Search Results (169)

This paper presents a multi-parameter characterisation of the atmospheric composition at the RADO-Bucharest observatory, a regional WMO-GAW and ACTRIS facility in southeastern Europe, by anchoring recent observations within a multi-annual baseline (2015–2024). This paper utilises data from multi-wavelength active remote sensing, high-resolution near-surface speciation, and modelling to evaluate complex urban and transboundary processes across different aerosol and clouds regimes. The year 2025 was marked by an atmospheric perturbation, which was characterised by a quantifiable departure from the decadal climatology. Using Z-score analysis, this perturbation was identified as a seasonal decoupling: an atypical reversal in the vertical particle-size distribution occurred in the free troposphere between July and October, while a transition toward a high-absorption aerosol regime was recorded near the surface during the winter months. These shifts are quantified by a significant drop in the columnar Single Scattering Albedo and a systemic increase in high-troposphere Lidar ratios exceeding 70 ± 12 sr, indicating the presence of advected combustion products aloft. At the surface level, chemical speciation measurements recorded an overall increase in wintertime particulate mass concentrations alongside elevated levels of More-Oxidized Oxygenated Organic Aerosol (MOOOA) compared to previous years, reflecting an intensified accumulation of aged, processed emissions during the January–February period. Furthermore, this paper documents cloud vertical structure and phase occurrence, highlighting a persistent seasonal stratification. The application of unified inversion frameworks is demonstrated through case studies of smoke and mineral dust, applying the Generalized Retrieval of Atmosphere and Surface Properties (GRASP) algorithm to retrieve vertically distributed aerosol microphysics through lidar–photometer integration. Maintained under rigorous quality assurance protocols, these results demonstrate the value of continuous, multi-instrumental profiling to quantify transboundary perturbations and improve regional energy budget representations. Full article

In this work, we present a case study mainly from Raman lidar observations and radiosonde data for a site in the European Arctic to analyse how aerosol grows into a purely water-containing fog. We use a relatively high resolution (85 s, 7.5 m) for the lidar evaluation and use an inversion technique to derive aerosol size distributions before fog formation, in the fog and directly after fog disintegration. We find predominantly accumulation-size particles before and after the fog event. In the fog layer, this mode splits up into two branches: one is hydrophobic, and the other roughly doubles its size within 85 s but does not grow into typical cloud droplet sizes. We find very high lidar ratios > 100 sr for 532 nm. We use Mie calculations to confirm that such high lidar ratios are to be expected for micron-sized droplets if the imaginary part of the refractive index is not negligible. Further, we analyse how lidar ratio and colour ratio change as a function of cloud optical depth and derive that the formation and disintegration of several short-lived fog events at the same altitude seem to be symmetrical. Finally, we employ Koehler theory to show that the droplets should have grown into super-micron sizes. As this has not been observed clearly, the time for particle growth into cloud regime under low turbulence conditions takes longer than 85 s. Full article

Reliable prediction of the Indian Summer Monsoon (ISM) requires accurate representation of large-scale circulation features such as the Low-Level Jet (LLJ) and Tropical Easterly Jet (TEJ), yet sparse observational coverage over the Indian Ocean region continues to limit model validation and improvement. This study validated horizontal line-of-sight (HLOS) winds from the ESA Aeolus satellite, carrying the first spaceborne Doppler wind lidar (ALADIN), against high-resolution radiosonde observations at Gadanki (13.5°N, 79.2°E) during 2019–2021, with spatial intercomparisons against reanalysis datasets extended through 2022. Observation days were classified into clear-sky and cloudy-sky conditions using infrared brightness temperature to assess Aeolus retrievals under varying cloud regimes. Rayleigh-clear retrievals showed strong agreement with radiosondes (correlation coefficient = 0.95, bias = 0.11 m s⁻¹), while Mie-cloudy retrievals performed notably weaker (correlation coefficient = 0.35, bias = 3.40 m s⁻¹). Agreement improved with altitude, with the highest correlation (0.97) observed in the upper troposphere–lower stratosphere (UTLS). HLOS wind differences between Aeolus and radiosondes generally remained within ±2 m s⁻¹. The vertical structure and intensity of the LLJ and TEJ derived from Aeolus agreed most closely with radiosonde observations, followed by ERA5, MERRA-2, and NCEP-2 reanalyses, in order of increasing deviation. Spatial deviations were small relative to ERA5 and MERRA-2 but substantially larger relative to NCEP-2. These findings demonstrate that Aeolus provides reliable HLOS wind measurements for characterizing the vertical structure and seasonal evolution of Indian Summer Monsoon circulation, while supporting the evaluation of atmospheric reanalysis datasets over observationally sparse regions. Full article

