The use of NDVI in agriculture is no longer a new idea. As a result of numerous publications demonstrating the capacity of this measure to assist vintners in their technical decision-making, the number of NDVI maps providers has increased considerably in recent years. Given the many choices available, winegrowers may encounter difficulties in evaluating the various services offered to them. In addition, images from multiple vendors or separate dates are difficult to compare and reconcile with other data sources. The Fruition Sciences team regularly receives questions about the correlation between NDVI maps and field measurements. In this article, we offer an overview of the problems encountered and how to avoid them.
Aerial NDVI limitation #1: ground calibration of maps
One of the most frequent problems we have encountered is the confusion around the NDVI calibration. Although NDVI is very useful in capturing the variability of a vineyard, the measurement can be affected by a large number of factors. According to Johnson et al (2003) these factors include: soil brightness, atmospheric turbidity and canopy structure. Two NDVI images of the same plot taken at two different, or even two consecutive, dates may appear totally divergent because of a different cloudiness. According to the ICM (Integrated crop management) of Iowa, most NDVI providers do not provide calibrated maps. This explains the difficulty coming from comparing NDVI maps when taken at different dates. This lack of calibration represents a real challenge for winegrowers who are trying to understand how the spatial variability of their vineyards evolves from one year to the next, or within the season, and what impact their cultivation practices have on the vine vigor.
Figure 2: NDVI map of a vineyard (source)
How do you calibrate a NDVI map?
To calibrate maps, it is important to use ground invariants whose reflectance or index is well known. These invariants can be natural (roads, buildings etc.). For example, Johnson and Scholasch (2005) used a pedestrian spectroradiometer to measure the reflectance of four soil types (one asphalt road, one gravel parking lot and two concrete surfaces) and thus calibrate their NDVI measurements. It is also possible to use “target panels” whose NDVI index is perfectly known from laboratory measurement. Those panels have to be placed on the ground before the flight to be seen while the picture is taken (as it is offered by “l’avion jaune”).
Aerial NDVI limitation #2: pixel size and between vine row effect.
Contrary to proxi-detection (sensors embedded on tractors), remote sensing (aerial images) necessarily takes into account the NDVI index of inter-rows in viticulture. Depending upon the picture resolution and winegrower management strategy for ground cultivation practices, aerial images will be more or less ”distorted” by the influence of inter-row. In other words, the pixel value seen in the picture will reflect more or less strongly the pixel value corresponding to the vine only. Figure 3 illustrates how inter row directly influences the vine NDVI value. In fact, when the pixel size resolution is greater than one meter, it is impossible to identify pixels that correspond only to the vine. The pixel value will capture the vine row as well as the inter row. Thus, the NDVI value corresponding to the pixel will be proportional to the reflectance of each covered area, vines and soil.
Figure 3: Influence of different surfaces reflectance on pixel NDVI value (Fruition Sciences)
When constructing an NDVI image, it is important to understand that the value of each pixel is also influenced by the adjacent pixels. An NDVI image with a 1 meter resolution (pixel area of 1 * 1m) is a map where each pixel value corresponding to the vine row is influenced by the pixels of the inter-row, and vice versa.
When the NDVI image is being generated you have to consider the following effects coming from contrasting soil management strategies:
- if cover crop has been removed to maintain bare soil conditions between vine rows, then the contrast between a vine row pixel and an inter-row pixel soil is large since the NDVI value of a bare soil is close to zero. Consequently, there will be a great effect of the bare ground area onto the NDVI value corresponding to vine pixel.
- if a cover crop is maintained between vine rows, then the contrast between a vine row pixel and an inter-row pixel is small, since the NDVI values corresponding to a vine and to a cover crop are very similar. In fact, the more vigorous is the cover crop, the smaller is the difference between vine NDVI value and cover crop NDVI value. In such vineyard conditions, it is very difficult to distinguish areas corresponding to cover crop NDVI from areas corresponding to vine NDVI.
Figure 4 illustrates how adjacent pixels can influence vine pixel value.
In conclusion, it is strongly advised to avoid comparing blocks with cover crop with blocks with bare soil between rows.
Figure 4 : influence of inter-row pixel onto the vine row pixel (Fruition Sciences)
Does this mean that the only solution for vineyards with cover crop is proxi-detection?
No. Today with a fairly fine resolution (pixel size < 0.5 m), it is possible to identify vine rows by automatic image analysis. Once vine rows have been detected, classification processes can discriminate between cover crop and vine so that the picture only keeps pixels values corresponding to vine leaves. As a result, following such treatment, the NDVI map only displays the NDVI index of the vines as it shown in Figure 5. Of course, the finer the resolution, the more reliable is NDVI the map produced.
Figure 5: Aerial NDVI map after removal of between vine row area (L’avion jaune / Fruition Sciences)
What questions should you ask your NDVI image provider?
Fruition Sciences has compiled a list of questions to ask your NDVI map supplier:
- What is your experience with my case study? (is your objective aiming at monitoring are fertilization, leaf area development, detection of diseases, etc…?)
- What is the spatial resolution required in my case study?
- How do you calibrate your maps?
- What are the potential biases with the measurement?
- How do you correct these biases?
Next, we will discuss the main viticultural applications from NDVI.