A Comparative Study between Vis/NIR Spectroradiometer and NIR Spectroscopy for the Non-Destructive Quality Assay of Different Watermelon Cultivars

Abstract

In light of the great technological progress in non-destructive quality detection methods, sweetness is no longer the essential parameter in evaluating watermelon quality. There is an aspiration to determine physicochemical quality characteristics to enable us to select the best cultivars, agricultural practices, and harvest dates. In the present work, three different watermelon cultivars (Lady, Galander, and Style) were harvested at three consecutive harvest times. Two pieces were taken from each watermelon sample, one from the middle (part A) and the other from the ends (part B), to track the intensity of quality parameters inside the watermelon. Parts A and B were subjected to Vis/NIR spectroradiometer (475:1075 nm), near-infrared spectroscopy (NIRS) (950:1650 nm), and high-performance liquid chromatography to assess the physicochemical quality. Calibration and prediction models were conducted using partial least squares regressions (PLS). The results indicated that the harvesting time significantly influenced the color and chemical parameters. Quality parameters concentrations markedly degraded towards late harvest. The highest concentrations of quality parameters were sighted for the middle zone (part A), especially in the Galander cultivar. Spectroradiometer achieved the best coefficient of prediction (R²_P) ≃ 0.88 and 0.81 attached with the lowest value of the standard error of prediction (SEP) ≃ 0.03 and 1.06 for chroma (C*) and yellowness index (YI). However, the findings showed the superiority of the NIRS compared to the Vis-NIR method. The highest R²_P was achieved by values 0.92, 0.91, 0.90, 0.89, 0.85, and 0.85 for lycopene, total carotenoids, vitamin C, β-carotene, γ-content, and TSS, respectively. It could be concluded that the NIRS has the ability to monitor the maturity development and determine the harvest dates practically and reliably.

1. Introduction

Watermelon (Citrullus lanatus) is a world-popular fruit which is often consumed fresh. This global reputation of watermelon fruit is a result of its high nutritional value. According to the statistics of the Food and Agriculture Organization (FAO), the total world production of watermelon was 118.413 million metric tons. The watermelon crop was listed in Hungary’s agricultural production as one of the main vegetables, with production data of 214.19 to 187.77 K Metric Tons from 2005 to 2019 [1]. Watermelon belongs to the Cucurbitaceae family and consists of different cultivars worldwide, including disorder-resistant and seedless cultivars [2,3]. Watermelon quality is characterized by its high content of moisture, sugars, vitamins, and minerals. Furthermore, watermelon provides a wide range of antioxidants such as carotenoids (mainly lycopene and β-carotene), phenols, vitamins, and amino acids [4,5,6,7,8,9,10]. Quality parameters of watermelon, such as sweetness, firmness, color, and citrulline, are enhanced gradually during the maturity stage and the process of physiological maturation [11,12,13,14]. However, a lot of studies highlighted that the levels of physicochemical quality parameters of fruits and vegetables are strongly influenced by genetic and external factors such as agro-technical processes, environmental conditions, ripening stage, harvest, and post-harvest manipulations [3,6,10,15,16,17,18]. Watermelon is characterized as a non-climacteric fruit and needs to be harvested at an appropriate maturity level to obtain an optimum maturity stage [12]. The major methods for monitoring and measuring the quality of fruits are destructive and based on human labor or instruments for firmness, colorimetric, and chromatographic quantification. In addition to their destructive nature, these methods are laborious, time-consuming, require well-trained technicians, and require samples preparation with hazardous and expensive reagents [19,20,21]. Moreover, some traditional non-destructive inherited methods used by farmers in determining the maturity of watermelon are reported. For instance, experienced watermelon growers listen to the tapping sounds on the watermelon surface, observe the watermelon rind patterns, and take into account the weight of the watermelon. They integrate this information to estimate the watermelon maturity level to determine the optimum harvest time [2,12]. Hence, inaccuracy in estimating watermelon quality can lead to lower competitiveness in the fruit market worldwide. Likewise, [22] mentioned that the conventional methods for quality determination had proven their inability to achieve food quality and safety, and thus a failure to achieve food security. Therefore, there is an urgent need to develop intelligent, non-destructive, fast, real-time, and accurate methods for monitoring quality parameters to improve the accuracy of the quality determination. Acoustic analysis, optics, X-ray imaging, ultrasonics, near-infrared spectroscopy (NIRS), Raman spectroscopy, hyperspectral imaging (HSI), magnetic resonance (MR)/magnetic resonance imaging (MRI), and optical coherence tomography are popular modern non-destructive techniques and widely investigated for estimating fruit quality [22,23,24,25,26,27,28,29,30,31,32,33,34,35]. Among these methods, both spectroscopy and hyperspectral imaging (HSI) methods within the Vis-NIR wavelength band have proven their effectiveness and applicability in laboratories and production lines in measuring quality parameters. Moreover, it can be used for a wide range of products and to measure several quality parameters for the same product, allowing for a real-time decision based on applied wavelength data [11,36,37,38,39]. Few reports highlighted the use of NIRS to measure and evaluate the quality of watermelon. Most of these NIRS reports concerning watermelon quality were focused on color, sugar content, vitamin C, and antioxidants, especially lycopene and β-carotene. Color is described as one of the significant aspects in determining the quality of watermelon [40]. Tristimulus colorimeter values of fruit tissue, especially red color, have been positively correlated with lycopene content in tomatoes [41,42,43], watermelon [44], and grapefruit [45]. The objective of the present work was to monitor the changes in major physiochemical quality parameters of three watermelon cultivars grown simultaneously and harvested at three different times using HPLC and two modern non-destructive methods based on Vis-NIR wavelength for the quality parameters assessment and determination of the best harvest time.

2. Materials and Methods

2.1. Samples

Three watermelon cultivars Lady, Galander, and Style were used in the present study. The cultivars were cultivated on a private farm in Gödöllő, eastern Hungary, and harvested after 120 days of cultivation at three consecutive harvest times (1st, 2nd, and 3rd week) starting in the first week of August 2020. Four watermelon replicate samples were chosen from each cultivar at each harvesting time. All samples were verified that they were healthy, intact, and free of deformities, and their weight ranged from 3 to 5 kg. Two samples were taken from the flesh of watermelon fruit, one from the middle zone of the fruit (part A), while the other was taken from the end (part B), where the fruit is linked to stem. For the determination of vitamin C, the flesh was taken immediately and extracted, as described later. The remaining flesh was homogenized in a small warring blender and kept at −20 °C until analysis of carotenoids.

2.2. Assay Color Property

Immediately, after picking two parts (A&B) of watermelon flesh for each sample, the color was estimated based on the L*a*b* color model using Sheen Instruments Ltd., Kingston-Upon-Thames, UK. L*, a*, and b* parameters were used to calculate the hue (h*), the chroma (C*), the whiteness index (WI), and the yellowness index (YI) of the watermelon according to Equations (1)–(4) [46,47].

$$C^{*}=\sqrt{{\left(a^{*}\right)}^2+{\left(b^{*}\right)}^2}$$

(1)

$$h^{*}={\tan }^{-1}\left(\frac{b^{*}}{a^{*}}\right)$$

(2)

$$WI=100-\sqrt{{\left(100-L*\right)}^2+a{*}^2+b{*}^2}$$

(3)

$$YI=\frac{142.86b*}{L*}$$

(4)

where L* refers to the luminosity or lightness component, a* (intensity of red (+) and green (−)) and b* (intensity of yellow (+) and blue (−)) are the chromaticity coordinates. Color parameters L*a*b* measurements were performed in triplicate for each watermelon part.

2.3. Assay Soluble Solids Content (SSC)

Once the homogeneous juice was obtained for each watermelon flesh part, the total soluble solids content (TSS) was immediately estimated by a digital handheld refractometer (A. KRÜSS. OPTRONIC, Hamburg, Germany, Model: DR 201-95), operating at room temperature through placed a few drops of watermelon juice on a clean and dry refractometer prism and TSS was acquired as Brix (%) value. All samples were measured in triplicate, and the mean value was used in a subsequent calculation.

