Abstract
Background and goals Monitoring vineyards with remote sensing tools is challenging due to the site specificity and difficulty of accurately scaling the technology across large regions. To overcome these challenges, this study aimed to understand how a time series of remote sensing vegetation indices is influenced by vineyard design, vine age, and vineyard floor management practices.
Methods and key findings We examined Sentinel-2 time series data over a 5-yr period from over 1000 vineyard blocks covering more than 10,000 ha across California. Our analysis revealed a strong annual effect and a significant impact of vine trellis-training systems. Vine age was particularly relevant for blocks younger than 3 yr and older than 25 yr, while factors such as variety and row distance (ranging from ~2 to 4 m) were less significant. We also found that remote sensing vegetation indices calculated from the top of the canopy were less relevant for vines grown on the vertical shoot-positioned trellis compared to vines grown on other trellis systems.
Conclusions and significance These findings help define key vineyard characteristics that influence the normalized difference vegetation index and potentially other commonly used vegetation indices. They provide new insights into the factors that must be considered when using remote sensing data across heterogeneous sets of vineyard blocks, as well as the characteristic seasonal pattern for each factor. This work paves the way for large-scale vineyard monitoring using satellite-based vegetation indices.
Introduction
Over the past few decades, remote sensing technologies and digital image processing techniques have revolutionized crop production, including perennial crops such as grapevines. These technologies provide vineyard managers with detailed information about their crops that was previously unavailable or not accessible at scale (Hall et al. 2002, Usha and Singh 2013, Matese and Di Gennaro 2015). By enabling the identification and monitoring of plant behavior in vineyards, remote sensing technologies have been successfully applied to detect various forms of stress, including water stress (Acevedo-Opazo et al. 2007, Romero et al. 2018, Lopez Fornieles et al. 2021), nutrient stress (Comparetti and Marques Da Silva 2022), and biotic stress (Mohite et al. 2018, Wang et al. 2022). Remote sensing has also proven valuable in estimating plant biomass (Leolini et al. 2023), yield (Ballesteros et al. 2020, Sams et al. 2022), soil properties (Vaudour et al. 2019), and evapotranspiration (Semmens et al. 2016, Melton et al. 2022, Volk et al. 2024). These findings have primarily been derived by calculating various indices from spectral data, with the most commonly used index being the normalized difference vegetation index (NDVI) (Tucker 1979).
The European Space Agency’s Sentinel-2 satellite mission produces multispectral high-frequency imagery, acquired every 5 to 6 days, that has a spatial resolution ranging from 10 to 60 m/px depending on the wavelength (10 to 20 m/px for the bands commonly used for agriculture applications), and is available on a global scale at no cost to the user. Access to more advanced remote sensing data such as the Sentinel-2 imagery has provided an opportunity for large-scale vineyard monitoring (Devaux et al. 2019, Laroche-Pinel et al. 2021, Mucalo et al. 2024).
Despite their success in monitoring plant physiological status or estimating plant production, remote sensing studies in vineyards are often highly site- and time-specific. Vineyard characteristics (e.g., plant age or grape variety), vineyard design (e.g., trellis and training systems, row and vine spacing, and plant density), and local management practices (e.g., interrow floor management or canopy manipulations) also influence the surface reflectance and resulting vegetation indices. As a result, it is challenging to make direct comparisons between vegetation indices calculated for new vineyards and the normal patterns of these indices for established vineyards. For this reason, further research is needed to understand how vineyard characteristics influence remotely sensed vegetation indices, to develop accurate benchmarks for vineyard growth and health assessment.
To address this challenge, this study investigated how vine age, grape variety, vineyard design, and vineyard floor management practices affect vegetation index time series. It focuses on the NDVI, which is widely used in agriculture for its ability to capture crop vigor, as well as the normalized difference infrared index (NDII), which provides complementary information on plant water content and water status. Specifically, this study aimed to determine the relative importance of each factor and compare them to the natural annual variability of remote sensed NDVI time series to determine whether a particular seasonal pattern may be defined by one or more vineyard characteristics. Analyses are based on data from 1014 commercial vineyard blocks in California, for which Sentinel-2 data have been collected for five or more growing seasons and for which detailed information about region, trellis type, grape variety, and vine age is available.
