Flood influence analysis using satellite imagery

The flood that occurred on July 15th, 2021, was a serious disaster whose impact reached lands beyond the country. On July 14th, the Royal Netherlands Meteorological Institute constituted a great risk of the Limburg province being overflown (a.k.a. red warning) due to heavy rain. However, right before the rain actually started, they gave a less pessimistic estimation of the danger.

The flood had the 5th level of danger. It hit the province the same day, causing electricity outages due to equipment failure. 100 millimeters (3.9 in) in a day and 200 millimeters (7.9 in) over three days of precipitation has fallen over this period. Rescued citizens reported a lack of even necessary provisions such as food supplies.

Figure 1. Visualization of flooded areas in the investigated region

The Copernicus Emergency Management Service (CEMS) has published a short report examining the areas suffering from heavy rains. The river Maas has received an enormous amount of water and on top of the heavy rain caused river levels to rise downstream. CEMS also made available the geospatial information regarding this event. We can observe in blue on Figure 1 flooded areas. Other shapes on the picture are occupied lands by local town infrastructures and buildings.

As of 21th of July, the inhabitants of the area were still subjects of the evacuation order, and the consequences of the disaster were not removed. In the aftermath of this, more than 2000 houses were considered uninhabitable. Even more disastrous, it turned out to be for Germany and Belgium.

In the image, you can see one of the cities and several fields that suffered from flooding.

Picture 2. City and fields are suffering from flooding.

There is little to be done in the precise aftermath of the disaster from a satellite image processing perspective but study a peculiar climate phenomenon. There is almost constant cloud cover overcast for nearly two months after the event and nearly a month before. Such a phenomenon ought to be studied by disaster response projects and climate scientists.

Effects on the agricultural sector

According to the EU farmers’ association COPA-COGECA, this event disrupted the agricultural sector to such a degree that harvest outage was expected, the most severely affected being still Germany and Belgium. The flood caused major soil erosion. Fruit and vegetable plants were practically considered to be lost.

Besides open fields, Gartenbau-Versicherung, a German horticulture company, reports many greenhouses were flooded over 3 meters with water, destroying any hope for at least some harvesting.

“It’s too early to say [for Europe], but what is grown in the province of Limburg is lost. The main production area at this time of the year is the German Pfalz area. How big the losses are, until now, there is no indication. There will be a problem for sure, but how big, nobody knows.”, reports Nico Veldhoen, commercial director at Staay-Hispa Papendrecht, an influential fruit and vegetable agroholding of Netherlands.

According to the World Meteorological Organization, there are some benefits of flooding on ecosystems in general, and, depending on the specific soil composition, it can increase soil fertility and enhance positive biodiversity. However, each area’s specifics have to be taken into account.


We decided to perform the analysis of flooded areas using SoilMate. As the research matter regarding recent European flood proceeds plot boundary detection model has been applied to the flooded area in the Netherlands.

We started our research by detecting flooded parts inside the provided area of interest. We used Sentinel-1 to detect floods. Unfortunately, no satellite imagery data covered all investigation areas. We used two captures: the 12th of July (three days before the flood) and the 18th of July (three days after).

To analyze the influence of floods on crops, we used Sentinel-2 data.

It is worth emphasizing that the closest dates before and after the disaster were quite far in time; that is Sentinel-2 tiles with little cloud percentage found are of 1st of June and 25th of August, while heavy rain occurred on 15th of July. However, there was a semi-cloudy image on the 21st of July.

A few concerns have to be pointed out as well:

  1. As usually is the case, fields of European countries are considerably small in area; when visualized, each field is highly pixelated
  2. Some fields were almost unaffected by the flood. For others, the actual reason and nature of anomalies can be hypothesized due to the flood; there is still uncertainty.

In Figure 3, we can see the TCI (True Color Image) of the investigated area for 2 dates: the 1st of June (the date before the flood) and the 25th of August (the date after a flood). Rasters images are located in 31UGS tile of the Sentinel-2 satellite. The total area of analysis is about 80 sq. km. The data was obtained from SoilMate using the TCI_NDVI option on the selected area.