Soil erosion is recognised as one of the most pervasive forms of land degradation across the Western Balkans (WB), a 208,052 km² region for which its erodible substrates, contrasting climates, and long history of land-use pressure produce highly variable erosion regimes. This study delivers the first regional-scale assessment of soil erosion susceptibility for the entire WB through a cloud-based implementation of a modified Erosion Potential Method (EPM) in Google Earth Engine (GEE). The computation was performed at 30 m resolution and aggregated across 9524 catchments derived from the EU-Hydro database, with the dimensionless erosion coefficient (Z) serving as the primary susceptibility metric. The modelled patterns reveal a strong north–south asymmetry in erosion intensity: The highest country-level coefficients were obtained for Albania (0.47), North Macedonia (0.38) and Montenegro (0.37), while Serbia (0.29) and Bosnia and Herzegovina (0.28) recorded lower regional means. About 19.8% of the WB area falls within catchments classified as highly to very highly susceptible, with such basins concentrated along the Adriatic-facing mountain front and across the Aegean headwaters. Validation was carried out at three complementary levels, including point-based statistical validation on 1379 field- and orthophoto-verified points distributed across the WB, which yielded an AUC = 0.828. Although the proposed framework should not be considered a substitute for fine-scale, field-based modelling, it provides a transparent, reproducible, and scalable basis for regional screening, catchment prioritisation, and the design of soil-conservation and land-management interventions across the WB. Full article

Riming and aggregation are critical ice-phase microphysical processes in winter clouds, but their overlapping signatures and dynamic transitions pose challenges for conventional single-frequency radar detection. We introduce a novel gradient-based identification method using ground-based triple-frequency dual-polarization radar observations. By analyzing vertical gradients of triple-frequency radar variables, rather than their absolute values, we discern these microphysical processes through physically based thresholds that reflect particle growth regimes. This approach captures subtle spatiotemporal variations in riming and aggregation that conventional threshold methods would miss, particularly in resolving layered riming-aggregation transitions. The dynamic gradient-based method demonstrates enhanced physical consistency and adaptability near process boundaries, thereby improving the tracking of ice-particle evolution. These advances provide a pathway to refine microphysical parameterizations and enhance high-resolution snowfall forecasting. Full article

The transformation of a diffuse molecular cloud into a star necessarily increases the entropy of the universe, chiefly through the radiation emitted as gravitational binding energy is released. We present a compact, fully closed-form thermodynamic model of this process: the Sackur–Tetrode equation gives the entropy of the initial cloud and, generously, of the stellar material itself, while the released gravitational potential energy is converted into a radiation-entropy term $S_{\mathrm{rad}}=\Delta E_{\mathrm{pot}}/\left(2T_{\mathrm{eff}}\right)$, the factor of one-half following from the virial theorem for a self-gravitating star in hydrostatic equilibrium. For a solar-type star we obtain $\Delta S\simeq 1.9\times {10}^{37}\mathrm{J}{\mathrm{K}}^{-1}$, consistent with independent literature estimates of stellar and interstellar entropy. Extending the calculation across the main sequence (O through M) gives $\Delta S\propto M^{0.71}$, rising from $1.2\times {10}^{37}\mathrm{J}{\mathrm{K}}^{-1}$ for a $0.3M_{\odot }$ M dwarf to $2.0\times {10}^{38}\mathrm{J}{\mathrm{K}}^{-1}$ for a $20M_{\odot }$ O star. We then map the full $(M,R,T_{\mathrm{eff}})$ parameter space to locate the locus of $\Delta S=0$—the formal boundary of thermodynamic feasibility for a single monolithic collapse—and show that every real main-sequence star lies deep in the entropy-producing region, with the boundary itself displaced to radii and masses far outside the stellar regime. Applying the same closed-form model to representative red giants, supergiants, white dwarfs and neutron stars (not as a model of their true formation, but as a diagnostic of how compactness controls radiative entropy production) shows that $\Delta S$ is set primarily by the compactness $G{M}^2/\left(R{T}_{\mathrm{eff}}\right)$ of the final configuration, so that degenerate remnants—if they were assembled by a single collapse from a diffuse cloud—would be substantially larger entropy sources than main-sequence stars, while extended giants are comparatively modest ones. The same closed-form machinery gives direct access to a full thermodynamic feasibility map, something that would otherwise require a large grid of numerical simulations to reconstruct, and we compare our results throughout with the current literature on stellar and cosmic entropy rather than with ad hoc benchmarks. Full article

