Introduction
In Brazil, dairy and beef cattle livestock production is mainly pasture based, which contrasts with practices in many other countries, in which most animals are fed in feedlots. Brazil has a pasture area of approximately 180 million hectares, comprising both native and cultivated pastures (Parente and Ferreira, Reference Parente and Ferreira2018), which accounts for a major proportion of Brazilian agricultural land.
In recent years, there has been an increase in stocking rate in Brazilian pasturelands, as a consequence of pasture intensification, which has led to improvements in livestock production (Martha Junior et al., Reference Martha Junior, Alves and Contini2012). In intensive systems, quantifying forage mass and its nutritive value is essential with respect to determining carrying capacity, supplementation, and management strategies. Pasture growth depends on post-grazing stubble mass and residual leaf area in rotational stocking systems (Parsons et al., Reference Parsons, Johnson and Harvey1988), and the determination of these characteristics makes an important contribution to decision-making on pasture management (Wachendorf et al., Reference Wachendorf, Fricke and Astor2017; Pezzopane et al., Reference Pezzopane, Bernardi, Bosi, Crippa, Santos and Nardachione2019).
In addition to quantifying forage mass for the determination of nutritive value, information relating to parameters such as crude protein and nitrogen contents can assist farmers in balancing animal diets. Furthermore, pasture nutritive value is related to animal production and is essential for high-performance dairy production (Starks et al., Reference Starks, Zhao, Phillips and Coleman2006; Pullanagari et al., Reference Pullanagari, Yule, Tuohy, Hedley, Dynes and King2012).
The productive characteristics of pastures can be estimated through both direct and indirect methods. Direct methods, based on the cutting and weighting forage samples, are more laborious and challenging to apply over large areas (Sanderson et al., Reference Sanderson, Rotz, Fultz and Rayburn2001). Moreover, such methods can be prone to errors, attributable to the high spatial variability of forage mass. The sources of these errors include differences in grazing behavior, forage management, random distribution of animal excreta, and other factors related to soil and topography (Ogura and Hirata, Reference Ogura and Hirata2001; Pullanagari et al., Reference Pullanagari, Yule, Tuohy, Hedley, Dynes and King2012).
The use of nondestructive methods to estimate forage mass can be more profitable for integrating the spatial variability of this variable, thereby facilitating more accurate estimates and better decision-making on pasture management (Serrano et al., Reference Serrano, Shahidian and Marques Da Silva2016a; Pezzopane et al., Reference Pezzopane, Bernardi, Bosi, Crippa, Santos and Nardachione2019). The use of indirect methods enables the rapid and relatively inexpensive acquisition of adequate information for large areas, indicating forage yield variability and thus improving farmers’ decision-making (Serrano et al., Reference Serrano, Peça, Marques Da Silva and Shahidian2011; Wachendorf et al., Reference Wachendorf, Fricke and Astor2017).
In recent decades, several indirect methods have been developed for estimating forage mass, among which are the measurement of pasture height using a ruler (Gardner, Reference Gardner1986) or rising plates (Scrivner et al., Reference Scrivner, Center and Jones1986; Laca et al., Reference Laca, Demment, Winckel and Kie1989; Mannetje, Reference Mannetje2000), whereas other more complex approaches include the use of electronic meters (e.g., capacitance probes) (Serrano et al., Reference Serrano, Peça, Marques Da Silva and Shahidian2011; Trotter et al., Reference Trotter, Schneider, Lamb, Edwards and McPhee2012; Serrano et al., Reference Serrano, Shahidian and Marques Da Silva2016b).
