Journal: Communications Earth & Environment

Article Title: High-resolution imagery and neural networks link post-tsunami land cover changes to population health and well-being

doi: 10.1038/s43247-026-03396-0

Figure Lengend Snippet: Map of the area of Aceh Province, Indonesia, for which high resolution imagery is available for four time points (June 2004, late December 2004, July 2007, February 2009). The area is 289 km 2 and contains 164 administrative areas (villages or municipalities, outlined in black). Zones are designated by proximity to the shore (red is within 4.3 km, orange is 4.3–8.3 km, yellow is 8.3 km or more). The map inset shows the area in relation to the northern end of the island of Sumatra. Sources: Esri, Tom Tom, Garmin, FAQ, NOAA, USGS, © OpenStreetmap contributors, and the GIS User Community.

Article Snippet: To measure contextual changes caused by the tsunami and by subsequent reconstruction we implemented a variation of the prominent DeepLabV3+ convolutional neural network (CNN) trained to produce pixel-level segmentations from high-resolution Quickbird satellite imagery that we obtained through a digital imagery grant from MAXAR.

Techniques: Northern Blot

Journal: Communications Earth & Environment

Article Title: High-resolution imagery and neural networks link post-tsunami land cover changes to population health and well-being

doi: 10.1038/s43247-026-03396-0

Figure Lengend Snippet: A Quickbird imagery for a neighborhood in Banda Aceh, Indonesia, before and at multiple time points after the tsunami. B segmentations produced by our CNN. Classifications are other (grey), agriculture (dark green), water (dark blue), rubble (brown), foundation (orange), beach (yellow), cloud (white), road (light blue) and building (magenta). Destruction is visible in the images from December 2004 and August 2005. Over time, structures are rebuilt. Satellite images courtesy of the DigitalGlobe Foundation (now MAXAR).

Article Snippet: To measure contextual changes caused by the tsunami and by subsequent reconstruction we implemented a variation of the prominent DeepLabV3+ convolutional neural network (CNN) trained to produce pixel-level segmentations from high-resolution Quickbird satellite imagery that we obtained through a digital imagery grant from MAXAR.

Techniques: Produced

Journal: Communications Earth & Environment

Article Title: High-resolution imagery and neural networks link post-tsunami land cover changes to population health and well-being

doi: 10.1038/s43247-026-03396-0

Figure Lengend Snippet: A Quickbird imagery for the city of Banda Aceh, Indonesia, before and at multiple time points after the tsunami. B displays the segmentations produced by our CNN. Classifications are other (grey), agriculture (dark green), water (dark blue), rubble (brown), foundation (orange), beach (yellow), cloud (white), road (light blue), and building (magenta). Destruction is visible in the images from December 2004 and August 2005. Clouds obscure much of the image from 2008. Over time, structures are rebuilt and increase as a share of landcover. Satellite images courtesy of the Digital Globe Foundation (now MAXAR).

Article Snippet: To measure contextual changes caused by the tsunami and by subsequent reconstruction we implemented a variation of the prominent DeepLabV3+ convolutional neural network (CNN) trained to produce pixel-level segmentations from high-resolution Quickbird satellite imagery that we obtained through a digital imagery grant from MAXAR.

Techniques: Produced

Journal: Communications Earth & Environment

Article Title: High-resolution imagery and neural networks link post-tsunami land cover changes to population health and well-being

doi: 10.1038/s43247-026-03396-0

Figure Lengend Snippet: Flowchart of individuals who were eligible to participate in the 2005 and 2009 STAR surveys because they were members of households located (in 2004) in one of the 43 survey enumeration areas for which high resolution satellite imagery is available of four points in time. Of the 2,947 eligible individuals 610 died in the tsunami. Others could not be interviewed after the tsunami or died between the 2005 and 2009 interview.

Article Snippet: To measure contextual changes caused by the tsunami and by subsequent reconstruction we implemented a variation of the prominent DeepLabV3+ convolutional neural network (CNN) trained to produce pixel-level segmentations from high-resolution Quickbird satellite imagery that we obtained through a digital imagery grant from MAXAR.

Techniques: