Pasture ecosystems at low latitudes (<10°) show little variation in photoperiod and temperature but exhibit important differences in other climatic factors, such as rainfall. In these areas, depending on the climate in which they are located, tropical forages have limited growth from a climatic point of view, usually due to water availability (Santos et al., 2013). Therefore, close to the equator, the water regime dictates plant growth both in a humid tropical climate and a semi-arid climate. Such weather types are present at lower latitudes on four of the seven continents. Humid tropical climates are observed in South American countries, such as those of the Amazon region, as well as in the central region of the Congo in Africa, and the Indonesian islands of Asia, all of which have humid tropical forests constituting the main biome. A semi-arid climate can be observed in northeastern Brazil, some African countries such as Ethiopia, Kenya, Somalia, and Tanzânia, and in northern Australia (Peel et al., 2007).

Forage plant growth and consequently, canopy productivity are the result of the genotype and its related environment (Durand et al., 1991; Simeão et al., 2021). Environmental factors refer to the edaphoclimatic conditions of plant cultivation, including aspects related to soil (texture, density, and fertility) and climate (temperature, humidity, and photoperiod). The water demand of plants depends mainly on their metabolic requirements, which are linked to characteristics such as stomatal conductance, transpiration rate, and leaf area. These characteristics vary according to the stage of development (Tardieu, 2013). Water demand is also determined by factors such as leaf surface evapotranspiration, which is dependent on radiation, temperature, air humidity, wind speed, and leaf surface properties (Rind et al., 1990). In tropical pastures, evapotranspiration, mean temperature, and solar radiation influence total forage accumulation, leaf accumulation, tiller population density, and nutritional value (Lage Filho et al., 2021; Macedo et al., 2021; Tapia et al., 2021). As for aspects related to the soil, nitrogen (N) is the most important nutrient in tissue flow, and its assimilation may be limited by a water deficit (Onillon et al., 1995). Therefore, when climate conditions are favorable and nitrogen supply is adequate, studies in the literature support that the growth of tropical grasses, especially of the genus Megathyrsus, will be rapid, as there will be an increase in regrowth vigor and a reduction in the interval between grazing (Oliveira et al., 2020).

Cultivated tropical pasture ecosystems represent the main food source for many herds worldwide (Silva et al., 2013). Understanding the relationship between the agronomic characteristics of the grasses that make up such systems, and the agrometeorological conditions of these regions would increase the knowledge of the interactions between these factors in pasture ecosystems. Therefore, multivariate analyses of the factors related to climatic influences exerted on plants and the dynamics of growth and biomass production can provide important information through a systemic view of the ecosystem (Yeater et al., 2014; Araújo Júnior et al., 2021). In addition, the exploration of data using multivariate analysis can contribute to research with direct modeling applied to plant growth (Qiu et al., 2016), although little research has been previously conducted on the prediction of forage accumulation in tropical conditions.

Regions located close to the equator, such as humid and semi-arid tropical regions, may have climatic factors that can distinguish them, such as variables related to humidity and the water regime (Alvares et al., 2013). Production systems in these drier places use technologies such as irrigation (Araújo Júnior et al., 2021), which is often not required in humid tropical regions. Thus, agrometeorological characterization and analysis of agronomic variables of forage plants grown in different regions can provide valuable information on how different growing conditions interact with their environment, and how these conditions can be distinguished. Conducting trials under different climates helps in understanding the climatic influences on the growth and development of forage crops, and contributes to modeling studies involving climate action in tropical forage grasses that are highly responsive to change, such as those of the genus Megathyrsus.

This study aimed to understand the growth dynamics of grasses of the genus Megathyrsus under different growing conditions in humid and semi-arid tropical regions, and to answer the following questions: How are the different growing conditions characterized in relation to the indices that group agronomic and agrometeorological variables? How do agrometeorological variables relate to the agronomic variables measured under these growing conditions?, and How can discriminant functions be described that can distinguish cultivation conditions regarding forage species evaluated in the humid and semi-arid tropical regions of Brazil?

The relationship between agrometeorological and agronomic variables in tropical grasses has been little explored using multivariate approaches. A correct understanding of these interactions will help identify the most efficient use for tropical grass management according to region as will direct-to- research studies on grazing management and tropical grass growth modeling. The importance of this knowledge was confirmed in our study by the formation of indices that summarize the variable information into two groups based on agrometeorological variables, both by principal component and canonical correlation analysis. One group included variables related to energy supply in terms of radiation and temperature, and the other by water and nitrogen availability factors. The Megathyrsus sp. in different climatic conditions was distinguished mainly by the availability of water. This corroborates our hypothesis that even in regions located close to the equator, such as the Amazon and semi-arid regions of Brazil, there are climatic factors capable of discriminating them in terms of productive potential (Silva et al., 2019a).

PC1 (Figure 3) shows that the cultivation conditions with Guinea grass in Igarapé-Açu (GG.IGA.15 and GG.IGA.17-18) had greater data dispersion than the other trials. Climatic variability during the trials may explain the main cause of this difference. The tests in the semi-arid region involved a greater control of climatic conditions through water supply irrigation. In this region, radiation and temperature variability were not as pronounced; water supply through irrigation ensures continuous water availability, and there is no significant variability in actual evapotranspiration (Graham et al., 2016).

