Relation of Ground-Sensor Canopy Reflectance to Biomass Production and Grape Color in Two Merlot Vineyards

Site description.

This study was undertaken in two commercial Merlot vineyard blocks located in Goumenisa (northern Greece) in the summer of 2003 and 2004. Vines were trained on a bilateral cordon with two fixed pairs of foliage wires and were spur-pruned. Vine spacing was 2.2 m between rows and 1.3 m within rows. The two fields differed in soil properties and topographic features. The Tzianas field had a coarser soil with greater inclination and associated erosion problems than the Kalvakis field (Table 1⇓). As a result, root growth was restricted in the top 30 cm of the soil in the upland positions of the Tzianas field. These differences appear to be, to a large extent, responsible for differences in productivity between the two fields. Vines continued to grow postveraison in the Kalvakis vineyard, whereas shoot growth ceased before veraison in the Tzianas vineyard. Because of vigorous and prolonged growth, Kalvakis vines were subjected to multiple toppings even after postveraison that removed a portion of annual vegetation. Furthermore, Kalvakis soil was not tilled from 2003 onward in an attempt to enhance grape quality by controlling vine vigor through weed competition for water and nutrients. In the Tzianas vineyard, a single topping took place just before veraison (July) and the soil was surface tilled for weed removal.

Fertilizer was applied before the growing season in February or March. In accordance with common practices by the producers, both fields received the same amount of N (31.5 kg ha⁻¹), K (75 kg ha⁻¹), and Mg (25 kg ha⁻¹) throughout in the form of ammonium sulfate and K-Mg sulfate in 2003. The rate of N, P, and K application was increased at the eroded top of the Tzianas vineyard in 2004, whereas the respective rates remained more or less the same to those of 2003 in the Kalvakis vineyard. Other management practices were similar in the two vineyards in terms of planting, irrigation, and plant protection products (Table 1⇑). However, Tzianas grapes were heavily infested by downy mildew (Plasmopara viticola L.) in the second year of the experiment.

Rainfall occurred early in the growing season (April to May) with increased incidence in the second year. As the growing season progressed, the weather was steadily hot and dry. During this period the vines depended solely on drip irrigation for their water supply.

Field sampling and analysis.

Within each vineyard, 32 sampling positions were randomly selected in 2003 to represent the entire vineyard and each position consisted of four consecutive vines along the rows. The number of sampling positions was reduced in the second year of the experiment (2004) by retaining the first 25. At harvest, all bunches per sampling positions were collected and weighed. From those, 200 berries or 2 berries per bunch from each sampling position were selected, sealed in plastic bags, and stored in a freezer (−18°C) until analysis. Total phenolic (% absorption) and anthocyanin content (mg/L) of grapes were measured following Glories (1978) and Ribéreau-Gayon and Stonestreet (1965), respectively. Cane number and pruning weight per vine (2003) or per 4 consecutive vines (2004) were measured from the same sampling positions in December of both years.

The multispectral readings of the vine canopy were taken along the rows and in the sampling positions of the two fields at veraison (August) in 2003 and 2004 and postveraison (September) in 2004. Multispectral Crop Circle (Holland Scientific, Lincoln, NE) sensors were mounted in front of a tractor vehicle (Figure 1⇓), which traveled forward at a constant speed of 3.5 km h⁻¹ and measured reflectance integrated over intervals of 250 msec at midday over clear skies at four wavelengths (blue at 460 ± 10 nm, green at 550 ± 10 nm, red at 680 ± 10 nm, and NIR at 800 ± 65 nm). In 2003, one of the sensors viewed the top of the vine canopy vertically from a distance of ~0.5 m and a 0.25 m diam field of view. Another sensor had the same viewing angle, but with a rectangular mask over the optics that reduced the field of view by ~50%. Canopy reflectance measurements with the vertical field of view (masked sensor) were repeated in 2004 and were compared to measurements with an oblique side view of the canopy from the same distance. A third up-facing sensor was electronically coupled with the down-facing sensors in order to compensate for changes in irradiance. In a similar manner, a differential GPS (Ag 114 Trimble receiver, Sunnyvale, CA) was coupled to provide coordinates for the sensor readings with a precision of ±50 cm (Figure 1⇓). Data were recorded in a portable PC located in the interior of the vehicle. Sensor canopy reflectance was later extracted by aligning with the coordinates of each vine within each sampling position. In cases where NIR values were less than 0.35, they were interpreted as being dominated by soil reflectance and thus were removed from the data set. The normalized difference vegetation index (NDVI) was computed as NDVI = (NIR – red) / (NIR + red) and the wide dynamic range vegetation index (WDRVI) was computed as WDRVI = [(NIR*0.1) – red] / [(NIR*0.1) + red].