Accurate mapping of rubber plantations is essential for sustainable land management and forest conservation in tropical regions. However, existing methods face two major challenges: persistent cloud cover limits the effectiveness of optical remote sensing in tropical areas, and regional phenological heterogeneity hinders the transferability of fixed-parameter approaches. This study proposes a Phenology-Adaptive Rubber Plantation Mapping (PARM) framework that integrates Sentinel-1 SAR time-series data with optimally selected spectral indices through a cascading constraint architecture. The framework operates as a structurally coherent system wherein SAR-derived phenological anchors explicitly govern downstream optical analysis across three internally dependent modules. First, three key phenological nodes—leaf-off start (LOS), fastest greening point (FGP), and full canopy point (FCP)—are extracted directly from SAR VH-polarization backscatter time series, enabling cloud-independent extraction of phenological temporal anchors. Second, the Jeffries–Matusita (JM) distance, evaluated within SAR-constrained phenological windows, is employed to identify the optimal vegetation and water indices for each region from six candidate spectral indices. Third, a time-weighted Rubber Plantation Discrimination Index (RPDI) is constructed using the selected indices and locally extracted phenological nodes, thereby amplifying the coupled signals of canopy greenness and moisture dynamics during critical phenological transitions. The framework was validated in Hainan Island and Vietnam, two regions with contrasting phenological regimes, using a spatial-block partitioning protocol (leave-one-subregion-out combined with DBSCAN-based clustering) designed to prevent samples from the same plantation from occurring in both training and test subsets. Within the Dynamic World forest mask, PARM achieved overall accuracies of 92.04% and 91.24%, respectively (93.57% and 91.00% on the fully held-out Qionghai City and Gia Lai province subregions), with Kappa coefficients exceeding 0.81 in both regions, consistently outperforming schemes based on raw spectral bands, individual spectral indices or direct multi-source time-series stacking. Error structure analysis revealed that residual classification failures are primarily associated with landscape fragmentation, stand immaturity, and residual cloud contamination, delineating the generalizability boundaries of the framework. These results demonstrate that tightly coupling SAR-based phenological characterization with adaptive optical index selection through a cascading constraint architecture provides a reliable foundation for rubber plantation mapping in cloud-prone tropical regions. Full article

Dense partial-view LiDAR observations are attractive for outdoor perception, but limited overlap and viewpoint sensitivity make odometry and mapping less reliable than with spinning LiDARs. Many recent algorithms for this sensing regime are built as LiDAR-inertial odometry frameworks, whose localization and mapping performance can degrade or fail when the IMU state estimation becomes unstable. This paper presents a LiDAR-only framework for complex outdoor scenes using a factor-graph back-end. After denoising and motion compensation, the point cloud is projected onto a range image for ground, planar, edge, and line extraction. Pose estimation is strengthened by degeneracy-aware feature selection, while loop closing combines scan-based and path-based cues to handle partial-view revisits. Experiments in tunnels, urban roads, residential areas, and other challenging scenes show reduced drift and improved mapping consistency for dense limited-FoV LiDAR data. Full article

Claims of AI sovereignty are increasingly invoked but operational control remains uneven. Claims to control are made through national models, sovereign clouds, data localization mandates, and procurement rules; however, whether such claims translate into demonstrable control over how AI systems are run, inspected, and contested remains poorly understood. This paper introduces credible sovereignty, the gap between declared and demonstrable control in deployment, as a conceptual lens for analyzing AI governance to examine how this gap is opened and closed across infrastructure, data, and model supply chains. Using a PRISMA-guided social-science corpus and machine learning-based BERTopic modeling, validated through topic diversity and topic separation diagnostics and triangulated through close reading, the analysis identifies four governance logics through which sovereignty is contested: data infrastructure and legitimacy frameworks; techno-bloc diplomacy and infrastructure politics; European regulatory sovereignty; and community-driven sovereignty in the Global South. Across these logics, sovereignty is enacted less through national capabilities than through proxy mechanisms—certification regimes, procurement clauses, cloud governance, and deployment architectures—each carrying trade-offs between autonomy, dependence, and accountability. Rereading the corpus through an Antecedents–Decisions–Outcomes lens yields a testable research agenda: antecedents that push actors toward sovereignty seeking; design and governance choices that translate ambition into implementation; and outcomes—resilience, inclusion, accountability—against which sovereign AI programs should be assessed. This paper reframes sovereignty as a layered operational capability rather than a discursive claim and links computational synthesis to a normative construct that applies across jurisdictions and scales. Full article