2.4. HPLC Determination of Carotenoids and Vitamin C

To extract the carotenoid pigments, 10 g of watermelon flesh were taken and crushed in a crucible mortar with the addition of 1 g of ascorbic acid and quartz sand. To the macerate, 20 mL of methanol were added to bind the water. The methanol fraction was decanted into a 100 mL Erlenmeyer flask with a stopper. The residues were further crushed and extracted by the stepwise addition of 50 mL of a mixture of 1:6 methanol-1,2-dichloroethane. The extract was pooled with the methanol fraction. To increase the solubility of pigments in the less polar solvent 1 mL of water was added that assisted in separating the two phases. After mechanical shaking for 15 min, the two phases were separated in a separating funnel. The lower phase containing pigments dissolved in the less polar solvent was dried on anhydrous sodium sulphate and passed to a round bottom flask. The solvent was then evaporated under vacuum at 40 °C to dry using a vacuum-controlled evaporator (Ingots RVO-400). The residues were re-dissolved in 10 mL HPLC grade acetone before injection onto the HPLC column. Carotenoids were separated on a core C-30, 2.6 µ, 150 × 4.6 mm (accuser Thermo Scientific, Waltham, MA, USA) with gradient elution of Tert-butyl methyl ether (TBME) (A) in methanol containing 2% water (B) according to a newly developed protocol [48]. The gradient elution started with 100% B and turned to 30% A in B in 25 min, stayed isocratic for 5 min, and turned to 100% B in 5 min. The eluted carotenoid compounds were detected by DAD between 190 and 600 nm. Identification of carotenoids was based on a comparison of retention time and spectral characteristics with those of available standards such as lutein, β-carotene, and lycopene. In case no standard materials were available, the compounds were identified based on their mass determined by LC-MS/MS, spectral characteristics, and retention behavior as previously described in detail [48]. Quantitative determination of carotenoids was based on using β-8-apocarotenal as an internal standard spiked with the samples. For quantification, the area of each compound was integrated at the maximum absorbance wavelength. To extract vitamin C, 10 g of watermelon flesh were disintegrated in a crucible mortar with a quart of sand. To the macerate 30 mL of 3%, a metaphosphoric acid solution was gradually added with continuous crushing after each addition. The supernatant was quantitatively transferred to an Erlenmeyer flask with a stopper and subjected to ultrasonic force in a water bath ultrasonic device (Raypa, Turkey) for 2 min followed by mechanical shaking for 15 min and filtration through a Hahnemüehle DF 400–125 type filter paper. The filtrate was further cleaned up by passing through a Whatman 0.22 um cellulose acetate syringe filter before injection on the HPLC column. Vitamin C (L-ascorbic acid) was separated on an aqua Nataulis (Machary Nagel, Düren, Germany), 3 µ, 150 × 4.6 mm column with gradient elution of acetonitrile (A) in 0.01 M KH₂PO₄ (B). The separation started with 2% A in B, changed to 30% A in B in 15 min, stayed isocratic for 5 min, and finally turned to 2% A in B in 5 min. The separated compounds were detected by DAD between 190 and 400 nm. Identification and quantification of L-ascorbic acid were based on using of calibration curve of standard solutions. Under the used conditions, L-ascorbic acid had an absorption maximum at 262 nm, at which the area was integrated. A Hitachi Chromaster HPLC instrument consisting of a Model 5110 Pump, a Model 5430 Diode Array detector, and a Model 5210 autosampler was used. The separation and data processing were operated by EZChrom Elite software.

2.5. Spectra Acquisition

Two instruments were used to obtain the spectral data of all watermelon samples. The first instrument is a diode-array Perten DA7200, NIR analyzer-based spectrometer for the optical data collection. It consists of a stationary grating for wavelength dispersion and the diode-array detector type InGaAs, operating in the wavelength range 950–1650 nm for energy detection with 5 nm resolution. The homogenous watermelon samples were fitted in a dish with a 75 mm diameter. In a dark room, the DA 7200 rotates the sample dish during collecting spectral data on the full sample and is saved in reflectance mode. The reflectance spectrum for each sample was acquired as an average of 4 scans. In contrast, a spectroradiometer (ASD FieldSpec^® HandHeld 2™, Analytical Spectral Devices, Inc. (ASD Inc.) coupled with a contact probe of 10 mm spot size, tilting color LCD, integrated computing capability, large internal data storage, and laser targeting was used to perform all spectral measurements within wavelength range 325 to 1075 nm with accuracy ±1 nm. For each sample, 5 scans were acquired in reflectance mode, and their average was calculated.

2.6. Spectral Data Analysis

The watermelon samples were split into sets of calibration (70 samples) and external validation (26 samples). Spectral data analysis was carried out using the Unscrambler^® X program version 10.3 (CAMO Software AS., Oslo, Norway), a statistical software package for multivariate calibration. To assess watermelon quality attributes, a partial least squares regression (PLS) was used to build the calibration and prediction models. Full cross-validation was performed on the calibration samples to determine the optimal number of PCs and validate the models. With this method, one sample is left out from the calibration data set, and the model is calibrated with the remaining data points. Then, the value for the left-out sample is predicted, and the prediction residual is computed. The process is repeated until every observation has been left out of the calibration set once; then, all prediction residuals are combined to calculate the root mean square error of cross-validation (RMSECV). Selecting the best prediction models based on several parameters, including root mean square error of calibration (RMSEC), RMSECV, the calibration coefficient of determination (R²_C), and the optimal number of principal components (PCs) was selected based on the lowest value of the RMSECV for each quality parameter. The prediction results, such as prediction coefficient of determination (R²_P), standard error of prediction (SEP), and residual predictive deviation (RPD), were calculated according to the following equations to assess the accuracy of the prediction model performance. The best prediction model should have low values of SEP and high values of R²_P and RPD [49,50,51].

$$R_P^2=\frac{\left({{\displaystyle \sum }}_{i=1}^n{\left({\widehat{y}}_i-\overline{y}\right)}^2\right)}{\left({{\displaystyle \sum }}_{i=1}^n{\left(y_i-\overline{y}\right)}^2\right)}$$

(5)

$$\mathrm{SEP}=\sqrt{\frac{{{\displaystyle \sum }}_{i=1}^{n_p}{\left({\widehat{y}}_i-y_i-b\right)}^2}{n_p}}$$

(6)

$$b=\sqrt{\frac{1}{n_p}{{\displaystyle \sum }}_{i=1}^{n_p}\left({\widehat{y}}_i-y_i\right)}$$

(7)

$$RPD=\frac{SD}{RMSEP}$$

(8)

$$RMSEP=\sqrt{\frac{1}{n_p}{{\displaystyle \sum }}_{i=1}^{n_p}{\left({\widehat{y}}_i-y_i\right)}^2}$$

(9)

where $R_P^2$ is the prediction coefficient of determination; n is the number of observations in the data set; ${\widehat{y}}_i$is the predicted value of any quality parameter in watermelon sample number i; $y_i$ is the measured value of any quality parameter in watermelon sample number i; $\overline{y}$is the mean value for all samples; $n_p$ is the number of samples in the prediction set; b is the model bias, and SD is the standard deviation of the response variable.