A further indirect method that can be used to estimate forage mass is based on measurements of the interaction between solar radiation and the pasture canopy. This method uses radiometric sensors that measure visible radiation and near-infrared radiation reflected by the pasture canopy, which facilitates the estimation of forage mass over large areas through remote sensing, using reflectance data obtained by satellites (Hanna et al., Reference Hanna, Steyn-Ross and Steyn-Ross1999; Schellberg et al., Reference Schellberg, Hill, Gerhards, Rothmund and Braun2008; Yang et al., Reference Yang, Fang, Pan and Ji2009; Edirisinghe et al., Reference Edirisinghe, Hill, Donald and Hyder2011) or by proximal sensors (Albayrak, Reference Albayrak2008; Pullanagari et al., Reference Pullanagari, Yule, Tuohy, Hedley, Dynes and King2012; Serrano et al., Reference Serrano, Shahidian and Marques Da Silva2016a, Reference Serrano, Shahidian, Silva, Sales-Baptista, Oliveira, Castro, Pereira, Abreu, Machado and Carvalho2018; Pezzopane et al., Reference Pezzopane, Bernardi, Bosi, Crippa, Santos and Nardachione2019). Using this approach, data on vegetation indices can be derived from the relationships between canopy reflectance at different wavelengths.
One of the most classic vegetation indices is the normalized difference vegetation index (NDVI), which considers the reflectance of plant tissues in the red (670 nm) and near-infrared (760 nm) ranges of the electromagnetic spectrum (Rouse et al., Reference Rouse, Haas, Schell, Deering and Harlan1974). This index can be used to estimate agricultural production (Teal et al., Reference Teal, Tubana, Girma, Freeman, Arnall, Walsh and Raun2006; Amaral et al., Reference Amaral, Molin, Portz, Finazzi and Cortinove2015), leaf area index (LAI) (Cao et al., Reference Cao, Miaoa, Wanga, Huanga, Chenga, Khoslaa and Jiangaa2013), water deficit, nutrient content, and chlorophyll in plants (Gitelson et al., Reference Gitelson, Viña, Ciganda, Rundquist and Arkebauer2005; Zhao et al., Reference Zhao, Starks, Brown, Phillips and Coleman2007), as well as pasture degradation (Moleele et al., Reference Moleele, Ringrose, Arnberg, Lunden and Vanderpost2001). A limitation of the NDVI, however, is that the response curve often saturates at high chlorophyll and nitrogen (N) contents (Munoz-Huerta et al., Reference Muñoz-Huerta, Guevara-Gonzalez, Contreras-Medina, Torres-Pacheco, Prado-Olivarez and Ocampo-Velazquez2013).
Other indices were developed based on reflectance at the same wavelengths of NDVI, such as the Simple Ratio Index (SRI) (Tucker, Reference Tucker1979) and Modified Simple Ratio (MSR) (Chen, Reference Chen1996) or using the reflectance at 720 nm, such as the Normalized difference red edge (NDRE) (Rodriguez et al., Reference Rodriguez, Fitzgerald, Belford and Christensen2006) and Chlorophyll index (ChL) (Gitelson et al., Reference Gitelson, Viña, Ciganda, Rundquist and Arkebauer2005) indices. These indices are being used for several agricultural crops, such as wheat, rice, and coffee (Ciganda et al., Reference Ciganda, Gitelson and Schepers2009; Baker and Saweyer, Reference Baker and Saweyer2010; Prasad et al., Reference Prasad, Carver, Stone, Babar, Raun and Klatt2007; Martins et al., Reference Martins, Pinto, Queiroz, Valente and Rosas2020), but they have not been related to productive characteristics of tropical forages yet.
Laboratory methods for determining forage nutritive value tend to be expensive and time consuming, and results are generally available only after animals have consumed the forage. Furthermore, forage sampling is performed only at a few farm sites or paddocks because of the high cost of these analyses, which limits the monitoring of variability within the pastures (Zhao et al., Reference Zhao, Starks, Brown, Phillips and Coleman2007; Pullanagari et al., Reference Pullanagari, Yule, Tuohy, Hedley, Dynes and King2012). As alternatives, however, indices related to canopy reflectance are viable options for estimating parameters of nutritive value, such as crude protein content and digestibility (Starks et al., Reference Starks, Coleman and Phillips2004; Starks et al., Reference Starks, Zhao, Phillips and Coleman2006; Zhao et al., Reference Zhao, Starks, Brown, Phillips and Coleman2007; Ferner et al., Reference Ferner, Linstadter, Sudekum and Schmidtlein2015).