Materials and Methods
Study vineyards
A data set comprising observations from 1014 vineyard blocks distributed across California winegrape growing regions and covering a total area of 11,308 ha was used for the study. Observations for each site included grape variety, trellis system, training system, vine age, row spacing, and in some cases, floor management practices. The study encompassed 50 grape varieties, with 10 primary varieties (Cabernet Sauvignon, Pinot noir, Merlot, Syrah, Zinfandel, Petite Sirah, Malbec, Pinot gris, Chardonnay, and Sauvignon blanc) accounting for 84% of the block plantings.
The vineyards analyzed in this study feature a wide variety of training and trellis systems. The vines are trained with one or two cordons, which could be split either vertically or horizontally. Figure 1 provides an illustration of the types of trellis systems considered in this study. For simplicity, these systems were grouped into five main categories:
Quadrilateral cordon- Vines trained with two horizontally split cordons. The cordons are parallel, spaced 30 to 60 cm apart, and are positioned at a height ranging from 135 to 150 cm. Vine foliage grows freely and sprawls from the cordons into a double curtain, creating a natural canopy structure.
Bilateral cordon- Canopies trained on a single bilateral cordon or vertically divided with two bilateral cordons running in parallel, spaced 30 to 60 cm apart. Vine foliage grows freely and sprawls from the cordons, developing as a single curtain (or superposed double curtain) and creating a natural canopy structure. The first cordon is situated 137 to 157 cm aboveground, and when a second cordon exists, it is positioned between 137 and 168 cm.
Head trained- Head-trained vines cane pruned with vine foliage supported by one catch wire. Trunk height ranges from 76 to 114 cm, with the catch wire positioned at 127 to 165 cm and spaced 51 to 76 cm apart for shoot positioning.
Quadrilateral cordon shoot-positioned (SP)- Horizontally split canopy where two bilateral cordons run in parallel, spaced 30 to 60 cm apart, at a height of 135 to 150 cm. The shoots are held into a “V” shape by one catch wire positioned at a height of 117 to 142 cm. This wire is supported by a trellis structure that may consist of a metal crossbar (T-Stack structure), two metal bars forming a “V” shape, or a single bar shaped as a lyre (Elk-Horn), all mounted on the trellis poles.
Vertical shoot-positioned (VSP)- Canopies trained into a single cordon positioned at a height ranging from 53 to 122 cm. Shoots are held vertically by two or more levels of catch wires separated by less than 64 cm. The fully developed canopy resembles a vertical “wall” of vegetation, with a maximum height ranging from 142 to 218 cm.
Trellis system types included in this study.
Four primary winegrape production regions in California—North Coast, Central Coast, Northern Central Valley, and Southern Central Valley—were included in the study (Figure 2). Northern Central Valley blocks are located in the Delta basin, mostly in the Lodi and Clarksburg American Viticultural Areas (AVAs), and Southern Central Valley blocks are located in the San Joaquin or Tulare basins, around the Madera AVA. An overview of the types of blocks included in each region is included (Table 1).
Map of the California regions considered in this study, based on American Viticultural Areas.
Overview of the four California winegrape production regions and their corresponding vineyard blocks used for this study. VSP, vertical shoot-positioned; SP, shoot positioned.
All vineyards considered in this study were irrigated, and most used an interrow cover crop that persisted from winter to early summer, when it dried out due to lack of sufficient soil moisture. There was a clear distinction in vine capacity and productivity between vineyards located in California’s Southern and Northern Central Valley and those located in the North and Central Coasts. Vineyards in the Central Valley had larger canopies and greater yields than vineyards in the coastal regions. The warm climate and fertile soils support vigorous canopy growth, and vines are often trained to horizontally divided canopies with little to no shoot positioning. As a result, shoots frequently extend into the interrow area and in some cases, cover almost all the vineyard surface. In contrast, vineyards in coastal regions are typically less vigorous and produce lower yields. In these regions, which are cooler than the Central Valley, vineyards are commonly trained to VSP systems. The main characteristics of each growing region are provided in Table 1.