3. Results and Discussion

3.1. Color Analysis of Watermelon Cultivars

The red color is the prime optical quality attribute of ripe watermelon fruits. Therefore, an approach based on the color variations of samples during ripening to determine the optimal harvest time was followed. The evolution of color parameters L*, a*, b*, ΔE, a*/b*, C*, and h* of watermelon samples during different harvest times was investigated, and the results are represented in average ± standard deviation as in

Table 1. Illustrates Avg ± SD values of L*, a*, b*, ΔE, a*/b*, C*, and h* color parameters of Lady, Style, and Galander cultivars.

Harvest Time	Cultivar		L*	a*	b*	a*/b*	C*	h*
1st Harvest time	Lady	A	40.07 ± 1.56	30.30 ± 0.91	22.09 ± 0.80	1.37 ± 0.04	37.50 ± 1.1	36.11 ± 0.77
		B	39.68 ± 1.80	29.88 ± 2.05	22.06 ± 1.40	1.35 ± 0.04	37.14 ± 2.43	36.46 ± 0.77
2nd Harvest time		A	39.63 ± 0.75	30.08 ± 0.78	21.41 ± 0.88	1.41 ± 0.07	36.93 ± 0.84	35.46 ± 1.28
		B	39.51 ± 1.09	29.44 ± 0.50	20.72 ± 0.33	1.42 ± 0.03	36.00 ± 0.46	35.16 ± 0.61
3rd Harvest time		A	40.00 ± 0.63	30.28 ± 0.76	22.33 ± 0.76	1.42 ± 0.06	36.32 ± 1.26	36.30 ± 1.13
		B	39.95 ± 0.66	29.61 ± 1.22	21.49 ± 0.80	1.45 ± 0.03	37.09 ± 1.02	37.04 ± 0.53
1st Harvest time	Style	A	40.36 ± 0.31	30.95 ± 0.67	21.82 ± 0.60	1.42 ± 0.05	37.87 ± 0.62	35.20 ± 0.97
		B	41.18 ± 0.90	30.17 ± 1.00	21.81 ± 0.85	1.38 ± 0.07	37.23 ± 0.99	35.88 ± 1.33
2nd Harvest time		A	41.00 ± 0.34	32.02 ± 0.96	23.17 ± 0.35	1.38 ± 0.04	39.53 ± 0.87	35.92 ± 0.82
		B	41.17 ± 0.82	31.66 ± 0.90	22.72 ± 0.48	1.39 ± 0.10	38.98 ± 0.64	35.69 ± 1.16
3rd Harvest time		A	39.50 ± 0.55	29.40 ± 0.93	21.97 ± 0.66	1.34 ± 0.03	36.70 ± 1.07	36.79 ± 0.59
		B	39.00 ± 0.78	28.92 ± 0.82	21.34 ± 0.19	1.36 ± 0.04	35.94 ± 0.71	36.45 ± 0.72
1st Harvest time	Galander	A	38.76 ± 1.45	31.40 ± 0.59	21.32 ± 1.01	1.48 ± 0.09	37.96 ± 0.58	34.19 ± 1.54
		B	37.76 ± 1.23	30.14 ± 0.53	21.04 ± 0.64	1.43 ± 0.05	36.76 ± 0.62	34.93 ± 0.86
2nd Harvest time		A	39.93 ± 0.54	32.24 ± 0.56	22.41 ± 0.27	1.44 ± 0.04	39.27 ± 033	34.82 ± 0.77
		B	38.68 ± 1.03	31.61 ± 1.12	22.17 ± 0.26	1.43 ± 0.04	38.61 ± 1.03	35.07 ± 0.77
3rd Harvest time		A	39.66 ± 0.57	31.61 ± 1.12	21.51 ± 0.19	1.47 ± 0.05	38.24 ± 0.98	34.27 ± 0.87
		B	38.78 ± 0.98	30.89 ± 1.63	21.44 ± 0.74	1.44 ± 0.09	37.61 ± 1.36	34.82 ± 1.77

L*: lightness, a*: intensity of red (+ value) and green (− value), b*: intensity of yellow (+ value) and blue (− value), a*/b* color index; C*: chroma, h*: hue angle.

.

