Heavy rain in late July 2024 brought flooding to areas along North Korea’s Amnok River, also known as the Yalu. Satellite analyses document inundation around the lower river, including parts of the Sinuiju and Uiju area. A later research study used satellite-based rainfall and terrain data to simulate how floodwater moved.
The two records describe the event in different ways. A flood image shows what the sensor detected at an observation time or over a stated group of days. A simulation estimates water movement from inputs and assumptions. Reading the dates and methods together explains why their figures can differ.
- Sinuiju and Uiju are in the lower Amnok region near the border with China.
- The USDA report includes a 30 July to 3 August satellite flood composite.
- A modelled maximum and the water detected during a later satellite pass are different measurements.
USDA Foreign Agricultural Service — August 2024 flood and crop report
Choose a reading route
Place the event on a river map · Worked example: read three reports together · Frequently asked questions
Chapters 1–4 · Read the evidence
Chapters 5–7 · Test a comparison
Chapter 1 of 7
Place the event on a river map
Find the Amnok near Sinuiju and follow the channel towards the coast. Look at the low ground around the river and the position of Uiju upstream from the lower city reach. This locates the event before you start interpreting colours on a flood map. A map can show how areas relate to the channel without telling you how deep the water became.
High river flow can spread onto a floodplain when the channel cannot contain it. The spread depends on local topography and on conditions along the river. Water level and river discharge answer different questions. Water level describes the elevation of the surface at a location. Discharge measures how much water passes a point per unit of time.
Keep the date attached to the map. A map of the normal channel gives the baseline. An image from the flooding period helps show where water was detected beyond that baseline. A later image may record drainage, standing water or altered ground. Each provides evidence for a different stage.
UN Geospatial — North Korea country map
Chapter 2 of 7
What a five-day composite can show
The USDA Foreign Agricultural Service report used a five-day VIIRS flood composite covering 30 July to 3 August. It paired that information with a land-cover crop mask. A composite draws on multiple observations in a period. Read the legend and methods to see how those observations were combined before assigning a specific date to a patch on the map.
The flood product and the crop mask have different spatial resolutions. The report describes a 375-metre flood product and a 10-metre crop mask. Overlaying a finer crop layer does not give the flood observation the same fine resolution. A mapped overlap can help estimate exposure in the study area. Its precision still depends on the coarser input and the classification method.
The report also shows comparisons at a sample location using finer imagery for July and August windows. That gives readers a second view at a smaller scale. Check the dates, the area shown and whether clouds obscure the view before generalising the visible pattern.
USDA Foreign Agricultural Service — August 2024 flood and crop report
Contents · Previous chapter · Next chapter
Chapter 3 of 7
What the later model estimated
The 2025 paper by Joo-hun Kim and Yun-seok Choi used satellite-derived precipitation and global terrain data. It used separate modelling tools for flood discharge and inundation. The method helps answer when a peak may have occurred and where water could have spread given the inputs and assumptions.
The paper estimated that approximately 67% of the area it identifies as Sinuiju-si was inundated in its simulation. That figure belongs to the study’s model, geographic boundary and simulated maximum. It should not be presented as a direct census of households, an exact count of buildings, or a number for all of North Korea.
To interpret the percentage, first locate the study area and time step. Then check what counted as inundation. A different threshold or boundary can change the area reported. The same study may give a peak river flow at one point and a maximum extent later, because water takes time to spread.
Kim and Choi — 2025 flood modelling study and methods
Contents · Previous chapter · Next chapter
Chapter 4 of 7
Why flooded cropland and lost harvest are different measures
A crop mask shows land classified as cropland in the source dataset. A flood overlay identifies where that class and detected water overlap. It can suggest where crops were exposed. The amount lost depends on the crop, growth stage, duration of flooding and what happened afterwards.
An August image can help a reader observe how an area looked after the first flooding. Ground surveys or later production evidence can answer questions about recovery and output. The water map gives a physical signal. A harvest figure measures a different outcome.
USDA Foreign Agricultural Service — August 2024 flood and crop report
Contents · Previous chapter · Next chapter
Chapter 5 of 7
Worked example: read three reports together
Imagine a river chart gives a high water level, a satellite map shows flooded fields and a paper gives a modelled maximum extent. Label each with its date and location. Then put one sentence beside it saying what was measured or estimated. A water level is a point observation. A mapped flood extent is an area. A model output depends on its inputs and assumptions.
Write a short conclusion only after you have labelled the three. For example, you can describe the visible inundation and then say the model estimated a maximum for a named study area. If you are asked how many people were affected, seek a source that counts or estimates people and explains how it does so.
Kim and Choi — 2025 flood modelling study and methods
Contents · Previous chapter · Next chapter
Chapter 6 of 7
Use uncertainty as part of the map legend
In an optical image, cloud cover can hide the ground. A classification method can also confuse some dark or reflective surfaces with water if the surrounding evidence is weak. A good map legend or methodology explains how the analysts treated unusable pixels and how they recognised likely water. This matters as much as the colour chosen to display a flood.
Try drawing a simple map with three classes. Mark one area as water detected, one as no water detected under the method and one as unavailable. The last two classes answer different questions. A place hidden from the sensor during an observation may need another image. A place with a usable observation can still need field evidence if the question asks about depth or damage.
Use dates in the legend as well. A five-day composite can reveal water present across a window that a single pass would miss. It can also group observations taken at different stages of the flood. If you are trying to locate a peak, ask whether the product was designed to report the maximum for each pixel, a particular time or another summary of the observations.
USDA Foreign Agricultural Service — August 2024 flood and crop report
Contents · Previous chapter · Next chapter
Chapter 7 of 7
Check the boundary before comparing two totals
A flood number can refer to all mapped land, only mapped cropland or a particular city boundary. Write the geographic unit beside the number before comparing it with another. The USDA maps cover a wider area than a single city. The modelling paper reports a result for its own Sinuiju-si study area. A percentage for one cannot be applied to the other.
A useful example is the difference between 10 hectares of cropland exposed and 10 hectares of total area flooded. The first adds a land-use category. If both figures refer to the same time and map, the cropland could be part of the total. If the periods or boundaries differ, that relationship needs to be checked.
A short reconstruction of the event can keep these distinctions visible. Begin with the dated rainfall and geographic setting. Add what the satellite products show in their observation windows. End with the model result and its stated assumptions. A separate section can discuss damage, people or recovery only where you have sources that actually measure those outcomes.
This approach also explains why two published maps may both be useful even when they colour different areas. One may use a larger extent and coarser pixels. Another may focus on a smaller sample location with finer imagery. Read each as an answer to the question it was built to address.
Kim and Choi — 2025 flood modelling study and methods
Contents · Previous chapter · Questions
Frequently asked questions
Are Amnok and Yalu two rivers?
They are names for the same river.
Why can satellite and model estimates differ?
They may cover different dates and areas and use different definitions of inundation.
Does a flooded-cropland map tell you the final crop loss?
It shows estimated exposure under the map’s method. A crop-loss estimate needs additional evidence.
Sources and next reading
Sources
- USDA Foreign Agricultural Service — August 2024 flood and crop report
- Kim and Choi — 2025 flood modelling study and methods
- USGS — water levels, discharge and floodplains
- UN Geospatial — North Korea country map