Precision viticulture (PV) refers to the management of the variability within a vineyard, resulting in greater consistency with respect to factors such as yield and quality [1]. PV includes observing the factors responsible for any spatial variation and determining the potential for zoning or managing the variation [1]. Yield and grape quality are influenced by various physiological and environmental factors in the viticulture system such as canopy size, disease pressure, soil type, microclimate, sun exposure, nutrition availability, and water status [2,3,4,5]. Previous research in PV has aimed to identify the spatial variation of these factors in a single vineyard block, such that if the spatial variation can be identified, grapes can be managed differently in these subfield regions to maximize fruit quality and/or vine health or possibly harvest fruit into different lots based on this spatial variation [6,7].

In the environment, natural systems display periodic or structured variation in time and space (i.e., temporal or spatial dependence). The viticulture system is yet another example where spatial patterns can develop [8]. The quality and production of grapes can be significantly affected by the spatial variation of vineyard blocks [9,10,11]. In the Niagara grape-growing region of Ontario, Canada, significant glacial activity has created soils with a high degree of variability [12,13]. The water status of grapevines can be directly impacted by differences in soil hydrology in vineyard soils [14]. A drought can negatively affect the growth of individual vines and their photosynthesis [15,16] while an excessive amount of water can lead to a greater risk of disease [17], poor root growth [18], and leaching nutrients from the soil [19]. Vineyard soil moisture (SM) can differ significantly within a vineyard block [20] and differences in vine water status can be seen throughout the growing season [21]. Different soil types and vine vegetative growth impact the variations seen in vineyard water stress [2,22]. In addition to water stress, other metrics of plant health including chlorophyll levels can be affected by factors such as plant life cycle, air contamination, nutrient status, and pathogens [23]. Differences in vegetation size and structure can be attributed to many factors such as nutrient deficiency [24], water status [25], and disease infection [26].

Another stress facing plants in cool-climate grape-growing regions is cold weather exposure during dormant periods. Grapevines can vary in their winter hardiness, which refers to the cold acclimation process that is triggered by a reduced photoperiod and lower temperatures [27,28]. Winter injury may occur when the temperatures drop below the hardiness level of the vine. Depending on the severity of damage, the vine may be impaired in vegetative and reproductive growth, and suffer increased disease pressure as well as reduced yield [29,30,31]. Understanding the spatial variability of winter hardiness can lead to targeting management strategies to less winter-hardy areas, maximizing the effectiveness and response of the vineyard.

Another important source of vineyard variation is virus infection. Many vineyards around the world suffer economic losses due to virus infections [32]. Grapevine leafroll virus (GLRaV-3) is one such virus that commonly affects grape production in nearly every major grape-growing region [32,33]. According to a study conducted in Ontario, grapevine rupestris stem pitting-associated virus (GRSPaV) was the predominant virus in Ontario vineyard blocks, followed by GLRaV-3, with close to 50% overall infection [34]. There is clear evidence that GLRaV-3 negatively impacts the productivity and grape quality of the infected vines [35,36,37,38]. Grapevine red blotch-associated virus (GRBaV) is another common and detrimental virus in North America [39]. A recent study revealed its detrimental impacts on vine growth and grape quality, resulting in reduced vine size and yield, and reduced sugar, pigment, tannin, and yeast assimilable nitrogen in the berries while increasing berry pH and titratable acidity [40]. In most cases, grapevine viruses are detected in infected leaf tissue by amplifying the viral genome sequence with polymerase chain reaction (PCR) [41]. However, several diagnostic challenges exist with PCR, including that infected vines have uneven symptom distribution, so not all leaves may contain the virus at the same level; new infections are low in virus titre; and multiple viruses can infect grapevines at high rates, requiring multiple tests to detect all viruses present [33,34]. Sampling grapevines and testing is also time consuming and expensive. Since effective control and treatments for these grapevine viruses has not been developed yet, recommendations are to remove infected vines before they act as a source of transmission in the vineyard [42]. Thus, there is an urgent demand to develop an efficient method for the rapid and early detection of viruses.

