SPATIAL VARIABILITY OF CHLOROPHYLL CONTENT IN A TIFTON 85 BERMUDAGRASS PASTURE IN A TROPICAL REGION

Precision Agriculture techniques, such as the management of spatial variability of crop attributes, have been studied for several crops. However, few studies have been performed on Tifton 85 bermudagrass. Thus, this work aimed to analyse the spatial variability of chlorophyll content in a Tifton 85 bermudagrass production area, located in Seropédica, Brazil. A georeferenced grid was created to measure the chlorophyll content in two periods using a portable chlorophyll metre. Different geostatistical methods and models were evaluated in order to identify which had the best fit to analyze the spatial dependence of the chlorophyll content.The atribute was mapped based on interpolation by the ordinary kriging method. Therefore, kriging interpolation was used to create isoline maps, which were used to observe the spatial variability of the chlorophyll content. The methodology and maps generated proved to be of great value to the Tifton 85 bermudagrass producers.


INTRODUCTION
The use of Cynodon grasses in agriculture has grown exponentially due to the advantages they provide, especially in tropical and subtropical regions (HANNA; ANDERSON, 2008;BASEGGIO et al., 2015). In Brazil, Tifton 85 bermuda grass (Cynodon spp.) is predominantly used among ranchers (PEDREIRA et al., 2018). The extensive use of this forage by Brazilian ranchers is due to the advantages offered by this species, such as high hay productivity and tolerance to saline and sodic soils (FONSECA et al., 2006).
The monitoring of nutritional status and the management of Tifton 85 can be carried out by estimating the chlorophyll content (SILVA et al., 2011). This variable is directly proportional to the levels of some nutrients, mainly nitrogen. According to Barbieri Júnior et al. (2010), chlorophyll directly influences the amount of solar radiation absorbed by plants, and chlorophyll concentrations in leaves are closely related to photosynthetic rates and primary productivity.
For the proper management of a pasture area, according to Silva et al. (2011), it is possible to use a "chlorophyll meter" to directly measure the chlorophyll content in a plant and use the data obtained to indirectly estimate the need for nitrogen fertilization in the area (BARBIERI JÚNIOR et al. al., 2012), reducing sampling time and cost when compared to laboratory analyses.
The application of precision farming (PA) techniques, defined as agricultural practices based on information to manage the spatial variability of a crop (BERNARDI et al., 2016), combined with grass management, has the potential to increase productivity.
Within the PA, the use of GNSS (Global Navigation Satellite System) has a fundamental role in the application of remote sensing and geostatistics, which are fundamental techniques for the assessment of the spatial and temporal variability of attributes related to soils and plants of a crop .
For the processing of georeferenced data, geoprocessing is used, which include technologies for collecting, processing, analyzing and making available information with geographic reference (CÂMARA et al., 2004). Modeling via GIS makes it possible to merge these layers of information, expanding the ability to interpret data and assisting in decision making for the management of the production system (FILIPPINI ALBA, 2014).
Geostatistics is the study of a natural phenomenon, which can be characterized by the distribution in space of one or more variables, called "regionalized variables" (JOURNEL; HUIJBREGTS, 1978). The study of these variables aims to solve the estimation problems for those places where sampling was not performed (SANTOS et al., 2017).
As Brazil has large areas of pastures used for livestock or hay production and has had difficulties to expand the areas of planted pastures, the adoption of PA techniques is essential. PA tools were initially used in grain crops, but there are already several experimental and practical results indicating the potential for use in pastures (BERNARDI; PEREZ, 2014;BERNARDI et al., 2016).
The present study aims to evaluate the spatial and temporal variability of chlorophyll content in an area with Tifton 85 grass using geostatistics and kriging techniques to generate thematic maps representing the spatial variability of chlorophyll content in a grass pasture canopy -Tifton 85.

Study area
The study area is located in the Seropédica municipality, Rio de Janeiro state, Brazil, at geographical coordinates 22°44'38''S and 43°42'27" W SIRGAS 2000 and belongs to the Federal Rural University of Rio de Janeiro. The study field has a total area of 3.25 ha cultivated with Tifton 85 bermudagrass. The climate of theregion is classified as Aw, according to Köppen's classification (KÖPPEN, 1936), with mean annual precipitation of 1,213 mm and mean annual temperature of 23.9 ºC. Figure 1 shows the variation of temperature and precipitation during the study period.