The climate in coastal regions is Mediterranean, with mild, wet winters and warm, dry summers. Temperatures are often moderated by fog and cool ocean breezes. In contrast, the inland valleys have a warmer climate with hot, dry summers and more extreme temperature variations. Climate data during the study period are presented in Figure 3 and cumulative growing degree days from 1 April to 31 Oct (calculated using the Winkler index [Winkler et al. 1974]) are reported in Table 2.
Monthly weather data (average temperature and total precipitation) for the four regions considered in this study. Source: alifornia Irrigation Management Information System (CIMIS; https://cimis.water.ca.gov).
Growing degree days based on the Winkler index for each California winegrape production region and year examined in this study.
Image acquisition and filtering
A JavaScript code (Appendix 1) was used in Google Earth Engine to extract the NDVI and NDII time series for each vineyard block from 2019 to 2023. The script begins by loading all available Sentinel-2 surface reflectance products for the studied regions, after which a two-step filtering process is applied to the Sentinel-2 images. In the first step, cloudy scenes (those with more than 20% cloud cover) were filtered based on the cloudy pixel percentage information stored in the metadata of each image. In the second step, a built-in function (mask- S2clouds) was used to remove cloudy pixels in scenes with less than 20% cloudy pixel percentage. Two vegetation indices (NDVI and NDII) were then calculated for the clear pixels from near infrared (Band 8), red (Band 4), and short-wave infrared (Band 12) bands, as shown in Equations 1 and 2:
The NDVI and NDII layers were then clipped and averaged based on the block boundaries of the vineyards. Only pixels entirely within the boundaries were kept; others were excluded from the calculation. In the final step, the average NDVI and NDII for each vineyard were reported in a tabular database. Time series associated with vineyards that were removed or not yet planted were discarded. Additionally, blocks less than 3-yr-old were excluded from the analysis, except for illustration purposes when specified otherwise. The final data set consisted of a total of 4521 time series.
A final filtering was performed at the block level to remove remaining cloudy data. These outlier data were mostly isolated points with unusually low or unusually high NDVI in time series. To detect them and filter them out, a simple approach was performed. For each NDVI value, three indicators were computed (Figure 4):
d1, the difference between the current NDVI value and the previous NDVI value
d2, the difference between the current NDVI value and the next NDVI value
d3, the difference between the current NDVI value and the linear interpolation between previous and next NDVI values
Representation of the three intermediate values used to compute gap score and filter outlier values. A) Gap score is low and equal to d3; B) gap score is high and equal to d1. NDVI, normalized difference vegetation index.
The gap score was then obtained as the minimum of these three indicators:
Gap score was computed for each NDVI value in every available time series. By defining a threshold for an acceptable gap score, values above the threshold were removed. In this study, the threshold was empirically set at 0.1 to effectively remove remaining anomalous values while preserving all valid observations. This choice was based on the observation that a sudden increase immediately followed by a decrease (or vice-versa) of 0.1 NDVI within three consecutive observations typically reflects factors other than canopy changes in vineyards under normal vegetation dynamics. The average number of available images per block after filtering is provided in Supplemental Table 1.
After filtering, NDVI values were averaged over 2-wk windows to standardize all time series on the same temporal basis. The 2-wk window was chosen to reduce the impact of irregular time steps caused by filtered-out values and to smooth out differences between images acquired by the Sentinel-2A and 2B satellites. Each of these 2-wk average values for specific blocks was then considered as one observation.
Data analysis
For each vineyard, peak canopy NDVI and peak canopy NDII were defined as the median value of all the observations during a 6-wk period between early June and mid-July (week 23 to week 28). This time window was selected to capture the period of maximum canopy development, at a time when interrow cover crops no longer influence vegetation index values. The median was preferred as it offers a robust and representative estimate of vine vigor during peak development, minimizing the influence of outliers and accounting for variability in phenological timing across blocks.
For each trellis type and each AVA region, peak canopy NDVI and NDII density contours were calculated using the R ‘MASS’ package (Ripley et al. 2013). A type III analysis of variance (ANOVA) was performed to analyze the effect of each vineyard characteristic on the peak canopy NDVI of each block, using the R ‘car’ package (Fox et al. 2012). Type III ANOVA was chosen due to the unbalanced nature of the data set, as it allows for an unbiased estimation of main effects regardless of factor ordering. Although interactions were initially considered, they were not retained due to limited interpretability and insufficient representation of certain factor combinations.
