__________

This article is an original research article published in cooperation with the 23rd GiESCO International Conference, July 21-27, 2025, hosted by the Hochschule Geisenheim University in Geisenheim, Germany.

Guest editors: Laurent Torregrosa and Susanne Tittmann.

Introduction

Grapevine (Vitis vinifera L.) is an economically important crop worldwide, predominately cultivated in Mediterranean regions (Macedo et al., 2020). In these regions, precipitation is scarce and irregularly distributed throughout the year, and heat wave events are becoming more intense and frequent due to climate change (del Pozo et al., 2019). Under these climate conditions, irrigation management plays a significant role in maintaining or improving yield and berry quality (Santos et al., 2020). In this context, regulated deficit irrigation (RDI) has been widely implemented to optimise water application while improving water productivity and fruit quality (Jara Rojas et al., 2015; Ortega-Farias et al., 2012; Wang et al., 2019; Zúñiga et al., 2018). Several reports have shown that RDI strategies influence the morphological, physical, and biochemical characteristics of berries. RDI tends to increase sugar and phenolic content while reducing titratable acidity and moderately raising pH levels (Gambetta et al., 2020; Munitz et al., 2020). Rouxinol et al. (2022) suggested that water deficits can increase sugar concentration by reducing dilution in smaller berries; however, this response varies and depends on grape cultivars and climatic conditions. Additionally, the rise in temperature due to climate change has induced changes in phenological progression, berry development, and maturation dynamics. These changes affect berry composition, advance the veraison date, and increase sugar concentration, leading to harvests earlier than expected. Therefore, decision-making tools are essential for monitoring the evolution of grape composition, with the goal of optimising harvest time and maintaining wine quality, according to their specific wine level and commercial strategy (Gamboa et al., 2024).

The evaluation of berry ripening is typically performed using destructive methods that demand significant manual labour and skilled personnel to estimate the optimal harvest time (Manley, 2014). The non-destructive analysis of fruit maturity parameters (soluble solids, titratable acidity, and pH) provides a considerable advantage over traditional methods by preserving the sample, maintaining the fruit in its complete and intact state, and enabling repeated and continuous measurements of its chemical evolution (Daniels et al., 2019; Donis-González et al., 2020). Accurately and rapidly monitoring grape maturation is crucial for identifying homogeneous zones with the optimal harvest time, a process that remains difficult to predict or estimate (Basile et al., 2021; Chariskou et al., 2022; Ghozlen et al., 2010). Thus, the need for the adoption of non-destructive technologies is becoming increasingly important and steadily growing. Research has focused on developing non-destructive spectral reflectance (SR)-based tools for monitoring the ripening process and evaluating the internal chemistry of agricultural products (Goisser et al., 2021; Walsh et al., 2020). Due to their non-invasive approach, the SR-based tools allow for various measurements to be made on the same fruit sample during their physiological evolution, decreasing sampling time and increasing the number of samples available, which generates more representative information in real-time, without the need to destroy the sample (Costa et al., 2022; Piazzolla et al., 2013). In comparison with the current maturity monitoring, SR-based tools can carry out a greater number of measurements in a shorter period and in a non-destructive way, thus generating savings both in laboratory material as well as in handwork (Cao et al., 2010; Fernandes et al., 2015).