It was observed that L*, a*, and b* values in part (A) are often slightly higher than in part (B). However, a* and b* color parameters corresponding to redness and yellowness intensity for Galander and Style cultivars were higher than the Lady cultivar, most likely since Galander and Style cultivars were grafted with pumpkin plants. The color parameters increased during the first and second harvest time (HT) and then decreased gradually towards the third HT. This indicates that the second HT is the turning point of the color parameters for Style and Galander cultivars, but the Lady cultivar differed a little from them in this behavior. As a result, the highest values of a* and b* were 32.24 ± 0.56 and 23.17 ± 0.35 at the second HT for part A of Galander and Style, respectively. The lowest values of a* and b* were 28.92 ± 0.82 and 20.72 ± 0.33 at Style and Lady (part B) at 3rd and 2nd HT, respectively. The maximum value of the C* was recorded for the Style (part A) by a value of 39.53 ± 0.87 at the second HT, followed by Galander (part A) at the second HT, and Lady (part A) at the first HT. The WI and YI indexes measuring the intensity of white and yellow colors are very vital ripening indicators for harvesting watermelon fruits and determining the best harvest time. It is evident from Figure 1 that the maturity stages development of watermelon fruits has an inverse relationship with the WI. On the contrary, the YI increased directly with the development of maturity from the first to the second HT, and there was a decline in the value of YI at the third HT for the Style and Galander cultivars. As for the Lady cultivar, the YI value decreased as the harvest time was delayed, while the highest values of YI were recorded for all watermelon cultivars in part (A) as compared to part (B). Part (A) of the Galander cultivar approached the highest value of yellowness index 82.82 ± 1.79 at the second HT. In contrast, the highest values of YI for Style and Lady cultivars were 80.75 ± 1.90 and 79.65 ± 4.89 at the second and first HT, respectively.

3.2. Physicochemical Properties of Watermelon Cultivars

The evolution of the behavior of vitamin C, lycopene, β-carotene, total carotenoids, γ carotene, and TSS values of Lady, Style, and Galander cultivars are illustrated in Figure 2. It was noted from the physicochemical quality attributes results obtained for most watermelon samples that the concentration of these attributes in the middle zone (part A) is higher than its counterpart at the ends (part B). Among these quality attributes, almost all the samples analyzed in the first HT presented vitamin C content higher than the samples analyzed in the second and third HT for all watermelon cultivars. The highest values of vitamin C were found in part (A) of all cultivars, with a noticeable increase over the values of the part (B) taken from the ends of the watermelon at the first HT.

