Introduction

Bananas are an agricultural product with a production of around 120 billion tons in the world in 2020, with the largest cultivated areas in Asia (50.82%) and America (32.97%). The main producing countries in Asia are India, China, Indonesia, and the Philippines, while on the American continent, they are Brazil, Ecuador, Guatemala, Costa Rica, Mexico, and Colombia (FAO 2020). Currently, more than 1000 varieties of this species are produced in the world, but the most commercialized is the Cavendish (“Williams”) type, which represents about 47 percent of world production at approximately 50 million tons per year (FAO 2022). This variety has acquired commercial importance in production since the 1960s, as a proper alternative to the attack of banana Fusarium wilt that is mainly caused by Fusarium oxysporum f. sp. cubense Race 1 on the dominant susceptible variety, Gros Michel (Big Mike) (Stover 1962).

Other diseases, such as banana bacterial wilt, also known as “moko disease,” affect susceptible Cavendish cultivars. This is the most important bacterial disease in worldwide banana crops, generating losses that can reach 80 to 100% of the total crop yield (Yuliar and Toyota 2015). The dispersal ease, high virulence, number of hosts, lack of effective management practices, and the requirement for a long lie fallow period after eradication are factors related to the losses (Ramírez et al. 2015). The main control strategy in bananas involves early detection of infected plants, followed by the plants' elimination within a 5 m radius around the plantation and a 6-month quarantine establishment (Álvarez et al. 2008).

R. solanacearum race 2, a gram-negative bacterium of the proteobacteria and bacterial plant pathogens, causes moko disease, threatening global crop production (Agrios 2008). This aerobic bacterium is transmitted through the soil by the mobility granted by a polar flagellar tuft. It attacks the vascular system of the plant, causing vascular bundle obstruction, leading to dehydration, malnutrition, and eventually death (Vinatzer 2012). Although symptoms of the disease vary according to age, plant, transmission medium, affected organ, and bacterial ecotype strain (some strains cause less severe symptoms), the most common are yellowing of young leaves, wilting, and rotten fruit (Peeters et al. 2013). Besides, it is typical to see a color change from light brown to dark in the vascular tissues of the pseudostem, rhizomes, and leaf sheaths.

In general, this process quickly progresses in young plants, taking less than a week from the initial symptoms for the plant to die. In root and rhizome infections, the young leaves get yellow and wither first (Liberato and Gasparotto 2006). These symptoms are not homogeneous throughout the vertical structure of the plant, since the photophysiology of the leaves depends on the amount of available solar radiation (Stratoulias and Tóth 2020), which influences directly the electromagnetic irradiation of the plant sheet reflection. To this date, there are no effective control methods to prevent crop damage by infection, resulting in wide economic losses. Considering that timely detection of vascular plant diseases is fundamental, it is desirable to use nondestructive methods for productive units. A detection system for the disease before visible symptoms in the plant is necessary for the success of the control measures since the permanent monitoring of the bacteria in soil and water sources is an inefficient process due to low densities of the pathogen and irregular spatial distribution in the field (Pradhanang et al. 2000). Among the alternatives, the detection by spectrometric techniques, used in the initial stages of propagation of plant material by direct reading of planting units, is useful in the identification of the primary disease focus (Franke and Menz 2007; Franke et al. 2009).

For these spectrometry-based techniques to be applicable, it is necessary to generate specific information on plant/pathogen pathosystems of spectral fingerprints typical of infectious processes as a primary input to feed spectral libraries used for disease identification. This study was carried out on bananas because of their economic importance and because it is a plant with widely known agronomic characteristics and is susceptible to a range of pathogens that allow its spectral response to be contrasted when affected by different types of stress (biotic and abiotic). The primary goals of this study were to: (1) use reflectance spectroscopy to characterize the spectral response of healthy plants infected with R. solanacearum and subjected to water stress in two banana cultivars, and (2) use reflectance spectroscopy in the 400–1000 nm range to detect vascular wilt in banana. This is the basis for early detection of bacterial wilt in banana cultivars and other future plants using reflectance spectroscopy.

