شناسایی رژیم اقلیمی حوضه دریای خزر با رویکرد داده‌کاوی: رهیافتی برای کشاورزی پایدار و سازگاری با تغییر اقلیم

Extended Abstract

Introduction

The Caspian Sea, the largest enclosed body of water on Earth, is crucial for the ecological stability, food security, and economic development of the surrounding countries. However, in recent decades, this unique ecosystem has faced increasing hydroclimatic stresses that threaten both environmental sustainability and regional resilience. The basin's complex topography and significant spatial and temporal climatic contrasts have led to diverse and unstable climatic patterns that traditional classification systems cannot accurately represent. Furthermore, the impact of large-scale teleconnection patterns, particularly the El Niño–Southern Oscillation (ENSO), has increased climatic variability across the region. In light of these challenges, the application of modern data-driven methods for objective climatic zoning has become essential. Integrating unsupervised machine learning algorithms that can uncover hidden patterns with high-resolution reanalysis data presents new opportunities to explore the basin's climatic structures and multiscale interactions. Therefore, this study utilizes a hybrid DBSCAN–K-Means framework, along with ENSO indices, to identify homogeneous climatic zones, examine the effects of El Niño and La Niña phases on their stability, and assess the practical implications of these changes for agricultural and water resource management in the Caspian Sea basin.



Methodology

The Caspian Sea, the largest enclosed body of water in the world, is situated between Asia and Europe, stretching approximately 1,200 km in length and ranging from 196 to 435 km in width. It displays considerable variations in depth, temperature, and salinity, leading to a complex network of water circulation and diverse microclimates throughout the region.

In this study, high-resolution ERA5 reanalysis data and ENSO teleconnection indices were collected for the period from 1964 to 2024, specifically focusing on the months of September to November. Key variables included air temperature, dew point, wind components, surface pressure, and total cloud cover. The data were preprocessed and normalized, with Principal Component Analysis (PCA) applied as needed to reduce dimensionality.

To identify homogeneous climatic regions, we employed a combined K-Means and DBSCAN clustering framework. The quality of the clustering was evaluated using the Silhouette coefficient and statistical tests such as ANOVA. This methodology facilitated the extraction of spatial clusters and the analysis of their responses to different ENSO phases.



Results and discussion

The combined DBSCAN–K-Means framework successfully identified nine distinct and meaningful climatic regions across the Caspian Sea basin. The optimal number of clusters was established by maximizing the silhouette coefficient (0.65) and minimizing the Davies–Bouldin index, which indicates strong internal cohesion and clear separation among clusters. These clusters reflect significant latitudinal and elevational gradients, capturing the basin’s topographic and climatic diversity. The northern clusters, including the Greater Caucasus and adjacent highlands, exhibited mean autumn temperatures of 10–12 °C, while southern and southeastern clusters recorded mean temperatures near 20 °C, highlighting the stark thermal contrast between the northern and southern parts of the basin.

The clusters also demonstrate high spatial coherence, aligning with known geomorphological and climatic features such as the Alborz and Caucasus mountain ranges, inland plains, and the Caspian Sea coastline. By incorporating spatial coordinates into the clustering process, we reinforced the consistency of the boundaries, ensuring they reflect natural features rather than arbitrary patterns. A statistical evaluation using one-way ANOVA confirmed significant differences among clusters in 2 m air temperature and surface pressure (p < 0.001), supporting the statistical and physical reliability of the findings.

Each climatic cluster displays a unique profile of temperature, pressure, and humidity. Northern highland clusters exhibit higher surface pressure and broader seasonal temperature ranges, while southern lowland clusters have higher temperatures, lower surface pressure, and a dominance of summer thermal lows. Coastal clusters, influenced by their proximity to the Caspian Sea, maintain higher relative humidity, greater cloud cover, and lower evaporation rates, whereas inland and mountainous clusters are drier with increased summer evaporation.

Analysis of seasonal dynamics revealed that northern clusters experience cold winters and hot summers, resulting in high annual thermal amplitude, while southern and coastal clusters showcase milder seasonal cycles. Humidity patterns also demonstrate clear seasonal contrasts, underscoring the influences of the sea, topography, and regional atmospheric circulation.

The integration of ENSO analysis with clustering results indicated that El Niño and La Niña phases significantly affect the frequency and extent of the clusters. El Niño expands warm–humid regimes in the south while reducing cold areas in the north, whereas La Niña produces the opposite effect, enlarging northern cold regions. These findings provide evidence of teleconnections between global climatic patterns and regional variability.

Overall, the DBSCAN–K-Means framework offers a coherent, interpretable, and statistically robust delineation of climatic regions. It establishes a scientifically grounded foundation for climate-informed agricultural planning, water resource management, and regional policy-making, especially in the context of global climate change and increasing anthropogenic pressures in the Caspian Sea basin.



Conclusion

By integrating machine learning algorithms with climatological insights, this study effectively identified complex, multi-scale hydro-climatic structures within the Caspian Sea basin. The DBSCAN–K-Means framework delineated nine distinct, statistically robust climate zones that reflect thermal, moisture, and pressure gradients, consistent with coastal-inland and elevational contrasts. Analysis of ENSO phases revealed that both El Niño and La Niña significantly influence the spatial patterns and intensity of climate regimes. These findings lay the groundwork for seasonal forecasting, adaptive water management, and climate-responsive agricultural planning. Overall, the framework presents a practical approach for climate adaptation policy, linking advanced data-driven analysis with actionable decisions at both regional and local levels.



Keywords:

Climate classification, atmospheric risks, hybrid clustering, spatiotemporal patterns





Funding

There is no funding support.



Authors’ Contribution

Authors contributed equally to the conceptualization and writing of the article. All of the authors approved thecontent of the manuscript and agreed on all aspects of the work declaration of competing interest none.



Conflict of Interest

Authors declared no conflict of interest.



Acknowledgments

We are grateful to all the scientific consultants of this paper