Advances in the classification of Panicum maximum Jacq. forage using machine learning algorithms based on morphogenic and structural characteristicsABSTRACT:
This study aimed to assess the effectiveness of machine learning (ML) models in classifying tropical forages based on their productive, structural, and morphogenic characteristics. Four ML models were utilized for this evaluation: Logistic Regression (REGL), Multilayer Perceptron (MLP), Decision Tree (REPTree), and Random Forest (RF). The findings revealed that the RF model outperformed the others in categorizing the six forages, yielding coefficients of 62.25% and 0.55% for CC and Kappa, respectively. Notably, the RF model demonstrated superior performance for both low and medium-growing forage plants, achieving 68.95% and 0.53% CC and Kappa for the low-growing group. However, its performance for medium-growing forage plants was marginally lower, registering values of 61.37% and 0.42% (CC and Kappa, respectively). In conclusion, the study provided evidence of the efficacy of machine learning techniques, particularly the RF model, in the classification of different-growing Panicum maximum forage plants. However, further research is needed in connection with the results obtained in this research, associating them with technological tools such as drones and advancing the identification of cultivars via remote sensing in the field.
Key words:
computational intelligence; machine learning;
Panicum maximum
; morphogenesis
RESUMO:
Este estudo teve como objetivo avaliar a eficácia de modelos de aprendizado de máquina (AM) na classificação de forrageiras tropicais com base em suas características produtivas, estruturais e morfogênicas. Para essa avaliação, foram utilizados quatro modelos de AM: Regressão Logística (REGL), Perceptron Multicamadas (MLP), Árvore de Decisão (REPTree) e Floresta Aleatória (RF). Os resultados revelaram que o modelo RF superou os demais na categorização das seis forrageiras, apresentando coeficientes de 62,25% e 0,55% para CC e Kappa, respectivamente. Além disso, o modelo RF demonstrou desempenho superior tanto para plantas forrageiras de porte baixo quanto médio, atingindo 68,95% e 0,53% de CC e Kappa para o grupo de porte baixo. No entanto, seu desempenho para as plantas forrageiras de porte médio foi ligeiramente inferior, registrando valores de 61,37% e 0,42% (CC e Kappa, respectivamente). Em conclusão, o estudo forneceu evidências da eficácia das técnicas de aprendizado de máquina, particularmente do modelo RF, na classificação de plantas forrageiras Panicum maximum de diferentes portes. No entanto, são necessárias pesquisas adicionais para aprofundar os resultados obtidos neste trabalho, associando-os a ferramentas tecnológicas, como drones, e avançando na identificação de cultivares por meio de sensoriamento remoto em campo.
Palavras-chave:
aprendizado de máquina; inteligência computacional; morfogênese;
Panicum maximum
INTRODUCTION
In the tropical regions of Brazil, two widely used genera, Brachiaria (Syn. Urochloa) and Panicum (Syn. Megathyrsus), offer a variety of cultivars suitable for different environmental conditions (JANK et al., 2011). These genera account for approximately 160 million hectares of cultivated pastures in the region (LAPIG, 2023).
The genus Panicum, particularly the species Panicum maximum, stands out for its high adaptability to different soil types and climatic conditions, combined with high productivity and good nutritional quality (VERAS et al., 2025), which results in high animal performance indices, both on an individual and per-area basis (EUCLIDES et al., 2017). Conversely, its cultivars are highly demanding, especially with regard to management; and are therefore, more suitable for intensive production systems (GURGEL et al., 2021). In addition, some cultivars, such as Massai, Zuri and Tamani, have gained prominence for use as mulch in integrated production systems (BUBLITZ et al., 2024). Therefore, due to its versatility and high productive potential, the genus Panicum plays a strategic role in modern agricultural production systems.
Distinguishing among these cultivars requires a detailed understanding of plant morphology and structural traits, as these characteristics directly influence forage growth dynamics, grazing efficiency, and animal performance, particularly among forages of the same species that often exhibit similar visual appearance (SIMEÃO et al., 2021). In tropical perennial forages, contrasting differences in structural attributes, such as tiller density, canopy architecture, and leaf-to-stem ratio, determine their response to grazing management and seasonal variations. Thus, evaluating these traits under different management practices throughout the growing seasons is essential, as they progressively shape sward structure and; consequently, guide cultivar selection, forage breeding programs (CASTRO et al., 2020), and the definition of appropriate grazing strategies in animal production systems.
The field of machine learning centers on the development of algorithms for obtaining reliable predictive information (CAMACHO-PEREZ et al., 2025). By extrapolating from observations, machine learning models can yield accurate predictions for a wide range of inputs (DIFANTE et al., 2024). It should be noted that different machine learning models may exhibit variable performance, a crucial consideration for the objectives of our study. While the literature on the use of machine learning for classifying forages remains limited, machine learning models are widely employed in agriculture, offering the potential to address numerous research challenges (PALLATHADKA et al., 2023).
The need exists for sensitive models capable of discerning cultivars from different genera and species within forage plants (BRITZ et al., 2022; OLIVEIRA et al., 2023). Leveraging machine learning techniques can facilitate genetic improvement programs and streamline the selection of cultivars based on morphological, anatomical, and molecular characteristics, thereby alleviating the substantial time and resource demands associated with manual diagnosis.
The majority of published studies on pastures focus on solving prediction problems aimed at estimating forage production (SANTOS et al., 2022; GUEVARA-ESCOBAR et al., 2023). In this context, calibrated models capable of discriminating different forage species or cultivars, combined with technological tools such as remote sensing, have been effectively used as pasture management tools by enabling the spatial and temporal monitoring of biomass availability, canopy structure, and vegetation vigor. These approaches support management decisions related to grazing intensity, pasture renovation, and forage allocation, particularly in large or inaccessible regions where conventional field assessments are limited (ADELABU et al., 2013).