Methodology

Study area and sample size

The test was carried out under greenhouse conditions at the facilities of the Banana Research Center (CENIBANANO) in Carepa municipality (Antioquia, Colombia). The banana plants Gros Michel and Williams cultivars were propagated in vitro in a commercial nursery. A completely randomized experimental design with four treatments (2 for each cultivar) and their respective controls was carried out on the following aspects: (1) 30 control Gros Michel plants, (2) 30 Gros Michel plants inoculated with R. solanacearum, (3) 30 Gros Michel plants under water stress, (4) 30 control Williams plants, (5) 30 Williams plants inoculated with R. solanacearum, and (6) 30 Williams plants under water stress. These treatments were started with three-month-old plants at the phenological stage 1030 on the BBCH scale, which is described like “three leaves completely open and the youngest leaf at candela stage 0” (Meier 2001). Plants were grown on bags filled up with 2 kg of peat mixture. Environmental conditions in the greenhouse during the experiment time were the following: 26 °C average temperature and 90% relative humidity. The plants were watered three times weekly.

Microorganisms used and inoculation of plants

CENIBANANO provided a strain of R. solanacearum Race 2 for the research. The initial isolation and pathogenicity of this strain were determined from the Gros Michel and Williams cultivars following described protocols for pathogenicity tests (Ramírez et al. 2020). The strains were seeded in triphenyl tetrazolium chloride (TTC) medium and grown under constant agitation for 24 h. The inoculum concentration was estimated using spectrophotometry, using as a reference a 1 × 108 cfu/ml−1 solution with an optical density of + 0.1 at 650 nm. For the plant infection a scalpel was used to make a cut at the base of the root of each plant, which was immediately inoculated with 15ml of a previously arranged inoculum suspension (Jie et al. 2009).

Data analysis and spectroscopy

The reflectance spectra were obtained with a USB2000+ portable spectroscope (Ocean Optics, FL, USA) in a range between 400 and 1100 nm, in which different parameters necessary for calibration were determined according to Marín-Ortiz et al. (2020): The integration time was 6 s, with an average of 5 readings per measurement and an “interval time” of 2000 μs. The measurements were taken under greenhouse conditions when the plants had five functional leaves (1030 development stage according to the BBCH scale) with the optical fiber on the adaxial surface of the leaf, obtaining three spectra for each leaf. Therefore, they were measured three pseudo-repeats from the top of the leaves of each of the plants, at 48-h intervals during the disease incubation period, per each sampling day 900 spectra were assessed.

The spectrum average values were calculated every 4 days, from the inoculation day until symptoms were expressed to visualize differences in the reflectance spectra. Specific wavelengths were identified to classify healthy plants infected with R. solanacearum and subjected to hydric stress using the RELIEF characteristics classification algorithm. This scanning was complemented with an analysis of principal components (PCA) to identify the variables with the greatest contribution to the total variation of the data during the entire disease incubation period classified with the RELIEF algorithm. For each sampling day, we performed a linear discriminant analysis (LDA), as a supervised classification method, of qualitative variables. This LDA allowed us to divide a priori the data collected for both varieties (Gros Michel y Williams) into three groups: plants inoculated with R. solanacearum, plants with water stress, and control plants. These analyses were made with the software R v.4.0.5 (RStudio Team, Boston, MA, USA).