Monitoring vegetation-index dynamics in tropical regions remains challenging due to persistent cloud contamination, landscape heterogeneity, and the lack of standardised and reproducible analytical workflows. This paper presents an open-source, fully reproducible end-to-end methodology for multitemporal vegetation monitoring using Sentinel-2 Level-2A (L2A) Bottom-of-Atmosphere (BOA) reflectance imagery. The methodology was applied to Military Grid Reference System (MGRS) tile T18NVL in the Colombian Eje Cafetero region (4.43°N–5.43°N, 74.91°W–75.90°W) for the 2017–2025 period. The proposed workflow integrates storage-efficient direct extraction of spectral reflectance from compressed Standard Archive Format for Europe (SAFE) archives using the Geospatial Data Abstraction Library (GDAL) /vsizip/ interface, per-pixel cloud and shadow masking based on the Sentinel-2 Scene Classification Layer (SCL), computation of Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Soil-Adjusted Vegetation Index (SAVI), and Normalized Difference Moisture Index (NDMI) spectral indices, a diagnostic Random Forest experiment based on threshold-labelled spectral classes, non-parametric Mann–Kendall trend analysis with Sen’s slope estimation, and external agreement assessment against European Space Agency (ESA) WorldCover 10 m 2020 and Google Earth Pro reference data. The methodology reduced per-scene I/O time by approximately 98% without additional disk overhead while retaining a median of 57.9% valid pixels under a mean scene cloud fraction of 35.4%. Mann–Kendall analysis detected no statistically significant long-term trend in any of the four vegetation indices. Seasonal NDVI peaks during September–November were consistent with the bimodal regional precipitation regime, supporting temporal coherence in the satellite-derived vegetation-index response. External agreement was low, with an Overall Accuracy (OA) of 20.2% against ESA WorldCover and 19.9% against Google Earth Pro, indicating systematic over-prediction of woody and mixed-canopy vegetation classes. These results show that single-date optical spectral indices are insufficient for reliable thematic separation of shade-grown coffee, secondary forest, and dense forest within heterogeneous tropical landscapes. The complete version-controlled codebase is publicly available to support methodological reproducibility and adaptation across data-scarce tropical regions. Full article

IoT gateway–cloud systems support large-scale sensing, monitoring, and control applications, but must operate under finite buffering, dynamic traffic demands, and service interruptions caused by gateway and cloud failures. These challenges can lead to backlog accumulation, congestion, and degraded service performance, making admission-threshold selection an important reliability-management problem. This paper investigates reliability-aware admission-threshold selection for finite-buffer systems with service interruptions, motivated by IoT gateway–cloud architectures. A finite level-dependent quasi-birth-and-death (LD-QBD) model is developed to jointly capture probabilistic admission control, finite buffering, and gateway–cloud failures. Exact stationary analysis yields a multidimensional performance-characterization framework based on effective service deliverability, congestion-regime probability, saturation probability, and soft normalized headroom. The admission threshold is shown to govern the trade-off between service deliverability and congestion protection under failure-induced backlog dynamics. To address this trade-off, two complementary threshold-selection paradigms are developed: a trade-off-driven weighted optimization approach and a constraint-based feasibility-enforcement approach. Numerical results show that the weighted formulation exhibits a well-defined knee point, whereas the constraint-based method produces reliability-aware threshold adjustments when operational constraints become active. The results further indicate that increasing traffic load or failure intensity generally requires more conservative admission policies. Although motivated by IoT gateway–cloud systems, the proposed framework is applicable to a broader class of finite-buffer service systems with unreliable resources. Full article

