Abstract
Background and goals In winegrape production, clusters are sampled multiple times per growing season to assess fruit maturation. The economic cost of implementing a previously-developed spatial sampling method using normalized difference vegetation index (NDVI) images to assess block variability was compared to the economic costs of random sampling (R20) and of sampling via the four corners approach (4C). The objective was to compare the travel distance to sample blocks between the spatial sampling protocol (NDVI3), R20, and 4C, and the time and cost associated with the required travel distance of each sampling method.
Methods and key findings Travel distances for each sampling method were calculated for six commercial vineyard blocks. The cost of required labor and travel was calculated for each travel distance. Travel distances per block for the R20, 4C, and NDVI3 methods ranged from 3.34 to 13.63 km, 0.68 to 2.41 km, and 0.36 to 1.58 km (all per sampling event), respectively. Total cost savings per sampling event using NDVI3 ranged from $5.54 to $32.40 when compared to R20, and from $0.33 to $4.61 when compared to 4C.
Conclusions and significance The cost savings of using NDVI3 for maturation monitoring appears relatively small compared to R20 and 4C, but considering that most blocks are sampled multiple times leading up to harvest, and that larger wine companies have thousands of acres to sample for fruit maturity, savings may become more substantial. We conclude that adoption of the NDVI3 sampling protocol for fruit maturity may substantially reduce the cost of production for winegrapes grown in large blocks, compared to R20 and 4C sampling.
Introduction
Grapes are the highest value fruit crop in the United States. They are produced on ∼366,000 ha with a total production value of over $6.5 billion, as found on the National Association of American Wineries website (https://wineamerica.org/economic-impact-study). However, agricultural producers report that a lack of availability in nonsupervisory crop and livestock labor has been a challenge over the past five years, as found on the USDA Economic Research Service website (www.ers.usda.gov/topics/farm-economy/farm-labor#recent). Furthermore, labor wages for farmworkers have risen at an average rate of 1.2% per year between 1990 and 2021, and >80% of nonsupervisory labor received a >6.0% increase in wages between 2021 and 2022, as found on the aforementioned USDA Economic Research Service website. Ongoing labor shortages combined with significant wage increases contribute to increased operating costs for growers and, in some cases, insufficient personnel to effectively manage their vineyards.
Substantial labor is required to monitor grape maturation (total soluble solids [TSS], pH, and titratable acidity [TA]); this is an ongoing vineyard task that typically takes place from veraison through harvest. Extension personnel advise that random samples of entire vineyard blocks should be collected every seven to 10 days, starting three to five weeks before harvest and, if necessary, increasing in frequency to every two to three days prior to harvest (Hillin 2019). In total, most vineyards should conduct three to nine sampling events per growing season, depending on the ripening rate and the grower’s goals (Hillin 2019). While this frequent sampling is a best practice for grape production, a lack of labor availability or increased operational expenses may lead growers to improperly collect the sample (i.e., non-random), reduce sample size, or possibly forgo rounds of sampling, leading to less accurate representations of actual ripeness in the vineyard and less informed harvest decisions.
Stratified sampling has been shown to result in fruit samples of composition similar to the typical industry standard of random sampling in Australian vineyards (Krstic et al. 2002), although analysis of the method conducted using LANDSAT images in 13 California vineyards suggest that the sampling locations did not always match the pixel distribution of the vineyard block (Meyers et al. 2020). Previous research has shown that satellite normalized difference vegetation index (NDVI) images processed with an algorithm to locate three consecutive pixels representing the left tail, center, and right tail of the NDVI pixel population could be used for sampling, to effectively represent the grape maturation (TSS, TA, pH) variation in a vineyard block (Meyers et al. 2020). The three consecutive pixels (NDVI3) could then be used to create spatially explicit sampling protocols to reduce the number of cluster collection sites within a vineyard block to a 90 m section of a single row, potentially reducing the travel distance when collecting samples, compared to random sampling (Meyers et al. 2020). Despite this advancement in grape maturation monitoring, the actual time savings and associated cost savings have not been assessed to determine whether adoption of this NDVI3 protocol may be warranted. This study sought to determine the travel distance, time, and related inputs when using the NDVI3 spatial sampling protocol, compared to random sampling of 20 locations in a vineyard block (R20) or the four corners sampling approach (4C). More specifically, this study analyzed the cost differences between the three sampling protocols in two distinct grapegrowing regions, New York and California.