• Download figure
• Open in new tab

Figure 1

Mobile system configuration with two downward-facing sensors, one upward-facing sensor, and a differential GPS.

Statistical analysis.

Data analysis included analysis of variance (general linear models) and correlation analysis. The fixed effects of the ANOVA model were vineyard, landscape position, year, and their interactions. Samples within landscape position and vineyard were used as a random effect. The LSD test was used to detect differences between means of the fixed effects at p < 0.05. All employed procedures are reported in the Statistical Analysis System (version 6, 1990; SAS Institute, Cary, NC).

Plant differences between vineyards.

The two vineyards differed in terms of plant development and grape properties. Annual biomass production, as expressed in terms of pruning weight, was affected by all main effects of the experiment (Table 2⇓). Pruning weight was significantly greater in the Kalvakis vineyard in both growing seasons (Table 3⇓). The multiple toppings at the vigorous Kalvakis vineyard, which removed a portion of annual vegetation, probably led to an underestimation of total growth. Underestimation of growth would result not only from the cutting of vigorous-growing shoots but also from the emerging laterals that did not have time to mature and dropped off before pruning weight was recorded in the winter. In the Tzianas vineyard, biomass production increased in the second year because of increased fertilization. Topographic analysis of the data indicated that biomass production was significantly lower in the eroded upland positions compared with the other areas of the vineyard during both years of the experiment (Figure 2⇓). Deeper soils, such as those in the footslope positions of the Tzianas vineyard, generally have greater water storage capacity and root growth, which in turn supports increased vine vigor (Smart and Robinson 1991). In the Kalvakis vineyard, biomass production was similar in all landscape positions except in the footslope positions where pruning weights were lower in the first growing season (Figure 2⇓). The spatially more homogeneous growth within the Kalvakis vineyard is explained by the absence of severe erosion problems (Table 1⇑).

• Download figure
• Open in new tab

Figure 2

Annual biomass production (mean + SE) of the Tzianas and Kalvakis vineyards by landscape position and growing season.

Total phenol and anthocyanin content of grapes varied between vineyards and growing seasons, but not between topographic positions within each vineyard (Table 2⇑). Phenol and anthocyanin content of grapes was two times higher in Tzianas than in Kalvakis in 2003. This difference may be attributed to the difference in biomass production between the two vineyards as the accumulation of grape phenols and vine growth has been reported to have an inverse relationship (Carbonneau 1996, Clingeleffer 1994). The differences in grape phenols and anthocyanins between the vineyards were reduced in the second growing season because there was an almost two-fold increase in the amount of these compounds in Kalvakis in 2004 (Table 3⇑). The increased grape phenolics in Kalvakis in 2004 may be partly explained by increased yield that resulted in a better vine balance by increasing the grape yield to pruning weight ratio. Values between 4 and 10 fall within the optimum range for grapevines to be considered well balanced and capable of producing high-quality fruit in single-canopy trellis systems (Kliewer and Dokoozlian 2005). The causes of between-season differences in crop yield, vine balance, and the degree to which these affect anthocyanin content in Kalvakis are not known. Implicated factors may include seasonal differences in sunlight exposure, berry temperature, and weed competition for water and nutrients. Sugar content of grapes (Brix), as an indicator of grape maturation, also increased in the second growing season in both vineyards.

Relation of canopy reflectance to biomass production.