Mobile platforms such as high-speed trains and unmanned aerial vehicles (UAVs) experience quasi-periodic variation in link quality as they move through a cellular base-station lattice, so the service rate of their on-board uplink buffer is itself time-periodic. We model this buffer as a periodic $M/M\left(t\right)/1$ queue whose service rate follows from a signal-to-noise-ratio (SNR)-to-rate map and construct an edge-resident controller that exploits this periodic structure for real-time quality-of-service (QoS) control. From a harmonic-balance (Fourier–Galerkin) solution of the periodic regime, the controller derives backlog and tail-probability indicators and uses them to drive admission, redundancy and handover decisions on the device. The method rests on a stability criterion and a quantitative error bound for the spectral truncation, under stated regularity and stability conditions, and is validated against Monte Carlo simulation along a ∼650 km geo-anchored corridor: on the periodic backbone, the solver matches simulation to within about $1.6\%$, and a coefficient-driven admission rule lowers the 99th-percentile delay by about $28\%$ relative to a reactive baseline at high load. On the full map-derived profile with aperiodic coverage gaps, the proposed proactive controller—spectral backbone admission combined with a radio-map look-ahead—attains the lowest mean and tail delay, about $27\%$ and $21\%$ below the reactive baseline and $54\%$ and $42\%$ below uncontrolled DropTail, with buffer overflow cut from $2.2\%$ to $0.1\%$, at a deliberate admitted-load cost (goodput ≈0.84 vs. $0.94$). An operation-count analysis indicates compatibility with sub-$100$ms control deadlines on a Cortex-A55-class system-on-chip. The controller runs on the device itself, without cloud or GPU, and the architecture is realised in a granted patent; end-to-end hardware benchmarking and an extension to non-Poisson traffic are left for future work. Full article

Artificial intelligence architecture is no longer adequately described by model size alone. Dense Transformers remain the reference architecture for language and multimodal reasoning, but production systems increasingly combine conditional computation, retrieval, memory, tools, verifiers, edge-cloud routing, observability and governance. This review makes three engineering claims. First, sparse Mixture-of-Experts models are currently the clearest capacity-scaling pattern, because they decouple total parameters from active per-token computation, although routing imbalance and distributed communication remain hard constraints. Second, state-space, recurrent and linear attention hybrids are best interpreted as attention-budgeting architectures: they reduce KV-cache and long-context costs, but do not yet displace dense attention in every reasoning regime. Third, JEPA-style latent world models change the learning objective from surface-token or pixel prediction to representation prediction, which is strategically important for perception and planning but still not a drop-in replacement for general language interfaces. To make the maturity claims auditable, this review uses a PRISMA-inspired search protocol, an explicit technology readiness rubric, quantitative comparison tables, hardware and memory-bandwidth analysis, deployment and reproducibility categories, and failure cases for RAG and agents. The main conclusion is that the optimal architecture is task- and constraint-dependent: small dense or hybrid models are often preferred for real-time edge inference, RAG and graph memory for changing enterprise knowledge, frontier dense or sparse models for difficult reasoning, and agentic workflows only when tool permissions, rollback, provenance and human oversight are engineered as first-class components. Full article

Portable biosensor hardware can now sustain continuous multimodal physiological acquisition at the edge, yet the analytical layer that converts raw signals into deployment-consistent inference remains the main bottleneck for practical embedded systems. This study addresses that bottleneck by presenting the machine-learning layer of the Real-time Cognitive Grid, the analytical companion to the previously reported hardware architecture, which equips a fixed-wiring biosensor assembly with real-time physiological-state classification through an asymmetric edge–cloud workflow. The proposed framework assigns analytical responsibility across tiers: a locked 17-feature schema comprising 5 EMG features, 6 EEG spectral features, 2 cross-modal features, 2 HRV features, 1 EOG feature, and 1 EEG quality indicator governs window-bounded inference on the Arduino Nano RP2040 Connect with an LDA edge artefact requiring approximately 716 B RAM, whereas the cloud tier supports public-dataset pretraining, hardware-aligned refinement, multimodal fusion, deployment comparison, and feature-importance analysis under the same schema contract. To evaluate analytical consistency across physiological diversity, five public repositories covering stress physiology (WESAD), affective EEG (DEAP), inertial activity recognition (PAMAP2), sEMG gesture decoding (EMG Gestures), and motor-imagery EEG (EEGMMIDB) were evaluated under subject-disjoint GroupKFold (k = 5) protocols. To test whether the same contract survives translation to the physical rig, the hardware branch was evaluated under session-disjoint GroupKFold across five bench-acquired sessions. Unimodal performance was strongest in sEMG- and IMU-dominant tasks, whereas multimodal fusion improved macro-F1 by up to 0.141 over the strongest unimodal baseline in WESAD and by 0.109 in PAMAP2. In the hardware branch, the deployed edge LDA artefact reached 0.9435 macro-F1 with 0.9470 accuracy, while the retained cloud Random Forest reached 0.8792 macro-F1 with 0.8799 accuracy; feature-importance analysis further showed that the final 17-feature branch was dominated by EMG descriptors, with EEG spectral terms contributing secondary support and hardware-exclusive variables remaining weak under the present bench regime. These results show that a compact multimodal sensing assembly can be elevated beyond passive signal capture into an intelligent portable biosensor that performs context-aware interpretation with minimal user intervention, supported by a reproducible analytical workflow that remains coherent across heterogeneous benchmark repositories, hardware-specific refinement, and microcontroller-class deployment, thereby establishing cross-session bench feasibility as a structured basis for future multi-subject wearable validation. Full article

