Utilizing Natural Language Processing for the Analysis of BMKG Decadal Atmospheric Dynamics Reports in 2025

Abstrak

The Meteorology, Climatology, and Geophysics Agency (BMKG) publish decennial reports that provide valuable insights into Indonesia's meteorological conditions and their temporal fluctuations. However, due to their narrative structure, conducting direct quantitative analysis is problematic. This study seeks to address this issue by using a transparent, repeatable natural language processing (NLP) method to identify temporal trends in climatic conditions favourable to the formation of acid rain. The collection contains 36 BMKG decadal atmospheric dynamics studies for 2025. The proposed approach entails gathering textual input, performing basic preprocessing (case normalization, character sanitization, space-based tokenization, and stop-word removal), and subsequently employing predefined keyword dictionaries for analysis. These dictionaries delineate weather conditions that either facilitate or inhibit the formation of acid rain. The scores for acid rain conditions are determined by the frequency of specific keywords, adjusted for the document's length. Subsequently, they are categorized into groups utilizing statistical thresholds derived from the mean and standard deviation of the adjusted scores. Non-parametric statistical tests are employed to examine temporal patterns with greater specificity. The findings indicate that normalized acid rain scores are elevated in the initial years of the decade, specifically 2025, before gradually declining until year-end. The Spearman rank correlation test reveals a statistically significant negative correlation between normalized scores and time (\rho = -0.494, p = 0.0022). The Mann–Kendall test indicates a significant downward trend (Z = -2.902). These results demonstrate that the climatic conditions responsible for acid rain occurred only temporarily, rather than year-round. The core element of this work is a straightforward, lexicon-based NLP approach that is easily understood, replicable, and applicable, and can transform narrative atmospheric reports into structured quantitative metrics. This is beneficial for research on atmospheric dynamics and environmental analysis with official written data.