Materials and Methods
Vineyards
Subject vineyard blocks were located in the Central Valley of CA and the Finger Lakes region of NY. Blocks were selected for the study based on size, to ensure a range of sizes (Figure 1).
Images from Google satellite base layer taken for New York and California blocks. R20, 4C, and NDVI3 collection paths are overlayed on each of the images. R20, random sampling of 20 locations in a vineyard block; 4C, four corners sampling; NDVI3, satellite normalized difference vegetation index images processed with an algorithm to locate three consecutive pixels representing the left tail, center, and right tail of the NDVI pixel population.
Satellite imagery
Landsat-8 NDVI images of 19 and 20 Aug 2022 for NY and 7 Aug 2022 for CA were downloaded from USGS Earth Explorer. NDVI is a ratio of red and near-infrared channels, and was calculated for each block. NDVI is widely used in viticulture, primarily to proxy the canopy greenness, providing insights into relative vineyard management zones. Spatial sampling protocol NDVI3 was calculated, and its pixel distribution was tested against the R20 distribution using the Kolmogorov-Smirnov (KS) statistics, as described (Meyers et al. 2020) (Table 1). The KS statistic compares two sample distributions using the absolute difference between the cumulative distributions. KS statistic p value > 0.05 can be interpreted as two distributions that are not significantly different from one another. In brief, R20 is 20 computer-generated locations within the block, which represent a random sample. NDVI3 is a 90 m-long path representing the low, medium, and high quartiles of the NDVI distribution of the block (Meyers et al. 2020). 4C is a commonly-used method in which the four corners of the block are sampled.
Vineyard block information reflecting block size, normalized difference vegetation index (NDVI) image date, and Kolmogorov-Smirnov (KS) value for NDVI3 for the six study blocks. NY, New York; CA, California; NDVI3, satellite NDVI images processed with an algorithm to locate three consecutive pixels representing the left tail, center, and right tail of the NDVI pixel population.
Collection paths
For consistency, each block’s starting and ending point was set to the northeast corner of the block. Thus, all sampling paths started and ended at the same point. The R20 sampling paths ran through the whole row to get to the next row, assuming that ducking under the in-vine rows was not possible, as samples were being collected from a vehicle (Table 2). R20 sampling paths started from the starting point, ran through the 20 different locations, and ended at the same point. 4C paths involved sampling from each of the four corners of the block, but sampling location was calculated differently, based on the smaller block sizes in NY. For CA vineyards, 4C sampling was conducted in the tenth row from each corner, with sampling at the fifth panel from the entry point of the row and the fifth panel from the exit point of the row. In NY blocks, the fifth row from each corner and the fifth panel from the row entry and exit point were sampled for the 4C method. NDVI3 sampling paths started from the same starting point and went directly through the row that needed to be sampled. These paths were manually drawn in QGIS software using advanced digitizing tools (https://plugins.qgis.org/plugins/DigitizingTools/#plugin-details). The distance required to travel through the sampling path of R20 was calculated using the total length of R20 in meters. For NDVI3, the length of the sampling path was multiplied by two, given that the sampling path would start from the starting point and end back at that point following the same route (Figure 1). For 4C, the distance was calculated to complete the box (Figure 1) and return to the starting point.
Assumptions related to the transportation method, sample collection, and variable costs used for calculating sampling costs of R20, 4C, and NDVI3. ATV, all terrain vehicle; CA, California; NY, New York; R20, random sampling of 20 locations in a vineyard block; 4C, four corners sampling; NDVI3, satellite normalized difference vegetation index (NDVI) images processed with an algorithm to locate three consecutive pixels representing the left tail, center, and right tail of the NDVI pixel population.