Similar to the town of the semi-arid region, Igarapé-Açu is located close to the equator (Figure 1). Consequently, the photoperiod, solar radiation, and temperature data did not vary substantially over the experimental period (Table 1). However, the variability observed in ETa, particularly in the GG.IGA.15 and GG.IGA.17-18 studies (Table 1), can be attributed to the cultivation conditions that included evaluations in both the rainy and the dry periods (Lage Filho et al., 2021; Macedo et al., 2021). In the MG.IGA.15 trial, the evaluations covered only the rainy season and the rainy-dry transition period.

PC1 also considered supply nitrogen, along with the water variables (Figure 3). The SN in Guinea grass trials in Igarapé-Açu showed greater variability due to the absence of nitrogen fertilization during the dry period, which was not required because of the lack of soil moisture (Kunrathm et al., 2018). The contrast observed by PC1 (Figure 3) was related to a greater dispersion of the variables related to water availability and SN than the energy supply and potential water loss variables, mainly in Igarapé-Açu. Based on this, we consider that statistical methods, such as principal component regression, can be potential alternatives for modeling studies with agrometeorological variables and soil properties, as observed in other studies, such as those by Zhou et al. (2021). These authors used principal component regression to relate climatic variables, soil properties, and plant characteristics to the spatial variability of the net exchange of CO₂ between land and atmosphere, using data from croplands, pastures, and forests in different regions. Dispersion of observations in the Amazon region was observed in the second principal component (PC2, Figure 3), in which there was greater participation of the Massai grass trial (MG.IGA.15). In this trial, no evaluations were conducted during the dry season; therefore, the component was generated with most of the variables from only the positive side of the axis (Figure 3). For empirical modeling purposes, the variability in growing conditions of the GG.IGA.15 and GG.IGA.17-18 trials are important for obtaining models that can be generalized in prediction processes (Andrade et al., 2015).

We obtained three main components for the agronomic variables. All experimental tests showed some variability in the first principal component (PC1; Figures 4A,B). This is because the variation in observations is due to the treatments adopted in the trials, whether using different doses of nitrogen fertilization (trials with Massai grass) or different management techniques (trials with Guinea grass). In pasture production systems, it is common to adopt different management techniques based on grazing goals, which will depend on the technological level, production system, and possible edaphoclimatic variation in the region (Silva et al., 2019b,c; Macedo et al., 2021). These factors are indispensable when considering the changeability of response variables that are relevant for monitoring the structure and productivity of pastures.

Not only agrometeorological conditions are important in understanding and predicting productive variables in pastures. Other factors such as management of mechanistic models that estimate biomass production, such as CROPGRO (Bosi et al., 2020b; Brunetti et al., 2021) and APSIM (Bosi et al., 2020a; Gomes et al., 2020), are also relevant.

The second principal component (PC2; Figures 4A,C) mainly showed the larger variability in relation to the GG.IGA.15 trial, because the data from this trial reflected canopies with high and low values of agronomic variables due to use of rest days ranging from 14 to 49 (Macedo et al., 2021). The low participation of LBA in this study is linked to the small variability of this factor in the observed data. It is likely that the accumulation of leaf blades between canopies with different intervals between defoliations did not vary substantially because of TPD and the number of leaves per tiller. Canopies with higher RP had more leaves per tiller and wider and larger leaves, but lower TPD. Canopies with lower RP had fewer leaves per tiller and thinner leaves, but higher TPD. In general, compensation for the number of tillers did not modify the LBA (Sbrissia et al., 2010).

The third principal component (PC3; Figures 4B,C) highlighted the effect of regrowth days. Except for some data for the GG.IGA.17-18 trial, the trials presented a similar distribution of observations for this component. Fixed regrowth periods were not considered, and evaluations occurred during the dry period; therefore, the time for the canopy to reach 95% light interception (experiment goal) was high in some cycles of the GG.IGA.17-18 trial (Lage Filho et al., 2021). This may be related to the increased variability of observations in this trial for this component. Using management goals based on light interception in periods in which there are no favorable conditions for plant growth can be challenging, which has been described in other studies (Carnevalli et al., 2006).

The existing linear relationships between the set of agronomic and agrometeorological variables can be represented by canonical variables, whereby the maximum possible canonical correlation is desired (Araújo Júnior et al., 2021). Therefore, the high canonical correlation (0.84) between the agrometeorological and agronomic variables demonstrates the close interaction between these two groups. This influence is related to the fact that water availability and SN contrast with factors related to potential water loss (ETo, SR, and Tmean). Thus, this directly affects the set of agronomic variables linked to forage canopy growth (FA, LBA, TPD, and CHI), which are important factors in herbivore production systems. These important factors contrast with DMA (Figure 5), which is the result of plant tissue death and is viewed negatively in intensive production systems (Carnevalli et al., 2006).