The relationship between pruning weight and canopy NDVI was best described by a quadratic regression (Figure 3A⇓). The relationships between NDVI, leaf area index (LAI), and aboveground biomass are known to be inherently nonlinear (Myneni et al. 1997). The NDVI measured using the ground sensors reached saturation under the higher biomass conditions of the Kalvakis vineyard (Figure 3A⇓), as has been reported for other crops evaluated using satellite images (Gitelson et al. 2002, 2003). However, the weaker correlation between NDVI and pruning weight in this vineyard may also be attributed to the multiple toppings of the shoots, which removed a significant portion of annual vegetation and exposed the sensors to light reflectance resulting from new vegetative growth. On the other hand, the NDVI values of the Tzianas vineyard were lower, with a broader range and a greater linearity to pruning weight (Figure 3A⇓) that reflected greater spatial variability in topographic features and growth patterns. Canopy spectral reflectance measured using airborne platforms has been correlated to vine biomass and yield in other studies (Lamb 2004). Strong linear correlations between NDVI values derived from the Ikonos satellite and ground measurements of LAI in different stages were found within a growing season in a Californian red wine vineyard (Johnson 2003). The linear nature of the relationship was partly attributed to low values of LAI. Similarly, linear correlation between airborne spectral reflectance (NIR/R) and pruning weight was obtained in a Californian Cabernet Sauvignon vineyard (Dobrowski et al. 2003). Pruning weight was underestimated at high canopy densities, but its relationship to this vegetation index remained constant over two consecutive seasons. Our data also indicated a constant relationship between pruning weight and NDVI over time (Figure 4⇓). Data points for 2003 and 2004 fell on the same regression line despite increased biomass production in the second growing season.

• Download figure
• Open in new tab

Figure 3

The quadratic relationship between biomass production and canopy NDVI (A) or WDRVI (B) for 2003 and 2004 at sampling points in the Tzianas (n = 55) and the Kalvakis (n = 56) vineyards. Each sampling point represents the average of four consecutive vines. All regression coefficients were significant at p < 0.05.

• Download figure
• Open in new tab

Figure 4

The quadratic relationship between pruning weight and canopy NDVI categorized by growing season at sampling points in 2003 (n = 62) and 2004 (n = 49). Each sampling point represents the average of four consecutive vines. All regression coefficients were highly significant (p < 0.05).

Although pruning weight explained most of the NDVI variability (r² = 0.83), the quadratic nature of the model was indicative of the saturation effect of NDVI at high biomass conditions in these vineyards. This appears to be a limitation that did not allow the distinction of differences in canopy NDVI for pruning weight greater than 1.3 kg per plant or 1 kg m⁻¹ row (Figure 3A⇑ and Figure 4⇑). Average pruning weights in the two fields were considerably less than 1 kg m⁻¹ row (Table 3⇑), a value above which growth is considered excessive (Dobrowski et al. 2003). The wide dynamic range vegetation index (WDRVI) was recently proposed in order to reduce saturation effects by extending the linearity of NDVI-related indices to biophysical parameters of vegetation such as biomass (Gitelson 2004). As compared to NDVI, WDRVI did not substantially improve linearity and increased the variability of the relationship between pruning weight and canopy reflectance in this study (Figure 3⇑).

Other operational parameters of the ground sensors were investigated to assess whether different arrangements of the optics (field of view and viewing angle) and spatial resolution of canopy reflectance could provide better estimates of biomass production. The highest correlations between canopy reflectance and pruning weight were obtained when the sensors viewed the canopy from above at veraison independent of the field of view (Table 4⇓). A comparison of spectral readings of the wider (unmasked sensor) and the narrower (masked sensor) field of view indicated close agreement at the high range of NDVI values in both vineyards in 2003 (Figure 5⇓). However, over the lower range of NDVI values there was greater scattering and the readings of the unmasked sensor were lower than those of the masked sensor. This was attributed to dilution by soil reflectance when the wider vertical viewing angle of the unmasked sensor was passing through less developed canopy and probably of LAI. Therefore, it appears that the nadir view of the masked sensor provided more reliable estimates of canopy reflectance. Compared with an off-nadir view of the canopy, the nadir view provided estimates of NDVI that were better predictors of biomass production in 2004 (Table 4⇓). It is not clear why the relationship between canopy NDVI and pruning weight was weaker postveraison (Table 4⇓).

• Download figure
• Open in new tab

Figure 5

The effect of masking the sensor optics in the two vineyards. Data points represent NDVI values for single-vine plants (n = 128).

