Constructing a large-scale landslide database across heterogeneous environments using task-specific model updates

Savinay Nagendra, Daniel Kifer, Chaopeng Shen

Abstract:

Recent small-scale studies for pixel-wise labeling of
potential landslide areas in remotely-sensed images using deep
learning (DL) showed potential but were based on data from very
small, homogeneous regions with unproven model transferability.
In this paper we consider a more realistic and practical setting for
large-scale heterogeneous landslide data collection and DL-based
labeling. In this setting, remotely sensed images are collected sequentially in temporal batches, where each batch focuses on images
from a particular ecoregion, but different batches can focus on
different ecoregions with distinct landscape characteristics. For
such a scenario, we study the following questions: (1) How well
do DL models trained in homogeneous regions perform when they
are transferred to different ecoregions? (2) Does increasing the
spatial coverage in the data improve model performance in a given
ecoregion? and (3) Can a landslide pixel labeling model be incrementally updated with new data, but without access to the old data
and without losing performance on the old data? We address these
questions by developing a mechanism for incremental training
of semantic segmentation models. We call the resulting extension
task-specific model updates (TSMU). A national compilation of
landslide inventories by the U.S. Geological Survey (USGS) was
used to develop a global database for this study. Our results indicate
that the TSMU framework can be used to aid in the creation of
new landslide inventories or expanding existing datasets, and also
to rapidly develop hazard maps for situational awareness following
a widespread landslide event.