Various studies have investigated the use of different wavelength ranges-visible (VIS, 400–700 nm), near-infrared (NIR, 700–1300 nm), and middle-infrared (SWIR, 1300–2500 nm)—to estimate soluble solids, titratable acidity, and pH. For example, Fernández-Novales et al. (2019) found strong correlations (R² = 0.90 to 0.98) between visible spectrum (400–700 nm) measurements and soluble solids in the cultivar Grenache. Giovenzana et al. (2013) reported significant relationships (R² = 0.7 to 0.8) between wavelengths in the VIS range (500–700 nm) and both soluble solids and titratable acidity in Cabernet-Sauvignon. Fadock et al. (2016) developed non-linear models using 350–850 nm wavelengths to estimate soluble solids, pH, titratable acidity, and anthocyanin content in Cabernet-Sauvignon, Cabernet franc, and Syrah, with R² values ranging from 0.30 to 0.90. Kemps et al. (2010) used wavelengths between 700–1600 nm to estimate anthocyanins in Syrah and Cabernet-Sauvignon (R² = 0.8–0.9) and identified wavelengths near 700 nm as effective for measuring sugar content (R² = 0.78–0.93). In addition, several spectral indices (SI) have been developed using wavelength ratios from the VIS, NIR, and SWIR regions to estimate pH, titratable acidity, and soluble solids. The absorbance difference index (I_AD), calculated from wavelengths of 560 and 640 nm, has been correlated with soluble solids (SS) and titratable acidity (TA) with R² between 0.9 and 0.8, respectively (Ribera-Fonseca et al., 2016). The photochemical reflectance index (PRI), calculated from wavelengths of 570 and 530 nm, and the normalised difference vegetation index (NDVI), obtained from wavelengths of 800 and 670 nm, have been evaluated for their relationship to SS, TA, pH, and other variables (Suarez et al., 2021). Similarly, Gamboa et al. (2024) conducted a study in which they evaluated the normalised difference spectral index (NDSI29), calculated from the 420 and 720 nm wavelengths, to estimate soluble solids (SS), titratable acidity (TA), and pH. The main limitation of SI-based models is the site-specific nature of the approach. The accuracy of these models can be affected by variables such as crop type, species, fruit morphology and physiology, water content, and the atmospheric conditions of the environment in which the models were developed (Suarez et al., 2021; Tardaguila et al., 2021). These dependencies impact the stability and consistency of SI-based models over time, limiting their broader applicability. As a result, local calibration and optimisation of models based on SI are essential for improving their accuracy and reliability. To address these issues, the contour map methodology has been proposed as an alternative to develop more robust indexes. This approach offers a graphical representation of a correlation matrix, allowing the identification of “hot spots” where optimal combinations of wavelengths can be found for estimating a specific variable of interest. This approach optimises the spectral resolution through an interpolation process, generating a spectral signature with a wavelength resolution at 1 nm intervals, which facilitates an exhaustive analysis to calculate and develop narrowband indices. For this purpose, the Savitzky-Golay (SG) filter can be applied to remove noise and random fluctuations in the spectral data (Pôças et al., 2015; Rallo et al., 2014). The contour map methodology has been applied in several agricultural studies across different ways to optimise the spectral indices described in the literature for estimating physiological variables. It has resulted in improvements in correlation coefficients through the development of new formulations for specific indices (Inoue et al., 2008; Tian et al., 2011; Yao et al., 2010). For Vitis vinifera L. cv. “Touriga Nacional” the relationship between pre-dawn leaf water potential and fifteen vegetation indices was evaluated. All indices were calculated using both their standard formulations and optimised versions (spectral resolution every 1 nm). For example, the photochemical reflectance index (PRI) increased its coefficient of determination (R²) from 0.39 to 0.82 based on a new combination of wavelengths (Pôças et al., 2015).

Limited scientific research has been reported on the use of contour maps to improve the accuracy of SI-based models for monitoring grape ripeness under varying water levels. This study aims to evaluate the application of the contour map optimisation method for determining spectral indices for non-destructive monitoring of soluble solids, titratable acidity, and pH in grapevines under four levels of water application. By employing contour maps, this research aims to identify specific wavelength combinations, allowing the creation of new spectral indices with improved sensitivity and stability during growing seasons.

Materials and methods

1. Description of the experimental site

The study was carried out in a commercial vineyard located in the Pencahue Valley (–35,43°, –71,80°; 104 m above sea level) of the Maule Region, Chile during the 2019/20 and 2020/21 growing seasons. This valley has a temperate Mediterranean climate with an annual rainfall of 562 mm and reference evapotranspiration (ETr) of 689 mm year^-1. The grapevine growing seasons used in this study (from November to March) were dry and hot, with an average daily temperature of 21 °C and ETr value of 6.4 mm day^-1.

The vineyard was planted in 2014 with Cabernet-Sauvignon vines grafted on SO₄ rootstock (Vitis berlandieri × Vitis riparia Selection Oppenheim 4). Vines were trained using a vertical shoot-positioned system (VSP) oriented in a north-south direction with a plant density of 2.3 m × 1.2 m, a canopy length of 1.2 m, and pruned using the Guyot method with two canes per vine (between 7 and 8 buds per cane). Water application was carried out weekly, using two drippers per vine with a flow rate of 2.5 L h^⁻1, located 0.6 m apart from each other across the irrigation line.