The ‘randomForest’ package (Liaw and Wiener 2002) in R was used to fit a random forest model with default parameters (500 trees and two variables considered at each split), predicting median late-season NDVI values based on vineyard characteristics. Although randomForest models aim to minimize prediction error and can sometimes assign high importance to spurious variables, this approach was used to assess the relative influence of each input variable within a data-driven framework that captures potential nonlinear interactions. The relative influence of each input variable was computed using measures such as Mean Decrease in Accuracy and Mean Decrease in Gini to rank the contribution of each variable. The importance of each feature is reported as the percentage of total importance.
Results
The distribution of NDVI and NDII values across vineyards, categorized by their trellis types and geographic locations, is shown (Figure 5). Each zone within the figure represents the kernel of the distribution, containing 75% of the vineyards from a specific geographic area or trellis type. Overall, NDVI and NDII values exhibit a clear correlation; blocks with higher NDVI values tend to also have higher NDII values, and vice-versa. Although NDVI and NDII measure different block properties, this correlation is logical because during the peak canopy period most of the water present in the scene is stored in the plant tissue, therefore water content and vegetation amount are closely correlated.
Density contours of observed normalized difference vegetation index (NDVI) and normalized difference infrared index (NDII) values by trellis type (A) and geographic area (B). NDVI and NDII are median values over a 6-wk time frame from early June to mid-July. Each shape represents the density kernel containing 75% of the vineyards of the group. SP, shoot positioned; VSP, vertical shoot-positioned.
Among the different trellis types (Figure 5A), VSP is associated with the lowest NDVI and NDII values, ranging from 0.3 to 0.6 and from 0.05 to 0.35, respectively. Head-trained, bilateral cordons and quadrilateral cordons with shoot positioning follow and span a broader range of values. Quadrilateral cordons are associated with the highest NDVI and NDII values, reaching up to 0.88 and 0.65, respectively.
Regarding growing regions (Figure 5B), a clear distinction is observed between the coastal (North Coast and Central Coast) and valley regions (Northern and Southern Central Valleys). The coastal regions exhibit relatively low NDVI and NDII values, whereas vineyards in the valley, which are more vigorous and produce greater yields, have larger canopies with higher NDVI and NDII values reaching up to 0.85 and 0.65, respectively. More broadly, the distribution for the Central Coast matches that of the VSP trellis type, and the distribution for Southern Central Valley and the rest of the valley matches that of the quadrilateral and bilateral cordon trellises, which are used predominantly in these areas.
Sentinel-2 NDVI time series averages are presented by grape variety, training and trellis system, and geographic area (Figure 6). Vineyards in the coastal regions are predominantly managed with shoot positioning (VSP, headtrained vines or quadrilateral cordon SP). On the other hand, Central Valley vineyards are more frequently associated with quadrilateral or bilateral cordon and sprawling canopies. NDVI time series are characterized by two main phases: an initial growth phase corresponding to vine canopy development lasting until June or July, followed by a plateau phase during which fruit develops and canopy size remains relatively stable. For coastal regions, the growth phase is not readily observable due to the interference of weeds and cover crops. Interestingly, for several varieties the mean NDVI during this growth phase even surpasses that of the summer period, when the vegetation between the vine rows dries. This is the case of most canopy types in coastal viticultural areas, while in other areas these two phases are more distinct. Despite the widespread presence of early-season cover crop and untilled soil in the vineyards of the Northern and Southern Central Valleys, vines in these locations achieve much higher midseason NDVI values that stand out from those potentially mixed with the cover crop signal. The timing at which the canopy reaches its maximum extent can also be noted, varying from mid-June in the Central Valley to the start of July along the coast, where the climate is more temperate. The end of season plateau value also varies, with the main influencing factor being the type of training and trellising systems employed. Values associated with VSP and head-trained systems are linked to lower plateau values (between 0.4 and 0.6), whereas vineyards with bilateral or quadrilateral cordons predominantly exhibit values between 0.6 and 0.8.
