Pedro Henrique Fontes dos Santos

Mr. Pedro Henrique Fontes dos Santos

orcid

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
5 March 2025 in OBSERVATÓRIO DE LA ECONOMÍA LATINOAMERICANA
Verified
Journal Article
Open access
PDF file
PSPedro Henrique Fontes dos Santos
Pedro Henrique Fontes dos Santos
Gianmarco Goycochea CasasGianmarco Goycochea Casas
Gianmarco Goycochea Casas
CSCarlos Pedro Boechat Soares
Carlos Pedro Boechat Soares
DBDaniel Henrique Breda Binoti
Daniel Henrique Breda Binoti
LLLucas Sérgio de Sousa Lopes
Lucas Sérgio de Sousa Lopes
HLHelio Garcia Leite
Helio Garcia Leite
Forest management relies on accurate growth and harvest stock estimations, typically obtained through forest inventories. This study evaluated sampling strategies for training Artificial Neural Networks (ANNs) to predict tree heights, aiming to minimize sampling effort and associated costs, and assessed the efficiency of using these predictions for site index classification. Data were collected from Pinus taeda L. spacing experiment in southern Brazil, encompassing measurements from 3 to 15 years of age. Four scenarios were evaluated: (a) adjusting height-diameter equations with varying sample sizes, (b) applying equations from one age to the next, (c) training ANNs with different sample sizes, and (d) using ANNs trained in one year for subsequent years. Results showed that ANNs, particularly with a sample of three trees per plot (Scenario c), provided superior height predictions compared to traditional height-diameter relationships, achieving higher correlation and lower error. While applying ANNs trained in one year to the next (Scenario d) showed reduced accuracy, Scenario (c) maintained high precision. The use of ANN-predicted heights for site index classification was also deemed viable. The study demonstrates that ANNs can significantly reduce the sampling effort required for forest inventories while maintaining accuracy in height estimation and site productivity assessment. The findings support the continued use of the established practice of measuring three trees per plot for ANN training, optimizing both cost and precision in forest inventory processes.

23 February 2026 in Forest Science
Verified
Journal Article
Open access
PDF file
Gianmarco Goycochea CasasGianmarco Goycochea Casas
Gianmarco Goycochea Casas
Zool Hilmi IsmailZool Hilmi Ismail
Zool Hilmi Ismail
PSPedro Henrique Fontes dos Santos
Pedro Henrique Fontes dos Santos
LRLeonardo Ippolito Rodrigues
Leonardo Ippolito Rodrigues
TATassius Menezes Araújo
Tassius Menezes Araújo
HLHelio Garcia Leite
Helio Garcia Leite
Accurate prediction of forest growth and yield is essential for strategic planning in intensive plantation management. This study evaluates whether unsupervised anomaly detection can be used as a systematic data-quality layer in stand-level growth and yield modeling, after standard consistency checks have been applied. We used a multi-regional continuous forest inventory of hybrid Eucalyptus urophylla × Eucalyptus grandis plantations in Minas Gerais, Brazil (6,553 measurements from 1,749 permanent plots in three regions). Four unsupervised methods, Isolation Forest, One-Class SVM, Local Outlier Factor and a dense autoencoder, were applied within each region to identify multivariate anomalies in stand age, volume, basal area and dominant height. Using Isolation Forest with a contamination rate of 0.10, approximately 10% of records were flagged as anomalous, but this corresponded to the complete removal of 2–4% of plots, while most plots either retained all measurements or lost only a subset of them. We then calibrated nine nonlinear machine learning models to predict stand volume at the second measurement (V₂) under two scenarios: using the full dataset and using the anomaly-filtered dataset. Gradient Boosting and other tree ensembles achieved the best performance in both cases (R 2 ≈ 0.86–0.88, relative RMSE ≈ 13–15%), and differences in accuracy between full and filtered datasets were small (changes in R 2 < 0.03 and in relative RMSE < 1 percentage point). A sensitivity analysis across contamination levels from 0.05 to 0.20 confirmed that a 10% threshold offers a practical compromise, removing a consistent set of multivariate extremes while preserving plot coverage and predictive performance.

26 October 2026 in Forests
Verified
Journal Article
Open access
PDF file
Gianmarco Goycochea CasasGianmarco Goycochea Casas
Gianmarco Goycochea Casas
Zool Hilmi IsmailZool Hilmi Ismail
Zool Hilmi Ismail
AAAbdulaziz S Alnori
Abdulaziz S Alnori
SSSilas Tadin Sardinha
Silas Tadin Sardinha
PSPedro Henrique Fontes dos Santos
Pedro Henrique Fontes dos Santos
HLHelio Garcia Leite
Helio Garcia Leite
Ground robots are proposed for inventory in commercial eucalypt plantations, where inter-row corridors leave a platform about one metre of lateral room for error. This study asked, entirely in simulation with parameters plausible rather than calibrated to a platform or stand, when controller tuning changes the outcome. Three tuning algorithms (finite-difference descent, Bayesian optimisation and covariance matrix adaptation evolution strategy) and two robust controllers (sliding-mode and active disturbance rejection) were compared across two commercial spacings and a factorial ablation of state estimation and terrain disturbance. In 1440 simulated missions, the success rate was 0.999. Intermittent satellite positioning consumed 18% of the lateral error budget and underlay every stem contact, whereas terrain disturbance cost 18% of traversal time and 2% of the budget; a filter matched to the plant left this unchanged. Tuning the algorithm and control law moved clearance by at most 2% of the budget. Over 64,512 missions under amplified disturbance, breakdown was predicted by integral action rather than the stability margin, and an anti-windup clamp showed this effect to be mostly integrator windup. Within this model, safety is bounded by localisation rather than the control loop; the magnitudes depend on the simulated parameters and are hypotheses for field testing.