The REGL model estimates the probability that a specific input corresponds to one of the known classes (SHALEV-SHWARTZ & BEN-DAVID, 2014). The MLP represents an artificial neural network model, while the REPTree model constitutes a decision tree algorithm that applies a pruning method to minimize classification errors (NAHAR & ARA, 2018). Finally, RF known for its high accuracy and ability to process large datasets with multiple characteristics, comprises several decision trees and represents a powerful model commonly employed for classification and regression tasks (BREIMAN, 2001; CHEN et al., 2016). Therefore, considering the growing emphasis on the study of ML for the evaluation of different models applied to the classification of forage plants based on morphogenetic (LEMAIRE & CHAPMAN, 1996; GASTAL & LEMAIRE, 2015) and productive characteristics (DIFANTE et al., 2008; DIFANTE et al., 2011) it is essential to evaluate and test this method in the context of forage plants.
The primary objective of our research encompassed the evaluation of machine learning models’ performance in classifying tropical forages, taking into account their productive, structural, and morphogenic characteristics.
MATERIALS AND METHODS
The data utilized in this research stem from an experiment conducted in a greenhouse at Embrapa Gado de Corte, located in Campo Grande, Mato Grosso do Sul, Brazil.
The greenhouse experiment was conducted over a total period of 162 days, from September 2021 to February 2022 (FRONTADO et al., 2025), under natural photoperiod conditions, with temperature and relative humidity monitored but not artificially controlled. Plants were grown individually in plastic pots containing 2.55 dm³ of soil and arranged in a randomized block design. The study evaluated six Panicum maximum (syn. Megathyrsus maximus) forage plants, classified into two growth groups: low-growing (BRS Tamani, PM422, and PM408) and medium-growing (BRS Zuri, PM414, and PM406). The classification into low- and medium-growing groups was based on inherent plant stature and canopy structural traits reported in the literature (VERAS et al., 2025). BRS Tamani, classified as low-growing, is characterized by short stature, high tiller density, smaller leaves, and a high leaf-to-stem ratio, favoring management flexibility. BRS Zuri is a medium-stature cultivar with high forage accumulation, broad leaves, vigorous growth, and good resistance to leaf spot, being well suited to intensive and well-managed production systems.
To ensure plant uniformity, thinning was performed 15 days after emergence, leaving five uniform plants per pot. Each pot consisted of one experimental unit, with three replicates per treatment. Forage management was based on a fixed regrowth interval of 28 days, and plants were harvested accordingly throughout the experimental period. Productive, morphogenetic, and structural variables were evaluated, including leaf appearance rate (LAR), leaf elongation rate (LER), stem elongation rate (SER), final leaf length (FLL), leaf lifespan (LLS), tiller population density (TPD), number of living leaves (NLL), and phyllochron (PHY) (Figure 1), following the methodology described by Lemaire and Chapman (1996). Morphogenetic variables were obtained through periodic measurements during regrowth, while structural variables and forage production were assessed at each harvest by collecting all aboveground biomass, followed by drying and weighing to estimate dry matter yield. The harvests were conducted at 28-day intervals, totaling five harvests during the experimental period. The cutting residue height was set at 15 cm above ground level for short-stature cultivars and 20 cm for medium-stature cultivars.
Schematization of variables selected for classification via machine learning. Testing Machine Learning models.
Four machine learning models were employed for analysis: REGL, MLP, REPTree, and RF. The efficacy of the models was assessed using the metrics of correct classifications (CC) and Kappa Coefficient.
The tested models, listed in table 1, included REGL, MLP, REPTree, and RF. The ML models analyzed the data separately based on three distinct groups. Forage classification was conducted using stratified cross-validation with 10 k-fold and 10 repetitions, resulting in 100 runs for each model. Model parameters were set to the default configuration of the Weka 3.8.6 software, utilizing an AMD Ryzen 5 CPU with 8 GB RAM. The performance of the algorithms in classifying P. maximum foragers was verified using accuracy metrics (Table 2).
The accuracy metrics, including the percentage of CC and Kappa, were employed to assess the effectiveness of each model in categorizing forage based on growing size, both individually and collectively. To compare the performance of the machine learning (ML) models, boxplots containing the assessed configurations and the results of the Scott-Knott test at a 5% significance level for CC and Kappa were generated. The grouping of the means via Scott-Knott was conducted using the SISVAR software, while the boxplots were created using the ggplot2, gridExtra, and corrplot packages within the R software for Windows.
RESULTS
Cultivars classified as medium-growing showed higher values of LER, SER, and FLL compared with low-growing cultivars. In contrast, low-growing cultivars exhibited greater TPD, and, in some cases, longer leaf lifespan LLS. LAR and NLL were similar between the two growth groups, with limited variation among cultivars within each group (Table 3).
The classification of P. maximum forages used two metrics, CC and Kappa, to evaluate the performance of the ML algorithms. Among the six P. maximum forages, the RF algorithm showed the best performance for both CC (62.25%) and Kappa (55%) (Figure 2). The MLP and REPTree algorithms yielded intermediate values for correct classification, with MLP at 54.72% and 46% for CC and Kappa, and REPTree at 51.90% and 42% for CC and Kappa, respectively. The REGL technique produced the lowest values for CC (50.40%) and Kappa (41%).
Boxplots illustrating the CC and Kappa for the classification of six forages using REGL, MLP, REPTree, and RF. The means annotated with the same letters indicate no statistically significant difference from each other based on the Scott-Knott test at a 5% significance level.
