Spatial and Temporal Variability of Vegetation Indices with Industrial Tomato Yield

,


Introduction
Tomatoes (Solanum lycopersicum L.) are a widely cultivated crop and one of the most consumed vegetables due to their nutritional value, particularly the antioxidant compounds lycopene and carotenoid, which have been associated with health benefits (Colanero et al., 2020;Martí et al., 2016;Mozos et al., 2018;Sousa et al., 2020).
Crop yield can vary across seasons, locations, and even within the same field due to management practices and environmental conditions (Kayad et al., 2022).Monitoring this variability is crucial for farmers, landowners, and insurers to make informed decisions (Kayad et al., 2022;Oliveira et al., 2018).Soil, as a key factor in agricultural production, exhibits considerable variability in tropical regions, with diverse physical and chemical characteristics that influence crop growth and productivity (Amorim et al., 2022).Spatial and temporal variability of soil properties significantly impact crop yield, making it important to understand this variation and its relationship with crop performance for precision management practices (Kayode et al., 2022;Zhou et al., 2020).
Precision agriculture offers technologies to manage field variability (Finkenbiner et al., 2019), with geotechnologies such as remote sensing proving to be powerful tools for this purpose (Medorio-García et al., 2020;Wang et al., 2019).Through remote sensing, imagery is acquired and analyzed using specific algorithms to generate suitable information layers for decision-making and optimizing input rates (Kayad et al., 2022).Geostatistical methods enable the quantification of spatial variation, revealing the spatial structure, randomness, and correlation of samples, facilitating spatially irregular sampling for optimal interpolation (Zhou et al., 2020).
The chlorophyll content is an indicator of vegetation photosynthesis, which directly influences the growth and yield of crops, and is therefore an indicator of their health (Yuan et al., 2022).The SPAD (Soil Plant Analysis Development) chlorophyll meter has been widely used to quickly analyze the status of chlorophyll and nitrogen in leaves, indirectly estimates the chlorophyll concentration in a time-saving non-destructive way (Mehrabi & Sepaskhah, 2022).Leaf chlorophyll content can also be determined by the destructive method of solvent extraction in the laboratory.
The Normalized Difference Vegetation Index (NDVI), derived from multispectral images, is an effective indicator for assessing vegetation status and quantified attributes.Remote sensing images provide composite images for subsequent interpretation and analysis (Huang et al., 2021).According to Barros et al. (2020), NDVI is relevant because it can provide a view of the condition of vegetation, through images generated by remote sensors, usually satellites, drones or ground sensors.Furthermore, if the NDVI values are well below the expected value, it shows that there is an anomaly that must be checked.Chen et al. (2021) describe that NDVI data derived from Landsat satellites are pertinent resources for vegetation monitoring.Understanding what may be occurring is a precondition for developing effective strategies for greater productivity.Big data availability in agriculture, supported by free remote sensing data and sensor-based information, offers opportunities to understand field variability and implement site-specific precision agriculture practices (Toscano et al., 2019).
Considering the aforementioned, this study aims to evaluate the influence of spatial and temporal variability of vegetation indices on industrial tomato yield.It seeks to establish relationships between field-measured nitrogen content, remotely sensed vegetation indices, and the comparison of ground-based and orbital NDVI for industrial tomato cultivation.

Experimental Area Location and Characterization
The study was carried out at the Barcelos Farm in the city of Anápolis, Goiás State, Brazil.It lies at an average altitude of 995 m and geographical coordinates of 16º 25' 57.62"South latitude and 48º 50' 29.61" West longitude.According to the Köppen classification system, the region has an Aw type climate (i.e., Tropical Rainforest Climate).
The area has a low-slope topography, and the soil is classified as a Dark Red Latosol with a clayey-loamy texture (Teixeira et al., 2017).
The industrial tomato crop, cultivar NS 901, was established on a 55-ha area (Figure 1), irrigated using a center pivot system.The seedlings were transplanted in single rows with a spacing of 1.0 m between rows and 0.36 m between plants, resulting in a population density of 36,000 plants ha -1 .The area was fertilized before seedling transplanting with 1100 kg ha -1 of 04-30-16 N-P-K formulation.Topdressings were performed with 150 kg ha -1 potassium chloride (KCl), 170 kg ha -1 ammonium sulfate (NH 4 ) 2 SO 4 and 5 kg ha -1 potassium nitrate (KNO 3 ), split into three applications.

