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

ABSTRACT
The Caspian Sea basin is considered one of the regions sensitive to climate change due to its prominent role in ensuring Iran’s food security, economic development, and environmental sustainability. Topographic irregularities, land cover diversity, and proximity to a vast water body have led to the formation of complex and heterogeneous climatic patterns within this basin. Therefore, the precise identification of spatiotemporal climate structures in this region is an essential step toward climate-adaptive management and the planning of water and agricultural resources. This study employs a mixed-methods analytical framework, integrating principal component analysis (PCA) for dimension reduction with K-means and DBSCAN (Density-based spatial clustering of applications with noise) clustering techniques, in order to reveal latent climatic patterns. ERA5 reanalysis data from 1964 to 2024, along with seven key climatic indicators, were used as inputs to the model. Initially, principal component analysis extracted the dominant climatic features and reduced the dimensionality of the data. Subsequently, clustering algorithms divided the basin into distinct subregions that clearly demonstrate the influence of latitudinal gradient, elevation above sea level, and distance from the coast. Validation utilizing the silhouette and Davies–Bouldin indices, along with analysis of variance (ANOVA), confirmed the statistical significance of the distinction among these clusters. Furthermore, correlation analysis with ENSO indices (El Niño–Southern Oscillation and the Southern Oscillation Index) revealed that the El Niño phase is associated with warmer and wetter conditions in the southern half of the basin, whereas the La Niña phase exacerbates colder and drier conditions in the northern parts. The findings serve as an analytical framework for climate zoning, monitoring atmospheric risks, optimizing agricultural calendars, and sustainable management of natural resources in the Caspian basin.
Extended Abstract
Introduction
The Caspian Sea basin, as the largest enclosed body of water in the world, plays an essential role in ensuring food security, preserving biodiversity, and supporting agricultural and industrial activities in the adjacent countries. However, in recent decades, this unique ecosystem has come under increasing hydroclimatic pressures, including marine heatwaves, water level fluctuations, shoreline erosion, eutrophication, and the accumulation of heavy metals. The complex topographic structure and significant climatic contrasts within this basin, together with its high sensitivity to global atmospheric systems such as the El Niño–Southern Oscillation (ENSO), have given rise to spatiotemporal heterogeneities that traditional classification systems (e.g., Köppen–Geiger) are unable to fully represent. Prior studies have chiefly relied on linear and station-based statistical methods or classical classifications, and there exists a significant methodological gap in identifying homogeneous climatic regimes based on multidimensional data without structural assumptions at the scale of the entire basin. Consequently, this study aims to employ a hybrid data mining framework to identify homogeneous climatic regions, investigate the influence of ENSO phases on their stability, and provide practical applications for sustainable agriculture and water resource management in the Caspian Sea basin.
 
Methodology
This study utilizes ERA5 reanalysis data with a spatial resolution of approximately 31 kilometers for the period 1964 to 2024, concentrating on the months of September to November. Seven key climatic variables were employed, including air temperature at 2 meters above ground, dew point temperature, wind components (U and V) at 10 meters above ground, surface pressure, mean sea level pressure, and total cloud cover. Furthermore, ENSO indices, encompassing ENSO and the Southern Oscillation Index (SOI), were obtained from the Climate Prediction Center of the National Oceanic and Atmospheric Administration (NOAA) of the United States. The data were normalized via the min-max method, and geographic coordinates (longitude and latitude) were added to the feature vector to enhance spatial coherence. PCA was applied to reduce dimensionality and eliminate correlations, and the four principal components were retained that accounted for more than 78% of the data variance. In the clustering stage, the DBSCAN algorithm (with parameters ε = 0.9 and MinPts = 50) was first applied to remove noisy points (approximately 12% of the data), and its output was then transferred as the initial input to the k-means algorithm. Clustering quality was assessed using the silhouette index (coefficient of 0.65), the Davies–Bouldin index (0.58), as well as MSE and EAI indices. The statistical validity of the clusters was confirmed through one-way analysis of variance (ANOVA) and the chi-square test to examine the association with ENSO. Additionally, a 3×3 modal filter was applied for spatial smoothing and to enhance spatial continuity.
 
Results and Discussion
The hybrid DBSCAN-K-means framework successfully identified nine climatically distinct and physically meaningful regions across the Caspian Sea basin. The optimal number of clusters was determined by maximizing the silhouette coefficient (0.65) and minimizing the Davies–Bouldin index. The Analysis of variance confirmed significant differences among the clusters in terms of 2-meter temperature and surface pressure (p < 0.001). The spatial pattern of the clusters is clearly governed by latitudinal and elevational gradients: the northern clusters (including the Caucasus and adjacent highlands) exhibit average autumn temperatures of 10–12°C, whereas the southern and southeastern clusters experience temperatures approaching 20°C. Based on their dominant climatic characteristics, these nine clusters can be aggregated into three physical types: northern-mountainous (characterized by lower summer temperatures, higher surface pressure, and stable snow cover); coastal-humid (characterized by thermal moderation due to the sea, high relative humidity, and greater cloud cover); and inland-semi-arid (characterized by higher summer temperatures, distance from moisture sources, and intensified evaporative processes). Seasonal variability analysis indicates that the northern clusters show a high annual thermal range (cold winters and hot summers), whereas the coastal clusters display a more moderate seasonal cycle. The analysis of the association with ENSO revealed that during the El Niño (warm) phase, the frequency of warmer clusters in the southern half of the basin upsurges, with positive temperature anomalies of 1–2°C observed, while during the La Niña (cold) phase, the cold northern clusters expand and lead to the intensification of frost events. The chi-square test confirmed a significant relationship between ENSO phases and the frequency of climatic zones (p < 0.01).
 
Conclusions
This study, by mixing unsupervised machine learning algorithms with ERA5 reanalysis data, successfully identified nine homogeneous climatic zones in the Caspian Sea basin, reflecting the complex atmosphere–ocean and topographic interactions. Statistical validation confirmed the high internal coherence and significant differentiation of these zones. The analysis of the ENSO effect demonstrated that El Niño is associated with the expansion of warm and humid conditions in the south, whereas La Niña is associated with the intensification of cold and dry patterns in the north. From a practical perspective, the findings can serve as a basis for designing adaptive cropping calendars, optimizing water resource allocation at the sub-basin scale, and developing early warning systems for climate hazards. The proposed framework supports the transition from traditional static planning to dynamic, climate-adaptive planning in the Caspian Sea basin. It is recommended that forthcoming research studies incorporate additional variables such as soil moisture, actual evapotranspiration, and socio-economic indicators into the model, so that more comprehensive decision support systems can be developed for resilience to climate change.
 
Funding
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
 
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
The authors declare no conflict of interest.
 
Acknowledgements
We are grateful to all the scientific consultants of this paper.