The Amazon Rainforest is highlighted by the global community both for its extensive vegetation cover that constantly suffers the effects of anthropic action and for its substantial biodiversity. This dataset presents data of meteorological variables from the Amazon Rainforest region with a spatial resolution of 0.001° in latitude and longitude, resulting from an interpolation process. The original data were obtained from the GoAmazon 2014/5 project, in the Atmospheric Radiation Measurement (ARM) repository, and then processed through mathematical and statistical methods. The dataset presented here can be used in experiments in the field of Data Science, such as training models for predicting climate variables or modeling the distribution of species.

The application of Machine Learning (ML) to Species Distribution Modeling (SDM)—a critical task for biodiversity monitoring and wildlife management—has expanded significantly. To consolidate the state-of-the-art, this paper presents a systematic literature review of 201 articles published between 2017 and 2024. Our analysis reveals a clear and increasing trend in the adoption of AI and ML for SDM, which is predominantly framed as a classification problem. We identify Random Forests (RF) as the dominant technique, appearing in 52.2% of reviewed studies, followed by Maximum Entropy (MaxEnt) and Boosting-based models. These models consistently demonstrate high predictive performance, with the Area Under the Receiver Operating Characteristic Curve (AUC-ROC) being the primary evaluation metric. The review also reveals a significant taxonomic bias, with the Animal kingdom (62.7% of studies) and the Chordate phylum being the most common subjects. This work provides a definitive benchmark of current methods, performance standards, and application trends, offering valuable guidance for researchers and practitioners in ecological informatics.

The convergence of Machine Learning (ML) and ecology has provided powerful tools for Species Distribution Modeling (SDM). However, the application of complex algorithms to presence-only data often lacks methodological standardization and interpretability. This study proposes a comprehensive and standardized workflow that integrates robust pseudo-absence generation strategies with Explainable Artificial Intelligence (XAI) to enhance the reliability of SDMs. We evaluated four distinct pseudo-absence selection methods—Random, Spatially Constrained, Environmentally Dissimilar, and a Three-Step Methodology—across different classifiers, utilizing a dual-metric framework that prioritizes the False Negative Rate (FNR) for consistency and the True Skill Statistic (TSS) for discriminatory power. The workflow was validated through a case study involving Tyrannidae and Thraupidae bird families in the Amazon, incorporating anthropogenic variables from the GoAmazon campaign. Our results demonstrate that the choice of pseudo-absence strategy is a critical determinant of model performance, with the Three-Step Methodology combined with Support Vector Machines achieving the optimal balance between sensitivity and specificity. Furthermore, the application of post-hoc XAI techniques, including SHAP and LIME, confirmed the ecological validity of the models by revealing biologically meaningful relationships between species distribution and environmental predictors, such as atmospheric pollutants. This work provides a reproducible pipeline that mitigates the closed box nature of ML models, offering a transparent and effective tool for biodiversity conservation and environmental planning.

Machine Learning Models are widely used in Computational Ecology. They can be applied for Species Distribution Modeling, which aims to determine the probability of occurrence of a species, given the environmental conditions. However, for ecologists, these models are considered as "black boxes", since basic Machine Learning knowledge is necessary to interpret them. Thus, in this work four Explainable Artificial Intelligence techniques - Local Interpretable Model-Agnostic Explanation (LIME), SHapley Additive exPlanations (SHAP), BreakDown and Partial Dependence Plots - were evaluated to the Random Forests classifier for Coragyps atratus in the Amazon Basin region. It was found that the SHapley Additive exPlanations technique and Partial Dependence Plots are able to improve the explainability of the model.

Explainable AI (XAI) techniques enhance ML interpretability but often require technical expertise. We propose a multi-agent architecture that improves LLM-generated explanations through structured reasoning and contextual retrieval. Our system integrates web search, Retrieval-Augmented Generation (RAG), and XAI outputs via specialized agents. Experiments on the Adult dataset show that our approach outperforms standard LLM explanations by 7% in Context Awareness. Additionally, we validate LLM as a Judge, achieving over 80% correlation with human evaluations. Different LLMs, including OpenAI’s GPT-4o and GPT-4o-mini, were tested, highlighting the effectiveness of multi-agent systems in ML explainability.

This dataset presents data of aerosol, trace-gases and meteorological variables from the Amazon Rainforest region, resulting from an interpolation process. The original data were collected from the GOAmazon 2014/15 project, from the Atmospheric Radiation Measurement (ARM) repository.