The maximum values of vitamin C (108.49 ± 3.23 and 106.24 ± 2.20 µgg⁻¹) were recorded for the Galander cultivar parts (A and B). The Lady and Style cultivars ranked second and third with values of 103.60 ± 1.49 and 97.03 ± 7.24 µgg⁻¹ for part (A) and 103.41 ± 3.89 and 94.47 ± 6.30 µgg⁻¹ for part (B). The values of vitamin C deteriorated with the delay in harvest time at all cultivars, where the lowest value was found at the second HT by a value of 82.02 ± 1.60 µgg⁻¹ for Style cultivar part (A). This was also the case for Galander and Lady cultivars with values of 93.32 ± 3.97 and 96.64 ± 2.17 µgg⁻¹ part (B). These results agree with [6], who concluded that the values of vitamin C varied from 119.7 to 204 mgkg⁻¹ (fw) for five different watermelon cultivars. This deterioration in vitamin C levels is probably due to it being an unstable molecule when exposed to many factors such as temperature, oxygen, light, and pressure for long periods [52]. The other possible reason may associate with differences in genotype, the ripening stage at harvest, and environmental factors, as reported by [16,53,54]. It is clear from Figure 2 that harvest time significantly influenced lycopene, β-carotene, and total carotenoid contents. For all watermelon cultivars investigated, the amount of lycopene, β-carotene, and total carotenoids on a fresh weight basis markedly decreased with the delay in harvest time from first to third. The highest rate of lycopene synthesis and accumulation in the flesh tissue of the ripe watermelon was found in samples from the first HT for all watermelon cultivars. These results agree with both [6,16,18,44], who reported that the levels of phytonutrients and the antioxidant activity of fruits and vegetables are highly influenced by the genotype, environment, ripening stage, harvest, and post-harvest manipulations. The level of lycopene during the first, second, and third HT were varied for all investigated watermelon cultivars from 103.25 to 54.2 µgg⁻¹. At the first HT for all watermelon cultivars parts (A and B), the lycopene showed a sharp, high value and then markedly decreased in fruits harvested later towards the second and third HT. The prominent increase in the lycopene content value occurred in the Galandar cultivar at first HT, reaching the highest average content of 94.75 ± 5.89 and 90.41 ± 5.14 µgg⁻¹ for parts A and B, respectively, followed by Style and Lady cultivars with 89.84 ± 6.28, 86.72 ± 3.03 and 89.84 ± 12.95, 81.14 ± 5.12 µgg⁻¹ for parts A and B. These values of lycopene were higher than those determined by [6,44], who determined the highest amount of lycopene in different watermelon cultivars at the red-ripe stage ranging from 44.5 to 64.5 mgkg⁻¹ (fw) and from 47.3 to 68.6 µgg⁻¹, respectively. Additionally, [11,55,56,57] reached the lycopene concentration of fresh and puree watermelon varied from 2.65 to 151.75, 19.9 to 80.7, 33.8 to 113.4 mgkg⁻¹, respectively. The changes that occurred in the total carotenoid content showed a similar tendency to that of lycopene. The highest average total carotenoid content was achieved at the first HT with values 109.88 ± 6.88 and 106.5 ± 6.93 µgg⁻¹, in parts A and B, respectively, for the Galander cultivar, while Style and Lady cultivars ranked second and third with the highest mean values of 107.71 ± 4.39, 100.74 ± 2.25, and 101.85 ± 10.51, 99.51 ± 3.22 µgg⁻¹ for parts A and B, respectively. As regards the content of β-carotene it gradually increased through the first and second HT, then decreased in the third HT for all watermelon cultivars, with Galander being the richest cultivar containing 21.07 ± 2.23 and B ≈ 20.11 ± 0.89 µgg⁻¹ of β-carotene at the second HT, followed by Style and Lady part (A) ≈ 18.54 ± 1.19 and part (B) ≈ 17.18 ± 1.18 µgg⁻¹), and part (A) ≈ 17.56 ± 0.64 and part (B) ≈ 17.54 ± 2.35 µgg⁻¹, respectively. These results were higher than the results obtained by [6,14], who found a range of 0.19–9.39 and 1–2.1 mgkg⁻¹ (fw) for β-carotene. As concerns the values measured for γ-carotene, and TSS concentrations, there were changes in all watermelon cultivars at first, second, and third HT. A slight gradual decrease in the values of γ-carotene and TSS was observed with harvesting time delayed from the first to the third. At the first HT, the highest average of γ-carotene content was achieved with values of 1.40 ± 0.10 and 1.41 ± 0.11 µgg⁻¹, in parts A and B, respectively, for Galander, followed by Style and Lady, which contained 1.27 ± 0.26, 1.27 ± 0.14 part (A) and 1 ± 0.06 and 0.99 ± 0.08 µgg⁻¹ part (B), respectively. As for TSS, the Galander cultivar distributed the highest level at the 1st HT (10.15 ± 0.17 and 10.11 ± 0.19 Brix % for watermelon parts A and B, respectively). Style and Lady cultivars reached the maximum mean values at the first HT with the values of 10.06 ± 0.05, 10.02 ± 0.03, and 9.88 + 0.28, 9.84 ± 0.10 Brix % for watermelon flesh parts A and B, respectively. The results of TSS agree with the results of [11,58], who determined 9.18 to 9.24 Brix % in mature and over-mature watermelon.

3.3. Reflectance Spectra Analysis of Watermelon Cultivars

Figure 3 show on the panels’ A, C, and E, the mean raw reflectance spectrum of Lady, Galander, and Style cultivars parts (A and B) within the Vis-NIR range from 475 to 1075 nm.

It shows that the Vis-NIR spectra are quite similar, with some distinctive peaks and valleys revealing the quality attributes of the tested samples. Lady, Galander, and Style, three spectral groups, show wide differences in the Vis range 475–700 nm, most likely due to differences in the concentration of the red pigment, mainly lycopene. This led to a decrease in the reflectance in the visible band associated with a strong absorption peak noted at 560 nm. These results are consistent with what has been stated earlier [58,59] that the reflectance of carotenoids is in the ranges of 550–590 nm and 620–630 nm. After 560 to 925 nm, reflectance values became higher, and the maximum reflectance was measured between 645 and 700 nm at all samples. The highest reflectance was noticed around the peak at 650 nm, which is associated with chlorophyll absorption according to the previous study [60,61]. These variations between samples may be due to the cultivar, flesh parts (A and B), and the harvest time. In contrast, in bands from 780 to 1075 nm, there was a clear absorption peak of about 970 nm due to absorption by water and carbohydrate bonds. Additionally, it corresponds to the strong absorption band of water at 960–990 nm, which includes fruit quality components as mentioned earlier [62,63]. It is clear from the Vis-NIR curves that there are clear differences between the watermelon cultivars and the flesh parts (A and B). Watermelon cultivars, according to consecutive harvest dates (first, second, and third), may be distinguished by the reflectance spectra values, where the average reflectance values decreased from the first to the third HT. The Vis-NIR results obtained agreed with both studies [64,65]. In contrast, the mean NIR original reflectance spectra curves ranging from 950 to 1650 nm of Lady, Galander, and Style cultivars parts (A and B) were represented on the right-side B, D, and F Figure 3. It was evident that the spectra of the NIR wavelength showed a high correlation in the bands 1050–1150 and 1400–1500 nm, corresponding to the O–H bonds, second and first overtones that related to sugars and water. Furthermore, there was a distinguished peak for water at 1460 nm; this is due to the first O–H overtone and O–H combination band as described by [66]. The NIR bands 1150–1400 and 1500–1650 nm, third and first overtone, are compatible with the C–H bonds of carotenoids. These results agree with other studies [11,63,67,68,69] where they found similar absorption bands in their investigations of watermelon, tomato, passionfruit, apricot, and olive, respectively.

