Flood early warning systems depend on decisions made before a flood arrives. Where should gauges be placed? Which villages need earlier alerts? Where should evacuation routes, shelters, communications and preparedness planning be prioritised?
Those decisions are difficult when planners only have broad flood information. A hazard map can show where flooding may occur. Early warning system planning also needs to understand where buildings, villages and vulnerable communities are exposed.
A flood susceptibility, exposure, vulnerability and risk assessment was developed across nine provinces of Lao PDR. This study was produced by Geoneon under the ACER-SEA project, Addressing Climate and El Niño-Related Risks in Southeast Asia, for the Swiss Agency for Development and Cooperation (SDC) and People in Need (PIN). The work also drew on ADPC’s village-level flood vulnerability data, which was used to connect modelled flood exposure with community vulnerability and risk.
Across the nine Lao PDR provinces of Attapeu, Bolikhamxai, Champasak, Khammouan, Louangphabang, Salavan, Savannakhet, Xaignabouly and Xekong, this study analysed more than 2.75 million buildings. It was found that 310,403 buildings, or 11.3% of the assessed building stock, fell into moderate to extreme flood exposure categories, equivalent to modelled flood depths above 0.5 m. At village level, the concentration was even sharper: 30 villages had at least 90% of buildings in those moderate to extreme exposure categories.
The aim was not to forecast a specific flood event. It was to create a screening-level evidence base that can help identify where flood exposure is concentrated and where early warning and preparedness investment should be prioritised.
Flooding is one of the most severe and recurrent natural hazards in Lao PDR. Monsoonal rainfall, extensive river systems, floodplains and rapid land-use change all contribute to flood risk across the country.
For early warning system planning, the challenge is not only understanding where water may go. It is understanding where flooding would affect people, buildings and communities most severely.
This distinction matters because flood susceptibility alone does not show where action is most urgent. A floodplain may experience modelled inundation, but if few buildings or communities are exposed, it may be a lower early warning priority than a village where many buildings intersect with flood depths above 0.5 m.
That is why this assessment connects modelled flood behaviour to building footprints, village boundaries and vulnerability data: to identify not only where flooding is plausible, but where it is most likely to affect people and preparedness planning.
This work moves from flood modelling into early warning prioritisation by connecting four related concepts.
Flood susceptibility shows where flooding is plausible under a severe rainfall scenario.
Exposure shows which buildings and villages intersect with modelled flood depths.
Vulnerability identifies where communities may be more affected by flooding.
Risk combines exposure and vulnerability into a single prioritisation layer.
Fig 1: Workflow overview chart for calculation of Exposure and Risk Indices and extraction of Vulnerability Index.
Together, these layers provide a more useful basis for early warning system planning than a flood map alone.
A susceptibility surface can identify where water may accumulate. A building-level exposure layer shows which structures are affected. A village-level vulnerability index adds social context. A combined risk index helps identify where preparedness, mitigation and response planning should be prioritised.
The assessment produced a 30 m flood susceptibility map across the nine target provinces, then used building footprints and village boundaries to calculate exposure at multiple scales.
More than 2.75 million buildings were assessed. Each building was assigned a modelled exposure height based on the mean flood depth within its footprint, then classified into an exposure index from negligible to extreme.
This provides a level of spatial detail that is directly useful for preparedness planning. Instead of only identifying broad flood-prone areas, the workflow identifies where exposed buildings are concentrated and how those patterns aggregate to villages, districts and provinces.
The assessment also incorporated ADPC’s village-level vulnerability index. This allowed exposure and vulnerability to be combined into a 0–6 risk index, supporting more targeted prioritisation.
The flood simulation used SynxFlow, an open-source, GPU-accelerated hydrodynamic model. SynxFlow solves the full two-dimensional shallow water equations, making it possible to simulate how water moves across terrain, river channels and floodplains under severe rainfall conditions.
In practical terms, the model asks: if a severe rainfall loading is applied across the landscape, where does water flow, where does it accumulate, and how deep could it become? To answer that question, the model uses several types of spatial data:
Terrain data defines the shape of the land.
Water polygons help represent rivers, lakes and reservoirs.
Land cover data informs surface roughness, which affects how quickly water moves.
Rainfall data provides the synthetic event used to drive the simulation.
The main inputs included FABDEM 30 m terrain data, Overture water polygons, Sentinel-2 land cover, CHIRPS rainfall data, Overture building footprints and ADPC vulnerability data.
The Mekong and Xe Khong watersheds were modelled separately. This reduced computational cost and allowed the smaller Xe Khong domain to be used for method iteration before scaling the approach across the broader assessment area.
Fig 2: Map of adjusted Mekong watershed (grey), Xe Khong watershed (blue) and Lao PDR's provinces (red).
The primary model output was a flood susceptibility map representing the maximum modelled water height reached during the simulation.
Fig 3: Flood susceptibility map.
This flood-depth surface was then intersected with building footprints. For each building, the mean flood height within the footprint was calculated. That value became the building’s exposure height.
The exposure height was then converted into a 0–6 exposure index:
negligible exposure for depths up to 0.01 m,
very low exposure for depths above 0.01 m and up to 0.2 m,
low exposure for depths above 0.2 m and up to 0.5 m,
moderate exposure for depths above 0.5 m and up to 1 m,
high exposure for depths above 1 m and up to 2 m,
very high exposure for depths above 2 m and up to 4 m, and
extreme exposure for depths above 4 m.
Fig 4: Building exposure.
This building-level information was then aggregated to villages by calculating the mean exposure index for buildings within each village boundary.