Results

Spectral response characterization of two banana varieties during the bacterial wilt incubation period

The visible symptoms of the disease became evident on day 12 with the onset of chlorosis, for the Gros Michel and Cavendish varieties, indicating that the incubation period of Ralstonia is completed at this time under the environmental conditions of the experiment. This section presents the reflectance spectra of banana leaf plants inoculated with R. solanacearum, control plants, and those subjected to hydric stress, during the bacterial wilt incubation period (Fig. 1). The Williams cultivar inoculated with R. solanacearum and subjected to hydric stress showed no variation in reflectance in the visible range for the first 10 days after infection (dpi) (Fig. 1a–c), but both treatments generated opposite spectral response at approximately 12 dpi (Fig. 1d). After the incubation period, plants inoculated with the pathogen started to express the disease symptoms (12 dpi). In the first instance, the plants inoculated with the pathogen increased the reflectance in the range of 550–700 nm (green, yellow, and red) from 0% up to ~ 12% after the incubation period, while those subjected to water stress showed a decrease of up to 5% in their reflectance at the green peak (550 nm). The control plants only had noticeable changes in the range of the near infrared (NIR) measured at 12 dpi, decreasing approximately 10% of reflectance in both cultivars. Regarding the measured infrared range (700–1000 nm), the Williams clones presented a rapid decrease in reflectance from maximum 75% to minimum 12% at 12 dpi, for the plants infected and subjected to water stress, respectively. The Gros Michel clones did not show significant variation in the Vis range in the treatments carried out, but they showed a slower decrease in the measured NIR, from 62% (0 dpi) to 25% at 12 dpi; however, it is important to note that the plants subjected to water stress in this clone did not significantly vary their average reflectance spectrum in the measured NIR range in comparison with healthy plants (Fig. 1e–h).

Fig. 1
figure 1

Spectral response from leaves of healthy plants (green line), infected with R. solanacearum (red line), and subjected to water stress (blue line), of the Gros Michel (a–d) and Cavendish (“Williams”) (e–h) banana cultivars during the incubation period of banana bacterial wilt

Full size image

Spectral response characterization of two banana varieties during the bacterial wilt incubation period

To detect Ralstonia infection in the asymptomatic period, starting from the identification of relevant specific wavelengths, a binary classification of the healthy leaves and the diseased plants was performed. The RELIEF algorithm was used to separate the healthy and the diseased plants, reducing the measured information spectrum to obtain specific wavelengths. Day 0 is the baseline weight value of the wavelengths, where the value was near 0 for the measured range in all treatments for both cultivars (Fig. 2). This means that no wavelength discriminates treatment groups that day, and thus, we can state that the plants start homogenously in terms of spectra. The specific wavelengths relevant for R. solanacearum infection (biotic stress) were 525, 595, 710, and 770 nm, clarifying that the 510–700 nm, 700–750, and 770 nm ranges were especially relevant after 4 dpi in clone William (Fig. 2a). In the Gros Michel clones, the RELIEF algorithm assigned greater weights to wavelengths at 430, 605, 705, 765, and 965 nm, with greater relevance also from 4 dpi (Fig. 2b).

Fig. 2
figure 2

Specific relevant wavelengths for classification of plants with vascular wilt for R. solanacearum during the incubation period and water stress of Cavendish (a, c) and Gros Michel (b, d) banana clones

Full size image

The weights of the specific relevant wavelengths for the classification of plants with water stress of the Cavendish variety (Fig. 2c) were greater than those presented by Gros Michel (Fig. 2d), with a specific wavelength standing out at 430 nm and the spectral ranges 500–600 nm and 770–1000 nm. Otherwise, only wavelengths at 500 and 740 nm were selected for the Gros Michel variety (achieving weights > 0.1).

The PCA shows that wavelengths 710, 770, and 965 nm allow the discrimination of R. Solanacearum-inoculated plants from healthy plants at 8 dpi. To test the specificity of these wavelengths, a check was made on the data from day 0, confirming that at the beginning of the experiment, these wavelengths do not have a discriminatory capacity for the treatments. These wavelengths allow us to distinguish Williams clones infected with R. solanacearum from healthy plants with a 96% accuracy at 8 dpi and 100% accuracy at 12 dpi (Fig. 3a, b). Following the same pattern, the Gros Michel plants with bacterial wilt could be distinguished from the healthy plants 98% of the time at 8 dpi and 100% of the time at 12 dpi (Fig. 3c, d).