3.4. Spectral Data Analysis

Partial least-square regression (PLSR) was applied to the spectral data collected of watermelon fruits within the range Vis-NIR (475–1075). In addition, the calibration, prediction models, and data validation were executed for watermelon quality attributes such as lycopene, total carotenoids, β-carotene, γ-carotene, and some color attributes represented in a*, b*, c*, h*, a*/b*, and YI as shown in

Table 2. Illustrates the predictive efficiency within the range 475–1075 nm of calibration models for watermelon chemical and color quality attributes.

Quality Attributes	SD	N. PCs	Calibration		Cross-Validation		Prediction
			R2C	RMSEC	R2CV	RMSECV	R2P	SEP	RPD
Lycopene	11.98	9	0.84	4.87	0.71	6.77	0.71	6.81	1.77
Total carotenoids	9.38	6	0.80	4.16	0.68	5.33	0.68	5.36	1.76
Vitamin C	8.59	3	0.74	3.72	0.68	4.19	0.68	4.21	2.05
TSS	0.26	3	0.64	0.15	0.56	0.17	0.56	0.172	1.50
β-carotene	4.22	6	0.87	1.53	0.78	1.98	0.78	1.99	2.13
γ-carotene	0.28	4	0.85	0.12	0.8	0.13	0.8	0.14	2.15
a*	1.48	3	0.71	0.76	0.65	0.85	0.66	0.82	1.74
b*	0.83	3	0.47	0.62	0.31	0.70	0.32	0.71	1.19
h*	1.18	3	0.68	0.66	0.62	0.73	0.62	0.74	1.61
a*/b*	1.96	3	0.62	1.21	0.52	1.37	0.52	1.38	1.43

SD; standard deviation; PCs, principal components; R²_C, calibration coefficient of determination; RMSEC, root mean square error of calibration; R²_CV, cross-validation coefficient of determination; RMSECV, root mean square error of cross-validation; R²_P, coefficient of prediction; SEP, standard error of prediction; RPD, residual predictive deviation.

. For chemical attributes, the best R²_C and R²_P were 0.85, 0.8, and 0.87, 0.78, with the lowest RMSECV (0.14 and 1.98 µgg⁻¹) for γ-carotene and β-carotene. Lycopene and total carotenoids ranked second with values of 0.84, 0.71 and 0.8, 0.69, and the lowest RMSECV of 6.77 and 5.33 µgg⁻¹, respectively. The prediction model of lycopene was compatible with the finding of a previous study [64,70] regarding the lycopene of tomato processing juice and watermelon, where the coefficient of prediction (R²) of 0.75. In contrast, this result of the lycopene prediction model was lower than what [55] had obtained in predicting the lycopene content of watermelon puree with a predicted value R² of 0.97 and a SEP of 3.4 mgkg⁻¹ within a visible wavelength of 500–750 nm. Finally, the minimum R²_C and R²_CV were achieved at vitamin C and TSS of 0.74, 0.68, and 0.64, 0.56 at the RMSECV 0.17 µgg⁻¹, and 4.19%, respectively. The finding of TSS was less than those found by [71] for the detection of SSC of melon (Hami) within VIS-NIR wavelength by R_p value of 0.91 with RMSEP = 0.95 °Brix. As well [71,72] obtained high values of R² = 0.86 and 0.70 for the SSC of watermelon. Additionally, the prediction model of TSS was a slight decrease from what was found by [64] in estimating the SSC of processing tomato fruit juice, where the coefficient of determination for cross-validation was R² = 0.72 with RMSECV 0.50 °Brix. While in the color models, the best coefficient of determination for R²_C and R²_p was achieved at 3 PCs. The highest R²_C and R²_P were noted at C* and YI parameters by values 0.9, 0.88, and 0.85, 0.81, with the lowest value of SEP 0.03 and 1.06, respectively, as shown in Figure 4. The minimum value of R²_C and R²_P for the color attribute was recorded at the b* parameter and was 0.47 and 0.31, with the lowest value of SEP at 0.71.