Cluster collection
The study assumed the sample size was the same for all three methods in terms of the number of clusters collected (Meyers et al. 2020). Differences in the amount of time to collect clusters of different varieties on different training systems were assumed to be negligible (Table 2).
Vehicle
Sample collection transportation was assumed to be conducted with a gasoline all terrain vehicle (ATV) with a fuel economy of 20 mpg (8.5 km/L). The operating speed of the ATV in this study was 11.3 km/hr (7 mph), which was based off the upper end of desirable tractor operation speeds between 8.1 to 11.3 km/hr (5 to 7 mph), as found on the Farm Energy website (https://farm-energy.extension.org/selecting-engine-and-travel-speeds-for-optimal-fuel-efficiency/). It can reasonably be assumed that transportation speeds may be slower or faster depending on the vineyard layout (row spacing, vine vigor, etc.), slope, and transportation method (tractor, ATV, walking, etc.), however, 11.3 km/hr was selected as a moderate speed for this study (Table 2). For additional context, the time required to sample each block while walking (rather than driving) was calculated at a speed of 4.83 km/hr.
Gasoline
This study assumes the fuel source for the ATV was regular unleaded gasoline. Gasoline prices for NY and CA were based off the NY State Energy Research and Development Authority’s weekly average motor gasoline prices (https://www.nyserda.ny.gov/Energy-Prices/Motor-Gasoline/Weekly-Average-Motor-Gasoline-Prices) and the U.S. Energy Information Administration’s weekly retail gasoline and diesel price reports (https://www.eia.gov/dnav/pet/pet_pri_gnd_dcus_sca_w.htm), respectively. Gasoline prices from the latest published price reports prior to the collection of satellite images were used in this study. The price data for NY and CA were collected from the reports published on 15 and 1 Aug 2022, respectively (Table 2).
Variable costs
Rates for labor, capital recovery, insurance, and lube and repair were determined. Labor rates used in the calculation of labor costs were assumed to be $23.10 and $17.50/hr in CA (Murdock et al. 2022) and NY (Davis et al. 2020), respectively. The capital recovery rate, the annual depreciation and interest on capital investment of the ATV, was assumed to be $2.05/hr (Table 2). This rate was determined using the formula ((Purchase Price – Salvage Value) × (Capital Recovery Factor)) + (Salvage Value × Interest Rate), which accounts for the time value of money (Murdock et al. 2022). The ATV insurance rate was assumed to be charged at 0.886% of the average value of the ATV over its useful life, which is estimated to be $0.01/hr (Table 2) (Murdock et al. 2022). The lube and repairs rate was determined by multiplying the total hourly operating cost of the ATV by the hours per acre (Murdock et al. 2022). In this study, the estimated lube and repairs cost was $0.89/hr (Table 2).
Results
The NDVI3 sample paths were shorter than the R20 and 4C sample paths in all studied blocks; R20, 4C, and NDVI3 travel paths ranged from 3.35 to 13.63 km, from 0.68 to 2.41 km, and from 0.35 to 1.58 km, respectively. As a result, the expected time of completion for the NDVI3 sample paths was less than both the R20 and 4C sample paths, and the cost estimates to complete the sample path routes were also lower (Table 3). Total cost savings using NDVI3 compared to R20 and 4C ranged from $5.54 to $32.40 and from $0.33 to $4.61 per sampling event, respectively, depending on the block (Table 3).
Variable costs per sampling event of cluster sampling in each trial block using R20, 4C, and NDVI3 sampling protocols. The sum of the variable costs per sampling event for each sample path is included and is used to calculate the total transportation cost of sampling using each method in each vineyard block. CA, California; NY, New York; R20, random sampling of 20 locations in a vineyard block; 4C, four corners sampling; NDVI3, satellite normalized difference vegetation index (NDVI) images processed with an algorithm to locate three consecutive pixels representing the left tail, center, and right tail of the NDVI pixel population.