Fig. 3
figure 3

Histograms for the first linear discriminant dimension of Cavendish (“Williams”) (a, b) and Gros Michel (c, d) banana clones infected with R. solanacearum and their respective controls, at 8 dpi (a, c) and 12 dpi (b, d). The wavelengths selected to perform the discriminant analysis were 710 and 770 nm for the Cavendish, and 705, 765, and 965 nm for the Gros Michel clones

Full size image

Discussion

This research found differences in the specific characteristics of the Vis–NIR ranges for the reflectance spectra of healthy plants of the Gros Michel and cv. Williams varieties. These results agree with other authors’ reports, which evidence a variation in the spectral response of different banana varieties. The low reflectance values have been found on the green peak (~ 550 nm) in the visible range (~ 400–740 nm) with average values between 5 and 20% and the infrared valley (~ 740–1100 nm) varies between 40 and 80% (Sinha et al. 2020). It is widely accepted that the low reflectance in the Vis region and its variation in plant leaves depend on the content and concentration of photosynthetic pigments, mainly chlorophylls, carotenoids, xanthophylls, and anthocyanins (Sims and Gamon 2002).

Plants' leaves undergo physiological changes during their development, varying the concentrations of photosynthetic pigments and spectral responses that can be used for the application of narrow-band indices related to plant biochemistry (Fernandes et al. 2020). Accordingly, it is important to consider the position of the plant leaf when making comparisons by reflectance spectra obtained from different studies. In the experimental design carried out, the banana leaf photosynthesis rate is mainly driven by its intrinsic adaptations (Siles et al. 2013), considering that the effects of the variation of the microclimatic conditions surrounding the plants’ leaves of both varieties could be “negligible.” A lower reflectance (hence, higher absorbance) was obtained at day 0 of leaf 3 in the Vis range for dpi's Gros Michel variety (~ 5%) for these experiments. These values were reached in the Williams variety only up to 24 dpi, indicating slower development speed and lower chlorophyll concentrations for this variation.

Evidence for the internal reflection mechanism in the 700–1000 nm spectral range in plant leaf tissues is quite strong from before the 1970s (Knipling 1970). To explain this mechanism, we have to consider that water fills the air cavities and forms a continuous medium of liquid phase throughout the leaf. When leaves wilt during dehydration and senescence, many of the reflective interfaces are eliminated as the internal air space is reduced and the cell walls draw together, which is called “far infrared” (IFR). Consequently, there is little internal absorption for these tissues, but there is high reflectance in the leaves of healthy plants. For example, about 60% of the radiation is scattered upwards through the incidence surface in healthy plants of the two varieties studied, but it rapidly decreases to values of between 20 and 25% in the plants infected with R. solanacearum.

Previous studies confirm that R. solanacearum has a quick expansion mainly in the root and stem of the plant (Liberato and Gasparotto 2006). Its movement is limited to the upper tissues, possibly due to the thickening of the membranes and/or the accumulation of electron-dense materials in the vessels and parenchymal cells (Nakaho et al. 2000). The low (or null) densities of the pathogen in the leaf xylem affect the indirect spectral response. Perhaps this explains the slow increase in reflectance at the Vis range during the disease incubation period in both varieties, being more evident in the Williams variety at the end of said period. Otherwise, reflectance in the 700–1000 nm range sharply dropped from 70–75% to 25% in the two varieties of plants inoculated with R. solanacearum. The same pattern was observed in the average spectrum of leaves of var. Williams that were subjected to water stress, confirming that the model of infrared collapsing in the mesophylls of the diseased plants since dehydration is generated due to the plugging of the vascular plant bundles.

