New deep-learning map reveals scale of forest and mangrove loss in post-war Vietnam and Laos, and offers a template for future land-use studies

The research, led by Dr Philipp Barthelme in the School of GeoSciences, leverages advanced deep-learning image segmentation to translate topographic map sheets into digital land-use maps at 4 m and 30 m pixel resolutions. The resulting maps distinguish 10 different land-use classes and deliver extremely high accuracy: 98.8% for Laos and 98.6% for Vietnam. 

By comparing these historical reconstructions with more recent land-cover data, the team uncovered dramatic changes since the mid-20th century.

Forest cover has declined significantly across both countries over the past half-century. In Laos, it fell by around 18.2%, while in southern Vietnam forest loss reached 25% by 1990. Vietnam’s mangrove ecosystems were especially affected, shrinking by 36.8% by 1996.

Figure 5. (a) Change in forest cover per map sheet between topographic maps (1963–1973) and GLC_FCS30D data from 1990. (b) Main transition for map sheets with aggregate forest loss, for example shrubland here indicates that the most common pixel-to-pixel transition from forest in the baseline topographic map sheet to non-forest in the 1990 GLC_FCS30D data across the full map sheet area was from forest to shrubland. Map sheets that did not differentiate between forest and brushwood were excluded.

Because the maps are spatially explicit and includes precise geographical location information, the team could not only quantify change over time, but pinpoint where it occurred, distinguishing between northern and southern regions, and between uplands and lowlands. This revealed patterns that overall national averages alone would have obscured.

The results show that land-use transformations differed markedly across the two countries. In the north of both Laos and Vietnam, areas that were once forested largely gave way to shrubland. Further south, however, particularly in Savannakhet Province in Laos and across the Southeast of Vietnam, forest loss was more commonly followed by the expansion of cropland.

The findings offer a much clearer picture of how conflict, development and shifting land-use policies have reshaped ecosystems across Southeast Asia over the last half-century. By creating a reliable baseline of past landscapes, the research gives environmental scientists, conservation organisations and policy makers a powerful new tool for estimating long-term changes such as carbon loss or impacts on biodiversity, work that has often been limited by gaps in historical data.

This research not only supplies a critical, spatially explicit baseline for assessing post-war land-use change in Southeast Asia, but also establishes a scalable and transferable methodology which can be used on a global scale.

The study demonstrates the potential of archived paper maps, with the team’s approach showing that with modern deep-learning techniques, these resources can be digitised and analysed at scale to recover historical land-use information. This opens the door for similar reconstructions in other regions, offering a blueprint for global-scale studies of environmental change.

The authors suggest that historical map archives – an often overlooked source of environmental knowledge – hold enormous potential for understanding how landscapes have changed over time. When this rich information is unlocked with modern deep-learning techniques, these collections could transform the way researchers study long-term land-use patterns and environmental change, offering insights into how ecosystems have changed and adapted over decades.

View full paper and read more about the methods and results:

Large-scale historical land use mapping in Vietnam and Laos using military topographic maps. Philipp Barthelme*, Eoghan Darbyshire, Dominick V Spracklen and Gary R Watmough. 2025 Environ. Res. Lett. 20 124014 https://iopscience.iop.org/article/10.1088/1748-9326/ae1bbe

Investigating UXO in South East Asia – an exciting trip for Philipp | Blog post