Indirect methods for estimating forage mass and nutritive value must be calibrated, as the relationships between these variables and those used to estimate them are dependent on plant species, phenological stage, live:dead mass ratio, and pasture management (Hirata, Reference Hirata2000; Serrano et al., Reference Serrano, Peça, Marques Da Silva and Shahidian2011; Serrano et al., Reference Serrano, Shahidian, Silva, Sales-Baptista, Oliveira, Castro, Pereira, Abreu, Machado and Carvalho2018; Pezzopane et al., Reference Pezzopane, Bernardi, Bosi, Crippa, Santos and Nardachione2019). However, for tropical pastures of Urochloa brizantha ‘Marandu’, there are not calibrations of this type.
For the purposes of the present study, we hypothesized that it would be possible to determine relationships between canopy reflectance and pasture productive and nutritive value parameters under different management conditions. To verify this hypothesis, we sought to estimate the forage mass, crude protein content, and N in live forage of Marandu palisadegrass cultivated under different management conditions, based on vegetation indices calculated from canopy reflectance data obtained using a proximal canopy reflectance sensor.
Material and Methods
Experiment description
The forage productivity data used in this study were obtained from an experiment with Urochloa (syn. Brachiaria) brizantha (Hochst ex A. Rich.) Stapf ‘Marandu’, conducted in São Carlos, SP, Brazil (21°57′42″S, 47°50′28″W, 860 m a.s.l.). The climate of this location is classified as Cwa (Köppen), with two well-defined seasons: one between April and September that is cool and dry with an average temperature of 19.9°C and 250 mm rainfall and the other between October and March that is warm and wet with an average temperature of 23.0°C and 1100 mm rainfall. The soil of the experimental area is classified as a Dystrophic Red-Yellow Latossol (Calderano Filho et al., Reference Calderano Filho, Santos, Fonseca, Santos, Primavesi and Primavesi1998).
The pasture was sown in January 2015, after soil tillage and fertility correction, and was uniformly cut at 15 cm above the soil surface in April 2015. Four treatments were used for the experiment: no irrigation or N fertilization (Rainfed_0N), irrigation but no N fertilization (Irrigated_0N), fertilization with 50 kg N ha−1 per cycle but no irrigation (Rainfed_50N), and fertilization with 50 kg N ha−1 per cycle and irrigation (Irrigated_50N). The treatments were assessed in plots measuring 5 × 3 m, with three replicates per treatment, in a split-plot experimental design with the irrigation treatments as the main plots and the nitrogen treatments comprising the subplots.
From July 2015 to November 2016, forage at a height greater than 15 cm was assessed in 14 growth cycles, varying in length from 28 days in the wet season to 42 days in the dry season. At the end of each cycle, the forage was uniformly cut to a stubble height of 15 cm above the soil surface. After each cut, plots allocated to the Rainfed_50N and Irrigated_50N treatments were fertilized with 50 kg N ha−1 (ammonium sulfate), totaling 550 kg N ha−1 year−1. Stubble forage below 15 cm was assessed in four growth cycles (September 2015, November 2015, February 2016, and June 2016).
The chemical characteristics of the soil before starting the experiment and in its last cycle are presented in Table 1.
Table 1. Soil chemical characteristics (0–20 cm) of the experimental area, before starting the experiment (June 2015) and in its last cycle (October 2016), in São Carlos, Brazil
* Soil sampling in the whole area, before the start of the experiment and in its last cycle. O.M. = organic matter; CEC = cation exchange capacity; S.B. = sum of bases; V = base saturation.
Meteorological variables were monitored using a weather station installed near the experiment site, under standard conditions. The data obtained were used to calculate the reference evapotranspiration (ET0) using the Penman–Monteith method (Allen et al., Reference Allen, Pereira, Raes and Smith1998) and the water balance (Thorthwaite and Mather, Reference Thornthwaite and Mather1955). In the irrigated plots, sprinkler irrigation was applied when the readily available water, considered 20 mm, had been consumed, according to the water balance. Irrigation amounts were calculated to increase the soil water content to the level of field capacity.