In the results for low-growing forages for classifications by the CC and Kappa metrics (Figure 3). The RF technique demonstrated higher values for forage classification with CC at 68.95% and Kappa at 53%. The MLP model provided CC values of 63.66% and Kappa of 45%, which were lower than RF and higher than REPTree. The REPTree performance showed decreased values compared to RF and MLP, with CC at 60.31% and Kappa at 40%. The lowest values for the CC and Kappa metrics were observed in the REGL technique, with 53.37% and 30%, respectively.
Boxplots for CC and Kappa in the classification of low-growing forage plants (BRS Tamani, PM422 and PM408) in each ML model: REGL, MLP, REPTree and RF. Means followed by the same letters do not differ from each other by the Scott-Knott test at 5% probability.
The results of classifying medium-sized forage plants using the CC and Kappa metrics are shown in figure 4. The RF technique performed best with CC (61.37%) and Kappa (42%). The MLP model showed intermediate values of CC (58.71%) and Kappa (38%), which were no different from REGL, which obtained CC (57.53%) and Kappa (36%). The lowest CC (52.62%) and Kappa (28%) values were observed in the REPTree technique.
The boxplots illustrate the CC and Kappa for the classification of medium-growing forage plants (BRS Zuri, PM414, and PM406) in each machine learning (ML) model: Logistic Regression (REGL), Multilayer Perceptron (MLP), REPTree, and Random Forest (RF). Instances where the means share identical letters denote a lack of statistically significant differences, as determined by the Scott-Knott test at a 5% probability level.
The table in figure 5 shows the confusion matrix for classifying different forages. The RF model demonstrated the best ability to distinguish between classes out of all the models considered, achieving an accuracy of 62% with 559 correct instances. Within this model, the genotypes PM422 (111), PM408 (99), PM414 (94), and the cultivar BRS Zuri (100) had the highest number of correct classifications (Figure 5D).
Confusion matrix for correct classifications of six P. maximum foragers testing the Logistic Regression-REGL (a), Multilayer Perceptron-MLP (b), REPTree (c), Random Forest-RF (d) models.
In contrast, the MLP and REPTree models attained intermediate accuracy values of 54% and 53%, respectively. The MLP model recorded the highest correct instances for the cultivars BRS Zuri (103), PM422 (102), and PM414 (93), while the REPTree model achieved the highest correct instances for the cultivars BRS Zuri (95), BRS Tamani (89), and the genotypes PM422 (88) and PM408 (87), totaling 489 and 477 correct classifications for the six forages, respectively (Figure 5B-C). Conversely, the REGL model demonstrated the lowest accuracy of 50% and the lowest number of correct classifications (455), with the highest correct instances for the genotypes PM414 (100), PM408 (81), PM422 (75), and the cultivar BRS Zuri (96) (Figure 5A).
Furthermore, the REGL model exhibited the highest number of incorrect instances, most notably with the genotype PM406 being confused with PM414 and the cultivar BRS Zuri, both occurring 46 times each (Figure 5A). Similar instances of confusion included the cultivar BRS Tamani, which was confused with PM408 in 45 cases (Figure 5A). The MLP model exhibited the greatest confusion with the cultivar BRS Tamani and PM408, occurring 52 times (Figure 5B). The REPTree model’s highest classification confusion was observed between the genotype PM406 and the cultivar BRS Zuri (56 instances) (Figure 5C). Lastly, the RF model displayed the greatest confusion, including instances of BRS Tamani and PM408 (45), PM408 and BRS Tamani (39), and PM406 with the cultivar BRS Zuri (38) (Figure 5D).
The confusion matrix analysis in figure 6 shows that the RF model had the highest capability to distinguish low-growing P. maximum forages, with an accuracy of 68% (307 correct instances). The PM422 and PM408 genotypes had the highest number of correct instances for this model (Figure 6D). MLP and REPTree had intermediate accuracy values of 63% and 61% (287 and 275 correct instances, respectively). The MLP had the highest correct instances in the PM422 genotype (122) and the BRS Tamani cultivar (88), whereas for REPTree, they were in the PM422 genotype (114) and the BRS Tamani cultivar (90) (Figure 6B-C, respectively). Conversely, the REGL model showed the lowest accuracy of 52%, with the highest correct instances found in the PM422 (100) and PM408 (80) genotypes (Figure 6A). The BRS Tamani cultivar had the highest number of incorrect instances or classification confusion with the PM408 genotype, with 47, 45, 34, and 44 incorrect instances across REGL, MLP, REPTree, and RF, respectively. Conversely, the PM408 genotype was confused with the BRS Tamani cultivar in 37, 61, 55, and 32 incorrect instances across REGL, MLP, REPTree, and RF, respectively.
Confusion matrix for classification of low-growing forage plants of P. maximum testing the Logistic Regression-REGL (a), Multilayer Perceptron-MLP (b), REPTree (c), Random Forest-RF (d) models.
Figure 7 shows the confusion matrix for medium-growing P. maximum forage classification. As can be seen, the MLP algorithm exhibited the highest discriminatory performance in identifying forages with an accuracy of 62% (282 correct instances). The algorithm showed the highest accuracy in the BRS Zuri cultivar (103) and the PM414 genotype (90) (Figure 7B). RF and REGL algorithms yielded intermediate accuracy values of 60% and 56% (274 and 255 correct instances, respectively). Among these, the RF algorithm achieved the highest correct instances in the BRS Zuri cultivar (99) and the PM414 genotype (97), while the REGL algorithm showed the highest correct instances in the PM414 genotype (99) and the BRS Zuri cultivar (94) (Figure 7D-A). Notably, the REPTree algorithm demonstrated the lowest accuracy of 49%, with the highest correct instances in the PM414 genotype (87) and the BRS Zuri cultivar (85) (Figure 7C). In terms of incorrect instances or confusion, the PM406 genotype exhibited the highest number of instances with the BRS Zuri cultivar, totaling 49, 33, 56, and 38 for REGL, MLP, REPTree, and RF, respectively. Similarly, the PM414 genotype demonstrated a significant number of incorrect instances with the PM406 genotype, with 35, 43, 39, and 40 for REGL, MLP, REPTree, and RF, respectively.