Multiple regression analysis is a particular case of canonical correlation that can be used to understand which agrometeorological variables are important in the composition of empirical models for estimating productive variables, such as biomass production and tiller density (Qiu et al., 2016; Mathieu and Aires, 2018). In this case, these productive variables would be more related to data on water availability, particularly ETo, WI, and SN. High temperature is an important growth factor for C4 cycle grasses that is related to the increased metabolic activity of plants; however, water stress can be a harmful factor for plant development (Prasad et al., 2011; Mathieu and Aires, 2018). In forage plants, this may be linked to increased plant material death in contrast to biomass accumulation, as observed in the canonical agronomic variable. This was mainly due to the greater variability of the data in the environmental conditions during the conduct of the tests, reflecting the lack of control of the water supply of the crop in the GG.IGA.15 and GG.IGA.17-18 tests (Figure 5). The aspects related to the management of the GG.IGA.15 study, considering regrowth periods that ranged from 14 to 49 days, caused this cultivation condition to present high variability in the canonical agronomic variable which leads to the high variability of forage canopy structural conditions (Macedo et al., 2021).

The variability of observations from these assays, both by principal component and canonical correlation analysis, is related to the response characteristics of grasses to environmental conditions. Unlike many other species, such as those of the genus Urochloa (syn. Brachiaria), grasses of the genus Megathyrsus are more responsive to edaphoclimatic conditions and highly productive when these conditions are favorable, so they are used in intensive production systems under grazing (Pontes et al., 2016). Discriminant function 1, which had the greatest power to separate the studies, discriminated the trials mainly by location (Figure 6). In other words, the difference between the climatic regions (humid tropical and semi-arid) promotes a greater power of distinction than the studied cultivars (Guinea and Massai grass).

It is possible to visualize the difference between experiments GG.PEN.03 and GG.PEN.05–06 (Table 3) despite having the same locations. This was likely due to climatic differences between the times the tests were conducted, mainly with regard to temperature, which was 3.52°C higher in the period from 2005 to 2006 compared to the experimental period in 2003 (Table 1). Regarding the tests in Igarapé-Açu, the temperature difference was only 0.65°C between GG.IGA.15 and GG.IGA.17-18. There was a difference between these two studies regarding DF1, which did not occur between GG.IGA.15 and MG.IGA.15 (Table 3), which that took place in the same year. Thus, the importance of climate-related variables in the composition of DF1 is evident. As both Pentecoste trials were adequately supplied with water through irrigation, the effect of temperature change may have been the main factor in discriminating these trials (Table 2). Temperature influences the acceleration of the physiological processes of the plant, resulting in greater forage accumulation, stem elongation, and dead material (Ivory and Whiteman, 1978). DF1 was more clearly observed by the greater effect of Tmean and SA, since the effect of temperature on stem elongation is well evidenced in grasses (Liu et al., 1998; Pierre et al., 2011; Yang et al., 2014).

The discrimination of the trials regarding the forage cultivar was based on the discriminant function 2. This function accounted for only 19.20% of the variation in the separation power of the studies. It is possible to observe the strong influence of agronomic variables on this discriminant function, mainly regarding the effects of TPD, FA, DMA, and LBA. Massai grass dominated for these variables because it presented higher TPD and biomass production compared to Guinea grass (Veras et al., 2020). In contrast, Guinea grass presented greater values in height increment, as this cultivar is larger than Massai grass. It is possible to see here the negative relationship between canopy height and TPD for tropical grasses observed in recent studies (Macedo et al., 2021; Xiliang et al., 2021), in which canopies with greater height tend to have lower TPD. Unlike DF1, DF2 did not differ in the Guinea grass trials in Igarapé-Açu at different times (Table 3). The highest values for the MG.FOR.09 study showed the strong production capacity of the Massai grass cultivated under conditions of high temperature and radiation in low-latitude regions (Table 1). Associated with this, the high availability of water explained the stronger positive correlation of WI with DF2 than the other variables (Figure 6).

Through a multivariate approach with experimental test data, we show that in production systems based on the cultivation of grasses of the genus Megathyrsus in regions located near the equator, despite the similarities in energy supply in terms of radiation and temperature, water availability, and nutrient supply are the determining factors for biomass accumulation. Therefore, these factors should be prioritized in future studies modeling forage biomass accumulation. However, in tropical pastoral ecosystems, forage plant management also had a determining effect on the accumulation of total biomass and its components, which should be considered in studies of relationships between productive variables and pasture characterization.

The data analyzed in this study is subject to the following licenses/restrictions: Dataset will be used in future modeling research by the team. Requests to access these datasets should be directed to anibal.cr@ufra.edu.br.

VM: conceptualization, methodology, formal analysis, investigation, data curation, and writing—original draft. NL, AC, ML, RS, and JC: investigation, resources, data curation, and writing—review and editing. CF, MC, and AR: conceptualization, resources, writing—review and editing, supervision, and project administration. All authors contributed to the article and approved the submitted version.

This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior—Brazil (CAPES; Finance Code 001).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.