Normalized difference vegetation index (NDVI) time series averaged by variety, trellis system, and main area in California, from mid-March to mid-October. The 10 main varieties are represented (eight red and two white); other red and white are aggregated. Line width represents the number of time series averaged. SP, shoot positioned; VSP, vertical shoot-positioned.
Figure 7 illustrates how factors other than geographic location, grape variety, or trellis system affect the NDVI time series. To capture the effects of these individual factors, the most represented grape variety (Cabernet Sauvignon) was selected, along with region–trellis combinations that provided a broad range of values for the tested variables. Among them, vine age appears to be a significant determinant of NDVI values (Figure 7A). Vineyards less than 3 yr of age with developing canopies exhibit lower peak NDVI values. Vines reach peak vigor at 5 to 6 yr of age, after which NDVI values gradually decline.
Normalized difference vegetation index (NDVI) time series for Cabernet Sauvignon in two California geographic areas (Lodi [A, B] and the North Coast [C, D]) with two specific trellis systems (bilateral cordon [A, B] and vertical shoot-positioned [C, D]). Results are averaged by A) age of the vines, B) weeds and cover crop management, C) row spacing, and D) year.
The early-season time series are significantly influenced by the presence of cover crops, weeds, or both, and by the approach to the resulting vegetation cover management (as shown in Figure 7B). On available data, a tilled soil corresponds to NDVI values ~0.25, but with weeds or cover values, early NDVI can increase to 0.5 or even to 0.6 in some areas. This pattern fades between June and July as the interrow vegetation dries out.
Plant density also influences the NDVI values observed in mixed pixels from Sentinel-2 data (Figure 7C). Within the same trellis system, a decrease in row spacing, leading to a higher number of vines per unit area, results in higher NDVI values. This is attributed to the increased vine-to-soil ratio (green fractional cover) as the distance between rows decreases.
Furthermore, NDVI values show substantial year-to-year variability (see Figure 7D). Years with sufficient rainfall in winter, such as 2019 and 2023 (Figure 3), provide conditions favorable for vine canopy development, while other years with low off-season precipitation, such as 2021, are associated with significantly reduced NDVI values.
A comprehensive ANOVA (including all available locations) of factors affecting peak canopy NDVI is presented (Table 3). The analysis reveals significant effects from all considered variables, including AVA, trellis type, grape variety, year, age, and row spacing. Due to the size of the data set, the ANOVA can confidently detect even negligible effects; all p values are extremely low (1 × 10-9 or lower), indicating statistical significance for all factors. However, by looking at the F value, their relative importance varies. The year effect emerges as the most substantial, followed by region, age, and trellis type. Grape variety and row spacing show only a marginal influence on peak canopy NDVI.
Analysis of variance of the factors affecting peak canopy normalized difference vegetation index, based on all available time series. Sum sq, sum of squares; DF, degrees of freedom; AVA, American Viticultural Area.
A random forest model predicting peak canopy NDVI values based on vineyard data explains up to 77.4% of the variability (result not shown), compared to 63.0% for ANOVA. The relative importance of variables, as determined by the random forest, highlights the dominant role of the region (31%), followed by trellis type (21%), year (14%), row spacing (14%), vineyard age (12%), and grape variety (8%).
In addition, the same ANOVA was applied to one area at a time. In the Central Valley, trellis type stands out as the dominant factor affecting peak canopy NDVI, significantly more so than other variables (Table 4). The year effect is also notable, while other factors contribute marginally.
Analysis of variance of the factors affecting peak canopy normalized difference vegetation index in the Central Valley of California. Sum sq, sum of squares; DF, degrees of freedom.
For the coastal region of California, some of the general trends that were observed in Table 3 are echoed in the data presented in Table 5. The year effect is again dominant, followed by age and trellis type. Grape variety and row spacing have weaker effects, similar to the data shown in Table 3. This similarity suggests that despite the comparable vineyard areas between the valley and the coast, the higher number of blocks along the coast create parallels with the general ANOVA results shown in Table 3.
Analysis of variance of the factors affecting peak canopy normalized difference vegetation index in coastal California. Sum sq, sum of squares; DF, degrees of freedom.