Confusion matrix for the classification of medium-growing forage plants of P. maximum, testing the Logistic Regression-REGL (a), Multilayer Perceptron-MLP (b), REPTree (c), and Random Forest-RF (d) models.
Regarding the importance of attributes (input variables), the cluster analysis of the six forage plants (Figure 8) showed that TPD, FLL, and NLL were the most important attributes for classification. These same attributes were identified in the same order of importance for the low-stature cultivar group. For the medium-stature cultivars, these attributes also stood out as the most important; however, they appeared in a different order of importance.
DISCUSSION
The results indicated that the differences observed in the morphogenetic and structural variables (LAR, LER, SER, FLL, LLS, TPD, NLL, and PH) are primarily associated with plant growth habit. Medium-growing genotypes generally exhibited higher stem and leaf elongation rates, greater final leaf length, and longer phyllochron, reflecting a growth strategy characterized by greater structural development. In contrast, low-growing materials showed higher rates of tissue renewal, expressed by greater leaf appearance and elongation rates, shorter final leaf length and phyllochron, and higher tiller population density, which are typical of canopies with faster turnover and greater plasticity.
In addition, the variability observed among cultivars can be attributed to the different management strategies adopted throughout the experimental period (FRONTADO et al., 2025), which directly affected leaf emergence, expansion, and senescence dynamics. Management-induced changes alter the balance between tissue growth and renewal, leading to distinct morphogenetic and structural responses even within the same growth group, highlighting the strong interaction between genetic traits and management conditions in shaping canopy morphogenesis (CARVALHO et al., 2026).
Plants of the panicum genus present structural variability promoted by changes in morphogenic characteristics (DIFANTE et al., 2008; DIFANTE et al., 2011). Based on the findings of this study, it is evident that the Random Forest (RF) algorithm exhibits the highest accuracy in classifying cultivars and genotypes of P. maximum forages, irrespective of the growing size. This observation is supported by VILAR et al. (2020), who noted RF as the algorithm with the greatest potential in identifying different soil covers in the context of agroforestry systems.
Similar results showcasing the superior performance of the RF algorithm have been reported in various studies involving different crops and purposes. Examples include discrimination of soybean cultivars (SANTANA et al., 2024a; TEODORO et al., 2024), sorghum (SANTANA et al., 2024b), coffee ALVES et al. (2022) and corn LEE et al. (2021).
Upon evaluating the performance of the algorithms, it is evident that when the growing sizes were assessed together (medium and low), there was a noticeable decline in performance. This decline could be attributed to the substantial variability in the data pertaining to the structural and morphogenetic characteristics between the low and medium-growing groups. For instance, PACIULLO et al. (2016) observed a significant difference of 52 tillers between the Massai cultivar (low-growing) and the Tanzânia cultivar (medium-growing), which likely influenced the performance in this group. The higher performance in the metrics evaluated and accurate instances in the confusion matrix indicate that the RF model is a practical option for classifying low-growing tropical P. maximum forage plants. Additionally, the model has displayed efficacy in plant and disease classification scenarios (RAMOS et al., 2020).
The improvement in the models’ accuracy with independent evaluation of different growing sizes was unexpectedly significant given the decrease in the number of observations (LÓPEZ et al., 2021). However, the analysis of forage plants by separating low and medium growing sizes (Figure 5 and Figure 6) allowed for grouping materials with similar characteristics, leading to reduced variability in input data and improved results. This likely contributed to better performance in classifying low-growing forage plants.
The data in the confusion matrix for RF (Figure 5) showed that the lowest confusion percentage was between the BRS Tamani cultivars with BRS Zuri and the PM406 genotype. This indicated that the model effectively distinguishes between different forages based on their structural and morphogenic characteristics, particularly for low and medium-growing forages. A similar finding was reported in the TPD variable by VERAS et al. (2020), where the difference between the two cultivars was an average of 128 tillers.
In figure 6 (low growing), it is apparent that the PM408 genotype shares similar structural characteristics with the BRS Tamani cultivar. The RF algorithm presented the greatest confusion in the classification of the two forages in all ML techniques evaluated. This is expected given that the two forages have tiller population density and number of live leaves similar to those of the Massai cultivar, and different from medium-growing forages, such as the BRS Zuri cultivar (VERAS et al., 2020).
Figure 7 (medium growing) reveals that the MLP had a higher percentage of confusion in the characteristics between the PM406 genotype with the BRS Zuri cultivar and the PM414 genotype. This indicates that the forages have similarities, and hence the algorithm had greater difficulty in discriminating. The notable performance of MLP only in the medium-growing group can be attributed to the structure of the confusion matrix analysis, which identifies error patterns in the discrimination of forages. However, it’s noteworthy that the Kappa coefficient does not have this feature (SILVA & PEREIRA, 1998) but it considers the observed agreement by the possibility to predict and is reliable when the objective is only to evaluate the agreement. Given the discrepancy in the results of the two analyses and the objective being related to the classification of forages, it can be inferred that the MLP is the model with the best performance to be used with medium-growing forages.