Plant sampling
Forage above the stubble height (15 cm) was assessed at the end of each growth cycle by cutting the forage within a rectangular frame (0.5 × 1.0 m). The forage samples thus obtained were weighed, and a subsample was taken to determine dry matter content and for morphological separation (leaf, stem, and dead material). The leaf fraction was used to determine leaf area using an LI-3100C leaf area meter (Li-Cor, Lincoln, NE, USA). LAI was calculated using the data obtained for leaf area and leaf mass and the sampling area (0.5 m2). The dry matter content of each morphological fraction was calculated using their masses before and after drying in an air-forced dry oven at 60°C until a constant weight was reached. Using these data and those of the total fresh mass collected in 0.5 m2, we estimated forage mass, expressed in terms of kg ha−1.
Leaf and stem fractions were milled separately to determine crude protein content (CP). The samples were analyzed based on the Fourier Transform–near infrared (FT–NIR technique using an NIRFlex N-500 spectrometer (Büchi, Flawil, Switzerland) equipped with a polarization interferometer. These measurements were performed using a specifically developed calibration model (R2 = 0.944) calibrated for Urochloa species and cultivars. CP was assessed only in those cycles when both forages above and below the stubble height were assessed (September 2015, November 2015, February 2016, and June 2016). Forage N concentration was determined by dividing the CP value by 6.25 (Ball et al., Reference Ball, Collins, Lacefield, Martin, Mertens, Olson, Putnam, Undersander and Wolf2001), and N in live forage was estimated by multiplying forage live mass by its N concentration.
Canopy reflectance measurement
Measurements of pasture canopy reflectance were performed using an ACS-430 Crop Circle canopy sensor (Holland Scientific, Lincoln, NE, USA), which is an active multichannel proximal sensor that measures reflected radiation in the red (670 nm), red edge (724 nm), and near-infrared (760 nm) fractions of the radiation spectrum (Holland Scientific, Reference Holland2013). The reflectance was measured at the beginning (immediately after the pasture was uniformly cut at the stubble height) and at the end (when the pasture was assessed) of each growth cycle. Additionally, weekly measurements were performed during the growth cycle. Reflectance measurements were performed at 0.7 m above the pasture canopy, with 150 readings per plot and the operator walking at a constant speed.
Equation development
The relationships between the vegetation indices and forage variables were determined using the data collected in all the plots, but only during the cycles in which the stubble mass was also quantified (September 2015, November 2015, February 2016, and June 2016). This data set provided representation of a wide range of canopy characteristics related to management and seasonal differences. The pasture variables used to assess these relationships were total forage mass, leaf + stem mass, leaf mass, LAI, CP, and N in live forage.
Five vegetation indices, called Normalized difference vegetation index (NDVI) (Rouse et al.,Reference Rouse, Haas, Schell, Deering and Harlan1974), Normalized difference red edge (NDVI_RE) (Rodriguez et al., Reference Rodriguez, Fitzgerald, Belford and Christensen2006), Simple ratio index (SRI) (Tucker,Reference Tucker1979), Chlorophyll index (ChL) (Gitelson et al.,Reference Gitelson, Viña, Ciganda, Rundquist and Arkebauer2005), Modified simple ratio (MSR) (Chen, Reference Chen1996), were selected and calculated using the following equations:
$${\rm NDVI} = {{NI{R_{760}} - Re{d_{670}}} \over {NI{R_{760}}{\rm{\;}} + {\rm{\;}}Re{d_{670}}}}$$
$$NDV{I_{RE}} = {{NI{R_{760}} - R{E_{720}}} \over {NI{R_{760}}{\rm{\;}} + {\rm{\;}}R{E_{720}}}}$$
$$SRI = {{NIR\_760} \over {Red\_670}}$$
$$ChL = {{NI{R_{760}}} \over {{\rm{\;}}R{E_{720}}}} - 1$$
$$MSR = {{\left( {NI{R_{760}} - Re{d_{670}}} \right){\rm{\;}} - {\rm{\;}}1} \over {{{\left( {NI{R_{760}}{\rm{\;}} + {\rm{\;}}Re{d_{670}}} \right)}^{0,5}}{\rm{\;}} + {\rm{\;}}1}}$$
The average of 150 readings from each plot and for each selected vegetation index was calculated. The relationships between data of pasture canopy reflectance (vegetation indices from equations 1 to 5) and pasture variables (total forage mass, leaf + stem mass, leaf mass, LAI, CP, and N in live forage) were used to develop exponential or linear regression models. Analyses were performed using Microsoft Excel. The determination coefficient (R2) values obtained for these regressions were used to select the best index for each pasture variable. The standardized residuals method was used to eliminate outliers, and values lower than −2 and greater than +2 were removed.