Fig 5: Building exposure per village.
Vulnerability was assigned using ADPC’s village-level vulnerability index, which ranges from 1 to 5. Each building received the vulnerability value of the village in which it was located. Exposure and vulnerability were then normalised, averaged and converted into a 0–6 risk index.
The assessment found that 310,403 buildings, or 11.3% of the assessed building stock, fall into moderate to extreme flood exposure categories (representing modelled flood depths above 0.5 m).
The distribution of exposure was not even.
At the province level, Champasak recorded the highest number of buildings in moderate to extreme exposure categories, followed by Attapeu, Louangphabang and Xaignabouly.
Fig 6: Number of buildings by Flood Exposure Index categories 3-6 by province.
In relative terms, Salavan and Savannakhet had the highest proportions of buildings within their provinces classified in these higher exposure categories.
Fig 7: Percentage of buildings by Flood Exposure Index categories 3-6 by province.
At the district level, Sanamxay, Luangprabang and Samakkhixay recorded the highest numbers of exposed buildings. In relative terms, Lamarm, Khounkham and Ta Oi stood out, with large shares of buildings falling into moderate to extreme exposure categories.
At the village level, exposure became even more concentrated. The village of Mitsumphun in Attapeu had the highest number of buildings in moderate to extreme exposure categories. Seven of the ten villages with the highest exposed building counts were located in Attapeu Province.
Fig 8: Flood exposure at the village level.
The assessment also identified 30 villages where at least 90% of buildings were classified in moderate to extreme exposure categories.
This village-level concentration is important for early warning system planning because it shows where preparedness measures may need to be highly targeted.
Exposure identifies where buildings intersect with modelled flood depths. Vulnerability adds the human context needed for prioritisation.
A village with high flood exposure is important. A village with high flood exposure and high vulnerability is a stronger priority for preparedness, early warning and response planning.
Fig 9: Flood susceptibility and building exposure.
By combining exposure and vulnerability into a single risk index, the assessment provides a way to identify where flood impacts may be most severe.
This does not replace local knowledge or detailed site assessment. Instead, it creates a screening-level evidence base for identifying where further attention, data collection and preparedness work should begin.
The model was checked against the 2019 Xe Khong flood event using satellite-derived flood extents.
For this verification run, the model was forced with daily CHIRPS rainfall data from 28 August to 10 September 2019. The simulated maximum inundation was then compared with UNOSAT Sentinel-1 flood extent data for the event.
The comparison showed broad spatial agreement between the modelled inundation and the satellite-derived flood extent. However, the model generally overestimated the extent of inundation.
Because the satellite-derived flood map had not been validated in the field, a formal accuracy assessment was not carried out. A pixel-by-pixel comparison could have mixed model error with remote-sensing detection limitations.
Instead, the verification provides a useful sense-check. It shows that the model reproduced broad flood patterns, while also highlighting the need to interpret local modelled extents carefully.
The practical value of the assessment is that it connects flood susceptibility to decisions about early warning system design.
The outputs can help identify where to act first, including provinces, districts and villages with large numbers or high proportions of exposed and vulnerable buildings.
They can also support sensor siting. Flood susceptibility maps can be used to identify upstream reaches that drain into exposed building clusters, helping guide where river gauges may be most useful. Where exposed clusters span multiple tributaries, the outputs can support planning for redundant sensors and backup communications.
The exposure bands also provide a potential basis for tiered alert thresholds. Depth categories such as above 0.5 m, above 1 m, above 2 m and above 4 m can help structure warning levels, community messaging and activation triggers.
For preparedness planning, the outputs can help prioritise evacuation routes, shelters, critical facilities, and drills in the identified hotspots.
The main capability upgrade is moving from flood susceptibility maps to a practical flood risk planning layer.
The assessment is designed as a screening-level stress test for prioritisation and early warning system planning.
It is not a real-time flood forecast. It is not a regulatory zoning product. It is not a substitute for higher-resolution local modelling, engineering design or field survey.
Several limitations are important.
The model uses 30 m FABDEM terrain data. While appropriate for large-scale modelling, this resolution can overestimate flood extent near rivers, especially where actual channels are narrower than the grid.
The modelling also relies on Overture water polygons to represent channels and water bodies. This helps capture rivers that may not be fully represented in the terrain data, but it can simplify channel geometry and may artificially increase exposure estimates near narrow streams.
The rainfall forcing is based on a synthetic 99th-percentile CHIRPS rainfall event applied over four days. This is suitable for screening-level susceptibility, but it is not a probabilistic return-period analysis and does not include future climate projections.
Local daily extremes can also be higher than the selected rainfall forcing, meaning local flooding and flash-flooding potential in small headwater catchments may be underestimated.
These limitations do not remove the value of the assessment. They define how it should be used: as a prioritisation layer for early warning system planning, and as a foundation for further refinement where higher-resolution terrain, local surveys and validation data become available.
The strongest value of this work is not only that it maps flood susceptibility. It connects modelled flood behaviour to buildings, villages and vulnerable communities.
That connection makes the outputs useful for planning. It helps show where flood exposure is concentrated, where vulnerable communities may be affected, and where early warning and preparedness investments should be prioritised.
For Lao PDR’s nine assessed provinces, the result is a first building-level, near-national-scale flood susceptibility and risk evidence base that can support disaster risk reduction, adaptation planning and early warning system development.
As better elevation data, local validation, gauge observations and updated vulnerability information become available, the assessment can be refined further. But even at screening level, it provides a practical starting point for more spatially targeted flood preparedness.