Conventional methods of detecting vineyard variability involve direct ground surveys by well-trained personnel or specialized equipment. Limitations associated with these methods range from being time consuming and labour intensive, to requiring elaborate field and laboratory procedures and expertise, and differences in measurement accuracies depending on the variable under investigation [24,43]. For these reasons, conventional detection methods of vineyard variability are often not viable for processing large numbers of plant samples due to the time and labour involved in their proper execution [43]. Despite the utility of traditional ground surveys, there remains a high demand for an efficient method for rapid detection of plant stress [44].

A solution to this demand could be the use of remote sensing technologies. Remote sensing technologies are used to measure electromagnetic energy reflected by a surface, such as vegetation or soil, to obtain information about the object [45]. The majority of remote sensing applications involve measuring the reflected radiation compared to the transmitted or absorbed radiation from a plant’s surface [46]; however, plant leaves can also emit energy via fluorescence or expulsion of heat [47]. Photosynthesizing plants require external energy in the form of sunlight to execute their photosynthetic activity. Biological pigments absorb energy from photons of sunlight, which carry energy proportional to their radiation frequency. These pigment molecules are responsible for the colour that we visually perceive from plants, as they interact with specific wavelengths in the visible range of the electromagnetic spectrum, absorbing particular wavelengths for energy harvesting and reflecting all others [48,49,50]. Due to the role of pigment molecules in plant photosynthesis and productivity, pigment content in leaves is a good indicator of plant health and photosynthetic activity [48]. It has been demonstrated that the photosynthetic capacity of a plant is impacted by abiotic stresses such as water stress and nutrient availability [15,48,51], and that the quantities and ratios of pigments in a leaf can be altered in response to changes in photosynthetic capacity [49]. Stresses such as water, nutrients, and viral infection have demonstrated physical symptoms associated with changes in leaf colour and patterns, indicative of changes in their pigment content. These changes would further impact the specific wavelengths of light being absorbed and utilized by the plant, and those being reflected [52,53,54,55,56]. There have been several techniques developed to identify changes in plant health using wavebands composed of EM spectrum data [57]. As an example, the normalized difference vegetation index (NDVI) is based on calculations transferring individual wavelength data into the ratio between near-infrared and red reflectance in each band [58]: NDVI = [(near infrared) − (red)]/[(near infrared) + (red)]. Numerous other remote sensing indices can be calculated from the green, red, red-edge, and NIR regions of EM reflectance to detect vegetative growth and potential vine stress including that due to water, nutrients, and disease [59,60,61,62]. In addition, a thermal sensor is a useful tool in precision viticulture for investigating plant stress levels, as it measures the heat released by plants, primarily to detect water stress as leaf temperature increases when water stress increases [63]. The use of remote sensing thermal imagery in deficit irrigation has also been investigated [64]. Using thermal imagery, it was possible to discern differences in leaf water potential resulting from different irrigation treatments and acquisitions taken under full canopy growth and dry conditions were most effective for monitoring deficit irrigation [65].

Remotely piloted aircraft system (RPAS) flights at low altitude allow the collection of remote sensing imagery at high spatial resolutions, in the range of one centimetre/pixel [66]. Several promising approaches to mapping leaf water potential [25,67] and vegetative growth [68] have been suggested. Vineyard canopy characterization by remote sensing is an improvement over manual characterization of the canopy, ensuring high levels of accuracy, efficiency, and reliability [1,24]. The existing research shows that remote sensing can detect vineyard variability, but there is still a limited literature on its ability to detect certain variables, such as winter hardiness and virus infections. The correlation of remote sensing data to crop quality has been shown to vary across vineyards and vintages; thus, research on different varieties and site-specific methodologies should be investigated [69,70,71,72].

The primary objective of this research was to evaluate remote sensing data including several vegetation indices and thermal data collected from RPAS flights to improve vineyard management. To accomplish this objective, remote sensing data along with measured viticultural data (vine performance and health) were used to determine vineyard spatial/temporal variability and to elucidate relationships between variables.