Equation assessment
Having selected the best vegetation index for each pasture variable, we assessed the utility of the equations generated for total forage mass, leaf and stem mass, leaf mass, and LAI. For this purpose, we used the forage accumulation data obtained from 10 growth cycles. These observed data were compared with the estimated forage accumulation, which was obtained by subtracting the estimated mass (or value for LAI) at the beginning of the growth cycle (stubble) from the estimated mass (or value for LAI) at the end of that cycle (precut). The equations developed for CP and N in live forage were not assessed because these variables were not measured during the cycles used for equation assessment.
The performance of equations was statistically evaluated using the following indices and errors.
-
a) Linear regression between the observed (O) and estimated (E) values of each variable and the respective R2 values
-
b) The Willmott (Reference Willmott1981) agreement index (d), which quantifies a model’s accuracy:
$$d = 1 - {{\mathop \sum \nolimits_{i = 1}^n {{\left( {{O_i} - {E_i}} \right)}^2}} \over {\mathop \sum \nolimits_{i = 1}^n {{\left( {\left| {{E_i} - \overline O} \right| + \left| {{O_i} - \overline O} \right|} \right)}^2}}}$$
-
c) Root mean square error (RMSE):
$$RMSE = \sqrt {\left[ {\left( {{1 \over n}} \right)\mathop \sum \limits_{1 = i}^n {{\left( {{O_i} - {E_i}} \right)}^2}} \right]} $$
-
d) Relative error:
$${\rm{Relative\;error\;}}\left( {\rm{\% }} \right) = 100{\rm{\;}} \times {\rm{\;}}{{\left( {{1 \over n}} \right)\mathop \sum \nolimits_{i = 1}^n \left| {{E_i} - {O_i}} \right|} \over {\overline O}}$$
The best equations for estimating total forage mass and N in live forage were used to estimate, respectively, the time series of total forage mass and N extraction by pasture during the growth cycles. The N extracted by the pasture was calculated by subtracting the N in live forage at the beginning of a cycle from that at the end of each treatment.
Results
Equation development
A descriptive analysis of the pasture variables is presented in Table 2. Total forage mass, considering all the treatments and all the growth cycles, ranged from 2,201.7 to 11,058.3 kg ha−1, with an average value of 6,186.4 kg ha−1 and coefficient of variation of 31.1%. Leaf + stem mass, leaf mass, and LAI also showed high variability, with a higher coefficient of variation than total forage mass (Table 2). CP content varied from 6.3% to 17.2%, with a coefficient of variation of 22.0%, and N in live forage also showed high variability.
Table 2. Descriptive statistics for total forage mass, leaf + stem mass, leaf mass, and leaf area index of Urochloa brizantha ‘Marandu’ observed under the different management systems used in the study
* n = number of samples; Min = minimum value; Max = maximum value; Average = average value; SD = standard deviation; CV = coefficient of variation.
The performance of the vegetation indices in estimating the pasture variables is presented in Table 3. For all pasture variables, the linear equations showed R2 values higher than those of the exponential equations. For total forage mass, leaf + stem mass, leaf mass, LAI, and N in live forage, the SRI index showed higher R2 values, with the best performance being that for leaf + stem mass (R2 = 0.794) and the worst for LAI (R2 = 0.696). For CP, the ChL index achieved the highest R2 value (0.314).