Figure 3. TCI rasters, left — 1st of June, right — 25th of August


To detect flooded areas inside aoi, we compared remote-sensing data from Sentinel before and after the flood. On Figure 4, you can see red areas — the most water detected there after the flood.

Figure 4. Flooded areas. Red Line — investigation area.

We performed field boundary detection; this process was used in the Field boundaries change detection article. Field boundary detection is performed using a neural network processor inside SoilMate, which can help farmers and governments to analyze field usage.

We compare fields before and after the flood. Plot boundary detection pipeline yielded 1503 and 1357 fields for images made on the 1st of July and 25th of August, respectively. As you can see, the number of detected fields decreased, so we can assume that some fields were ruined and merged for satellite imagery. In such a way, we can recognize the scale of the disasters.

According to the received data, we made a few hypotheses (for each of them, there are a few examples shown as 5 images: 1, 2 — true color before/after, 3,4 — field shape before/after, 5 — an intersection of field areas):

  1. Some fields have patches of water remaining after the flood or have been shrunk by lake overflow. On Figure 5, you can see examples of such fields.
Figure 5. Partly flooded fields.

2. On Figure 6, some erosion-like phenomena can be observed for some fields.

Figure 6. Fields with erosion

3. Some areas changed their shape and/or content as a consequence of other events, not related to flood (or not directly related). Examples are shown on Figure 7.

Figure 7. Changes in field boundaries due to other events

Here you can see that some fields remain water even a month after the flood.

We calculated the number of flooded fields three days after the flood using a map with flooded areas. We detected 51 fields flooded after the disaster. On Figure 8, you can see flooded (blue) and non-flooded fields (red).

Figure 8. Field boundaries with flooded fields (blue).

Our goal was to determine whether the plants were stressed from the flood. SoilMate provides functionality to detect plant stress in each field. Plant stress is recognized as an anomaly of NDVI (Normalized Difference Vegetation Index). The NDVI is a dimensionless index that describes the difference between visible and near-infrared reflectance of vegetation cover and can estimate the density of green on an area of land. We calculated and compared the mean plant stress of flooded fields with fields not affected by the flood. On figure 9, you can see the distributions of mean plant stress.

Figure 9. Mean plant stress distribution on the 21st of July. Orange — flooded fields, blue — non-flooded fields

As you can see, the distributions are completely different. Flooded fields are mainly 2 types — no stress and more stressed than others. That is because of the fact that flooded fields have water on them, so the NDVI is nearly the same, or they are more stressed due to the water presence.

After achieving such a distribution, we decided to compare the same fields in a month (25th of August). The distribution obtained there is on Figure 10. Here we can see that distributions are nearly the same. Comparison of distributions was performed using the Kolmogotov-Smirnov test and was successfully passed with a threshold 0.05. However, we can see that, in general, there are more fields with greater plant stress.

Figure 10. Mean plant stress distribution on 25st of July. Orange — flooded fields, blue — non-flooded fields


  1. Plot boundary analysis using SoilMate suggests the influence of floods such as localized small water bodies even a month after the flood.
  2. There are soil degradation occurrences observed. Although such a disaster is the most likely explanation for water erosion, additional evidence is needed to conclude that the flood event was the only reason for its emergence.
  3. Local water bodies (lakes) seem to be overflown even after a month after the flood.
  4. Sentinel-1 imagery can help in the exact detection of flooded areas and fields.
  5. Plant stress can be a metric to determine water on the field.
  6. Plants can normalize their growth in a month after the flood.




AI-powered automation tool for collecting analytics data from agricultural fields all around the world!

Love podcasts or audiobooks? Learn on the go with our new app.

Recommended from Medium

Can Data Science Boost Up Your Career?

Funding — or Defunding — the Police

A line of police cars. Photo by thall/Getty Images

How to write a results section

Market Report: Allston Average Rent Prices

Cryptocurrency Prediction with LSTM

Pandas on Databricks via Koalas, a Review

The Future Of Analytics

Sharable Jupyter Notebooks!

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Soil Mate

Soil Mate

AI-powered automation tool for collecting analytics data from agricultural fields all around the world!

More from Medium

Google Earth Engine Makerspace: Community Commons in action

Types of satellite networks used in IoT solutions

Suitability Study for Coffee Arabica cultivation in relation to climate change.