Georeferencing of the Area
Before planting, a sampling grid was established using a Garmin ETREX global positioning system (GPS) device with geographic coordinates (GSW84), and geographic coordinates were collected at the ends and center of the pivot, defining the arrangement of the sampling grid.The area was composed of a sampling grid with 61 georeferenced points spaced at 90 x 90 m intervals (Figure 1).

Data Collection
Crop vegetation indexes were collected 37, 53, and 69 days after transplanting (DAT).It was aligned with the satellite overpasses and intervals of nitrogen top-dressing (16 days).Industrial tomato yield data were collected at the harvest point (108 DAT) in August of the 2017 crop season.The data were collected at each georeferenced point.

Soil Plant Analysis Development (SPAD) Index
The Soil Plant Analysis Development (SPAD) index was determined by taking three readings using a portable chlorophyll meter (ClorofiLOG® model CFL 1030), which provides total chlorophyll values with the Falker Chlorophyll Index (FCI).At each sampling grid point, the SPAD index value comprised the average of readings taken on ten leaves from ten industrial tomato plants.The readings were taken on the central region of leaves in the upper third of the plants between 10:00 a.m. and 2:00 p.m. (Fontes & Araújo, 2007;Singh et al., 2017).

Laboratory Determination of Foliar Nitrogen (FN) Content in Industrial Tomatoes
Foliar nitrogen (FN) contents were determined on the same ten leaves used for SPAD.After SPAD determination, the leaves were removed from each plant, stored in pre-labeled plastic bags, and sent to the analytical chemistry laboratory at the State University of Goiás to determine the NF concentration.
The samples were dried in a forced-air oven at 75 ± 3°C until reaching a constant weight (four and a half hours).Then, the samples were ground using a Wiley-type mill with a 0.5 mm sieve, and a fraction of 0.5 g of the dried samples was weighed.For NF extraction, the samples were analyzed using the micro-Kjeldahl method (Kjeldahl, 1883).
After the NF extraction process, the calculation was performed to obtain the percentage of total nitrogen present in the leaf (Badr et al., 2016;MAPA, 2013;Souza et al., 2014), according to Equation 1.
In which: %TN -percentage of total nitrogen; V -milliliters of 0.1 mol L-1 HCl solution used in titration; N -theoretical normality of 0.1 mol L -1 hydrochloric acid solution; M -mass of the sample in grams.
The FN content analysis was performed 37, 53, and 69 DAT of the tomato seedlings.

Acquisition of Radiometric Data for Industrial Tomato Crop
Radiometric readings were obtained using a spectroradiometer with a passive sensor from Ocean Optics, model USB 2000+RAD.The device was used to obtain reflectance data within the spectral range of 400 to 900 nm at a 0.34 nm resolution.It was connected to a portable computer, where the data were stored.The device was installed on a metallic support, maintaining a height of 1 m above the crop canopy (Singh et al., 2017).
For this study, the values of reflectance corresponding to the wavelengths of Landsat 8 satellite bands were used.
In which: NDVI -Normalized Difference Vegetation Index; IR -average value of near-infrared reflectance in Band 4; R -average value of red reflectance in Band 3.
The reflectance values were obtained as the average of three readings taken around each sample point within a 1.5 m radius.The readings were taken between 10:00 a.m. and 2:00 p.m. due to the consistent solar radiation.The readings were performed 37, 53, and 69 DAT of the tomato seedlings.

