Legacies of the Vietnam War: the long-term effects of bombing and herbicide spraying on post-war land use/cover in Southeast Asia

Abstract

The Vietnam War left a deep and lasting legacy across Southeast Asia, where bombings, herbicide spraying, and other military activities disrupted ecosystems and livelihoods. During the conflict, over 7.5 million tons of bombs were dropped, and more than 74 million litres of herbicides were sprayed across Vietnam, Laos, and Cambodia, creating widespread environmental damage and contamination. Decades after the conflict ended, these impacts persist, with about 20% of the land still contaminated by unexploded ordnance (UXO) and the long-term health effects of dioxin exposure from herbicides still affecting thousands of people. Despite these ongoing challenges, incomplete and imprecise historical records on the locations of bombing and spraying activities hinder efforts to map the exact locations of remaining UXO and residual dioxin. At the same time the long-term consequences of this contamination on both the environment and local community livelihoods remains poorly understood.

The aim of this thesis was to determine the long-term impacts of bombing and herbicide spraying on post-war land use/cover changes (LUCC) in Vietnam and Laos. To do so I had to first understand which areas were affected by the bombing and spraying and if declassified military records were reliable sources. Additionally, to analyse post-war changes, I had to determine the initial land use/cover (LUC) at the time of the war. I therefore present my work in three results chapters about (i) detecting Vietnam War bomb craters using declassified U.S. spy satellite imagery (Chapter 2); (ii) mapping historical LUC in Vietnam and Laos using declassified military topographic maps (Chapter 3); and (iii) determining the long-term impacts of bombing and herbicide spraying on post-war LUC using panel regression models (Chapter 4).

In Chapter 2 I developed machine learning methods to automatically detect bomb craters in declassified very-high resolution (0.6-1.2 m) KH-9 satellite imagery taken during the Vietnam War. The models achieved an overall F1-Score of 0.61 and predicted more than 500,000 bomb craters across two study areas covering parts of Vietnam, Laos and Cambodia. I found that the detected craters were positively correlated with target locations available from declassified U.S. bombing records but provided more precise information on the impact locations of bombs.

In Chapter 3 I utilized topographic maps (1:50,000) created by the U.S. Army Map Service during the Vietnam War (1963-1973) to create detailed historical LUC maps of Laos and Vietnam. I compared multiple model architectures on the manually labelled training data, with the UNet++ achieving the best performance. The resulting maps, produced at 4 m and 30 m resolutions, include 10 LUC classes and achieved high overall accuracies of 98.8% for Laos and 98.6% for Vietnam on separate test sets. Analysis of the maps revealed forest cover losses of 18.2% in Laos and 25.0% in southern Vietnam (below 17°N) by 1990 and a 36.8% reduction of mangrove forests in Vietnam by 1996.

Finally, in Chapter 4, I combined declassified bombing and herbicide spraying records with the historical LUC maps created in Chapter 3 and existing LUC products derived from remote sensing data between 1990 and 2020. I aggregated the data across grid cells covering Laos, South Vietnam, and North Vietnam and used panel regression models to determine the long-term impacts of bombing and herbicide spraying on post-war forests and agricultural land, controlling for a comprehensive set of confounding variables informed by a directed acyclic graph. The results indicated that both bombing and herbicide spraying caused substantial losses of natural forests in the immediate post-war decades, with the strongest impacts in South Vietnam, where bombing was estimated to have reduced natural forest cover by as much as 25.8% by 1990. Over time, however, these negative effects declined, while the same exposures became increasingly associated with an expansion of plantation trees. In South Vietnam, the long-term effects of bombing resulted in an estimated 66.6% increase in the area of plantation trees between 1990-2020, with smaller effects in North Vietnam and Laos. Bombing also drove cropland expansion across all regions in the immediate post-war period, though these effects had largely disappeared by 2020. Together, these results indicated that wartime exposure to bombing and herbicide spraying was an important driver of Vietnam’s forest transition that started in the 1990s. This suggests that conflicts can reshape both forests and agricultural landscapes for many decades through long-lasting, path-dependent impacts.

With global conflict events almost doubling in the past five years, it is important to remember that the impacts of armed conflicts persist for decades after the fighting stops. My results indicate that bombing and herbicide spraying during the Vietnam War resulted in long-lasting, path-dependent, and regionally varying changes in land use/cover across Vietnam and Laos, including short-term losses of natural forests, short-term expansion of non-rice croplands, and long-term expansion of plantation trees. Moreover, this thesis demonstrates how historical data, combined with advanced machine learning techniques, can support evidence-driven mine action and guide more effective post-conflict interventions that address environmental legacies. Implementing such data- and evidence-driven approaches should therefore be an urgent priority to reduce the long-term negative impacts of past and current conflicts, both in Southeast Asia and beyond.