Canopeo and GreenSeeker applications as tools to support tropical pasture management

ABSTRACT: This study determined whether Canopeo and GreenSeeker measurements in Megathyrsus maximus can estimate plant height, dry matter mass, morphological components, and content of crude protein and neutral detergent fiber at different days of growth. Five plots of 5 × 25m M. maximus grass were defined: subplots of 1×1m were evaluated every three days, in which the pasture shows 3, 6, 9, 12, 15, 18, 21, 24, 27, and 30 days of regrowth. The subplot was evaluated for canopy height and mass accumulation. The canopeo index (CI) obtained at a higher height was lower than those obtained at a smaller height. Higher measurement height increased the normalized difference vegetation index (NDVI) relative to 0.10 m. The highest indexes were observed since 18 d of regrowth. Except for the CI evaluated at 0.10 m of height, the indexes were not correlated to the chemical composition of the forage. The CI and NDVI were positively correlated to plant height, dry matter mass, and leaf index, whereas both were negatively correlated with stalk index. Thus, lower evaluation heights for CI and NDVI can be a good predictor of forage height. Values of 0.83 and 85.8 for NDVI and CI, respectively, indicated an appropriate time to start the grazing of M. maximus.


INTRODUCTION
Pastures represent the most important source of livestock feeding for ruminants (BELLA et al., 2004).In this way, livestock systems need an accurate estimate of the biomass, canopy height, and nutritional value of the pastures to optimize pasture management, stocking rate, and animal productivity (CARDOSO et al., 2020).The vegetative mass is a function of canopy height (CH) and dry matter (DM) density (TUCKER 1980;MACHADO et al. 2002, FRICKE & WACHENDORF, 2013).Destructive sampling is the most accurate method to determine pasture availability (JÁUREGUI et al., 2019).Different methods have been developed to measure pasture production in the last two decades: rising plate meter and capacitance meter stick (SANDERSON et al., 2001).Traditional non-destructive methods have shown outstanding results in estimating green biomass locally (BELLA et al., 2004).However, with Ciência Rural, v.53, n.6, 2023.Campana et al. new technologies being developed, non-destructive sensing techniques can be used to evaluate pasture biomass, making it convenient for farmers.
Optical sensors that detect absorbed and reflected lights have been used to detect the qualitative differences among materials (KENYON, 2008).GreenSeeker ® is one of these sensors that uses lightemitting diodes in the red (650 nm) and NIR (770nm) (CRAIN et al., 2012).The reflectance reading is calculated by an internal microprocessor, obtaining the normalized difference vegetation index (NDVI).The NDVI can be correlated with forage nitrogen uptake (FREEMAN et al., 2007).ANDERSSON et al. (2017) used a combination of NDVI and falling plate height index to estimate pasture biomass.CAMPANA (2017) also observed a linear increase of NDVI related to nitrogen doses and biomass yield in tropical pastures.
The Canopeo ® mobile phone application is an image analysis tool developed in the MATLAB programming language (Mathworks, Inc., Natick, MA).The app can access the phone camera and analyze and classify all pixels in the image or video using color values in the red-green-blue system (PATRIGNANI & OCHSNER, 2015).It turns the color green into white with a smooth manual adjustment and produces the percentage of white pixels in a given frame (YELLAREDDYGARI & GUDMESTAD, 2017) that correspond to the pixels that meet the selection criteria (green canopy; PATRIGNANI & OCHSNER, 2015).Measuring the green color of an image's background by counting green pixels using digital image analysis may represent biomass (YELLAREDDYGARI & GUDMESTAD, 2017).
The Canopeo mobile device application was used to measure soybean canopy cover and showed promising results when compared with light interception measurements.The app showed to be very fast and valuable for this parameter (SHEPHERD et al., 2018). CHUNG et al. (2017) used the same application to evaluate sorghum biomass and concluded that the application could replace hand data collection of plant height, considering it as an easily accessible high-throughput phenotyping tool for quantifying biomass.
This evidence indicates a potential for Canopeo ® and GreenSeeker ® to be used as a tool to estimate biomass, light interception, and nitrogen of forage crops provided that proper calibration is done.Pastoral farming requires fast, reliable, and farmer-friendly methods to determine pasture availability, which is crucial to increasing the farm's profitability (JÁUREGUI et al., 2019).The present study hypothesizes that Canopeo and GreenSeeker, regardless of height measurement, provide similar estimates, which is positively associated with the biomass and chemical composition of M. maximus tropical pasture.The present study determined whether Canopeo and GreenSeeker measurements in M. maximus can estimate plant height, DM mass, morphological components, crude protein, and neutral detergent fiber at different days of growth.