Remote sensing data can identify vineyard variation with temporal stability for certain viticultural variables including water status, virus titre, and canopy size. Even though some sites displayed highly patchy or striped spatial patterns, temporal consistency of spatial patterns was observed in the maps of the remotely sensed NDVI in most of the vineyards, as well as the spatial patterns of vine size and soil moisture, over time.

The relationships between thermal data and many variables (SM, Gs, and LT50) did not show temporal stability or consistent relationships across sites. However, half of the research sites indicated a positive correlation between thermal data and leaf ψ. As previously shown in the literature, thermal imagery can be used to correlate with vine water status since it directly measures canopy temperature, which is related to stomatal closure [74]. Interestingly, four of six sites in 2016, which was a particularly hot and dry growing season, showed inverse correlations between thermal data and vine size, while only two sites in 2015, a year with ample precipitation, showed the same correlation (Table 2). This could demonstrate that higher grapevine water stress is a constraint for vegetation in drought-like growing seasons [2].

The NDVI was most consistently correlated with vine size at all six sites in both years. This is consistent with the literature that reports the efficacy of remote sensing data for detecting vegetative growth such as leaf area coefficient [75], total canopy [76], and the health of vegetation [2]. Plant canopies are largely influenced by the environment and stresses [77]. Thus, grapevine condition and level of stress may be reflected in variations in vegetative growth. Additionally, yield-to-canopy size ratios are often used to diagnose fruit quality indirectly based on balanced pruning [78]. By using NDVI data, grape growers can determine variations in vegetative growth to support them in monitoring vineyard balance [79]. A precise projection of canopy size may allow vine vegetative and reproductive balance to be altered through cluster thinning during the growing season and afterward through pruning tactics [80].

Another interesting observation from this research was an inverse correlation between NDVI and soil moisture (SM). Remote sensing could detect variations in SM at three out of the six sites and showed temporal stability throughout the years. SM values tended to be higher at these sites and there was visual evidence of water accumulation in some spots, mainly caused by heavy clay soil with insufficient drainage. Interestingly, the NDVI was positively correlated with SM in 2015 and 2016 at the site 4 vineyard, located in the Lincoln Lakeshore sub-appellation, with relatively well-drained soil types (glacial till soils with high sand and stone content [81]). There is a possibility of using remote sensing to detect SM for sites where water drainage issues or known high clay content [82] can negatively impact vegetative growth [18,19]. There were fewer sites that showed a consistent correlation between NDVI and leaf ψ; they exhibited a consistent positive correlation at only two sites. With leaf conductance (Gs), the relationship with NDVI was also inconsistent, with three sites displaying positive correlations in 2016, but not showing any stability over time. To examine the feasibility of using remote sensing data for vineyard water management, it is necessary not only to establish a correlation between remote sensing data and vineyard water stress but also to survey vineyards’ soil profiles and their soil drainage capacities. Another variable showing an inconsistent relationship with NDVI was winter hardiness, exhibiting a negative correlation, with temporal stability only at site 3. Three out of the six sites showed significant negative correlations (95% confidence interval, p ≤ 0.05) between NDVI and LT50 in 2016, which had a colder dormant season (Appendix A Figure A1), while two vineyards had a correlation in 2015. More winter-hardy vines with low LT50 values were observed at all six vineyard sites in 2016. Water status [83] and pruning weight [84] can influence cold acclimation, but the lack of relationships observed in this research could be explained by the low vineyard variations in LT50 values. There could also be an impact of the dormant season minimum temperature on its detection by remote sensing.

Interestingly, site 3 was the only site where the NDVI was correlated to all measured viticulturally important variables, with these relationships displaying high temporal stability over the two years of study. There were three distinct cultural practices at site 3 that differed from other vineyards. First, the inter-row management at site 3 was topsoil cultivation while the other sites planted cover crops between the rows. The inter-row spectral reflection in the data was not removed for this study, so the NDVI data extracted from the vineyards could have been affected by this reflection [85,86,87]. Secondly, pruning practices and the training system in this vineyard were distinct from other vineyard sites. A canopy-based NDVI measurement can be greatly affected by cultural practices such as pruning and training systems [88]. The spur-pruned cordon systems with stronger horizontal profiles in this site could be more appropriate for remote sensing with an RPAS. There would need to be further research on the effect of training systems on the performance of remote sensing data. Lastly, the application of heavy copper spray was observed through the growing season at site 3. The heavy copper spray created a blue residue on the leaves when the RPAS flight was conducted to obtain the NDVI data. There should be more research on the impact of spray regimes on the performance and timing of remote sensing data collection.