2. Experimental design

The experimental design was a randomised block design (RBD), with blocking based on soil depth due to the presence of a hardpan in certain repetitions. The experiment consisted of four treatments with four replicates. Each replicate included six complete rows; however, harvested fruit was obtained exclusively from the two central rows. For the T0 and T1 treatments, water application corresponded to 100 % and 70 % of evapotranspiration (ETa) during the entire growing season, respectively. Water application for T2 was based on 50 % ETa from fruit set (FS) to veraison (V) and 100 % ETa from V to harvest (H); whilst for T3 it corresponded to 30 % ETa from FS to V, and 100 % from V to H.

3. Measurements of midday stem water potential (Ψx)

Ψx measurements were taken from two healthy, mature leaves per replicate, selected from the middle third of the canopy and exposed to sunlight. The leaves were wrapped in plastic film and aluminium foil to prevent gas exchange and light absorption. After two hours, they were detached from the branch, and Ψx was measured using a Scholander pressure chamber (model 600, PMS Instrument Company, Oregon, USA) (Fulton et al., 2014; Scholander et al., 1965).

4. Berry sampling

From veraison to harvest, shaded and sun-exposed clusters were randomly sampled at morning hours (7:30 to 8:30 local time). Samples from the 2019/20 season were collected from January 22 to March 12 (harvest), while those for the 2020/21 season were taken from January 27 to March 24 (harvest). Every week, two hundred berries per replicate were harvested to measure soluble solids, titratable acidity, and pH.

5. Determination of soluble solids (SS), titratable acidity (TA) and pH

The 200 berries were manually ground or pressed to extract the most homogeneous juice possible, trying to obtain only liquid without solid parts. This liquid was used for SS, TA, and pH measurements according to the standard methodologies proposed by the OIV (OIV, 2025). SS was estimated using a refractometer (model Pla -1, TW, Taiwan, China), performing a wash with demineralised water between each measurement, until reaching 0 °Brix. TA was measured by acid-base titration using 0.1 N NaOH and bromothymol blue as an indicator (Kontoudakis et al., 2010). To obtain pH values, a portable pH meter (Extech, model 246072, Bogota, BTÁ, Colombia) was used, performing periodic calibrations between each treatment and repetition.

6. Reflectance spectrometry measurements

The spectral reflectance (SR) measurements of the berry samples were taken using a portable hyperspectral spectroradiometer (HR-1024i, Spectra Vista Corporation, New York, NY, USA) connected to an optical fibre with a field of view (FOV) of 25°. The instrument can measure wavelengths between 350 and 2500 nm with intervals less or equal to 1.5, 3.8, and 2.5 nm in the range of 350–1000, 1000–1890, and 1890–2500 nm, respectively. Berries were placed on Petri dishes (9 cm in diameter) at 20 cm from the optical fibre. To measure SR, berries were exposed to direct light from halogen bulbs with a dichroic reflector inside a completely sealed black box. In this case, the dishes were used solely as a physical base to hold the berries, without being part of the active measurement area, so its influence on the spectral reflectance of the sample was negligible. Additionally, a reflective panel (white reference) (Spectralon, Panel, North Sutton, NH, USA) was placed between each measurement to obtain reference measurements at 100 % reflected light (Figure 1).

7. Optimisation of spectral information

A customised MATLAB code was developed by the Digital Agriculture group at Stellenbosch University (South Africa) to create contour maps that analyse the spectral information and associate it with the chemical variables measured at the laboratory. The code automatically resamples the spectra using a Savitzky–Golay filter that generates an interpolation process with a resolution of 1 nm, which is necessary for calculating narrow-band indices (Pôças et al., 2015). Moreover, the code performs the matrix calculations of indices and correlations generating the contour maps associated with each variable.

8. Development of new spectral indices

Simple ratio indices (SR), which are defined as the quotient of the reflectances at two different wavelengths, highlight subtle variations in reflectance that might be difficult to discern in individual bands. The development of new indices was done using the contour map technique based on the principle of SR. This methodology involves testing various combinations of reflectance values within a given range VNIRS (350–2500 nm) with precision of 1 nm against a variable of interest through linear regressions. The result is a heat map where each pixel represents the determination coefficient (R²) of a specific combination wavelength. Based on the above, the R² with the best value was adjusted to determine the optimised index that allows the variable in question to be better defined.

