How to design a vineyard sampling strategy accounting for spatial variability?
One of the main challenges for a winegrower before harvest is : “where is it best to sample?” and “how many locations do I need to sample to get a reliable estimate?”
In this blog we discuss how to design a sampling scheme to best capture vineyard spatial variability. The plant health index used in this example is plant water status measured at predawn, but the same approach can be applied for other indexes involving plant or fruit composition. Interpretation of those indexes will directly drive vineyard decisions such as harvest, differential irrigation, differential fertilization, cluster thinning or pruning. With more vineyard data becoming available, scientists have made strides towards understanding spatial variability to better monitor plant and fruit indexes. In a previous blog post, we discussed recent results helping winegrower mapping vineyard water status variations. Today we discuss how to implement those results in practice to improve vineyard performances monitoring.
Why is it challenging to optimize your sampling design?
Gathering field data to characterize a vineyard is by no means simple. Due to spatial and temporal variations, traditional reference measurements can be too costly and time consuming to give a comprehensive picture of vineyard performances. Thus, there is a compromise to find between reducing the number of measurements to lower the cost of data collection while getting enough measurements to get a reliable and representative sampling. To address this question, a team of french researchers led by Herrero-Langreo (here) compared two stratified sampling methods, called Kennard & Stone (KS) and Surface Response (SR), with an ideal sampling method called “the reference model”.
- In the reference model, vines are systematically sampled every 200 square meter (ie. a square 45 feet long) to give the most precise picture of vineyard performance. Obviously, the reference model, with so many sampled vines is cost prohibitive and would not be implemented in a production context. (see figure 1)
- The principle of the 2 stratified sampling strategies is to rely on fewer measurement points while using the relationship between a plant index (like predawn water potential) and a proxy data represented by a map. Vineyard maps can be obtained directly from plant based measurements or indirectly, for instance from aerial pictures.
What have the researchers found?
In 2017, Herrero-Langreo and al. (here) analyzed the offset between the vineyard water status obtained from the reference model (ie. the most complete information) with an estimate of vineyard water status obtained from the stratified sampling methods. By varying the number and position of sampling sites for each stratified sampling methods authors could quantify the quality of the estimates according to the sampling size. The authors could give recommendations to assist winegrowers finding the best compromise regarding how many sampling sites and where to position them.
Technically, authors studied 1) what was the the minimum number of sampling sites needed to represent the reference model with accuracy; 2) what was the optimum location of sampling sites to maximize the accuracy. An acceptable level of accuracy means that the “error” between the reference model and the “estimation” from the stratified sampling is small enough to have no impact on practical decisions.
Figure 1: Locations of sampling sites for the reference model (from Herrero et al, 2017) Each red dot represents an area of 0.02 Hectare
Overall, both methods showed a high level of accuracy compared to the reference model. Authors came up with 2 practical conclusions
- If 100 sampling sites are needed to describe the vineyard with a reference model, the same accurate picture can be achieved with 18 stratified sampling sites only.
- Stratified sampling methods yield better estimates when sampling points are located at the most contrasted locations of the vineyard.
Practical take home for the winegrower
To optimize your vineyard sampling strategy in a production context here is how you could do:
#1: analyze the spatial structure of your vineyard.
For that you can use maps obtained from:
- Direct plant measurement: describing variations of trunk circumference, pruning wood accumulation (Physiocap), color accumulation (Multiplex). Such plant based maps are useful because they represent plant ability to store carbohydrates or to regulate activity during ripening.
- Indirect measurement: like a soil map or an aerial picture to generate indexes like NDVI.
#2: position sampling sites at the most contrasted locations highlighted by your maps
#3: track plant and fruit indexes variations at those very same locations throughout the season
#4: use the different maps obtained during the season to extrapolate point measurements over the whole vineyard.
We hope this information will help you implementing a more accurate sampling strategy before harvest.
Credit cover picture : Oaks vineyard
Share this post :