Discussion
Trellis systems and geographic location significantly influence peak canopy NDVI (Figure 5), findings supported by the ANOVA results (Tables 3 to 5). This effect is particularly pronounced in the Central Valley (Table 4), where trellis systems strongly influence canopy architecture. Different trellis systems alter the spatial distribution and density of vine foliage, which affects light reflection and absorption by the canopy. VSP systems create vertically extended canopies, forming a wall of leaves. In contrast, quadrilateral or bilateral cordon trellis systems distribute the canopy horizontally, increasing ground coverage from an aerial perspective. These horizontally-oriented canopies not only cover a larger area, but due to their architecture, progressively extend into the interrow space as the canopy grows. This results in a gradual increase in the vegetation signal over the interrow or mowed cover, a behavior that contrasts with VSP systems.
With VSP trellises, canopy porosity decreases as the vine develops, leading to saturation in NDVI values at specific growth thresholds. This saturation effect, well documented in the literature, occurs because NDVI becomes insensitive to additional layers of leaves once the canopy reaches a certain density. For instance, similar saturation effects were reported in fully developed canopies with a leaf area index exceeding 2, beyond which NDVI no longer responded to further increases in leaf layers (Gamon et al. 1995). In addition, canopies associated with VSP trellises tend to occupy a smaller proportion of the mixed pixels in satellite images; VSPs leave more room for the ground, reducing the canopy signal captured from above. The flat or even declining NDVI trends observed early in the season for VSP systems (Figure 6) can also be explained by early-season growth dynamics, where cover crops or weeds may produce higher NDVI values, obscuring the signal from the vine canopy. Although the early-season canopy development could provide useful information for vineyard management, this is contingent on the absence of other vegetation in the scene, such as in vegetation-free interrow soils. This suggests that satellite-derived vegetation indices may have limited utility for VSP, particularly during peak canopy conditions. Ground-based proximal sensing tools may offer more accurate and actionable information for vineyard managers. Adjusting the image acquisition angle or using new hyperspectral sensors that can distinguish between the canopy and the cover crop spectral signatures could also resolve this issue and improve canopy monitoring, though their generalized use still remains a challenge because of cost and complexity of data.
The effects of vine age, row spacing, interrow vegetation, and annual variability exemplified in Figure 7 can be observed in Figure 6 through the range of values of peak NDVI. The effect of interrow vegetation is particularly evident early in the season, where NDVI values can become significant, even surpassing the peak NDVI of vines observed in June and July in coastal regions. The presence of cover crops can hinder the effectiveness of low spatial resolution detection tools, as the interrow space may contribute more to the photosynthetically active biomass than the vines themselves. This phenomenon is influenced by differences in soil types and precipitation patterns. In the Central Valley, soils generally have lower water retention capacity (sandy or sandy-loam) and receive less winter and spring precipitation compared to the coastal regions. Consequently, the water available to cover crops is limited in both duration and quantity, causing them to dry out or be mowed earlier. Due to insufficient data, the ANOVA does not quantify the impact of interrow vegetation; however, this effect has been documented in the scientific literature (Matese et al. 2015, Khaliq et al. 2019, Abubakar et al. 2023, Williams et al. 2024). The ANOVA does confirm the significance of the remaining variables (Tables 3 to 5). Annual variability emerges as a primary driver of NDVI variability, as highlighted by the general ANOVA (Table 3). This factor is particularly pronounced in the coastal regions (Table 5), where viticultural practices are quality-oriented and heavily influenced by vintage years. In contrast, the Central Valley (Table 4) shows less sensitivity to annual variability due to its yield-focused management systems, which rely on extensive irrigation and consistently warm climates. The strong influence of age on NDVI shown in Figure 6 may relate to the presence of diseased, dead, or missing vines in older vineyards.