9. Development of models and validation

The development of linear models was through with season 2019/20 and validation was carried out by analysing the 2020/21 season by calculating the ratio of the estimated to observed values, root-mean-square error (RMSE), mean absolute error (MAE), and index of agreement (I_a) (Mayer & Butler, 1993; Willmott et al., 1985).

where N is the total number of observations, Pi and Oi are the estimated and observed values, respectively, and $\overline{O}$ is the mean of the observed values. In addition, the ratio of observed to estimated values was evaluated using the t-test at a 95 % confidence level.

Finally, ANOVA tests were conducted to determine the significant effect of irrigation treatments on SS, TA, and pH. In this case, Tukey’s multiple comparison tests were used at 95 % confidence using RStudio (R: The R Project for Statistical Computing, 2023).

Results

1. Measurements of stem water potential

The irrigation treatments resulted in a wide range of midday stem water potential (Ψx) values across the two seasons (Figure 2), ranging from –0.79 to –1.6 MPa in 2019/20 and from –0.65 to –1.50 MPa in 2020/21. These values indicate that the vines experienced water stress levels ranging from none to severe, based on the vine water deficit thresholds defined by van Leeuwen et al. (2009). Additionally, the irrigation treatments had a significant effect on nearly all analysed dates. In general, Ψx values were consistently higher in treatment T0 compared to T3. The most pronounced differences between these treatments were observed during the final week of January and the initial week of February. Subsequently, after irrigation restarted, Ψx values progressively increased across all treatments until harvest. At this phenological stage in the 2019/20 season, Ψx values in T3 were significantly higher than those in T0 and T2, which showed similar values. In contrast, during the 2020/21 season, T0, T2, and T3 presented comparable values, all of which were significantly higher than those of T1. The reason for this difference was that T1 was continuously maintained at only 70 % ETa replacement throughout the trial and never reached 100 % ETa.

2. Analysis and optimisation of spectral information

Based on spectral information obtained from Cabernet-Sauvignon under RDI in the 2019/20 and 2020/21 seasons considering all the treatments with different irrigation regimes, contour maps were generated in a wavelength range of 350–2500 nm with a 1 nm interval with respect to each of the variables individually: soluble solids, titratable acidity, and pH (Figure 3). The individualisation and optimisation of the wavelengths allowed for a more targeted analysis of all possible combinations within contour maps, facilitating the evaluation of their performance across both seasons. The maps display the coefficients of determination (R²) between the maturity parameters and all possible combinations of two wavelengths, generated by a matrix that covers the entire spectral range along the horizontal (λx) and vertical (λy) axes. The results are visualised using a colour bar, where redder areas indicate wavelength combinations that produce stronger linear models with the maturity parameters. To identify common patterns across seasons for each maturity parameter, a visual selection criterion was applied. This process led to the identification of three common zones—also referred to as “hot spots” or clusters—where R² values were ≥ 0.8. These zones, highlighted with white squares, were recognised as consistent and persistent across seasons and were subsequently analysed individually.

Based on the above, the cluster patterns for soluble solids, titratable acidity, and pH showed similar trends across both measurement seasons: a small cluster in the VIS region (400–700 nm) (cluster 1), a more pronounced cluster in the NIR region (700–1200 nm) (cluster 2), and another in the SWIR region (1300–1700 nm) (cluster 3).

