Earth

In semi-arid Mediterranean regions like the Middle Seybouse Watershed (Northeastern Algeria), the accurate spatial prediction of environmental groundwater potential is vital for sustainable resource management and risk mitigation. This study develops a high-resolution mapping framework by integrating multi-source geospatial data managed by a geographic information system (GIS) with optimized ensemble machine learning algorithms to identify the groundwater potential zones map. Initially, 18 environmental conditioning factors covering topographic, climatic, and hydrogeomorphologic elements were evaluated for data redundancy through dimensionality reduction techniques including Pearson correlation and multicollinearity diagnostics. The research identified a refined set of 10 independent predictors: elevation, land use/land cover (LULC), Topographic Wetness Index (TWI), Topographic Position Index (TPI), roughness, curvature, rainfall, drainage density, soil texture, and soil type, all maintaining a variance inflation factor (VIF) of less than 3.15 to ensure both model simplicity and stability. Four advanced machine learning algorithms, including random forest (RF), support vector machine (SVM), light gradient boosting machine (LGBM), and a stacking ensemble, were subsequently trained and validated using an 80/20 data split. Among these, the LGBM model emerged with a high predictive performance, achieving an area under the curve (AUC) score of 0.944, closely followed by the stacking ensemble with an AUC of 0.943. The comprehensive 30 m-resolution potential map, which is classified into five classes (very low, low, medium, high, and very high), offers a robust, data-driven resource for regional stakeholders. This mapping exercise not only provides a transparent and replicable methodology, but also effectively connects machine learning outputs with geomorphological insights in intricate hydrological settings, thus supporting sustainable environmental management and decision-making in the region.

Anomalous ozone loss in the Northern Hemisphere during winter–spring periods has been increasingly observed in recent decades. Regular monitoring of stratospheric gases involved in ozone-depleting processes is essential for predicting the appearance of ozone mini-holes and the growth in UV surface irradiance. The St. Petersburg State University site located on the border of mid- and high latitudes is equipped with the Bruker IFS 125HR spectrometer for trace gas measurements and the UV and VIS Ocean Optics HR4000 spectrometers for total ozone column (TOC) measurements. The harmonization of TOC series from three instruments made it possible to analyze TOC variability during rapid changes in air masses. A decrease in ozone in the 8–18 km (−0.41 ± 0.33% yr⁻¹) and 18–25 km (−0.17 ± 0.15% yr⁻¹) layers, as well as in stratospheric HNO₃ (−0.31 ± 0.18% yr⁻¹), is observed for the 2009–2026 period, along with an increase in HF (+0.41 ± 0.20% yr⁻¹) mainly due to changes in atmospheric dynamics over the past decade. An observed decrease in CCl₃F (−0.43 ± 0.06% yr⁻¹), CCl₂F₂ (−0.71 ± 0.04% yr⁻¹), ClONO₂ (−1.14 ± 0.82% yr⁻¹), and HCl (−0.31 ± 0.16% yr⁻¹) indicates a decrease in the chlorine-containing gases in the atmosphere. A comparison of the EMAC and SOCOLv3 simulations with observations for the common 2009–2019 period showed that ozone variability, including trends, seasonal variations, and rapid winter–spring changes, is described by the models within the measurement uncertainty limits, whereas for HCl, ClONO₂, and HNO₃, the quantitative discrepancy between the simulations and measurements exceeds the corresponding measurement uncertainty.

The consequences of tropical forest cover loss for regional hydrometeorology and global ecosystem function represent one of the most pressing environmental challenges of our time. The Yucatán Peninsula (YP) has experienced unprecedented deforestation pressure over recent decades, yet the effects of vegetation loss on precipitation dynamics remain poorly understood and uncharacterized. This study examines the 30-year relationship (1993–2022) between land-cover change and precipitation trends in the YP using satellite-derived coverage indices from LANDSAT and monthly precipitation data from the CHIRPS dataset. We applied multi-scale buffer analysis, combined with Spearman’s ρ statistic for precipitation trend detection and Pearson correlation for annual and seasonal time series. The results reveal a persistent positive correlation between two-epoch mean vegetation biomass and precipitation availability, consistent with regional water recycling mechanisms. Using directional (upwind-sector) buffers and a pre-comparison-period NDVI baseline, extended to 250 km, Adjcor_sp_ΔNDVI, the partial correlation between vegetation loss and precipitation trend, net of NDVI baseline and water surface proportion, is season- and scale-dependent rather than a single degradation signal: negative (wetter-with-loss, opposing the hypothesis) at small scales for Annual (≤55 km, not significant) and at small-to-medium scales for MAM (≤95 km), and positive (drier-with-loss, the hypothesized direction) at large scales for Annual and MAM and throughout for JJA and SON. Under Tjostheim’s coefficient, the association is significant across most of the tested range for every series (108 of 135 cells, 80.0%, unchanged after Benjamini–Hochberg FDR correction): consistently in the hypothesized direction for JJA (24/27 scales, 15–250 km) and SON (20/27, 35–250 km); predominantly in the opposite direction for MAM (20 of 26 significant scales, ≤95 km) with a hypothesis-consistent minority at its largest scales (6/26, 125–250 km); and sign-dependent on scale for Annual (15/27, all hypothesis-consistent, 60–250 km) and DJF (23/27, 12 opposite-direction at small scales and 11 hypothesized-direction at large scales); this threshold marks the magnitude at which forest cover loss, driven by agriculture, urbanization, and infrastructure development, disrupts the vegetation–precipitation association. Seasonal decomposition under this revised method shows the hypothesized (positive) direction throughout both summer and fall, with summer’s signal detectable from a shorter buffer radius (15 km) than fall’s (35 km), consistent with direct, locally triggered convection during the wet-season core giving way to a longer-range moisture-recycling pathway by fall. Mean ΔNDVI does not differ between federal protected areas and the rest of the study area, a balance between coastal-ANP forest loss and weak-signal interior recovery; independent land-cover classification shows a net gain within protected areas that this vegetation-index comparison alone does not detect. These findings provide a quantitative, multiscale basis for conservation policy, highlighting priority zones for conservation intervention.