MATERIALS AND METHODS
The trial was performed in the Grupo de estudos e Trabalhos em Agropecuária (GETAP), Universidade Federal de São Carlos (UFSCar), from December 2017 to February 2018.Before the trial, five plots of 5 × 25 m of a long-term managed M. maximus grass were defined.The pasture of each plot was mowed at 40-cm height 30, 24, 18, 12, and 6 days before the first sampling.
The first sampling was performed on December 09 th , 2017.Sampling from each plot was performed every three days throughout the season.After 30 days of regrowth, the plot pasture was mowed at a height of 0.40-cm, and another cycle began.It resulted in ten samples for each plot in each cycle.Moreover, it was evaluated in three subsequent cycles in the present study.The experimental period finished on February 19 th , 2018.

Plant height, Canopeo Index, NDVI, and destructive methods
Plot pasture was evaluated during each sampling day using non-destructive methods, such as plant height, CI, and NDVI, and destructive methods, such as "cut and weight."First, the pasture height was evaluated using a graduated ruler.The Canopeo was free obtained at Google Play Store ® .Then, in the same plot, images were taken before cutting using a mobile phone to obtain CI and NDVI measurements using a GreenSeeker ® crop sensing system (Trimble, CA, USA) at 16:00 h.Pictures were taken at 0.10, 0.20, 0.30, 0.60, and 0.90 m above the plants' canopy height using a monopod with a scale (Figure 1).Canopeo evaluates fractional green canopy cover (CI) based on color ratios in the picture.GreenSeeker ® was used at 0.10, 0.20, and 0.30 m above canopy height.The GreenSeeker ® sensor captures incident and reflected light from plants at 660 ± 15 nm (red) and 770 ± 15 nm (NIR).Five points were evaluated in each parcel.Normalized difference vegetation index was calculated according to the following equation (MULLA, 2013): Pasture biomass (i.e., accumulated biomass above 0.40-cm height level) was determined by cutting 1 m 2 from each same plot (PENATI et al., 2005).Fresh mass was weighed, and a sample was used to assess the morphological composition, which was evaluated by the manual separation of the leaf lamina and stem.After these evaluations, each sample was dried for 72 h in a forced-air oven at 60 °C and ground in a knife mill (SL-31, Solab Científica, Piracicaba, Brazil) through a 1-mm sieve.Moreover, the samples were analyzed for DM (method 950.15,AOACInternational, 2016), crude protein (CP; N × 6.25; method 984.13,AOAC International, 2016; Kjeldahl method), and neutral detergent fiber (aNDF-NDF) using α-amylase, without the addition of sodium sulfite, and expressed including residual ash ( VAN SOEST et al., 1991).

Data analysis
The subplot measure was considered independent and defined as the experimental unit.Different height measurements were considered as measurements in the same experimental unit.Therefore, CI and NDVI were evaluated using the PROC MIXED of SAS 9.4, considering the following statistical model: and , in which Y ijk is the observed value of the dependent variable; µ is the overall mean; T i is the fixed effect of time (i = 1-10) and the random error associated with each sample (j = 1-75); H k is the fixed effect of height (k = 1-3 for NDVI and 1-6 for CI); e ijk is the experimental error; N stands for Gaussian distribution; σ ω 2 is the variance associated with the random error of experimental units; NRM stands for approximately normal multivariate distribution; and R is a matrix of variance and covariance due to repeated measures in the same experimental unit.CS, CSH, AR (1), ARH(1), TOEP, TOEPH, UN, and FA(1) matrixes were evaluated considering the Bayesian method.
Plant weight, mass, and chemical composition data were analyzed considering the following model: with , in which Y ij is the observed value of the dependent variable; µ, T i , and N were previously defined; and e ij is the experimental error and the variance associated with residual for each time of plant regrowth.Differences among the evaluation times were studied using Fisher's means test at 5% of probability.
Pearson correlations between the indexes (Canopeo and NDVI) obtained at different heights and plant information (plant height, DM mass, morphological and chemical composition) were obtained using PROC CORR.Significant correlations were studied with simple linear regression using PROC REG.For all statistical analyses, a 5% level of significance was considered.

CI and NDVI
Height and age of plant interaction showed no significant effect (P = 0.99) on the CI and NDVI (Table1).The CI obtained at higher heights (0.60 and 0.90 m) was lower (P ≤ 0.05) than that obtained at smaller heights.Conversely, higher height measurements (0.20 and 0.30 m) increased (P ≤ 0.05) the NDVI relative to 0.10 m.Days after the harvest affected (P < 0.01) both indices (Figures 2 and 3).The lowest (P ≤ 0.05) indices were observed at 3 and 6 days of growth.The highest (P ≤ 0.05) indices were observed since 18 days of regrowth, whereas intermediary values were observed between 9 and 15 days.

Forage mass, height, and chemical composition during growth
Plant height and DM mass increased (P ≤ 0.05) almost linearly up to 30 after harvest (Figures 4 and 5).At 18 days of growth, the plants showed 2920 kg ha −1 of available forage and a height of 0.903 m.Throughout the cycle, time showed no effect (P ≥ 0.36) on the plants' CP and NDF content (Figure 6).
However, at a higher age (21 and 30 days), the plants had a higher (P ≤ 0.05) DM content than the younger ones.The highest NDVI was observed since 18 days of regrowth.The NDVI value was 0.83, CI was 85.8, and CH was 0.903 m at 18 days of pasture regrowth.