The analysis of the importance of attributes was considered in the model with the highest performance observed. The TPD variable stood out within the input set as the most important attribute in the classification, regardless of whether the sizes were separated. This variable is a structural characteristic of great importance as it impacts the persistence and productivity of the pasture (LARA & PEDREIRA, 2011).
An important aspect of this study is that the evaluated forages received varying doses of phosphorus (P) and limestone (Ca), which resulted in improved responses of the observed variables. The classification performance found in both medium and low-growing forages may be attributed to the increase in soil fertility due to different doses of limestone (HUNT JR et al., 2011). The algorithm used data from the input variables under conditions of low and high P availability, with increasing doses of limestone, which could be relevant for the adoption of classification models (ZHU et al., 2020). The classification of different cultivars on a larger scale, such as in plot experiments and/or under grazing, may likely be associated with soil fertility (ZHANG et al., 2020).
The data used as input variables to test the models yielded intermediate performances, which can be replicated, and other variables could be added with other cultivars such as Andropogon gayanus, Brachiaria brizantha, Brachiaria decumbens, Hymenachne amplexicaulis, Panicum maximum, Pennisetum purpureum, Pennisetum setaceum, to name a few. Additionally, it could assist in the formulation and calibration of more complex models with the addition of other variables, such as vegetation indices, under different experimental conditions, in plot experiments, and under grazing, also being useful for monitoring and/or discriminating pastures in remote areas in the different biomes in Brazil.
CONCLUSION
Upon comprehensive analysis, it has been ascertained that the application of morphogenetic and structural variables for the classification of Panicum maximum through machine learning is indeed viable. The discrimination of cultivars considering the different forage sizes is more accurate. The use of morphogenic and structural variables to classify cultivars using machine learning is feasible. The RF algorithm can be used to discriminate cultivars with greater accuracy in all the data sets evaluated, with slightly greater accuracy for the low-growing cultivars evaluated. The traditional Logistic Regression classification model provided the worst classification results. Therefore, the use of machine learning techniques is promising for differentiating Panicum maximum cultivars using variables that are easy and quick to obtain, such as the morphogenic and structural variables measured here.
ACKNOWLEDGMENTS
The authors thank the Embrapa gado de corte, Universidade Federal de Mato Grosso do Sul, through the Programa de Pós-graduação em Ciencia Animal, the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (Finance code 001) for the Support of the Fundação de Apoio ao Desenvolvimento do Ensino, Ciência e Tecnologia do Estado de Mato Grosso do Sul. (FUNDECT).
REFERENCES
-
ADELABU, S. et al. Exploiting machine learning algorithms for tree species classification in a semiarid woodland using RapidEye image. Journal of Applied Remote Sensing, v.7, n.1, 2013. Available from: <Available from: https://doi.org/10.1117/1.JRS.7.073480 >. Accessed: Apr. 01, 2025. doi: 10.1117/1.JRS.7.073480.
» https://doi.org/10.1117/1.JRS.7.073480.» https://doi.org/10.1117/1.JRS.7.073480 -
ALVES, M. C. et al. The role of machine learning on Arabica coffee crop yield based on remote sensing and mineral nutrition monitoring. Biosystems Engineering, v.221, p.81-104, 2022. Available from: <Available from: https://doi.org/10.1016/j.biosystemseng.2022.06.014 >. Accessed: Apr. 01, 2025. doi: 10.1016/j.biosystemseng.2022.06.014.
» https://doi.org/10.1016/j.biosystemseng.2022.06.014.» https://doi.org/10.1016/j.biosystemseng.2022.06.014 -
BREIMAN, L. Random forest. Machine Learning, v.45, p.5-32, 2001. Available from: <Available from: https://doi.org/10.1023/A:1010933404324 >. Accessed: Apr. 01, 2025. doi: 10.1023/A:1010933404324.
» https://doi.org/10.1023/A:1010933404324.» https://doi.org/10.1023/A:1010933404324 -
BRITZ, R. et al. Spectral-based classification of plant species groups and functional plant parts in managed permanent grassland. Remote Sensing, v.14, n.5, 2022. Available from: <Available from: https://doi.org/10.3390/rs14051154 >. Accessed: Apr. 01, 2025. doi: 10.3390/rs14051154.
» https://doi.org/10.3390/rs14051154.» https://doi.org/10.3390/rs14051154 -
BUBLITZ, L. R. et al. Panicum maximum cultivars for use in integrated agricultural production systems in Cerrado biome soils. Grassland Science (Online), v.70, p.121-129, 2024. Available from: <Available from: http://dx.doi.org/10.1111/grs.12423 >. Accessed: Jan. 27, 2026. doi: 10.1111/grs.12423.
» https://doi.org/10.1111/grs.12423.» http://dx.doi.org/10.1111/grs.12423 -
CAMACHO-PEREZ, E. et al. Predicting dry matter intake in Pelibuey sheep using machine learning methods. Heliyon, v.11, p.e41913, 2025. Available from: <Available from: http://dx.doi.org/10.1016/j.heliyon.2025.e41913 >. Accessed: Apr. 01, 2025. doi: 10.1016/j.heliyon.2025.e41913.
» https://doi.org/10.1016/j.heliyon.2025.e41913.» http://dx.doi.org/10.1016/j.heliyon.2025.e41913 -
CARVALHO, A. A. et al. Morphogenesis, structure, and chemical composition of Paiaguás grass under different nitrogen doses and deferment periods. Plants, v.15, p.341, 2026. Available from: <Available from: https://doi.org/10.3390/plants15030341 >. Accessed: Jan. 27, 2026. doi: 10.3390/plants15030341.