Table 3. Coefficients of determination (R2) of the linear and exponential functions for the relationships between vegetation indexes obtained from crop circle ACS-430 and total forage mass, leaf + stem mass, leaf mass, leaf area index, crude protein content, and N in live forage of Urochloa brizantha ‘Marandu’ under different management systems
* NDVI = normalized difference vegetation index; NDVI_RE = normalized difference red edge; SRI = simple ratio index; ChL = chlorophyll index and MSR = modified simple ratio.
The linear regressions that showed the best R2 values for the relationships between vegetation indices and pasture variables are presented in Figure 1. The high variability promoted by the four selected management methods, the seasonality of pasture production, and the forage assessments at the beginning and end of the growth cycles contributed to producing a wide range between minimum and maximum values for all the evaluated pasture variables. The equation that used ChL for estimating CP (Figure 1E) showed a tendency to reach saturation of this index for CP values higher than 14%. For the other equations, which used SRI for estimating pasture variables, no tendency of saturation was observed.
Figure 1. Relationships between the simple ratio index (SRI) and total forage mass (A), leaf + stem mass (B), leaf mass (C), leaf area index (D), and N in live biomass (F), and between the chlorophyll index (ChL) and crude protein content (E) of Urochloa brizantha ‘Marandu’ under different management systems.
Equation assessment
During the equation assessment phase, wherein the pasture productive variables (total forage mass, leaf + stem mass, leaf mass, and LAI) were estimated using SRI, the equations showed R2 values of between 0.69 (leaf mass) and 0.74 (LAI) and relative errors ranging from 19% to 21% (Figure 2). Good equation performance can be seen in Figure 2, in which the relationships between the observed and estimated data are shown. In no case did we detect tendencies of over or underestimation for any of the variables or management conditions evaluated.
Figure 2. Relationship between observed values of forage mass (A), leaf+stem mass (B), leaf mass (C) and leaf area index (D) of Urochloa brizantha ‘Marandu’ and those estimated with the simple ratio index (SRI) under different management systems. Each point is an average of three replications. Vertical bars represent the standard error.
Estimated N extraction by the pasture
Figure 3 presents data for the water balance (Figure 3A), the time series for estimates of total forage mass (Figure 3B), and the estimated N extraction by the pasture (Figure 3C). The experimental period was characterized by climatic conditions similar to those expected for the region. Between cycles 3 and 7, water availability was sufficient to meet pasture demands. In contrast, a period of substantial water deficit occurred between cycles 11 and 13.
Figure 3. Climatological water balance (A), and total forage mass (B), and N extraction by pasture (C), estimated using the simple ratio index (SRI), for Urochloa brizantha ‘Marandu’ under different management systems, from August 2015 to November 2016.
The climatic conditions influenced the total forage mass and N extraction by pasture for different forage management practices. During the period between cycles 3 and 7, no significant differences were observed between different plot managements. Irrigation and N fertilization effects were only perceptible subsequent to cycle 9. During cycles 12 and 13, which were characterized by the highest water deficit, almost no growth was observed in the plots under the Rainfed_0N treatment. Throughout the experimental period, the estimated stubble forage mass was approximately 4,000 kg ha−1, and the estimated forage accumulation ranged from 4,000 to 5,000 kg ha−1, particularly with respect to treatments incorporating fertilization, between cycles 4 and 6 (wet and warm season).
Estimated N extraction by pasture varied in a manner similar to that of forage accumulation, with the highest values being observed in the cycles coinciding with spring and summer, reaching values close to 80 kg N ha−1. The highest differences in N extraction among treatments were observed between cycles 9 and 13 (dry period).
Discussion
The vegetation indices used in this study provided satisfactory simulations of pasture productive and nutritive value variables of Marandu palisadegrass grown under different management conditions. The potential utility of vegetation indices with respect to estimating pasture biomass has been established in numerous previous studies (Pullanagari et al., Reference Pullanagari, Yule, Tuohy, Hedley, Dynes and King2012; Serrano et al., Reference Serrano, Shahidian and Marques Da Silva2016a, Reference Serrano, Shahidian, Silva, Sales-Baptista, Oliveira, Castro, Pereira, Abreu, Machado and Carvalho2018) under several climatic conditions and using different forage species. For intensive pasture production systems, estimating forage yield and nutritive value is essential for ensuring production success.