The ANOVA results also indicate that other vineyard characteristics, such as grape variety and row spacing, have less pronounced but still statistically significant impacts on peak NDVI. Although grape variety does not play a major role in peak canopy NDVI, it might affect the timing of peak canopy development. This aspect could be a subject for further investigation. Similarly, factors such as age and AVA region also influence peak canopy timing, with coastal areas reaching peak NDVI 2 wk later than the Central Valley. Notably, Southern Central Valley NDVI values also peak slightly earlier than those in the Northern Central Valley. This analysis is challenging due to the inherent noise in NDVI time series, making it difficult to accurately detect peak dates for individual vineyard blocks.
Although this study provides valuable insights into the factors influencing peak canopy NDVI in vineyards, several limitations and considerations must be acknowledged. While the data set well represents commercial vineyards across California, it may not fully capture the diversity of viticultural practices in specific regions such as Northern California and the South Coast. Additionally, despite accounting for numerous factors, other influential parameters (e.g., fertilization practices and exposure to pests and diseases) that were not considered in this analysis could significantly affect NDVI time series. Given the size of the study area and the number of blocks observed, block properties and floor management practices were verified by many collaborators. Observer effects or omissions in updating certain data could marginally affect the results. Moreover, since grapevines are a permanent crop, their cumulative growth and development history should ideally be accounted for to provide a comprehensive understanding of NDVI variations.
Overall, this study highlights the complexity and significant variability of NDVI values observed across vineyards. This is reflected in the relatively modest predictive performance of the models used, which remained intentionally simple and were not fully optimized in terms of tuning and training. Further research should address the challenge of predicting expected NDVI values based on vineyard characteristics and management practices, by using more advanced and tailored methodological approaches. By enabling the comparison of expected vineyard performance to observed behavior, this approach facilitates the detection of abnormal events and stress, allowing for timely interventions to address issues and optimize vineyard health and productivity. The integration of remote sensing data with detailed vineyard management practices represents a promising pathway toward advancing precision and sustainability in viticulture (Bellvert et al. 2016, Devaux et al. 2019). Future studies could also investigate the applicability of these findings to other remote sensing platforms such as aircraft or unmanned aerial vehicle imagery (Sozzi et al. 2020).
Conclusion
This study highlights the variability of NDVI time series derived from Sentinel-2 imagery across California vineyards, leveraging a large data set covering more than 10,000 ha. It demonstrates how location, training system, and other vineyard characteristics explain variations in vegetation indices and canopy size. Analyses reveal a clear distinction between California vineyards grown in the coastal regions and those grown in the Central Valley. This difference is driven by climate and vineyard management practices, particularly the training and trellis systems used, which reflect different inherent levels of vine vigor and productivity. In both cases, significant annual variability is observed, underscoring the persistent vintage effect. Despite this variability, the effects of vineyard age, planting density, and other factors remain significant. Although satellite imagery and derived spectral indices are now widely accessible to growers, actionable tools remain limited due to the absence of benchmark values for the diverse vineyard conditions. Overall, this work illustrates the key factors that must be considered when using vegetation indices across heterogeneous vineyards and provides guidance on reference values and characteristic seasonal trends for different vineyard types, which may serve as a foundation for the development of decision-support tools at larger scales through remote sensing technologies.
Supplemental Data
The following supplemental materials are available for this article in the Supplemental tab above:
Supplemental Table 1 Average number of images after filtering, per block, for each year and winegrape production region examined in this study.
Appendix 1 JavaScript code used to extract and filter Sentinel data from Google Earth Engine.
Data Availability
Some data underlying this study cannot be shared publicly. However, the remaining data are available on request from the corresponding author.
Footnotes
The authors thank GALLO vineyard managers for providing vineyard characteristic data, and GALLO research assistants and interns for collecting or validating data from hundreds of vineyard blocks across California. This work would not have been possible without their contribution. The authors declare that they have no conflicts of interest.
Oger B, Aboutalebi M, Dokoozlian N, Sanchez L and Mar Alsina M. 2025. Vineyard design, vine age, and floor management practices affect Sentinel-2 NDVI time series analysis of California vineyards. Am J Enol Vitic 76:0760023. DOI: 10.5344/ajev.2025.25010
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 February 2025.
- Accepted July 2025.
- Published online September 2025
This is an open access article distributed under the CC BY 4.0 license.
References
Vol 76 Issue 2
Jump to section
Related Articles
Cited By...
More from this TOC section
Similar Articles