Labor costs (assuming driving for sample collection) were the highest variable cost and made up ∼80% and ∼83% of the total cost per sampling event in NY and CA, respectively, making this variable cost proportionally greater than all others (Table 3). The labor cost contributed a larger amount to the total cost per sampling event in CA due to the 32% higher labor rate, compared to the NY labor rate. The second highest variable cost, albeit insignificant to labor, was capital recovery, making up ∼9% and ∼7% of the total cost per sampling event in NY and CA, respectively. Gasoline was ∼7% of the total cost per sampling event in both NY and CA. Despite CA having higher regular unleaded gasoline prices, the difference in fuel cost was minimal in relation to other variable costs (Table 3), as sampling with an ATV requires low gas usage. Furthermore, lube and repairs made up ∼4% and ∼3% of the total cost per sampling event in NY and CA, respectively. Insurance, on a per sampling event basis, was found to be negligible, as the maximum cost per sampling event was $0.01 when using R20 sampling paths for CA1, CA2, and CA3, while the insurance cost for all other sampling paths was a fraction of $0.01 per sampling event (Table 3).
Time to complete sampling differed substantially when sample collectors were assumed to be walking. Since the paths were the same and only the speed of collection changed between walking and driving, the time to walk the sample paths at 4.83 km/hr was 233% greater than the time required to drive them at 11.3 km/hr. In some vineyard blocks, sample collection by walking may allow for shortcuts (i.e., ducking under rows) depending on the trellising, vine size, and vine phenology. When comparing costs of sample collection by walking or driving, it is important to note that the additional costs (Table 3) of gas, insurance, lube and repairs, and capital recovery for the ATV would not be required to analyze the cost of sample collection by walking.
Since the total cost per sampling event was directly related to the amount of time and distance traveled, greater cost savings using NDVI3 compared to R20 and 4C were associated with the larger blocks in CA, which saw greater reductions in sample path distances. However, larger block size did not necessarily lead to greater cost savings nominally or as a percentage. The NY blocks’ total cost savings generally increased nominally as block size increased, with a slight variation of $0.40 in total savings per sampling event between blocks of the same acreage, NY2 and NY3. Despite this nominal increase in savings correlated to block size, by using NDVI3 compared to R20, NY1 had a 1.79% greater percentage saving than NY2, which is a 0.13-ha larger block (Table 1). Similarly, the CA block’s total cost savings with NDVI3 did not correlate to block size as the smallest block. CA1 had the lowest total cost savings of the CA blocks, while the largest block, CA3 (36.69 ha), saw $5.91 less total cost savings per sampling event than the second largest block, CA2 (27.72 ha), compared to R20 (Table 3).
For the blocks in this study, cost savings as a result of adopting NDVI3 rather than R20 ranged from $16.62 to $98.07 and from $49.86 to $294.21 for three and nine sampling events, respectively (Table 4). Compared to 4C, cost savings using NDVI3 ranged from $0.99 to $13.80 and from $2.97 to $41.40 for three and nine sampling events, respectively (Table 4).
Total cost savings per block, using NDVI3 compared to R20 and 4C, assuming an average of three, five, and nine fruit maturity sampling events per season. CA, California; NY, New York; R20, random sampling of 20 locations in a vineyard block; 4C, four corners sampling; NDVI3, satellite normalized difference vegetation index (NDVI) images processed with an algorithm to locate three consecutive pixels representing the left tail, center, and right tail of the NDVI pixel population.
Discussion
On a per block, per sampling event basis, the cost savings from using NDVI3 (over random sampling) to monitor fruit maturation appears relatively small. When the cost of NDVI3 is compared to the cost of 4C, the cost savings is even smaller. However, it is important to consider the potential application of this protocol over the course of an entire growing season and on multiple blocks in a vineyard. Growers may sample anywhere from three to nine times per growing season (Hillin 2019), which would multiply their expected savings per season by the number of sampling events they conduct (Table 4). The economic benefit of utilizing this sampling protocol may be more evident for vineyards that cluster sample frequently, vineyards that have multiple blocks that are sampled separately, vineyards with larger blocks, and especially for businesses managing large acreages of vineyard blocks.