The NDVI index has frequently been used to estimate pasture yield under temperate climatic conditions and in studies on pasture degradation (Paruelo et al., Reference Paruelo, Epstein, Lauenroth and Burke1997; Rodriguez et al., Reference Rodriguez, Diaz-Ambrona and Alfonso2014; Li and Guo, Reference Li and Guo2010). In the present study, we found that the SRI index (Table 3) showed better performance than NDVI, which could be attributable to the occurrence of high biomass values (in excess of 8000 kg ha−1) in our assessments (Table 2 and Figure 1). Under these conditions, NDVI frequently shows saturation (Tucker, Reference Tucker1979; Hanna et al., Reference Hanna, Steyn-Ross and Steyn-Ross1999; Zhao et al., Reference Zhao, Starks, Brown, Phillips and Coleman2007; Muñoz-Huerta et al., Reference Muñoz-Huerta, Guevara-Gonzalez, Contreras-Medina, Torres-Pacheco, Prado-Olivarez and Ocampo-Velazquez2013), which does not occur when using SRI. The NDVI saturation occurs due to the greater absorption (less reflectance) in the red band (670 nm) with the increase in biomass. For SRI, this does not happen because there is greater reflectance in the 760 nm band with the increase in biomass (equations 1 to 5).
For a tropical climate, Pezzopane et al. (Reference Pezzopane, Bernardi, Bosi, Crippa, Santos and Nardachione2019) also obtained high correlations between vegetation indices and the productive parameters of Piatã palisadegrass (Urochloa brizantha ‘Piatã’). In this study, the authors assessed growth characteristics in full sun and shaded pastures, in the former of which, SRI was found to be the index showing the best relationships with pasture production, whereas for the shaded pasture, the best results were obtained using the MSR index. These findings are consistent with those reported by other authors, who have observed that the relationships between vegetation indices and productive characteristics vary according to pasture conditions and management, among other factors (Hirata, Reference Hirata2000; Serrano et al., Reference Serrano, Peça, Marques Da Silva and Shahidian2011, Reference Serrano, Shahidian, Silva, Sales-Baptista, Oliveira, Castro, Pereira, Abreu, Machado and Carvalho2018).
During the equation assessment phase (Figure 2), we observed no apparent tendency of over- or underestimation for the high or low values of the pasture variables. The RMSE value obtained for total forage mass in the present study (726.02 kg ha−1) (Figure 2A) was lower than that obtained by Serrano et al. (Reference Serrano, Peça, Marques Da Silva and Shahidian2011) and Serrano et al. (Reference Serrano, Shahidian and Marques Da Silva2016b), who used the capacitance method and vegetation indices to estimate forage mass in the Mediterranean region of Portugal. The relative errors of our estimates (Figure 2) were lower than those obtained by Pezzopane et al. (Reference Pezzopane, Bernardi, Bosi, Crippa, Santos and Nardachione2019) for a different cultivar of U. brizantha, which could be explained by the fact these authors conducted evaluations in a grazed pasture, and it is conceivable that the high variability in pasture conditions impaired equation validation.
Among the variables assessed in the present study, the best estimates were those obtained for leaf mass, leaf + stem mass, and N in live forage. Despite the high R2 values obtained for the total forage mass estimate using SRI (0.727), the results for this variable were inferior to those of the other productive variables, owing to the presence of dead material in the forage. Dead forage material contributes to reducing the efficiency of vegetation indices with respect to estimating forage mass, as has been observed by other authors for both temperate and tropical pastures (Todd et al., Reference Todd, Hoffer and Michunas1998; Flynn et al., Reference Flynn, Dougherty and Wendroth2008; Pezzopane et al., Reference Pezzopane, Bernardi, Bosi, Crippa, Santos and Nardachione2019). Similarly, the presence of dead material or conducting assessments during dry periods also contribute to lower estimates of forage mass when using alternative methods, such as the use of capacitance probes (Serrano et al., 2016, Reference Serrano, Shahidian, Silva, Sales-Baptista, Oliveira, Castro, Pereira, Abreu, Machado and Carvalho2018).