» https://doi.org/10.3390/plants15030341.» https://doi.org/10.3390/plants15030341 -
CASTRO, W. et al. Deep learning applied to phenotyping of biomass in forages with UAV-based RGB imagery. Sensors, v.20, n.17, p.4820, 2020. Available from: Available from: https://doi.org/10.3390/s20174802 Accessed: Apr. 01, 2025. doi: 10.3390/s20174802.
» https://doi.org/10.3390/s20174802.» https://doi.org/10.3390/s20174802 -
CHEN, W. et al. A comparative study of logistic model tree, random forest, and classification and regression tree models for spatial prediction of landslide susceptibility. Catena, v.151, p.147-160, 2016. Available from: <Available from: https://doi.org/10.1016/j.catena.2016.11.032 >. Accessed: Apr. 01, 2025. doi: 10.1016/j.catena.2016.11.032.
» https://doi.org/10.1016/j.catena.2016.11.032.» https://doi.org/10.1016/j.catena.2016.11.032 -
DIFANTE, G. S. et al. Can different cultivars of Panicum maximum be identified using a VIS/NIR sensor and machine learning? Agriengineering, v.6, p.3739-3751, 2024. Available from: <Available from: http://dx.doi.org/10.3390/agriengineering6040213 >. Accessed: Apr. 01, 2025. doi: 10.3390/agriengineering6040213.
» https://doi.org/10.3390/agriengineering6040213.» http://dx.doi.org/10.3390/agriengineering6040213 -
DIFANTE, G. S. et al. Morphogenetic and structural characteristics of marandu palisadegrass subjected to combinations of cutting heights and cutting intervals. Revista Brasileira de Zootecnia, v.40, p.955-963, 2011. Available from: <Available from: https://doi.org/10.1590/S1516-35982011000500003 >. Accessed: Apr. 01, 2025. doi: 10.1590/S1516-35982011000500003.
» https://doi.org/10.1590/S1516-35982011000500003.» https://doi.org/10.1590/S1516-35982011000500003 -
DIFANTE, G. S. et al. Dinâmica do perfilhamento do capim-marandu cultivado em duas alturas e três intervalos de corte. Revista Brasileira de Zootecnia, v.37, p.189-196, 2008. Available from: <Available from: https://doi.org/10.1590/S1516-35982008000200003 >. Accessed: Apr. 01, 2025. doi: 10.1590/S1516-35982008000200003.
» https://doi.org/Dinâmica do perfilhamento do capim-marandu cultivado em duas alturas e três intervalos de corte» https://doi.org/10.1590/S1516-35982008000200003.» https://doi.org/10.1590/S1516-35982008000200003 -
EUCLIDES, V. P. B. et al. Maintaining post-grazing sward height of Panicum maximum (cv. Mombaça) at 50 cm led to higher animal performance compared with post-grazing height of 30 cm. Grass and Forage Science, v.73, p.174-182, 2017. Available from: <Available from: https://doi.org/10.1111/gfs.12292 >. Accessed: Jan. 27, 2026. doi: 10.1111/gfs.12292.
» https://doi.org/10.1111/gfs.12292.» https://doi.org/10.1111/gfs.12292 -
FRONTADO, N. E. V. et al. Severity of leaf spots caused by Bipolaris maydis and Cercospora fusimaculans on Panicum maximum forages under phosphate fertilization and limestone. Agriculture, v.15, p.2506, 2025. Available from: <Available from: https://doi.org/10.3390/agriculture15232506 >. Accessed: Jan. 27, 2026. doi: 10.3390/agriculture15232506.
» https://doi.org/10.3390/agriculture15232506.» https://doi.org/10.3390/agriculture15232506 -
GASTAL, F.; LEMAIRE, G. Defoliation, shoot plasticity, sward structure and herbage utilization in pasture: Review of the underlying ecophysiological processes. Agriculture, v.5, n.4, p.1146-1171, 2015. Available from: <Available from: https://doi.org/10.3390/agriculture5041146 >. Accessed: Apr. 01, 2025. doi: 10.3390/agriculture5041146.
» https://doi.org/10.3390/agriculture5041146.» https://doi.org/10.3390/agriculture5041146 -
GUEVARA-ESCOBAR, A. et al. Estimating forage mass in a mixed grassland using machine learning, grassland management data, and satellite meteorological data. Revista Mexicana de Ciencias Pecuarias, v.14, n.1, p.61-77, 2023. Available from: <Available from: https://doi.org/10.22319/rmcp.v14i1.6162 >. Accessed: Apr. 01, 2025. doi: 10.22319/rmcp.v14i1.6162.
» https://doi.org/10.22319/rmcp.v14i1.6162.» https://doi.org/10.22319/rmcp.v14i1.6162 -
GURGEL, A. L. C. et al. The effect of residual nitrogen fertilization on the yield components, forage quality, and performance of beef cattle fed on Mombaça grass. Revista de la Facultad de Ciencias Agrarias, v.53, p.296-308, 2021. Available from: <Available from: http://dx.doi.org/10.48162/rev.39.029 >. Accessed: Jan. 27, 2026. doi: 10.48162/rev.39.029.
» https://doi.org/10.48162/rev.39.029.» http://dx.doi.org/10.48162/rev.39.029 -
HUNT .JR; E. R. et al. Remote sensing leaf chlorophyll content using a visible band index. Agronomy Journal, v.103, n.4, p.1090-1099, 2011. Available from: <Available from: https://doi.org/10.2134/agronj2010.0395 >. Accessed: Apr. 01, 2025. doi: 10.2134/agronj2010.0395.
