Rujm el-Hiri has long been considered one of the most enigmatic archaeological monuments in the Southern Levant. Variously interpreted as a funerary, ceremonial, or astronomical locale, it has been the centre of multiple studies spanning over more than 50 years. While traditionally viewed as an isolated protohistoric monument, our study reveals it as the most elaborate example of a widespread regional tradition of large, circular basalt stone structures. This study presents a comprehensive regional reassessment of these large circular stone structures in the basalt highlands surrounding Rujm el-Hiri, revealing over 30 previously undocumented examples within a 25 km radius. Utilizing high-resolution satellite imagery, geophysical modelling, and spatial analysis, we document a consistent architectural tradition characterized by concentric and radial basalt walls, often associated with dolmens, tumuli, and field systems. These structures exhibit similarities in design and landscape placement, frequently located near seasonal water sources and integrated within broader agro-pastoral land-use networks. Our findings challenge the view of Rujm el-Hiri as an isolated monument, instead situating it within a wider phenomenon of protohistoric monumental architecture in this region. This expanded dataset provides new perspectives on landscape organization and monumentality in the protohistoric southern Levant. The application of remote sensing techniques proves crucial in overcoming previous survey limitations, revealing a complex and interconnected archaeological landscape hitherto underappreciated.

Urban sustainability assessment is a crucial challenge in achieving balanced economic growth and environmental protection in modern cities. Traditional statistical evaluation methods often overlook spatial heterogeneity and environmental patterns that can be captured from remote sensing imagery. To address this limitation, this study proposes a multi-modal deep learning framework that integrates high-resolution remote sensing data with socioeconomic indicators for comprehensive urban sustainability assessment. Specifically, a Vision Transformer (ViT) is employed to extract fine-grained spatial and environmental representations—such as vegetation coverage, surface temperature, and built-up density—from Sentinel-2 satellite imagery, while a Graph Neural Network (GNN) models the spatial and economic dependencies between cities, enabling cross-modal and inter-city information fusion. The proposed ViT–GNN framework effectively captures both environmental and socioeconomic dynamics to generate a composite sustainability score. Experiments conducted on a dataset covering 120 major Chinese cities from 2019 to 2023 demonstrate that the model achieves an MSE of 0.012, an MAE of 0.071, and an () of 0.931, outperforming existing regression and CNN-based baselines. The results highlight that the model can accurately evaluate urban sustainability levels, providing an interpretable and data-driven tool for policymakers and planners to support sustainable urban development, resource allocation, and green policy formulation.

Abstract Extreme fire spread during the 2018 Camp Fire in northern California was driven by organized long‐range spotting tightly coupled to plume dynamics. Doppler radar and satellite observations reveal distinct regions of ember lofting and downwind fallout within the convective column, forming direct pathways for firebrand transport several (up to 10) kilometers ahead of the main fire front. These firebrands ignited dense clusters of new fires that merged into rapidly advancing lobes, producing abrupt surges in fire growth that exceeded expectations based on surface wind and fuel conditions alone. By integrating radar‐derived plume structure, infrared satellite imagery, and ground‐based ignition reports, this study provides one of the first high‐resolution, multi‐scale depictions of spotting behavior during an extreme wildfire. The reconstructed distribution of spot fire distances shows that most ignition occurred within 1–5 km of the fire front, but numerous events exceeded 5 km, including some at 8–10 km from active plume cores. These distances are notable relative to prior observational studies and highlight the capacity of organized plume structures to transport embers beyond conventional expectations. Spot fires were not random but aligned within coherent fallout zones shaped by plume dynamics and wind shear. These findings demonstrate that long‐range spotting was a dominant mechanism of Camp Fire spread and underscore the limitations of models that omit ember transport and fire–plume feedback. The results also highlight the potential of operational weather radar to identify active lofting and fallout regions in real time, offering a new observational basis for anticipating spotting‐driven acceleration.

Abstract. Widhah SF, Ambo-Rappe R, Adrianto L. 2026. Spatio-temporal distribution of seagrass extent in three zones of Spermonde Archipelago, Indonesia. Biodiversitas 27 (2): d270216. https://doi.org/10.13057/biodiv/d270216. Seagrass meadows are vital blue carbon habitats that support biodiversity, fisheries, and coastal defense, but human activities and environmental changes increasingly threaten them. This research examined how seagrass distribution shifted over time and space across Barrang Lompo, Badi, and Langkai Islands in the Spermonde Archipelago, Indonesia, representing inner, middle, and outer zones. Using supervised classification of Sentinel-2 satellite images from 2015 and 2025, validated by field observations, we identified distinct regional patterns of seagrass loss over a decade. Area measurements included 95% confidence intervals, and all observed declines were confirmed by field data. Over ten years, seagrass cover shrank by 12.53 hectares (30.69%) in Barrang Lompo, 3.91 hectares (56.02%) in Badi, and 12.23 hectares (26.86%) in Langkai. Major losses were noted in Barrang Lompo and Badi due to dense populations, intensive fishing, land-use change, and pollution, while oceanographic factors like strong currents, sediment buildup, and variable water quality mainly drove Langkai’s decline. These spatial patterns demonstrate the combined impacts of human stressors and environmental factors on seagrass decline and underscore the necessity for targeted conservation efforts, water quality improvements, and zone-specific fisheries management to protect blue carbon functions and ecosystem resilience in the Spermonde Archipelago.

