Landslide inventory of Puerto Rico after the 2022 Hurricane Fiona

Key Points:
- Individual landslide locations (n=486) were identified as points during field reconnaissance conducted from 29th of October to 9th of November 2022, following Hurricane Fiona in September 2022 in Puerto Rico.
- A portion of the point inventory was mapped as polygons (n=210) by using pre- and post-event high-resolution satellite imagery. The remaining points were obscured in the imagery by either clouds or tree canopy cover.
- A mask layer for the Area of Interest (AOI) was created to account for cloud coverage obscuring landslides in the satellite imagery during the landslide mapping process.

Research Overview:
A landslide inventory was developed following the 2022 Hurricane Fiona, which affected Puerto Rico from September 18-20, 2022. Landslide locations were identified as point features during a field reconnaissance conducted approximately one month after the event (October 29 to November 9). Subsequently, a comprehensive inventory was compiled using pre- and post-event optical imagery (~2m resolution) from the WorldView-1 and WorldView-2 satellites (MAXAR, Inc.).

Methodology:
Field reconnaissance was conducted in Puerto Rico approximately one month after Hurricane Fiona (2022) to assess the hurricane's impact on the island. The primary objectives of the deployment were to evaluate the extent of damage to infrastructure and communities and to document environmental changes, including erosion and landslides. Priority areas were determined by: a) the hurricane footprint, b) initial satellite mapping of landslides, and c) reports from news and social media. Due to access limitations and the duration of the field campaign, a total of 111 km of road length was surveyed, primarily in the southern and southwestern portions of the island. Within these areas, 486 landslides were documented as point features. GPS-based point recording was subject to a location error of approximately 3 meters. However, due to a GPS malfunction during part of the field campaign, some landslides could not be geolocated. These point-based locations were included in a separate file named (file: ExcludedLandPts) with their respective bounding boxes (file: BoundingBox) based on their approximate location. Following the field reconnaissance, we relocated the points collected from the road to the center of the landslide (file: Slides_pts_v1). Landslides that were not located along the road network were included in a separate file (file:Fiona_Observer_pts_v1).
Additionally, an inventory was created by using pre- and post-event optical imagery from the Worldview-1 and Worldview-2 satellites (MAXAR, Inc) (Table 1) covering an area of 4,445 km2. The mapping process involved identifying landslide polygons using visual analysis techniques on very high-resolution optical imagery. The imagery had a resolution ranging from 0.5 m to 2 m. Landslides in imagery data were identified by the loss of vegetation and/or changes in albedo and/or texture, resulting in a polygon layer (combined source and runout zones) created in ArcGIS version 3.2. These features are associated with newly exposed bedrock, scarp development, and the accumulation of landslide debris, all of which compose the identified features of landslide signatures after an extreme event. A mask layer was also created to exclude areas that were covered by clouds within the Area of Interest (AOI), and thus, mapping detection was not possible. The cloud mask layer represents 30% (1,320 km2) of the total mapping area.