While within the NIR domain range from 950 to 1650 nm, the R²_C, R²_P, SEP, and RPD of lycopene, total carotenoids, vitamin C, TSS, β-carotene, and γ-carotene for all watermelon samples are shown in Figure 5. All calibration models for chemical parameters showed an R²_C above 0.90. Lycopene and total carotenoids presented the same R²_C near 0.95, and the lowest SEP values were 3.58 and 2.88 µgg⁻¹, accompanied by prediction high values of R²_P up to 0.92 and 0.91, respectively. This finding is a slight increase from the conclusion reached by [11] of the quantified lycopene of red-flesh watermelon with R² = 0.87 and SECV = 15.68 mgkg⁻¹. At the same time, it is compatible with the prediction lycopene model of tomato juice processed by an R²_P value of 0.92 with SEP 55.03 µgg⁻¹ [24]. Furthermore, the results of NIR calibration and prediction showed a high level of accuracy in predicting vitamin C and β-carotene, which were sighted from the lowest values of SEP 2.34 and 1.4 µgg⁻¹, combined with the high values of the R²_P of 0.90 and 0.89, respectively.

These results are higher than those obtained by [73] measuring vitamin C in apples within the wavelength range from 400–2500 nm and obtained a good prediction model for vitamin C by value 0.80 with SEP 4.9 mg100 g⁻¹, as well as [24] obtained predictive coefficients for ascorbic acid and β-carotene of 0.82 (SEP = 504.27 µgg⁻¹), and 0.91 (SEP = 9.93 µgg⁻¹) for tomato, respectively. On the other hand, this β-Carotene finding is higher than that obtained by [11] with R² = 0.82 and SECV = 0.81 mgkg⁻¹ for intact red-flesh watermelon. Finally, calibration models for γ-carotene and TSS recorded a notable R²_C with values of 0.92 and 0.94. While, in prediction models, the errors for γ-carotene were similar to TSS, with SEP values between 0.12 µgg⁻¹, 0.1% for Y-carotene and TSS, respectively, and the same value of R²_P was 0.85. The TSS findings, which could be identified as sugar content, were in close agreement with [11], who quantified the TSS of red-flesh watermelon through online NIRS (900 to 1700 nm) with a prediction model R² of 0.84 and SECV = 0.8%. Additionally, it was compatible with the previous study [68,74] about the TSS of passion and melon fruits which found that the prediction models were 0.83 and 0.81, respectively. The R²_P of the TSS was higher and less than those found in a study by [75] regarding the prediction of the total sugar of guava and yellow passion fruit pulps where the R²val values were 0.72 and 0.90, respectively. In addition, it is higher than the predicted SSC of raw tomatoes with r² = 0.75 [76]. It is also less than the prediction model of the measured apple sugar content in the wavelength ranges from 400–2500 nm with an accurate prediction model of 0.94 with SEP 0.34% suitable for most applications [73]. Additionally, less than the dry matter content of date fruits with an R²_P of 0.91 [25]. The result of RPD is another significant factor in refereeing the quality of performing consistently well prediction models applied for monitoring and estimating quality parameters. The RPD factor achieved values of 3.49, 3.29, 3.27, 3.16, 2.55, and 2.5 for lycopene, total carotenoids, vitamin C, β-carotene, TSS, and γ-carotene, respectively. These results demonstrate the ability of NIRS technology to monitor and estimate quality parameters of watermelons and determine the optimal harvest time through high-accuracy prediction models. This high potency of prediction models is consistent with what [77] concluded in classifying the accuracy of prediction models, whereas the results of the RPD factor fall within the excellent range of both lycopene, total carotenoids, vitamin C, β-carotene, and the good range for TSS, and γ-carotene. The RPD value between 2.5 and 3.0 points to a perfect model, and values higher than 3.0 indicate that the prediction performance is considered excellent.

4. Conclusions

The physicochemical quality attributes evolution of the watermelon samples harvested at different times showed that the L*, a*, and b* values and chemical attributes in part (A) are often higher than part B, which was taken from the edges of the watermelon samples. Galander and Style cultivars were higher in a* and b* color parameters corresponding to redness and yellowness intensity than the Lady cultivar. All color parameters were higher at the first and second HT, and then decreased gradually over the third HT. It was found that the development of the maturity stages of watermelon fruits has an inverse relationship with the WI. On the contrary, the YI increases directly with the development of maturity from the first to the second HT for Style and Galander cultivars, while for Lady, the YI value decreased at the late HT. The Galander cultivar at part (A) achieved the highest yellowness index value of 82.82 ± 1.79 at the second HT. For all watermelon cultivars investigated, the amounts of vitamin C content, lycopene, β-carotene, and total carotenoids determined on a fresh weight base, markedly decreased with delay in harvest time. The results of the non-destructive methods of this comparative study proved that the Vis-NIR range 375 to 1075 nm gives the best prediction model for two-color parameters C* and YI with the highest R²_P based on the lowest value of the SEP, nearly 0.88 with 0.03 and 0.81 with 1.06, respectively. However, the findings showed a preference for the other NIRS method compared to the Vis-NIR in reliably estimating the physicochemical properties of watermelon. From the obtained results, it became clear that the preference for the NIRS method is applicable for monitoring and estimating the maturity state of watermelon and determining the optimum harvesting date.