Remote Sensing of Industrial Tomato
Remote sensing was performed by acquiring Landsat 8 satellite images corresponding to the days 06/05/2017 (37 DAT), 06/21/2017 (53 DAT), and 07/06/2017 (69 DAT), with a 16-day interval following the satellite overpass days.After the acquisition, the images were obtained from the National Institute for Space Research (INPE), and using the QGIS software, vegetation indices were calculated using the Normalized Difference Vegetation Index (NDVI) as the variable (Rouse et al., 1974;Zanzarini et al., 2013), according to Equation 3.

Productivity of Industrial Tomato Crop
The yield of industrial tomato crop was evaluated at the harvest time (108 DAT) in August of the 2017 crop season.
At each georeferenced point of the sampling grid, a representative 1-m² area was demarcated around the point, and all fruits within the demarcated area were collected.The collected fruits were weighed using a calibrated portable digital scale (Marine-type fisherman scale) with a precision of 0.01 g and a maximum capacity of 40 kg.After weighing, the sample weights were extrapolated to t ha -1 for all sampled points (Figueiredo et al., 2016).

Descriptive Statistics
Exploratory data analysis was conducted to identify outliers and assess the statistical distribution of the vegetation indices and industrial tomato productivity data.The critical limit for outliers was defined based on the interquartile range (IQR), calculated as the difference between the upper quartile (Q3) and the lower quartile (Q1).The upper limit for outliers was defined as (Q3 + 1.5 × IQR), and the lower limit as (Q1 -1.5 × IQR), where Q1 and Q3 are the first and third quartiles, respectively, according to the methodology adjusted by Bottega et al. (2013).

Geostatistical Analysis
To characterize spatial variability, geostatistical analysis was performed, where each evaluated attribute was analyzed individually to identify spatial dependence through calculation of simple semivariograms.The semivariograms were fitted based on the assumption of intrinsic hypothesis (Usowicz & Lipiec, 2017).
The mathematical model for the fitted semivariograms was selected based on the lowest sum of squared residuals (RSS), highest coefficient of determination (r²), and greatest degree of spatial dependence (SDD), according to Monteiro et al. (2017).
The semivariogram models for the studied variables were fitted using GS+ software, version 7. Once the spatial dependence of the studied variables was confirmed, ordinary kriging interpolation was used to estimate values at unmeasured locations, and thematic maps were built using Surfer 8 software.