Sampling grid
The study area was delimited using an SP-60 GPS receiver (Garmin, Kansas, USA). In this area, two sampling grids ( Figure 2) were delimited. Sampling grid 1 (A1-07/09/2014) was defined considering a 15 x 15 m spacing between samples, generating a total of 57 sampling points and a sampling density of 17.6 points ha -1 on the referenced date. Sampling grid 2 (A2-07/16/2014) was defined considering a spacing of 30 x 30 m, generating a total of 47 sample points (14.46 points ha -1 ). The geographical coordinates of the sampling points were also collected with the help of the GPS receiver.
The chlorophyll content was determined using a portable chlorophyll metre, model CFL 1030 (Falker Automação Agrícola, Porto Alegre, Brazil), operated according to the technical specifications provided by the equipment's manual. For the sampling, the first fully expanded leaf (in the direction from the end to the base of the tiller) was selected, which was fully exposed at the time of sampling, according to the methodology described by Barbieri Júnior et al. (2012).

Descriptive statistic analyses
A descriptive statistical analysis of the collected data was performed, evaluating the maximum, minimum, mean, coefficient of variation (CV), standard deviation (SD), and variance (Var) to observe variations in the collected dataset.

Geostatistical analyses
The spatial dependence of the chlorophyll content was analysed by fitting classical (MATHERON, 1965) Where N(h) represents the number of experimental pairs of observations, Z(x i ) and Z(x i +h) at locations x i and xi+h, respectively, separated by a distance h. The semivariogram is represented by the graph   to the (h), the coefficients of the theoretical semivariogram model are estimated: nugget effect (C 0 ), sill (C 0 +C 1 ) and range (a) (CRESSIE, 1992). According to Ferraz et al. (2012), the robust estimator of the semivariogram is less susceptible to the influence of the dataset values than the classical estimator. Thus, the robust estimator is described in Equation 2: This estimator considers that the differences Z(x i +h)-Z(x i ) are distributed normally for all pairs (x i +h, x i ). The denominator of the equation is the trend correction, and the square root transformation of the differences is presented, similar to the moments of a normal distribution (FERRAZ et al., 2012).
The semivariogram models were fitted with error minimization methods, namely, the Ordinary Least Squares (OLS) and Weighted Least Squares (WLS) methods, estimated using the classical and robust estimation methods and the Maximum Likelihood (ML) and Restricted Maximum Likelihood (REML) methods. For all error minimization methods, the spherical, exponential, and gaussian models were fitted to the experimental semivariograms.
The best fit models were selected based on the cross-validation method (ISAAKS; SRIVASTAVA, 1989), using the mean error (ME), root mean square error (RMSE), average standard error (ASE), and root mean square standardized error (RMSSE). The criteria for choosing the best fit model were the ME and ASE values closest to zero, the lowest RMSE value, and the RMSSE value closest to one (FERRAZ et al., 2012).
After fitting the semivariogram models, the Tifton 85 chlorophyll content was interpolated for non-sampled sites using the ordinary kriging interpolation method. Subsequently, the thematic maps were prepared for the two sampling dates.
Descriptive statistical analysis and geostatistical analysis were performed using the geoR library of the R (CORE DEVELOPMENT, 2014) statistical computer system (RIBEIRO JÚNIOR;DIGGLE, 2001).

Descriptive statistics
The measured chlorophyll contents have a CV above 10%, indicating variability among the data (Table 1). According to Gomes e Garcia (2002), a CV below 10% indicates homogeneity.
The CV, according to the classification of Warrick and Nielsen (1980) was intermediate, where the highest CV was observed for the sampling on 07/16/14 (40.3%). This variation can be attributed to the size of the sampling grid used during these periods.
The mean values of the two periods were similar (26.33 and 25.93). The maximum, minimum, and median values were lower for the sampling on 07/16/14, which was due to the plants being at a later development stage. As a consequence, the SD and Var were larger for this period, indicating a greater variability of the data in the area.
As noted, the data referenced to the mean, minimum and maximum values found in Table  1 shows the variation in values for this variable. However, knowledge of these measures alone cannot be used as the only way to identify a variable expression in an area. Therefore, a geostatistical analysis is carried out with the objective of verifying the spatial variability of a variable and, if identified, thematic maps are made that help to understand its behavior in the field.

