Product

Field variability measurement for fertiliser and seed input in Agriculture

5
mins watch
mins read
May 30, 2023

In the realm of modern agriculture, precision is paramount for maximising crop productivity. Farmers and agronomists are continually seeking innovative methods to optimise resource allocation and make data-driven decisions. One groundbreaking tool that has revolutionised field management is satellite imagery.

Satellite imagery provides a valuable overview of agricultural landscapes, unlocking insights into field variability. By leveraging satellite images, farmers can gain valuable information about soil quality, plant health, and overall field performance. In this article, we will explore how you can use satellite imagery to measure field variability for fertiliser and seed input, and create your own variable rate applications and prescription maps.

This article will provide you with a high level overview of the process to build the solution. To complement it, you can find a detailed Jupyter notebook ready to be run.

We randomly selected a field in the UK, which is growing some crops between March and July 2022. The images clearly show the growth and maturation of the crop, followed by the harvest.

Sentinel 2 images between March and July 2022

In our comparison, we want to look at several vegetation indices commonly used in Agriculture: NDVI (Normalized Difference Vegetation Index), LAI (Leaf Area Index) and NDRE (Normalized Difference Red Edge). If we take a date early in the season, we can compare the value of these indices for the same field.

For a single date, RGB, NVDI, LAI and NDRE visual

We can now start using the SpaceSense clustering function, to identify zones. This function enables you to cluster any satellite image with several parameters:

  • The number of output zones you want
  • Which dates should be included
  • Which bands or indices should be included
  • If you want to incorporate custom data (it can be any file that might be relevant: soil map, yield map, soil moisture map…)

We are going to run it for several experiments:

  1. A single day, with a single index
  2. A single day, with several indices
  3. Several days, with several indices
  4. A single day, with several indices and a custom file

1. A single day, with a single index

We will use the 21st of April for our single date, 6 zones, and NDVI for our index.

As shown in the result below, the vertical bands observed within the NDVI image are still clearly visible in the clustering result, and seem to cluster the zones efficiently.

Clustering with NDVI for a single date

2. A single day, with several indices

We keep the 21st of April for our single date, 6 zones, but we add LAI and NDRE to the NDVI.

As shown in the result below, we still see the vertical bands from the NDVI, but we also see an area that appeared in the bottom, which is visible in the LAI band. This clearly indicates that both indices were used in the result. The NDRE being pretty neutral, had little impact on the resulting cluster.

Clustering with NDVI, LAI and NDRE for a single date

3. Several days, with several indices

We took all the images from the season, 6 zones, and all the indices (LAI, NDRE and NDVI).

As shown in the result below, when taking the whole season, we see significantly less noise than with a single date, and it shows the long term differences in the field, and reveals structural variations within the field

Clustering with NDVI, LAI and NDRE for the whole season

4. A single day, with several indices and a custom file

We will use the 21st of April for our single date, 6 zones, all the indices (LAI, NDRE and NDVI) and a custom file, represented below

As you can see, this is a randomly generated file, just designed for an example. The image looks like static because each pixel value was generated randomly. This is designed to highlight the visual impact of that file on the clustering.

And sure enough, the result is clearly impacted by the added file. If replaced with a relevant file with actual field information, it will provide precious information

On the left the raw custom file, and on the right the clustering result with NDVI, LAI, NDRE and the custom file for a single date

And that is it! You now have all the elements to start building your own variable rate applications using your agronomic knowledge, to design something fully adapted for your crop, and your region. Have fun!

Want to benefit from geospatial
AI easily?

Connect with us

Check out our upcoming solutions that you can test

Check it out