Mapping tsunami deposits through a classification model based on satellite images

Abstract

Assessing tsunami risk requires knowledge of the potential inundation area, which can be inferred from the spatial distribution of tsunami deposits. However, field surveys of tsunami deposits are time-consuming and occasionally pose challenges, such as disturbance of sedimentary evidence by human and natural causes. Here, we propose a novel technique capable of mapping tsunami deposits using remote sensing, which was tested along a coastal stretch of central Chile following the tsunami of 27 February 2010. We trained a classification model using high-resolution satellite images from before (September 2004 and January 2005) and after (April 2010) the 2010 tsunami to map the sand deposit, yielding an overall accuracy of about 86%. Our satellite mapping of the deposit was validated with field observations in pits and eyewitness interviews conducted about a decade after the tsunami. The field data matched the model predictions by 88%. Likewise, our satellite mapping was also contrasted with the inundation area reported by previous post-tsunami surveys. The spatial distribution of the tsunami sand deposit inferred from our model reproduces a minimum inundation area, which was almost as extensive as the actual inundation area. Sand inundation ranged from 50 to 600 m inland, matching about 90% of water inundation. Both sand and water inundation were controlled by the land slope. Application of our technique to a satellite image from 11 years after the tsunami (May 2021) shows that the detection ability of the sand deposit was lost by about 86%, which is attributed to human intervention and masking by new soil development. Our results suggest that extensive tsunami deposits can be accurately mapped by a supervised classification model in a lesser time than that employed in field surveys.