Spatial distribution
The cross-validation criteria allowed the selection of the best semivariogram models for each data collection date (Table 2 and 3). The chlorophyll content values for the sampling on 07/09/2014 were estimated using the WLS-C method with a spherical model (Table 2 and Figure 3A). For the sampling on 07/16/2014, the variability of chlorophyll content was estimated using the ML method fitted with the exponential model (Exp) ( Table 3 and Figure 3B).    According to geostatistical analysis, the semivariogram modelling using the spherical model did not detect spatial variability in Figure 3, for the sampling on 07/09/2014, showing a pure nugget effect. Thus, the spatial variability of chlorophyll content in this area and for this sampling grid (30 x 30 m) is considered to be random, and according to Cambardella et al. (1994). The distances between samples must be reduced to detect the spatial variability in chlorophyll content. However, this result can also be attributed to field measurement errors. In Figure 3B, spatial dependence was detected, with a C 0 value of 61.02 for 07/16/2014.
The values differed in the two sampling periods, with 32.764 m on 09/07/2014 and 152.909 m on 07/16/2014. According to Ferraz et al. (2012), a sampling plan that considers the existing spatial variability would also have to consider the differences in the range found, always using values lower than the smaller range value to construct a more adequate sampling grid.
After fitting the semivariogram, chlorophyll values were estimated by ordinary kriging for the entire area. The thematic maps of the spatial distribution of chlorophyll content for the two periods are represented in Figure 3.
An analysis of the maps (Figure 4) reveals the temporal variation of the chlorophyll content during the study period. In the northern region of the area Figure 4B (07/16/2014), there was a reduction in the estimated chlorophyll content relative to Figure 4A (07/09/2014). Oliveira et al. (2010) found similar results for the chlorophyll content in June and July, when there is a reduction of the photoperiod, lower water availability, and lower net radiation (decreasing mean air temperatures), with a consequent reduction of the growth rates of the plants.
Another plausible explanation for the chlorophyll content variation can be attributed to the climatic factors and relief of the study site. The automatic weather station at Seropédica km 47 of the Brazilian National Institute of Meteorology (INMET) recorded that in the days before the 07/09/2014 sampling, there was a seven-day dry period, interrupted by 11.2 mm of rainfall on the eve of the first measurement (INMET, 2014) and that in the interval between measurements, there was another dry period. This accumulation of dry periods may have triggered senescence of the plants, thus affecting their development. This reduction in chlorophyll content in grasses caused by water scarcity was reported by Silva et al. (2006), where the reduction of chlorophyll was observed in grasses under increased water deficit.
The most significant decrease in chlorophyll content occurred in the northern part of the area, which is the highest region. This high slope reduces the water infiltration (OLIVEIRA et al., 2012) and storage rates in the soil. The centre-south part of the area presents a less irregular relief, which facilitates water storage by the soil and helps with crop maintenance. The thematic maps of chlorophyll content ( Figure  4) in conjunction with chemical soil attributes maps are important tools for the management of Tifton 85 bermudagrass, reducing the yield variability in the area during the production cycle.

CONCLUSIONS
• The chlorophyll content present in Tifton 85 grass evaluated in the two study areas showed heterogeneity by calculating the coefficient of variation obtained in the basic statistics. Through interpolation by ordinary kriging, it was possible to select the best method and best model to adjust the data and thus make the spatial variability maps of the chlorophyll content.
• The use of geoestistics associated with PA was an important tool to identify the spatial and temporal distribution of chlorophyll content in Tifton 85 grass in non-sampled locations, which will reflect in a better management for this crop.

DECLARATION OF INTERESTS
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.