NDVI needs to be replaced

Farm monitoring from space is not a new concept.

Satellite-based data has been used for many years to map and monitor vegetation across farms and has been valuable in providing insights not only to farmers, but also to commercial farm input companies, academia, and the finance sector.

But as the years progress, accuracy is becoming the point of concern for many users of satellite imagery.

That is why we, at Kawa, got our hands muddy and explored the possibilities of a new vegetation index, and have successfully made one, which is more accurate than the industry standard - NDVI.

The old, rusty NDVI

NDVI has been the go-to index to monitor plant activity, first used on satellite imagery back in 1973, using Landsat-1 imagery. NDVI is popular because of its simple equation and the direct relationship of Near-Infrared and Red Spectrum w.r.t vegetation.

But the simplicity of NDVI comes at a price. Not only is it a dated formula, that hasn’t changed since 1973, it comes with its own set of disadvantages like,

NDVI's love for soil and hate for crowded plants is responsible for false indications of plant activity.

  • Being an NIR and Red spectrum based index, NDVI values are highly affected by objects, other than plants, that operate in the NIR and Red spectrum.
  • During stages of low growth on ground, soil is more visible, and so  the plant activity measured by NDVI is higher, due to its inability to distinguish soil NIR reflection from plant NIR reflection.

When there is high growth of plants, they start to reflect extra NIR which they don't need for photosynthesis. NDVI thus gets easily saturated.

  • Dealing with this extra and compounding NIR is difficult for NDVI due to its linear nature from its simple equation, and it gets saturated when plants are at high growth, and cannot differentiate between  active plants and very active plants.

For many years, NDVI has been used in the agriculture domain by companies that want to understand how the farms of their customers are performing.

But I think the time has come to bid farewell to it and replace it with something newer, better and much more accurate.

And that is where the Kawa Vegetation Index comes into the picture. This is where it gets juicy.

Kawa Vegetation Index

After months of arduous research during which we read and analysed the math and pre-existing literature behind NDVI and similar indices, we were able to make a new index - KVI.

KVI, similar to NDVI measures plant activity, which is the photosynthetic activity of the plant.

But unlike NDVI, KVI has a much larger dynamic range, is only sensitive to plant reflected NIR, and is soil noise averse. KVI also does not get saturated.

KVI values increase only when there is true high growth and plant activity on ground, since it is able to compensate for the extra NIR reflectance during high growth levels.

The new KVI is bounded between 0 and 1, unlike -1 and +1 of NDVI, making it a much simpler and practically usable index. 0 is no plant activity, 1 is all plant activity. It really doesn’t get better than this.

NDVI gives false assurance of plant health

Since NDVI is quite highly affected by soil noise, during the low and mid growth stages NDVI values tend to be above 0.2-0.3 and consequently, it overestimates the plant activity on ground.

This tends to give false assurance to anyone monitoring the farms, and is not present onsite, that the farm is performing well.

It also leads to very similar trends of NDVI in a time series across all farms with the same crop type, making the differentiation between "good vs excellent" performance or "poor vs very poor" performance of farms very difficult to find and highlight.

An underlying but very important point to note, is that Photosynthetic activity does not always correlate to the health of plants, and the correlation between the two is very much time-series dependent.

Especially in the case of Cereal crops like Wheat, Rice, Corn, Soybean, Millets, etc. that stop photosynthesis at their final stages of Maturity to allow Photosynthates i.e., the grains to grow. The plants turn yellow/brown for multiple days or even weeks before they are harvested by farmers. In these cases, the loss of chlorophyll and photosynthetic activity does not indicate that the health of the plant has deteriorated; it only indicates a change in its "activity".

Hence we won't say that KVI gives plant health, it is simply a very accurate plant activity indicator, more so than NDVI.

On-ground tests - KVI vs NDVI

To understand how  KVI performs in comparison to NDVI, we performed a study on 2 plots of lands. To validate the findings, we collected on-ground crop- data from the farms - Crop type, greenness, height and phenological growth stage.

Plot 1

Here we compared KVI and NDVI for a "Corn" field that is at the "Stem Elongation" stage. Leaves had emerged, but there was spacing between the crops exposing soil to the satellites.

The Western/Left edge of the farm has other crops in different growth stage.

Field Image - Corn


  • Here, NDVI overestimated plant activity and was constantly above 0.25 in the ratio on both dates of observation.
  • However, KVI was comfortably at 0.1 to 0.15 indicating low plant activity on both dates of observation, removing all the soil reflected NIR and focusing only on the plant reflected NIR.

Plot 2

In this plot, we compared KVI and NDVI on "Millet" crops that are at the "Maturity" stage. The grains on top of each crop have begun to turn Yellow and were nearing Harvest time. The spacing of the plants ensured very little soil was visible to satellites above.

Field Image - Millet


  • NDVI on this farm was at 0.8+ on 2021-03-30 and 0.72+ on 2021-04-04
  • Measured KVI was at 0.6+ on 2021-03-30 and 0.5+, on 2021-04-04.

KVI was able to remove the excess NIR reflection from very mature crops and focus purely on the true activity levels, and more clearly identify the drop in greenery between the two dates.

The results had an excellent correlation to what we observed on ground and KVI was found to be much more truthful in its measurements compared to NDVI.

We focus on the plant, not the soil.

To help understand how big of a difference is the dynamic range of KVI vs NDVI, we have attached the slider below, showing an area with diverse crop types and growth stages.

KVI gives high values only on parts of the farmlands which have truly high plant activity, and low values on parts that have truly low plant activity.

NDVI overestimates both low growth and high growth and hence is not contrasting enough to get accurate insights on the farms.

Both the images have been given the same colormap sequence and a range of 0 to 1. Both images are of the same date.

Notice the over-saturated dark green farms in NDVI image, show more activity variance in KVI image. And notice how KVI fully exploits the 0 to 1 range.



What next?

We have released KVI as a vegetation signal in our agri-bundle API. The signal is screened and adjusted for clouds, fog/smog and cloud shadows to deliver accurate insights for a farm

Users can get KVI for anywhere in the world, via our API, in under 10 seconds!

We have several other agricultural offerings in the pipeline. Currently, we are focussing on crop classification and nutritional mapping, which we plan to release in the next 2-3 quarters.

To enable developers and PMs of companies of all sizes to test out our services, Kawa Space offers a pay-as-you-go model, removing the CAPEX dependency.

I would love to connect! Pick a spot at https://bit.ly/3wwYU2e and we can discuss our KVI, or just talk about space.

Written by - Sidharth Subramaniam
Sid is the GIS Associate at Kawa Space. He loves all things space and science. He likes to watch Formula One, listens to instrumental music, and dabbles in Bird Photography.