Pearson correlation between indices (NDVI and CI) and chemical composition
The CI measurements at different heights did not correlate (P ≥ 0.23) with the DM and NDF content of plants (Table 2).However, a positive correlation was observed (R = 0.32 and P = 0.04) between CI measured at 0.10 m of height and the CP content of pasture.Moreover, no correlation was observed (P ≥ 0.14) between CP content and CI measurements obtained at the highest heights (≥ 0.20 m).In addition, there was no correlation effect (P ≥ A-E Fisher means test at 5% of probability.Bars are standard error of means. Table 1 -Canopeo and Green Seeker (GS) index of Panicum maximum grass at different evaluation height (mean ± SE).

Pearson correlation between indexes (NDVI and CI) and height, DM mass and indexes of stalk and leaves of forage, and regressions
Both indices (NDVI and CI) showed a significant (P < 0.001) correlation with plant height, DM mass, and indices of stalk and leaves (Table 3).The correlations between indices and plant height range from 0.53 to 0.57.Higher height (≥0.30m) of CI measurement showed a higher correlation (0.54-0.56) than that observed using a lower height of measurement or NDVI (0.47-0.49).
The stalk index was more correlated with the CI (−0.42-−0.40)than the NDVI (−0.34 and −0.33).Finally, the NDVI showed a lower correlation   (0.33-0.34) with a leaf index than CI (0.39-0.44).In regression equations (Table 4), it is possible to observe that increased height of CI and NDVI evaluation showed a positive effect on intercept estimation and reduced slope estimation when estimating plant height, DM mass, and leaf index.

DISCUSSION
The was reduced with increasing height evaluation from the canopy, whereas the more considerable distance between the GreenSeeker device and canopy increased the NDVI values.Canopeo initially was tested with a camera kept at about 1.5 m from the top of the canopy using a 1.5 m monopod (PATRIGNANI & OCHSNER, 2015), but like any other measurement tool, it depends on the end user's operation and cannot compensate for some operational errors by the user.A limitation of Canopeo is the need to always keep the camera at an appropriate height above of the canopy (PATRIGNANI & OCHSNER, 2015).We decided to try different heights from the top of the canopy because tropical pastures can be naturally high.However, the NDVI value is obtained using the energy relation reflected light or irradiated in the infrared and red near the canopy.Contrary to our findings, MARTIN et al., (2012) observed that a distance between GS and canopy reduces the NDVI values.Lower heights are also more attractive for the user because less effort is required to hold, finding good NDVI values for a height of 31 cm (MARTIN et al., 2012).
Both indexes (NDVI and CI) showed a significant correlation with plant height, DM mass, and indexes of stalk and leaves, but the increase in height level leads to an increase in the intercept and a reduction in the slope for the CI and NDVI, showing a lower response of the model to predict grass height for both the CI and NDVI.Lower height assessments for the CI and NDVI can be a good predictor of forage height, and further research is needed to assess the different types of heights using the Canopeo and GreenSeeker device applications in the different classes of forage to produce more concrete conclusions.However, it is essential to emphasize that Canopeo is faster in calculating canopy cover percentage, can be easily applied in the field (SHEPHERD et al., 2018), and has a low-cost and easy operation (CHUNG et al., 2017).

CONCLUSION
Values of 0.83 and 85.8 for NDVI and CI, respectively, indicate an appropriate moment for the start of the grazing of M. maximus.Lower height measurements from the top of the canopy for CI and NDVI better predict forage height.Canopeo and GreenSeeker can be a powerful, accessible, and easy-to-use tool for farmers to facilitate tropical pasture management.

Figure 1 -
Figure 1 -Representation of height of evaluation.

Figure 2 -
Figure 2 -GreenSeeker index (NDVI) of Megathirsus maximum grass at different time of growth and evaluation height.

Figure 3 -
Figure 3 -Canopeo index of Megathirsus maximum grass at different time of growth and evaluation height.A-F Fisher means test at 5% of probability.Bars are standard error of means.

Figure 4 -
Figure 4 -Megathirsus maximum grass dry matter mass at different time of A-E Fisher means test at 5% of probability.Bars are standard error of means.

Figure 5 -
Figure 5 -Megathirsus maximum grass height at different time of growth.A-E Fisher means test at 5% of probability.Bars are standard error of means.

Figure 6 -
Figure 6 -Megathirsus maximum grass DM, CP, and NDF content at different time of growth.A-C Fisher means test at 5% of probability.Bars are standard error of means.

Table 3 -
Pearson correlation of CI and NDVI of Megathirsus maximum grass at different evaluation height with plants height, DM mass, and index of stalk and leaf.

Table 2 -
Pearson correlation of CI and NDVI index of Megathirsus maximum grass at different evaluation height with dry matter (DM), crude protein (CP), and neutral detergent fiber (NDF) content [R (Probability)].

Table 4 -
Regression of plant height, DM mass and leaf index of Megathirsus maximum in function of CI and NDVI evaluated at different height.