temporal classification stack. As a result, many fire maps from satellites are created after a fire is extinguished and show the ultimate extent of burned area. Recent developments suggest that it is possible to synthesize single-date classifications in a coherent time series, even if they contain smoke, clouds and haze. To address these mapping and monitoring challenges, we developed highly automated methods for synthesizing burned-areas classifications from multiple sources to map the hundreds of wildfires throughout their active phase for the 2017 fire season in BC. The advanced storage and processing capabilities of Google Earth Engine support near-term and retrospective disturbance detection across large areas using imagery from multiple sensors. We applied the Bayesian Updating of Land Cover Classifications (BULC) algorithm to merge sensor-specific burned-area classifications from a range of sources including Landsat-7 and -8, Sentinel-2, and MODIS (MCD64A1 burned-area dataset). BULC synthesizes the time-series of classifications by leveraging the prior knowledge of a pixel's state to inform the estimate of the burned area for each fire. BULC was used to merge single-date burned-area classifications in Google Earth Engine, producing a synthesized time-series stack with updated weekly burned areas for the 2017 fire season in BC. We classified fire growths using burn patterns, rates and forest characteristics for each fire in the 2017 fire season in BC. We used this novel time-series from BULC to explore the benefits of multi-source remote sensing for informing near-term active-fire analyses. Fire progression maps through time contribute to greater understanding of potential drivers of fire spread. Future research can apply this approach to create temporally dense fire-classification stacks to analyze active fires retrospectively and in near-real time.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFM.B24B..05C
- Keywords:
-
- 0414 Biogeochemical cycles;
- processes;
- and modeling;
- BIOGEOSCIENCESDE: 0439 Ecosystems;
- structure and dynamics;
- BIOGEOSCIENCESDE: 1640 Remote sensing;
- GLOBAL CHANGEDE: 1817 Extreme events;
- HYDROLOGY