Revisiting Earth’s Seasonality using Machine Learning Models

The seasonality of Earth has a profound impact on almost any aspect of life on our planet. Seasonality drives vegetation cycles, influences wildlife behavior, and shapes human health, society, and culture. Seasonality, traditionally defined by the equal-length Astronomical seasons uniformly applied across the Earth, provides a simple division but overlooks key weather patterns, latitudinal variations, and Climate Change effects. In this study we propose a data-driven approach to seasons based on unsupervised machine learning models. We develop an algorithm that clusters meteorological reanalysis data—temperature, precipitation, and relative humidity—into meaningful seasonal patterns. We build on this algorithm to objectively define seasons in every region globally, and analyze the effect of Climate Change on these clusters. The results demonstrate that seasonality in different regions of Earth is driven by different meteorological factors. We find that the duration of seasons varies significantly with latitude. For example, Summer lasts ~120 days at ±30° latitude but decreases by 7±1 days for every 10° closer to the poles. Conversely, Winter lasts ~125 days at ±30° latitude and extends by 13±2 days per 10° poleward. Our analysis reveals notable changes in the onset and duration of seasons driven by Climate Change; most notably, we find that summers have extended by 7±8 days at the expense of winters which have shortened by 8±11 days over the past 40 years, while the transition seasons have shifted accordingly. The observed shifts in seasonality highlight the rapid impact of Climate Change on Earth's systems, with profound consequences for ecosystems, agriculture, and society.