The capability of the ground sensors to scan the vineyards at a high spatial resolution was investigated in the first growing season when pruning weight was measured in single vines. The obtained data allowed the comparison of the relationship between canopy NDVI and pruning weight at three scales of resolution (Table 5⇓). The explained variability (r²) of the quadratic regression model increased with decreasing spatial resolution. At the high spatial resolution of single-vine plants, the greater unexplained variability may be attributed to canopy interference because of overlapping of branches in adjacent vine plants and possible errors of the coordinates provided by the differential GPS for the estimation of NDVI. Pruning weight was predicted with greater precision at the four-plant resolution (r² = 0.83, Table 5⇓). Although it was not possible to obtain satisfactory prediction of biomass production on a vine-by-vine basis, the implementation of site-specific management in these vineyards remains a realistic scenario by the estimation of plant properties and needs at the four-plant spatial resolution. At least, the obtained data might be used to define better management practices for the next growing season, such as crop-thinning strategies that optimize vegetative and reproductive balance (Dobrowski et al. 2003), provided that grape yield estimates are known. In the Tzianas vineyard, yield was not strongly correlated to canopy NDVI, but yield could be predicted with a reasonable degree of certainty from pruning weight data in the first growing season (r² = 0.54, n = 31, data not shown). More research is needed on the stability of these relationships over several seasons. If the strength of the relationship between canopy reflectance and pruning weight holds preveraison, then it will allow for the implementation of real-time spatial management of vineyards within the same growing season.

Relation of canopy reflectance to grape phenolics.

Previous studies suggest that fruit-quality indices may be inferable from vine canopy characteristics such as size, shape, or vigor (Smart and Robinson 1991). There is recent evidence of a quantitative relationship between canopy NDVI and maturity indicators of grapes, such as total soluble solids (Lamb 2004), or quality indicators, such as grape phenolics (Lamb et al. 2004). Phenolics are primarily contained in the grape skin and are an important determinant of wine quality. Anthocyanins are phenolic compounds specific to red grape varieties and contribute significantly to the color of wine (Lamb et al. 2004).

Our data demonstrated significant differences between the two vineyards in total phenolics and anthocyanins, but not in total soluble solids or Brix (Table 3⇑). Synthesis of these phenolic compounds is believed to be influenced by direct exposure to ultraviolet and blue light (Pirie and Mullins 1980, Jackson 2000). Therefore, high biomass or photosynthetically active biomass is expected to reduce direct or indirect exposure of fruit to sunlight and to reduce phenolic concentration (Mabrouk and Sinoquet 1998). However, shading clusters at prebloom did not limit anthocyanin content of grapes in two out of three seasons in a Shiraz vineyard (Downey et al. 2004). Further evidence implicates berry temperature as a major factor that regulates fruit responses for the synthesis of anthocyanins. Anthocyanins increased linearly as sunlight exposure and berry temperature increased to moderate levels (Bergqvist et al. 2001), but prolonged exposure of clusters to direct sunlight raised berry temperature to excessive levels and reduced berry color (Bergqvist et al. 2001, Spayd et al. 2002). Pruning weight explained a significant portion of the variability of anthocyanin content of grapes in the Tzianas vineyard in 2003. The linear relationship between anthocyanin content and pruning weight (r = −0.58, n = 29) or NDVI at veraison (r = −0.77, n = 27) was negative (Figure 6⇓) and supports the hypothesis of the link between biomass, microclimatic conditions, and berry color. Similar correlation coefficients were obtained between NDVI and grape phenolics and color in a Cabernet Sauvignon vineyard when NDVI values were extracted from airborne digital images (Lamb et al. 2004). The strength of their correlation increased at veraison in comparison to other developmental stages. The 5-m resolution pixels composed a mix of vine and interrow space whereas our spectral readings were direct measurements of the top of the canopy. The stability of the relationship between canopy reflectance and anthocyanin content needs further investigation, as the grapes of the Tzianas vineyard were heavily infested by the downy mildew (Plasmopara viticola L.) in the second year of the experiment. The infestation is suspected to have altered the spatial patterns in grape anthocyanins and their relationship to canopy reflectance (r = −0.45, n = 23, p = 0.03). On the other hand, the lack of correlation between NDVI and anthocyanin content of grapes within the Kalvakis vineyard may be attributed to the same factors that contributed to the weak correlation between biomass production and canopy reflectance, such as low spatial variability, high biomass conditions, and intense pruning practices.

• Download figure
• Open in new tab

Figure 6

The negative relationship between anthocyanin content of grapes and biomass production or canopy NDVI as extracted from the masked sensor in the Tzianas vineyard in 2003 (n = 27).

Despite these difficulties, the obtained strong relationship of canopy reflectance with biomass production and grape color in the Tzianas vineyard demonstrate the potential of proximal remote sensing for the spatial management of vineyards, specifically thinning practices or segmented harvest in order to produce a higher-value product. A standardization of the relationship among sensor reflectance, canopy biomass, and quality attributes of grapes over a range of vine varieties, soil types, and wider geographical areas will be necessary before this technology finds wider applications in site-specific management of vineyards.