Cloud and quality screening removes approximately 65% of daily TROPOMI tropospheric NO₂ pixels, creating structured data gaps that coincide with meteorological conditions driving pollution extremes. Standard gap-filling methods—kriging, Random Forests and other machine learning methods—act as statistical smoothers that systematically suppress extreme concentrations and ignore the Missing Not At Random (MNAR) character of cloud-induced missingness. Here we present a physically informed framework that treats urban NO₂ as a forced nonlinear dynamical system and reconstructs missing satellite observations through geometric navigation on a shadow manifold rather than statistical interpolation. The framework integrates five components: (i) Multivariate State-Space Reconstruction (MSSR) using multiview embeddings of continuous ground-based NO₂, O₃, and ERA5 meteorology, grounded in Stark’s forced-system embedding theorem; (ii) Short-Time Regime-Conditioned Convergent Cross Mapping (ST-RC-CCM) with a spatial-mismatch negative control for falsifiable causal validation; (iii) Inverse Probability Weighting (IPW) to correct the clear-sky sampling bias; (iv) trajectory-matrix denoising via Singular Spectrum Analysis (SSA) and Robust PCA; (v) topology-inspired fidelity metrics—Manifold Overlap Ratio (MOR) and Dynamic Trend Capture (DTC)—that penalize smoothing artefacts. The physical basis for this coupling is the shared dynamical history of surface and column NO₂: tropospheric NO₂ has a photochemical lifetime of 1–4 h near urban emission sources, comparable to the boundary layer mixing timescale, ensuring that surface and column concentrations are jointly governed by the same emission–photolysis–transport attractor. The planetary boundary layer height (PBLH), solar zenith angle (SZA), and surface O₃—all included as MSSR coordinates—are the dominant physical drivers of the instantaneous surface-to-column scaling, and their joint trajectory in state space constitutes the physically grounded basis for analogue selection. The framework is validated on a synthetic forced Lorenz-96 system, then applied to five European primary cities spanning contrasting regimes (Sofia, Milano, Stuttgart, Kraków, Hamburg) plus five N1 spatial-mismatch control stations (Plovdiv, Genova, Frankfurt, Warszawa, Berlin)—ten urban-background stations across four countries—with structured ablations (A0-A4V-A4K). Across >3600 evaluations, MOR_ext distributions for EDM and non-EDM methods are non-overlapping by a factor exceeding 5× (EDM minimum 0.59 vs. non-EDM maximum 0.10; median non-EDM MOR_ext ≤ 0.05 at every city × mask combination), while EDM achieves MOR_ext up to 0.915 (Milano Po Valley). Under a fair-comparison benchmark that withholds ground-level NO₂ from Random Forest, EDM’s RMSE advantage remains robust at a median of 3.9× (RF_FULL) and increases to 4.2× (RF_METEO), confirming that the performance gap is physical rather than an information artefact. A three-level temporal validation—within-window pseudo-cloud masking, cross-year transfer (full 2022 holdout and DJF 2023/24), and a COVID-19 out-of-distribution test—demonstrates robustness beyond standard train/test splits, with CCM library-length convergence confirmed for 60/60 ablations (p < 0.001) across all ten stations. Spatial-mismatch tests confirm local dynamical specificity at all five primary–control pairs (Δρ = 0.090–0.210), with seasonal modulation driven by orographic and synoptic mechanisms. These results establish manifold-based gap filling as a dynamically informative complement to statistical approaches, particularly in topographically confined, stagnation-prone basins where preserving extreme-event geometry is essential for exposure assessment. Full article