3. Development of new spectral indices

The analysis of each cluster allowed for the identification of the optimal wavelength combinations that produced the best linear models for each variable (Table 1). The best linear regressions were observed in cluster 3 with a spectral ratio of 1600 nm to 1412 nm (SI_SS) for soluble solids and 1617 to 1415 nm (SI_TA) for titratable acidity (TA), achieving R² values of 0.82 and 0.83, respectively. For pH, it was observed that cluster 1 produced the optimal linear model using a ratio of 694 nm to 697 nm (SI_pH), obtaining an R² of 0.95. Figure 4 shows the linear regression models using R1600/R1412, R1617/R1415, and R694/R697 ratios for predicting SS, TA, and pH, respectively, in grapevine under RDI treatments during the 2019/2020 growing season. In this case, the highest data dispersion was observed for SS between 22–25 °Brix (Figure 4a) and for TA between 3.5–7.5 g 100 mL^-1 tartaric acid (Figure 4b), but the models presented a good performance. Figure 4c illustrates that the points for pH were close to the regression line. Finally, the performance of the three linear regression models was not affected by the RDI treatments suggesting that models based on simple ratios (λx/λy) can effectively monitor maturity in grape growing under different irrigation levels. This stability indicates that these models are consistent under varying irrigation regimes, making them valuable tools for assessing grapevine maturity, regardless of the water stress conditions.

Table 1. Linear regression models for predicting soluble solids (SS), total acidity (TA), and pH based on new spectral indices (SI) in grapes under different levels of water application for the 2019/20 growing season. In addition, a and b represent the intercept and slope of the linear regression models, respectively.

Spectral Index	R2	a	b
SI SS = (R1600/R1412)	0.82	–184.99	211.54
SI TA = (R1617/R1415)	0.83	258.28	–255.51
SI pH = (R694/R697)	0.95	12.566	–8.22

The validation of linear regression models for predicting SS, TA, and pH based on the optimised spectral indices is indicated in Table 2. This table indicates that the I_a and R_oe values were between 0.79–0.94 and 0.93–1.05, respectively. The linear model based on SI_ss was the most accurate in estimating SS, with an RMSE of 1.15 °Brix and an MAE of 1.31 °Brix. In this case, the comparison plotted in Figure 5a shows that the points were very close to the 1:1 line with R_oe equal to unity at a 95 % confidence level. The worst agreement was found for the estimation of TA, with an RMSE of 1.60 g 100 mL^-1 tartaric acid and an MAE of 2.28 g 100 mL^-1 tartaric acid. A large dispersion of data points was observed, particularly for measured TA between 6–11 g 100 mL^-1 tartaric acid (Figure 5b). In this case, R_oe was different from unity, suggesting that the SI_TA-based model underestimated the TA with an error of 7 %. Additionally, the linear regression model based on SI_pH overestimated the pH with errors of 5 %. In this case, main errors were observed for measured pH greater than 3.0 (Figure 5c).

Table 2. Validation of linear regression models for predicting soluble solids (SS), total acidity, and pH based on spectral indices in grapes under different levels of water application for the 2020/21 growing season.

Spectral Index	RMSE	MAE	Ia	Roe	T-test
SISS = (R1600/R1412)	1.15	1.31	0.94	1.01	T
SITA = (R1617/R1415)	1.60	2.28	0.90	0.93	F
SIpH = (R694/R697)	0.10	0.08	0.79	1.05	F

The effect of four levels of water application on modelled and measured values of SS, TA, and pH in grapes are indicated in Table 3. This table indicates that the RDI treatments did not significantly affect the observed and estimated values of SS, TA, and pH for grape berries harvested from drip-irrigated vines under RDI. The measured values of SS, TA, and pH were between 21.9–24.4 °Brix, 6.4–7.5 g 100 mL^-1 tartaric acid, and 3.3–3.4 for pH, respectively. While the estimated values were 20.7–21.4 °Brix, 5.0–6.4 g 100 mL^-1 tartaric acid, and 3.6–3.7 pH units. Besides, neither the growing season nor the interaction between treatments and growing seasons had a significant effect.

Table 3. Effect of four levels of water application on observed and simulated values of soluble solids (SS), total acidity, and pH in grapes under different levels of water application for the 2020/21 growing season.

Treatment	SS (°Brix)		TA (g 100 mL-1 tartaric acid)			pH
	Observed	Simulated	Observed	Simulated	Observed		Simulated
T0	22.4	20.7	7.5	6.4	3.3		3.7
T1	22.4	21.4	6.4	5.0	3.3		3.6
T2	22.3	21.0	6.4	5.9	3.4		3.6
T3	21.9	20.9	6.6	5.7	3.3		3.6
Significance	n.s	n.s	n.s	n.s	n.s		n.s

Discussion

The results indicate that the application of regulated deficit irrigation (RDI) did not have a significant effect on ripeness parameters at harvest (DOY 70). These findings are consistent with those reported by Wang et al. (2019) who evaluated different irrigation regimes and observed that, although RDI reduces vine vegetative growth, it did not significantly impact grape composition. This similarity may be attributed to the fact that, in both studies, irrigation was restored during the final stages, allowing the vines to recover from water stress and enabling proper berry ripening. In contrast, Munitz et al. (2020) reported significant differences in grape composition because sustained water deficit, maintained until harvest, continuously limited soil water availability to the vines.