» https://doi.org/10.2134/agronj2010.0395.» https://doi.org/10.2134/agronj2010.0395 -
JANK, L. et al. Breeding tropical forages. Crop Breeding and Applied Biotechnology, v.11, p.27-34, 2011. Available from: <Available from: https://doi.org/10.1590/S1984-70332011000500005 >. Accessed: Apr. 01, 2025. doi: 10.1590/S1984-70332011000500005.
» https://doi.org/10.1590/S1984-70332011000500005.» https://doi.org/10.1590/S1984-70332011000500005 -
LAPIG. Atlas Digital das Pastagens Brasileiras. 2023. Available from: <Available from: https://pastagem.org/atlas/map >. Accessed: Feb. 27, 2024.
» https://pastagem.org/atlas/map -
LARA, M. A. S.; PEDREIRA, C. G. S. Respostas morfogênicas e estruturais de dosséis de espécies de Braquiária à intensidade de desfolhação. Pesquisa Agropecuária Brasileira, v.46, p.760-767, 2011. Available from: <Available from: https://doi.org/10.1590/S0100-204X2011000700012 >. Accessed: Apr. 01, 2025. doi: 10.1590/S0100-204X2011000700012.
» https://doi.org/10.1590/S0100-204X2011000700012.» https://doi.org/10.1590/S0100-204X2011000700012 -
LEE, D. H. et al. UAV, a farm map, and machine learning technology convergence classification method of a corn cultivation area. Agronomy, v.11, n.8, 2021. Available from: <Available from: https://doi.org/10.3390/agronomy11081554 >. Accessed: Apr. 01, 2025. doi: 10.3390/agronomy11081554.
» https://doi.org/10.3390/agronomy11081554.» https://doi.org/10.3390/agronomy11081554 - LEMAIRE, G.; CHAPMAN, D. Tissue flows in grazed communities. In: HODGSON, J. (Ed.). The Ecology and Management of Grazing Systems. Wallingford: Centre for Agriculture and Biosciences International, 1996. p.3-36.
-
LÓPEZ, O. A. M. et al. A review of deep learning applications for genomic selection. BMC Genomics, v.22, p.19, 2021. Available from: <Available from: https://doi.org/10.1186/s12864-020-07319-x >. Accessed: Apr. 01, 2025. doi: 10.1186/s12864-020-07319-x.
» https://doi.org/10.1186/s12864-020-07319-x.» https://doi.org/10.1186/s12864-020-07319-x -
NAHAR, N.; ARA, F. Liver disease prediction by using different decision tree techniques. International Journal of Data Mining and Knowledge Management, v.8, n.2, p.1-9, 2018. Available from: <Available from: https://doi.org/10.5121/ijdkp.2018.8201 >. Accessed: Apr. 01, 2025. doi: 10.5121/ijdkp.2018.8201.
» https://doi.org/10.5121/ijdkp.2018.8201.» https://doi.org/10.5121/ijdkp.2018.8201 -
OLIVEIRA, B. R. et al. Classification of tropical forage grass varieties under moderate and severe water stress using Naïve Bayes and Kernel Density Estimation. Trends in Agricultural and Environmental Science, v.1, p.e230006, 2023. Available from: <Available from: https://doi.org/10.46420/TAES.e230006 >. Accessed: Apr. 01, 2025. doi: 10.46420/TAES.e230006.
» https://doi.org/10.46420/TAES.e230006.» https://doi.org/10.46420/TAES.e230006 -
PACIULLO, D. S. C. et al. Morphogenesis, biomass and nutritive value of Panicum maximum under different shade levels and fertilizer nitrogen rates. Grass and Forage Science, v.11, p.590-600, 2016. Available from: <Available from: https://doi.org/10.1111/gfs.12264 >. Accessed: Apr. 01, 2025. doi: 10.1111/gfs.12264.
» https://doi.org/10.1111/gfs.12264.» https://doi.org/10.1111/gfs.12264 -
PALLATHADKA, H. et al. Impact of machine learning on management, healthcare and agriculture. Materials Today Proceedings, v.80, p.2803-2806, 2023. Available from: <Available from: https://doi.org/10.1016/j.matpr.2021.07.042 >. Accessed: Apr. 01, 2025. doi: 10.1016/j.matpr.2021.07.042.
» https://doi.org/10.1016/j.matpr.2021.07.042.» https://doi.org/10.1016/j.matpr.2021.07.042 -
RAMOS, A. P. M. et al. Random forest ranking approach to predict yield in maize with UAV-based vegetation spectral indices. Computers and Electronics in Agriculture, v.178, p.105791, 2020. Available from: <Available from: https://doi.org/10.1016/j.compag.2020.105791 >. Accessed: Apr. 01, 2025. doi: 10.1016/j.compag.2020.105791.
» https://doi.org/10.1016/j.compag.2020.105791.» https://doi.org/10.1016/j.compag.2020.105791 -
SANTANA, D. C. et al. Classification of Soybean Genotypes as to Calcium, Magnesium, and Sulfur Content Using Machine Learning Models and UAV-Multispectral Sensor. AgriEngineering, v.6, n.2, p.1581-1593, 2024a. Available from: <https://doi.org/10.3390/agriengineering6020090>. Accessed: Apr. 01, 2025. doi: 10.3390/agriengineering6020090.
» https://doi.org/10.3390/agriengineering6020090. -
SANTANA, D. C. et al. A New Approach to Identifying Sorghum Hybrids Using UAV Imagery Using Multispectral Signature and Machine Learning. Algorithms, v.17, n.1, p.23, 2024b. Available from: <Available from: https://doi.org/10.3390/a17010023 >. Accessed: Apr. 01, 2025. doi: 10.3390/a17010023.