Results and Discussion
According to descriptive statistics (Table 1), the different vegetation indices and productivity showed similar mean and median values, indicating a normal distribution of the data.This is supported by the skewness coefficients, which ranged from -0.68 to 0.54, indicating a low deviation from the mean.The variables exhibited negative kurtosis coefficients, ranging from -0.14 to -0.73, implying data flattening compared to the normal curve.The Soil Plant Analysis Development Index (SPAD), Foliar Nitrogen (FN) at the flowering stage (53 DAT), and Satellite-based Normalized Difference Vegetation Index (NDVI) at the early fruit formation stage (69 DAT) displayed a mesokurtic normal distribution.The remaining variables throughout the crop cycle exhibited a leptokurtic normal distribution.According to Sanquetta et al. (2014), kurtosis values below -0.263 and between -0.263 to 0.263 define a leptokurtic and mesokurtic normal distribution, respectively.
The coefficients of variation (CV) were classified following Ferreira Júnior et al. (2021).The vegetation indices variables were categorized as low CV when values were below 12%, while industrial tomato productivity was classified as medium with a value exceeding 12%.Similar results were found by Padilla et al. (2017) for vegetation indices in cucumber crops.
Foliar nitrogen content serves as an indicator of photosynthetic protein quantity and plays a significant role in understanding plant functioning and status (Jia et al., 2021).Foliar nitrogen analysis revealed mean values of 5.67%, 5.41%, and 4.40% at 37, 53, and 69 DAT, respectively, representing the nitrogen percentage (N) in the leaves of industrial tomato plants.Cheng et al. (2022) also observed a decrease in nitrogen concentration in cherry tomato plants as the growth period progressed.In the 2019 season, nitrogen concentration in plants ranged from 2.2%, 1.58%, to 1.44% at 34, 48, and 77 DAT, respectively.Lequeue et al. (2016) reported an average nitrogen content of 1.71% in tomato leaves.
The readings obtained for industrial tomato NDVI using a spectroradiometer showed mean values of 0.80, 0.78, and 0.65 at 37, 53, and 69 DAT, respectively, which aligns with the findings of Marino et al. (2014) and Fortes Gallego et al. (2015).Avotins et al. (2020) studied the NDVI for 6 tomato cultivars at plant heights, and observed that the values varied between 0.667 and 0.871.The results that the authors found show potential application in crop management by monitoring the quality parameters of tomato plant growth.
Regarding NDVI calculated from Landsat 8 satellite imagery for industrial tomato, the mean values were 0.48, 0.45, and 0.41 at 37, 53, and 69 DAT, respectively.Lykhovyd et al. (2022) observed an increase in average NDVI values for tomato crops from planting (0.17) to the fruit formation stage (0.69), followed by a decrease to 0.45 during fruit ripening.Maselli et al. (2020) observed, in tomato cultivation areas, an increase in NDVI (up to 0.7) until the moment of maximum biomass development.These findings were based on satellite imagery.
Vegetation indices displayed a reduction in nitrogen levels throughout the crop's development stages, particularly after 69 DAT, during the early fruit formation stage.According to Ronga et al. (2015), this stage corresponds to an increased nutrient demand by the plant.Guo et al. (2020) describe studies that showed that the concentration of N in crops reduces with increasing biomass.
The average productivity of industrial tomatoes was 80.0 t ha -1 .While these values were higher than those reported by Fortes Gallego et al. (2015) (66.0 t ha -1 ), they were comparable to the findings of Oliveira et al. (2018) (82.7 t ha -1 ) for the same crop.
Simple linear correlation analysis using Pearson's correlation coefficient (r) was employed to determine the correlation between productivity and vegetation indices of industrial tomatoes (Table 2).The Spectroradiometer-based Vegetation Index at 37 DAT (NDVI_Es_1) showed a moderate, positive, and significant correlation with NDVI_St_1 (0.454*).At 53 DAT, NDVI_Es_2 displayed a strong, positive, and significant correlation with NDVI_St_2 (0.551*), indicating a higher nitrogen concentration for vegetative development that can influence crop productivity.The readings at 69 DAT (NDVI_Es_3) and NDVI_St_3 exhibited a strong, positive, and significant correlation (0.704*).
The NDVI_Es and NDVI_St readings showed a strong, positive, and significant correlations throughout the crop cycle, indicating that both readings by through ground-based and orbital remote sensing images are significant to diagnose crop development.The wavelength reflectance for ground-based NDVI measurements in industrial tomatoes aligns with the values reported by Marino et al. (2014) and Fortes Gallego et al. (2015).