Although the average reflectance spectra in the Vis range in plants inoculated with R. solanacearum are similar to those seen at a glance, the RELIEF classification algorithm was able to show relevant wavelengths in this spectral range. The algorithm determines that wavelengths (or spectral range) are highly relevant if they can distinguish between healthy and infected plants. Based on this logic, it assigns a weight to each of them (Urbanowicz et al. 2018). The highest wavelength weights of different regions of the measured spectrum were highlighted in this research (Fig. 2), choosing radiation rates at 430, 525, 595, 605, 705, 710, 770, and 965 nm. These findings suggest that the concentrations of chlorophyll A (maximum absorption at 430 nm), chlorophyll B (second maximum absorbance peak at 430 nm), and possibly carotenes (maximum absorption between 490 and 570 nm) are decreasing during the pathogenesis process (Blackburn 1999; Sims and Gamon 2002). It was previously discussed that changes in leaf reflectance between 700 and 1000 nm are generally associated with light scattering and water content, and the specific spectral range around 710 nm is consistently related to different types of stress, whether biotic or abiotic (Carter 1993; Caturegli et al. 2020). After 6 days of infection, we were able to separate the banana plants infected with R. solanacearum from the healthy ones using these wavelengths, with correct classification percentages close to 100%. It is possible that, in the first days of the infection, the water balance, the concentration of photosynthetic pigments, and the cellular structure were not significantly affected by the spectral reflectance in the measured range. However, after 6 dpi, there are rapidly large changes in Vis and the measured NIR range (700–1000 nm) that can be explained by xylem vessels clogging (bacteria and polysaccharides), degradation of the cell wall of the conducting vessels and the adjacent parenchyma, as well as by the invasion of the phloem and the cortical tissue (Petrolini et al. 1986).

It is possible to improve the accuracy and efficiency of the spectroscopy reflectance-based diagnostic method for plant diseases by linking studies on changes in the spectral response for broader ranges (including the infrared near 750–2500 nm) with physiological, biochemical, and histological changes caused by pathogens on their respective hosts. The creation of symptom scales that can be additionally related to the spectral response from leaves at each stage of pathogenesis development could help standardize methodologies to quantify incidence and severity in crops of economic importance, making the results obtained by producers and researchers around the world more comparable.

Conclusions

Spectral characterization of plant varieties during their different stages of development is essentially a primary input to be obtained before starting any specialized applications, even in the early detection of plant diseases by spectroscopy reflectance. The spectral characterization for this research was carried out in the 400–1000 nm range of healthy plants, subjected to water stress, and infected with R. solanacearum from banana leaves of the Gros Michel and cv. Williams varieties. The healthy plants of both varieties showed a typically low reflectance in the visible range, but with the highest value at the green peak (~ 550 nm) in the Williams variety (~ 12%) compared to Gros Michel (~ 5%). In terms of the measured infrared range (700–1000 nm), both showed a range of values ranging from 60 to 70% for the longest time of the measurements, dropping to 50% after 12 dpi.

The average reflectance spectra of each plant variety infected with R. solanacearum during the disease incubation period were reported, exposing changes before the first 6 dpi. After the first week of inoculation, the healthy plants could be distinguished from those infected with R. solanacearum using wavelengths selected using the RELIEF algorithm, with a correct classification percentage ranging from 96 to 100%. It is remarkable the rapid decrease in reflectance between 700 and 1000 nm and the slight increase between 560 and 700 nm in water stress-submitted plants in cv. Williams, whereas water stress-submitted plants in var. Gros Michel maintain values close to healthy, suggesting a high-water stress tolerance for this variety.

The proposed methodology allows early detection of bacterial wilt by R. solanacearum and water stress in two banana varieties using spectroscopy reflectance. This technique constitutes an objective diagnosis and identification tool for diseases early detection in crops. Finally, preprocessing methods, feature classification algorithms, and multivariate analysis are efficient and indispensable tools to be used in hyper-dimensional arrays of spectral data in the Vis–NIR range to correctly classify healthy and infected plants with R. solanacearum in Gros varieties. Michel and Williams.