In the first principles derivation of NDVI, SAVI and EVI, we used theory and mathematics to show the strengths and weaknesses of each index. In this article we explore those same strengths and weaknesses via real world case studies.
We’ll look at:
The most notable takeaway from the case studies that follow is that EVI is superior to NDVI and SAVI in every scenario examined. This aligns completely with theory.
The photograph below shows a region of farmland in Alberta, Canada on April 21 2022. Notice several things. Firstly, nothing much is growing yet. There is however one field where crops are just starting to emerge. Finally, notice the variance of soil color. Due to moisture levels or soil type, some fields are noticeably darker than others. In the area outlined, there is a high number of fields that are a bit darker than most of the scene:
Let’s now examine this same area through the lens of NDVI. (Note, the visualization we use here and throughout will show NDVI on a grayscale going from black to white as NDVI goes from 0 to 1. The same will be true for SAVI and EVI.)
Firstly notice the NDVI for the one field with an emerging crop is higher (lighter) than anywhere else in the scene. This is what we would hope to see if NDVI is working as it should.
What we don’t want to see however, is the elevated NDVI in the encircled area. There is no vegetation here, but many fields in this area clearly have higher NDVI than other non vegetated areas. This is a “soil effect”.
NDVI is, unfortunately, sensitive to soil color. (The reason for this undesirable property is discussed in The Effect of Soil Color on NDVI.) As this example makes clear, wIth no other information, a practitioner will struggle to differentiate between real vegetation and soil effects.
SAVI and EVI are, by design, less sensitive to soil effects. (See Remedies for Soil Effects). The three images below show how NDVI, SAVI and EVI see the same area:
By virtue of the actual math of the formula, SAVI will always be lower than NDVI in practice. You can see this and its lower sensitivity to soil effects by comparing the NDVI visual to the SAVI visual.
EVI is theoretically better than both NDVI and SAVI in this early growth season situation. That too is visible: notice how the emerging crop is more pronounced in the EVI visual versus the SAVI visual while the sole effects are less stark.
Let’s look at the same area 32 days later. Visually it is clear that there is additional vegetation all over the scene:
NDVI captures this:
Notice however that NDVI is still “deceived” by soil effects. SAVI and EVI offer measurements that are more closely correlated with ground truth:
Let’s now look at the same area 35 days later. Visually, there is clearly more vegetation than 5 weeks prior:
NDVI has (correctly) increased substantially:
By this stage in the growing season, we are starting to see saturation in some areas. Recall from NDVI Saturation, that when an area gets sufficiently “green”, NDVI maxes out. By this point in the growing season that is happening for some fields.
By construction, (see Derivation of EVI), SAVI and EVI don’t saturate as early in the season as NDVI does:
Notice how SAVI is systematically darker almost everywhere than NDVI. This is because, by construction, SAVI is always less than NDVI. (See SAVI vs NDVI.) This of course ensures that SAVI saturates later than NDVI does. In the scenes above, you can clearly see NDVI is consistently “more white” (i.e. higher) than SAVI is. What you are seeing is NDVI approaching saturation ahead of SAVI.
Notice also how excellent EVI is here. it is not saturating and remains sensitive to vegetation. (Compared to SAVI which avoids saturation by just making everything a bit darker.) To see this, note the circled field on all three scenes. NDVI can barely detect the high biomass here. SAVI understates it. But EVI captures it.
In the late growing season, saturation becomes more pronounced. We’ll zoom in to study it closely:
Consider the large circular field in the left corner of each subscene. Notice two things:
By moving to the Amazon we can see a case study in extreme saturation. In the image below, we can see how dense the vegetation is, especially compared to the scene we just looked at in Alberta.
Here is the NDVI representation of the scene:
Not surprisingly, NDVI is basically saturated.
Notice that SAVI, and especially EVI, are able to detect the subtle variations in biomass that NDVI is blind to:
In this last study, notice the subtle changes through time during the late states of the growing season. NDVI is essentially useless here because it’s fully saturated. EVI, again, works best. While adjacent EVI images do look similar, there is a perceptible difference between each one. That same cannot be said for NDVI.
EVI is superior to NDVI and SAVI in every case study conducted there. We’ve created a visual showing NDVI, SAVI and EVI for the same location through time. Click on the image to see it in high resolution. All the concepts explored in this article can be seen in this single visual.
Sign up for the latest news, company insights, and Streambatch updates.