Despite these savings becoming more substantial at a greater scale, the cost assessment shows that there are potential economic benefits even for small vineyards with blocks as small as 2.75 ha. More important for some wine regions may be the scarcity of labor, regardless of the cost, and the resulting perceived need to reduce labor hours.
The blocks used in this study should only be considered as examples. The variability in savings among the blocks studied here is due to the physical attributes of the vineyard, as the NDVI3 algorithm creates a sample path based on the pixel distribution observed in the vineyard, which is inherently variable. Due to this variability, if the algorithm selects the three contiguous pixels near the starting point, the savings will be greater, while if it selects three contiguous pixels on the far side of the block, the savings will be less. Other factors can affect the sample path distance, particularly the starting point in the block. We randomly selected the northeast corner of each block for consistency, but sampling multiple blocks sequentially would likely not require the sampler to return to the original starting point each time. Blocks of similar sizes should not be assumed to have similar cost savings associated with adoption of the NDVI3 sampling protocol.
The cost of 4C sampling in the six blocks was quite low, but it is unclear whether 4C can accurately predict fruit maturation for the entire block. When tested in the Central Valley of CA, NDVI3 actually outperformed R20 by producing higher KS p values for 12 of the 13 commercial blocks (Meyers et al. 2020).
Greater cost savings using NDVI3 sampling protocols compared to R20 and 4C were associated with larger blocks, with the substantially larger CA blocks resulting in greater cost savings than the smaller NY blocks. Despite this general trend, larger block size did not necessarily indicate that greater cost savings should be expected, especially when block sizes are similar, as the location of the NDVI3 sampling pixels will ultimately determine how much savings the NDVI3 sampling protocol will provide.
One current drawback to the NDVI3 sampling protocol is the potential for a vineyard to be too small for there to be enough Landsat pixels (30 m × 30 m) in the satellite image; further work should focus on the possibility of using higher resolution NDVI images, such as those from Sentinel-2 (10 m × 10 m). For example, in the northeastern U.S., many winegrape vineyard blocks are <0.4 ha in area. Small vineyards are less likely to have three consecutive NDVI pixels that represent the distribution in a block, particularly when the pixels are large. Additionally, the NDVI3 protocol was developed in Central Valley vineyard blocks with uniform soil and management practices (Meyers et al. 2020). As increased soil variability may affect the NDVI response (Andre et al. 2012), the NDVI3 protocol should be tested for its ability to accurately predict fruit composition in vineyards with greater variability in soil.
Another drawback of utilizing this technology is the lack of availability of a user-friendly interface that would allow growers to conveniently use this NDVI3 sampling protocol; we are working to rectify this issue.
Conclusions
This study indicated that adopting the NDVI3 sampling protocol over random sampling for assessing fruit maturity can result in substantial cost savings over a growing season, primarily because of a reduction in labor (the result of a shorter travel path for NDVI3) and sampling time. Implementation of the NDVI3 sampling protocol for larger vineyard blocks has the potential to substantially reduce the cost of winegrape production in areas where labor shortages and costs are a concern.
Footnotes
We thank Jim Meyers for helpful discussions as well as our industry collaborators E.&J. Gallo Winery, Sheldrake Point Vineyards, and Fulkerson Winery.
Chock CKK, Trivedi MB and Vanden Heuvel JE. 2024. Spatial sampling of fruit maturity reduces sampling costs for winegrapes in California and New York. Am J Enol Vitic 75:0750012. DOI: 10.5344/ajev.2024.23057
By downloading and/or receiving this article, you agree to the Disclaimer of Warranties and Liability. If you do not agree to the Disclaimers, do not download and/or accept this article.
- Received August 2023.
- Accepted February 2024.
- Published online May 2024
This is an open access article distributed under the CC BY 4.0 license.
Vol 75 Issue 1
Jump to section
Related Articles
Cited By...
More from this TOC section
Similar Articles
