Build a deforestation monitoring solution using satellite imagery in 15 minutes

Deforestation is a critical issue that affects our planet's health and wellbeing. Monitoring deforestation is essential to understand its impact on the environment and to take necessary steps to prevent it. Satellite imagery is a powerful tool that can be used to monitor deforestation in real-time. With recent advancements in technology and the SpaceSense SDK, building a deforestation monitoring solution using satellite imagery is now more accessible than ever before. 

‍

In this article, we will show you how to create a deforestation monitoring solution using satellite imagery in just 15 minutes. The approach we present here is not designed to be implemented in an operational environment, but to give you the tools to generate initial results fast. At the end of the article we’ll also suggest ways to go deeper into more advanced solutions.

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.

To build your solution, we’ll use something called NDVI. NDVI, or Normalized Difference Vegetation Index, is a remote sensing technique that measures the amount of live green vegetation in an area based on the difference between near-infrared and visible light reflected by the plants. It is extremely common and useful for various vegetation-related projects. You can learn more about vegetation indices here. To make it simple, the more vegetation there is, the closer the NDVI value will be to 1. And if we have bare soil, the value will be close to 0.

‍

Our approach to measure deforestation is very simple. We’ll look at two images captured at a different time, and we’ll compare the values of NDVI for each pixel of the image. If the pixel value is similar, this means that there was no significant change in vegetation on the ground. If the value is very different, this means that something changed. So by getting the delta (the difference between two values) of the two NDVI values for each pixel, we are able to generate a new visualisation where the zones that changed are highlighted. We can either display a map that will show this for all pixels, or on which only displays pixels above a certain threshold value (for example a change of NDVI value of at least 0.4).

And here we have our first results, with the deforested areas highlighted in red

We have some good initial results, but we are now facing another issue: Clouds. Especially in tropical areas, you’ll often have at least a few clouds. By only using the method shown above, they are identified as “deforestation”, which is obviously not correct. To attenuate their impact, we can add a mask cloud which will discard the pixels that are clouded in the delta map. The results are not always perfect, but already provide significant improvement.

As you can see, the initial results are good, but can be improved. The seasonality is not taken into account in this solution, the clouds can still be an issue, and other events happening on the ground could trigger false positives. To improve the solution further, we suggest a couple of methods:

• Use a machine learning approach. If you have some knowledge of where some deforestation happened in the last few years in your region of interest, you can use this to create a training dataset, and then build an SVM model on top of it. If you want to know more about machine learning and satellite imagery, you can read this article
• Improve the threshold value. The 0.4 delta NDVI value has been created arbitrarily, based on general experience around vegetation. It could benefit from a more thorough analysis, and could be adapted based on the region and ecosystem
• Use other vegetation indices. Some other indices like the Ratio Vegetation Index (RVI) could provide better results than NDVI