The spatial maps of vegetation indices and industrial tomato yield were generated using spherical semivariogram parameters, which had better fit by kriging interpolation to estimate values in unsampled locations for characterizing spatial dependency (Figure 2).The maps of vegetation indices showed little variation throughout the industrial tomato crop cycle.The first reading at 37 DAT (Figure 2A) revealed that the southern, central, and northeastern regions exhibited higher SPAD index variation, with small areas ranging from 436 to 472, higher than in other regions.Figure 2D illustrates the spatial distribution of FN content in the crop, highlighting the southeastern region with low nitrogen content ranging from 4.1% to 5.3%, while the southwestern and central-western regions had FN content exceeding 5.9%.
The NDVI readings at 37 DAT using ground-based and orbital remote sensing are shown in Figures 2G and 2J, respectively.Both methods identified low NDVI values in the southwestern and central-western regions, ranging from 0.70 to 0.78 for NDVI_Es and from 0.41 to 0.47 for NDVI_St.
At the second vegetation index reading during the flowering stage (53 DAT), there was a prevalence of low SPAD index in the northwest, south, and southeast regions, ranging from 380 to 424 (Figure 2B).Other regions showed an increase in the SPAD index due to the second nitrogen top-dressing application.Figure 2E characterizes the spatial dependency of FN content, revealing low FN content in the southwestern, part of central-western, and southeastern regions, ranging from 4.5% to 5.3%.A small area in the southeastern region showed even lower FN content.
The NDVI readings at 53 DAT exhibited less spatial dependency in the western, central, and eastern regions, as shown in Figures 2H and 2K for ground-based and satellite-based measurements, respectively.These maps (Figures 2H and 2K) illustrate the central region's greater NDVI reduction between 37 and 53 DAT.
For the third vegetation index reading at 69 DAT, during the early fruit formation stage, the SPAD index map (Figure 2C) indicates that the western, northern, and southeastern regions exhibited lower values, ranging from 345 to 411.The spatial dependency of FN content (Figure 2F) featured higher values of 4.2% to 5.4% in the western, northern, central-western, and southeastern regions, while other regions had lower values than 4.2%.
The NDVI maps obtained from spectroradiometer and satellite imagery readings are represented by Figures 2I and  2L, respectively, with the central region showing lower NDVI values due to nitrogen fertilization in the industrial tomato crop, ranging from 0.54 to 0.64 and 0.35 to 0.41, respectively.
The NDVI maps of the different crop stages highlight the similarity among the regions and a strong correlation between spectroradiometer and satellite imagery readings (Table 2).The maps revealed a strong spatial dependency of the evaluated vegetation indices in the northern region.These results enable the characterization of management zones and the identification of regions that may influence crop development.
The readings across the different crop stages showed the spatial and temporal variability of vegetation indices in industrial tomatoes.The maps revealed a strong spatial dependency of the evaluated vegetation indices in the northern region.These results enable the characterization of management zones and identification of regions that may influence crop development.
The yield map of industrial tomatoes (Figure 2M) highlights the southwestern, northern, and northeastern regions as those with higher yields (from 84 to 108 t ha -1 ).When correlated with all vegetation index maps, we noted that the regions with higher yields also had better vegetation indices for crop development.Table 2 showed a correlation between yield and vegetation indices during the flowering stage of the crop (53 DAT).Overall, regions with higher yields also had higher vegetation indices.

Conclusions
The spatial variability of vegetation indices, foliar nitrogen content, and industrial tomato yield can be determined.
The temporal variability between vegetation indices and foliar nitrogen content indicates that the highest correlation with tomato industrial yield is reached at the flowering stage.
The correlation between vegetation index maps and crop yield allows us to identify regions with higher productivity potential.
Regions with lower normalized difference vegetation indices (NDVIs) coincide with reduced industrial tomato yields.Additionally, there is a positive correlation of 55% between NDVI obtained through terrestrial and orbital remote sensing during the flowering stage of the tomato crop.
Temporal variability between vegetation indices and foliar nitrogen content shows low correlation across the developmental cycle of the crop.
Based on the obtained results, we recommend determining foliar nitrogen content and vegetation indices during the flowering stage of the industrial tomato crop.

Figure 1 .
Figure 1.Representative image of the industrial tomato production area Note.The area enclosed in the highlighted semicircle represents the satellite image of the industrial tomato production area at Barcelos Farm with georeferenced points.Source: Google Earth, 2018.

Table 1 .
Descriptive statistics of vegetation indices and yield of industrial tomato crop

Table 2 .
Pearson's correlation analysis (r) of vegetation indices and yield of industrial tomato crop