This study determined that the application of the contour map methodology in the analysis of the full spectrum (350–2500 nm) allowed the identification of three spectral “hot spots”. They were located in the ranges of 400–700 nm (cluster 1), 700–1200 nm (cluster 2), and 1300–1700 nm (cluster 3) and are sensitive to variations in SS, TA, and pH. These regions showed significant correlations with the ripeness parameters measured in this study (R² > 0.8 in most of the cases). This high level of correlation was consistent in both measurement seasons (2019/20 and 2020/21). From the clusters, the optimal linear regression models to simulate SS, TA, and pH were obtained using SI_SS (R1600/R1412), SI_TA (R1617/R1415), and SI_pH (R694/R697), respectively. The selection of these spectral indices was made using a specific statistical criterion, which allowed us to analyse the reliability of the results and adjust the most accurate models in terms of the maturity parameters.

Previous studies have reported that the VIS-NIR range enables the establishment of significant correlations with various maturity parameters in grapevines. For instance, Fadock et al. (2016) used the 350–800 nm spectrum to develop PLS models, achieving strong correlations (R² between 0.58 and 0.89) for the estimation of soluble solids and pH in Cabernet-Sauvignon, Cabernet franc, and Syrah. Similarly, Pampuri et al. (2022) analysed the 450–860 nm spectral range in grapevines cv. Chardonnay uses a Cost-Effective Visible/Near Infrared optical system, obtaining significant correlations for the estimation of the SS/TA ratio and pH. These findings support the suitability of the VIS-NIR range for the non-destructive determination of key maturity parameters in wine grapes.

Regarding the following “hot spots” which cover the spectral range from 700 to 1200 nm (cluster 2), it is visually the broadest category. Statistical analyses of the SI found in this area were performed to evaluate their estimation capability. However, the models developed in this spectral region did not show optimal performance and stability in both seasons. In contrast, Ferrara et al. (2022) reported high R² values for the prediction of ripening parameters (SS, TA, and pH) in Primitivo grapevines using a portable NIR device operating within the 740–1070 nm range, based on PLS analysis and various pre-processing methods. However, their models exhibited very high RMSE values, particularly for titratable acidity (TA), which exceeded 5.06 g/L.

Finally, in the 1300 to 1700 nm spectral range (cluster 3), the analysis identified well-defined “hot spots” associated with the highest statistical performance of the developed models, demonstrating strong correlations with the SS and TA. This region falls within the shortwave infrared (SWIR) domain, which is particularly sensitive to overtones and combination bands of molecular vibrations associated with carbohydrates, water, and organic acids-key components that undergo significant transformations during berry ripening (Mejía-Correal et al., 2023; Siesler, 2016).

The relevance of this spectral range has been highlighted in previous studies, particularly by González-Caballero et al. (2010), who analysed the 380–1700 nm range and identified this region as crucial for the accurate prediction of SS, TA, and pH across different wine grape varieties using NIR-SWIR. Their results demonstrated high predictive accuracy, with R² values of 0.89 for SS, 0.87 for reducing sugars, and 0.69 for pH. This strong relationship is attributed to the interaction of these wavelengths with specific molecular vibrations linked to carbohydrate and water absorption, which are essential for the structural and biochemical modifications occurring in berries as sugar content increases and acidity decreases. In agreement with these findings, the present study determined that the SWIR spectral region (1400–1600 nm) exhibited the strongest correlations for SS and TA estimation by using SI_SS (R² = 0.82) and SS_TA (R² = 0.83), respectively. Moreover, the consistency between both studies reinforces the applicability of NIRS in the non-destructive assessment of grape ripeness, enabling the implementation of field monitoring strategies and optimisation of decision-making in wineries.