» https://doi.org/10.3390/a17010023.» https://doi.org/10.3390/a17010023 -
SANTOS, L. et al. Deep learning regression approaches applied to estimate tillering in tropical forages using mobile phone images. Sensors, v.22, n.11, 2022. Available from: <Available from: https://doi.org/10.3390/s22114116 >. Accessed: Apr. 01, 2025. doi: 10.3390/s22114116.
» https://doi.org/10.3390/s22114116.» https://doi.org/10.3390/s22114116 - SHALEV-SHWARTZ, S.; BEN-DAVID, S. Understanding Machine Learning: From Theory to Algorithms. Cambridge University Press, New York, United States, 2014.
-
SILVA, E. F.; PEREIRA, M. G. Avaliação das estruturas de concordância e discordância nos estudos de confiabilidade. Revista de Saúde Pública, v.32, n.4, p.383-393, 1998. Available from: <Available from: https://doi.org/10.1590/S0034-89101998000400012 >. Accessed: Apr. 01, 2025. doi: 10.1590/S0034-89101998000400012.
» https://doi.org/10.1590/S0034-89101998000400012.» https://doi.org/10.1590/S0034-89101998000400012 -
SIMEÃO, R. M. et al. Genomic selection in tropical forage grasses: Current status and future applications. Frontiers in Plant Science, v.12, 2021. Available from: <Available from: https://doi.org/10.3389/fpls.2021.665195 >. Accessed: Apr. 01, 2025. doi: 10.3389/fpls.2021.665195.
» https://doi.org/10.3389/fpls.2021.665195.» https://doi.org/10.3389/fpls.2021.665195 -
TEODORO, L. P. R. et al. Machine learning for classification of soybean populations for industrial technological variables based on agronomic traits. Euphytica, v.220, n.3, p.40, 2024. Available from: <Available from: https://doi.org/10.1007/s10681-024-03301-w >. Accessed: Apr. 01, 2025. doi: 10.1007/s10681-024-03301-w.
» https://doi.org/10.1007/s10681-024-03301-w.» https://doi.org/10.1007/s10681-024-03301-w -
VERAS, E. L. L. et al. Gas exchange and productivity responses of Panicum maximum cultivars to increasing soil potassium levels in a greenhouse study. Scientific Reports, v.15, p.40704, 2025. Available from: <Available from: https://doi.org/10.1038/s41598-025-24330-y >. Accessed: Jan. 27, 2026. doi: 10.1038/s41598-025-24330-y.
» https://doi.org/10.1038/s41598-025-24330-y.» https://doi.org/10.1038/s41598-025-24330-y -
VERAS, E. L. L. et al. Tillering and structural characteristics of Panicum cultivars in the Brazilian semiarid region. Sustainability, v.12, p.3849, 2020. Available from: <Available from: https://doi.org/10.3390/su12093849 >. Accessed: Apr. 01, 2025. doi: 10.3390/su12093849.
» https://doi.org/10.3390/su12093849.» https://doi.org/10.3390/su12093849 -
VILAR, P. et al. Object-based classification approaches for multitemporal identification and monitoring of pastures in agroforestry regions using multispectral unmanned aerial vehicle products. Remote Sensing, v.12, n.5, p.814, 2020. Available from: <Available from: https://doi.org/10.3390/rs12050814 >. Accessed: Apr. 01, 2025. doi: 10.3390/rs12050814.
» https://doi.org/10.3390/rs12050814.» https://doi.org/10.3390/rs12050814 -
VISHNOI, V. et al. Crop disease classification through image processing and Machine Learning techniques using leaf images. In: proceedings of the 1st international conference on advances in computing and future communication technologies, Bangalore, India. ICACFCT, 2021. 2021. Available from: <Available from: https://doi.org/10.1109/ICACFCT53978.2021.9837353 >. Accessed: Apr. 01, 2025. doi: 10.1109/ICACFCT53978.2021.9837353.
» https://doi.org/10.1109/ICACFCT53978.2021.9837353.» https://doi.org/10.1109/ICACFCT53978.2021.9837353 -
ZHANG, L. et al. H2O2, Ca2+, and K+ in subsidiary cells of maize leaves are involved in regulatory signaling of stomatal movement. Plant Physiology and Biochemistry, v.152, p.243-251, 2020. Available from: <Available from: https://doi.org/10.1016/j.plaphy.2020.04.045 >. Accessed: Apr. 01, 2025. doi: 10.1016/j.plaphy.2020.04.045.
» https://doi.org/10.1016/j.plaphy.2020.04.045.» https://doi.org/10.1016/j.plaphy.2020.04.045 -
ZHU, W. et al. Estimating leaf chlorophyll content of crops via optimal unmanned aerial vehicle hyperspectral data at multi-scales. Computers and Electronics in Agriculture, v.178, p.105786, 2020. Available from: <Available from: https://doi.org/10.1016/j.compag.2020.105786 >. Accessed: Apr. 01, 2025. doi: 10.1016/j.compag.2020.105786.
» https://doi.org/10.1016/j.compag.2020.105786.» https://doi.org/10.1016/j.compag.2020.105786
-
CR-2025-0173.R2
-
DATA AVAILABILITY STATEMENT
All the original contributions presented in this study are included in the article. Further inquiries can be directed to the corresponding author.
-
DECLARATION OF USE OF ARTIFICIAL INTELLIGENCE
We declare that this work did not use any artificial intelligence resources for the conception and writing of the manuscript.
Edited by
-
ASSOCIATE EDITOR:
Alessandro Dal’Col Lúcio (0000-0003-0761-4200)
All the original contributions presented in this study are included in the article. Further inquiries can be directed to the corresponding author.