The adequate utility of vegetation indices for estimating nutritive value parameters of grasses was confirmed by Zhao et al. (Reference Zhao, Starks, Brown, Phillips and Coleman2007) and Starks et al. (Reference Starks, Zhao, Phillips and Coleman2006). These authors, who examined the hyper-spectral responses of the forage canopy and their relationships with nutritive value parameters, found that the most detectable bands were in the red and infrared ranges, which are the same ranges as those measured by the sensor used in the present study. The simple relationship between near-infrared and red is also an index that has been found to best explain the nutritive value of the leguminous forage sainfoin (Onobrychis viciifolia Scop.) (Albayrak, Reference Albayrak2008).
By obtaining estimates of total forage mass under different management conditions (irrigation and N fertilization), it is possible to identify the productive responses of tropical grass to different environmental conditions (Figure 3B). In the present study, during the spring–summer period, particularly in the first experimental year (from cycles 2 to 7), characterized by high temperatures and water availability, we found that values obtained for the estimated total forage mass were similar among the different management treatments. A similar pattern has also been observed by other authors (Pezzopane et al., Reference Pezzopane, Santos, Evangelista, Bosi, Cavalcante, Bettiol, Gomide and Pellegrino2016; Bosi et al., Reference Bosi, Sentelhas, Huth, Pezzopane, Andreucci and Santos2020) in studies that evaluated the seasonality of tropical grass growth. During the cycles assessed in the present study, we also found that values obtained for N extraction by pasture were comparable among treatments, probably due to the high residual soil fertility at the beginning of the experiment. In this regard, soil analysis performed prior to commencing the experiment revealed organic matter levels of 45 g dm−3 in the 0–20-cm layer (Table 1). Subsequent to the second experimental year, however, we found that differences among the treatments with respect to forage accumulation and N extraction by pasture were more perceptible. Soil analysis conducted during this period indicated that organic matter content had declined to 35 and 40 g dm−3 for the treatments without and with N fertilization, respectively.
The high variability of the conditions under which Marandu palisadegrass was grown in the present study indicate that the equations we developed (Figure 1) can be used under a wide range of management situations. When used to calculate vegetation indices, proximal canopy reflectance sensors can provide reasonable estimates of the nutritive value of pastures in real time, thereby enhancing the efficiency of strategies employed for animal feeding, as reported by Pullanagari et al. (Reference Pullanagari, Yule, Tuohy, Hedley, Dynes and King2012). Although in the present study, we developed equations using data of pastures in experimental plots, if sensors are linked to a global positioning system (GPS), this technology can also be used to generate maps of productive characteristics and nutritive value of Marandu palisadegrass. In this case, according to Serrano et al. (Reference Serrano, Peça, Marques Da Silva and Shahidian2011), the generated maps can assist farmers in making decisions relating to their production systems.
Despite these potential applications, however, further studies should be conducted to enhance the robustness of equation calibration for grazed pastures, in which the pasture canopy structure would differ from that observed in the present study, mainly due to a high variability in pasture cover and an increase in the proportion of dead material.
Conclusion
In this study, we demonstrated that vegetation indices, calculated from reflectance values measured at 670, 720, and 760 nm using a proximal canopy reflectance sensor, have considerable potential for estimating productive and nutritive value variables of Marandu palisadegrass. Among the different vegetation indices evaluated, we found that the simple ratio index produced the best results in estimating these variables. Our findings highlight the considerable potential of the ACS-430 sensor of the Crop Circle equipment in monitoring temporal variations in the forage mass and nutritive value of the Marandu cultivar. By linking this sensor to a global positioning system, it could also be used to estimate the spatial variability of productive and nutritive value variables, and thereby represent a useful tool that would assist farmers in optimizing pasture management.
Acknowledgements
To National Council for Scientific and Technological Development (CNPq) for the fellowship granted to J.R.M. Pezzopane, AC.C. Bernardi and P.M. Santos.
Competing Interests
The author(s) declare none.