In general, SR model stability over time and generalisation problems are largely attributed to high biological variability, which may be influenced by various factors such as cultivar, site-specific growing conditions, cultural practices, training systems, harvest season, and fruit ripening stages (Bedbabis et al., 2014; Boselli et al., 2019; Zheng et al., 2020). The application of contour maps as a methodological tool addresses this issue by enabling the integration of a wide range of samples collected across different harvest periods (two or more seasons), including various developmental and ripening stages, thereby optimising model calibration.

The present study demonstrates that contour maps exhibit performance comparable to that of other methodologies based on advanced algorithms, such as Partial Least Squares Regression (PLSR), Random Forest Regression (RF), and Support Vector Regression (SVR). In this context, Kalopesa et al. (2023) developed predictive models using the full spectral range (350–2500 nm) to estimate soluble solids (SS), titratable acidity (TA), and pH in four wine grape varieties (Chardonnay, Malagouiza, Sauvignon blanc, and Syrah). The results indicated similar predictive performance for the different variables, with R² and RMSE varying depending on the variable analysed and the modelling phase. For SS, R² values ranged from 0.74 to 0.89 during model development, with an RMSE for °Brix between 1.39 and 1.91, while in validation, R² ranged from 0.64 to 0.85, with an RMSE between 1.44 and 2.36. For TA, R² values during modelling ranged from 0.67 to 0.83, with an RMSE between 2.06 and 2.85, whereas in validation, R² varied between 0.61 and 0.82, with an RMSE ranging from 1.66 to 2.22. Finally, for pH, the R² values during modelling ranged from 0.44 to 0.88, with an RMSE between 0.09 and 0.35, while in validation, R² fluctuated between 0.45 and 0.88, depending on the grape cultivar analysed.

Our study demonstrates the applicability of contour maps as an alternative method for extracting spectral features and identifying narrow and highly sensitive bands, which exhibit stable and specific performance in the estimation of various ripeness parameters. Our findings are consistent with prior research and support the usability of contour maps (Mariotto et al., 2013; Rallo et al., 2014; Yao et al., 2010). The linear models based on new spectral indices obtained from contour maps confirmed their ability to monitor the evolution of maturity components from veraison to harvest (Gamboa et al., 2024). Nevertheless, it is crucial to acknowledge that this method is susceptible to measurement noise and the presence of faulty sensors, which can result in inaccurate contour representations unless outliers are appropriately addressed (Meng et al., 2006). Soil water availability affects the ability of reflectance indices to estimate quality attributes in vineyards (Rodríguez-Pérez et al., 2022) and induces greater variability in the data, which could reflect the intrinsic variability present under field conditions. In this regard, further research is needed to evaluate the effectiveness of these models in different grape varieties and growing conditions, as well as to develop more reliable models for predicting the important maturity components of grapes.

A future direction for this research involves adapting the developed spectral indices for use in more accessible field-based systems. Specifically, evaluating the integration of these indices into low-cost sensors operating within the identified relevant spectral regions would be valuable. This could lead to the development of portable or ground-based devices for direct cluster-level measurements in practical vineyard conditions, advancing non-destructive monitoring tools in viticulture.

Conclusion

This study demonstrated that the optimisation of spectral indices through contour mapping significantly improves the accuracy and stability of non-destructive grape ripeness assessment. By analysing hyperspectral reflectance data (350–2500 nm), three optimal wavelength combinations were identified, enabling the development of robust spectral indices for predicting soluble solids, titratable acidity, and pH. Validation of these models in an independent season confirmed their stability under variable water availability conditions, suggesting their applicability across different production environments. The use of contour mapping not only optimised the selection of key wavelengths but also reduced the dependence on site-specific calibrations, increasing the generalisation capacity of the developed spectral indices. Finally, the application of contour mapping in the development of spectral indices constitutes a significant methodological advancement for non-destructive grape ripeness assessment.

Acknowledgements

This study was supported by the Chilean government through the projects FONDEQUIP (EQU-009CONICYT) and FONDECYT (N^o 1230500). The authors would like to thank Dr Alvaro Gonzalez (“Viña Concha y Toro” winery) for allowing us to set up the trials in the company´s vineyards. Also, to Scholarship to promote the internationalisation of master's degree programs, R.U. N^o 136/2019 by the University of Talca, Chile.