Título: An Integrated Wavelet DAG -CNLA -SCM Structure with Recurrent Neural Networks for Climate Vulnerabilities Prediction
Autores: Babatunde Ayinla, Danton Diego Ferreira, Felipe Oliveira e Silva, Ahmed Ali Abdalla Esmin
Resumo: Advancements in Artificial Intelligence (AI) have yielded several machine learning and deep learning models that highlight correlation while failing to differentiate cause-and-effect relationships. Current methodologies, such as Granger Causality (GC), solely identify temporal causal linkages in which causes precede effects, hence neglecting contemporaneous associations. This paper introduced the integration of Causal Network Learning Algorithms (CNLA), Structure Causal Models (SCM), and Long Short-Term Memory (LSTM) networks to forecast climatic catastrophes. A Directed Acyclic Graph (DAG) is constructed utilizing the Temporal Causal Discovery Framework (TCDF) technique, which employs attention-based Convolutional Neural Networks (CNNs) to ascertain causal relationships between temperature and carbon dioxide (CO). The study examines the enduring issue of causality versus correlation in the relationship between temperature and CO with regard to climate change. Results reveal Europe undergoes the fastest warming at 0.80°C per decade, with Eastern Europe, particularly Ukraine, experiencing extreme warming at 1.21°C per decade. Northern Europe exhibits an unusual wavelet maximum power of 5.596, indicating complex periodic patterns. Despite distinct CO-temperature correlations, regression analysis shows emissions account for merely 0.032% of temperature variability. Time lag studies verify emission-temperature connections persist beyond 1–5 year temporal scales, endorsing emission reduction measures while highlighting geographic climate inequities affecting low-emission regions like Africa disproportionately.
Palavras-chave: Causal Inference; Wavelet Analysis; Convolutional Neural Networks; Structure Causal Models; Causal Network Learning Algorithms; Artificial Intelligence; Granger Causality; Climate Change Prediction.
Páginas: 7
Código DOI: 10.21528/CBIC2025-1191067
Artigo em PDF: CBIC_2025_paper1191067.pdf
Arquivo BibTeX:
CBIC_2025_1191067.bib