When evaluating the possibility of applying remote sensing data to measure vineyard variability, the development of maps with reliable spatial patterns and potential zonal delineation is essential. In the Moran’s I analysis, viticulturally significant variables and remote sensing data were highly clustered, and zonal vineyard management such as selective shoot thinning or precise spraying based on the clustering across the vineyards may be feasible [1].

The production and quality of grapes are often negatively affected by virus infections in major grape-growing regions. In the absence of effective control or treatment for grapevine viruses, it is urgent to develop a rapid and early detection method, such as using remote sensing data, to detect virus infection. This research confirmed that the remote sensing NIR and red-edge peaks based on RPAS’s multispectral sensor data positively correlated with GLRaV-3 and GRBV infection. The two notable reflectance increases from the virus-symptomatic leaves in the red-edge and NIR spectra can possibly be explained by disorganization of the structure of the mesophyll cell by the accumulation of anthocyanins and carbohydrates from the plant defence response [89]. Internal leaf structure controls reflectance in the red-edge and NIR peaks and the leaf reflectance emerges from the mesophyll cells due to the orientation of the cell walls and the differences in the refractive cell wall and air of the intercellular spaces [90,91]. The cell reorientation results in an increase in the cell wall–air interface and multiple scatterings of radiation and leads to a higher reflectance level in the red-edge and NIR peaks [92,93].

This study demonstrates that the remotely sensed NDVI from an RPAS could be used to detect the variability within a vineyard for several agriculturally relevant variables. The variable that was most correlated with remotely sensed NDVI, and the most temporally stable, was vine size. The NDVI data can be used to project canopy size accurately and help growers determine an ideal balance of plant vegetation and reproduction. A meaningful inverse correlation with temporal stability between NDVI and soil moisture was also observed for multiple sites. There was evidence at these sites of considerable standing water in areas with inadequate water drainage, indicating that SM can be detected by remote sensing in areas with drainage concerns. Furthermore, only one site showed perfect correlation of the NDVI and all the viticulturally important variables with temporal stability. There were some distinct cultural practices at the site that differed from other vineyards; these may impact the usefulness of remote sensing. Further investigation of the efficacy of remote sensing in PV across vineyard cultural practices will promote accuracy and usefulness of remote sensing data. This research also demonstrated that the red-edge and NIR data had the highest correlation with virus infections, which may indicate their usefulness in confirming virus-infected vines. Future research on these indices for early virus detection with a multispectral sensor should be explored. Overall, the biological relevance of remote sensing data per site and per vintage should still be evaluated through ground truthing of sampled vines. In many cases, it was difficult to use remote sensing data to detect agriculturally significant variables across all sites and vintages.

We acknowledge cooperation and fruit donation from our collaborating vineyards in Niagara wine regions. We also acknowledge OMAFRA’s New Directions Program for providing funds for the project.

L.L.: conceptualization, data curation, formal analysis, project administration, writing—original draft, writing—review and editing. A.R.: conceptualization, methodology, project administration, resources, writing—review and editing. B.D.: data curation, writing—review and editing. A.S.: data curation, software. All authors have read and agreed to the published version of the manuscript.

The data presented in this study are available on request from the corresponding author because the data involves the geo-referenced site information of private landowners.

Author Leeko Lee was employed by the company Bluebottle Butterfly Inc. Author Adam Shemrock was employed by the company AirTech UAV Solutions Inc. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

This research was funded by Ontario Ministry of Agriculture, Food and Rural Affairs (OMAFRA)’s New Directions Program, grant number: ND2014-2456.