Atmosphere

This study assesses Land Use and Land Cover (LULC) and Land Surface Temperature (LST) variability in the semi-arid Aligarh District, Uttar Pradesh, India, using two Landsat 9 images acquired on 30 May 2022 (summer) and 24 December 2022 (winter). This research employs a semi-automated hybrid classification approach integrating Maximum Likelihood Classification and Object-Based Image Analysis with Landsat imagery to map seasonal LULC for summer and winter 2022. Five classes, namely, cropland, built-up areas, barren land, plantations, and water bodies, were identified with overall accuracies of 94.4% (Kappa = 0.93) in summer and 91.2% (Kappa = 0.89) in winter. On the summer date LST ranged from 26.48 °C to 46.40 °C and on the winter date from 13.11 °C to 23.85 °C. Inter-predictor Pearson correlation analysis revealed severe multicollinearity among NDVI, SAVI, NDBI, and NDWI, with inter-predictor correlations ranging from |r| = 0.855 to 0.994, forming a tightly redundant spectral cluster. NDVI and SAVI were consequently excluded from the final OLS model. The reduced model using NDBI, BSI, and NDWI explains 54.3% of LST variability in summer and 57.8% in winter. NDBaI was excluded from the regression models because it is computed from the thermal band that is also used to retrieve LST, whereas BSI uses reflective bands only. NDBI is the dominant predictor in both seasons, followed by BSI as a significant secondary predictor, confirming built-up surface intensity and bare soil exposure as the primary thermal drivers. In contrast, GWR substantially improved model fit (Summer: R² = 0.895, Winter: R² = 0.918) and reduced the residual spatial autocorrelation (Moran’s I from 0.404 to 0.028 in summer and from 0.299 to 0.033 in winter), showing its effectiveness in capturing spatial heterogeneity. LST patterns differed markedly between the two acquisition dates in association with LULC differences, confirming the utility of spatially adaptive modeling for climate-responsive planning. The findings provide evidence for the scientific basis of sustainable urban management and climate adaptation strategies, contributing to the implementation of Sustainable Development Goal 13 (Climate Action).

Hourly VOC observations collected in Shannan from September 2024 to September 2025 were analyzed using maximum incremental reactivity (MIR), positive matrix factorization (PMF), and backward trajectories. During the year-long analysis period, monthly median TVOC concentrations were lowest in May (52.65 µg m⁻³) and highest in October (208.52 µg m⁻³), whereas median ozone formation potential (OFP) was lowest in May (81.87 µg O₃ m⁻³) and highest in January (231.46 µg O₃ m⁻³). Isopentane had the highest median concentration, and acetaldehyde the highest species-level OFP. PMF of 8189 valid hours and 42 species resolved four factors accounting for 28.8%, 19.8%, 23.3%, and 28.1% of summed reconstructed mixing-ratio contributions. The factors were interpreted as traffic-related combustion (F1), mixed solvent-use and fuel emissions (F2), a halocarbon- and light-alkane-rich mixed factor (F3), and a fuel-related and OVOC-rich mixed factor (F4). During the study period, 7630 of 7949 PMF-valid hours had an assigned trajectory class; 319 remained unclassified. W trajectories comprised 356 h, 235 in winter. Season-adjusted W comparisons with pooled SSW and W/WNW hours showed positive conditional associations for F1 and F4, but not stable positive enhancement for F2 or F3. The 37-species PMF-reconstructed OFP showed weak in-sample agreement with observations (R² = 0.0283), limiting source-specific reactivity interpretation; the trajectory associations do not quantify regional emissions.

Wind energy forecasting in complex terrain is a critical challenge for reliable power grid integration. This PRISMA 2020-guided structured review examines 134 eligible studies (2018–2026) that apply deep learning (DL) and other neural-network methods to wind speed forecasting, wind power prediction, and wind field downscaling, with particular emphasis on complex terrain. Four methodological pillars are analyzed: automated deep learning (AutoDL/NAS), multi-source data fusion, uncertainty quantification with emphasis on conformal prediction, and explainable AI (XAI). Only 24 studies (18%) explicitly evaluate mountainous, valley, coastal, or urban settings. CNN-based models are the most frequent architecture family, whereas AutoDL/NAS and automated hyperparameter optimization appear in only 11 studies, and no NAS framework has been evaluated in complex terrain. Reported fusion gains are study-specific (from about 13% RMSE reduction to over 50% MAE reduction) and are not directly comparable across datasets, metrics, and baselines. Conformal prediction offers finite-sample coverage guarantees under exchangeability, yet its empirical reliability under terrain-induced non-stationarity has not been evaluated. XAI appears in only five DL studies and is mostly applied post hoc. No study combines three or more pillars. We identify ten challenges and propose ten future directions toward integrated, terrain-aware, uncertainty-calibrated, and interpretable forecasting frameworks.