Detection of slow-moving landslides using satellite imagery at the Glacier Bay National Park and Preserve

Abstract The increasing volume of presence‐only (PO) data generated by citizen science initiatives has greatly expanded biodiversity databases, but the statistical use of these data in species distribution models (SDMs) remains limited by strong sampling biases and the absence of reliable absence information. Existing approaches based on Poisson point processes, such as Maxent, provide powerful tools, yet rely on predefined features that restrict their flexibility and scalability. We introduce DeepMaxent, a new SDM framework that leverages neural networks to learn a shared, data‐driven feature extractor across multiple species while remaining grounded in the maximum entropy principle of Maxent, enabling efficient learning even on very large datasets with thousands of species. DeepMaxent uses a normalized Poisson likelihood, which models the probability of choosing each site given a species, to estimate species‐specific suitability surfaces directly from PO observations. In other words, the model predicts suitable locations for each species rather than predicting which species occurs at a given site. We evaluate DeepMaxent on two contrasting datasets: the National Centre for Ecological Analysis and Synthesis (NCEAS) benchmark, containing six small case studies designed to evaluate the impact of spatial sampling biases, and the much larger GeoPlant, dataset covering the whole of Europe. Using PO data for calibration and independent presence–absence data for validation, DeepMaxent consistently outperforms Maxent and leading deep learning‐based SDMs. Compared with Maxent, it achieves an area under the ROC curve of 0.768 versus 0.760 on NCEAS, 0.860 versus 0.823 on GeoPlant and enables the use of high‐dimensional data modalities, such as satellite images, for which Maxent is unsuitable. DeepMaxent combines the normalized Poisson formulation of Maxent with the learnable features, shared among species, of deep learning approaches. This results in better performance than either Maxent or previous deep learning methods, and lower compute requirements than single‐species SDMs, while the formulation makes the method compatible with the integration of survey data to further improve sampling bias correction.

The work presented herein proposes DisasterReliefGPT, a multimodal AI system for automation in the areas of crisis communication and post-disaster assessment. The system integrates three tightly coupled components: a vision module called DisasterOCS for structural damage detection in satellite images, a Large Vision–Language Model (LVLM) for enhanced visual understanding and contextual reasoning, and a Large Language Model (LLM) to produce detailed, clear assessment reports. DisasterOCS relies on a ResNet34-based encoder with partial weight sharing and event-specific decoders, coupled with a custom MultiCrossEntropyDiceLoss function for multi-class segmentation on pre- and post-disaster image pairs. On the benchmark xBD dataset, the developed system reaches a high score of 78.8% in identifying F1-damage, making correct identifications of destroyed buildings with 81.3% precision, while undamaged structures are found with a very high value of 90.7%. From a combination of these components, emergency responders can immediately provide reliable and readable assessments of damage that can be used to directly support urgent decision-making.

Abstract. This study presents a comprehensive, multi-year assessment of winter supraglacial lake drainages on the Northeast Greenland Ice Sheet, detailing cascading drainage events, in which drainage of one lake triggers a chain of subsequent drainages that often occur within days, examining links to melt-season conditions, and evaluating their potential impact on ice dynamics. Supraglacial lakes can drain rapidly, delivering meltwater to the ice-sheet bed, increasing basal water pressure, reducing friction, and accelerating ice flow. Such drainage events are well-documented across Greenland during the melt season using optical satellite imagery. Recent studies using satellite and airborne radar data reveal that many supraglacial lakes persist beyond summer and may also drain during winter, potentially affecting ice dynamics in a manner similar to melt-season drainages. Here, we use C-band synthetic aperture radar imagery from Sentinel-1 and RADARSAT Constellation Mission spanning ten consecutive winters (2014/2015–2023/2024), to detect winter lake drainages. We develop a normalisation method to integrate images from varying acquisition geometries, enabling high-temporal-resolution monitoring. Our analysis identifies 90 winter drainage events from 55 unique lakes, exhibiting substantial interannual variability, ranging from a maximum of 18 events in winter 2018/2019 to a minimum of four events in both 2020/2021 and 2021/2022. Drainages occurred most frequently in early winter, with decreasing frequency as winter progressed. Approximately half of the observed drainages were part of 13 cascading events, each involving two to seven lakes over distances up to ∼33 km. Comparisons with preceding melt-season conditions reveal negative correlations between winter drainage frequency and both melt-season intensity and melt-season drainage frequency. Ice velocity analyses over the ten-year period show no sustained seasonal or annual increases attributable to winter drainages, although isolated short-term increases (6–12 d) were observed.

Accurate cloud detection is an important preprocessing step for subsequent remote sensing data processing. Traditional threshold cloud detection methods have a complex process and require a large number of threshold tests. In recent years, deep learning has been widely applied to cloud detection. However, annotating training datasets for deep learning models typically requires substantial human effort and time investment. Consequently, there are few existing manually annotated cloud detection datasets, and MODIS cloud detection datasets are particularly scarce. To overcome this limitation, we proposed a cloud detection method that combines radiative transfer simulations with deep learning. We first produced a simulated cloud detection dataset using a radiative transfer model and some existing remote sensing products, and then proposed a neural network for training the cloud detection model. Compared with other deep learning models for cloud detection, our method has achieved satisfactory results on the simulated dataset overall. Furthermore, we conducted cloud detection experiments on real satellite imagery. For comparative analysis, we trained other deep learning models on a real satellite image dataset and compared their performance with that of models trained on our simulated dataset. The cloud detection results on real satellite images demonstrate that the models trained on the simulated dataset we proposed achieve performance comparable to those trained on real remote sensing datasets. Specifically, for MODIS data, we compared our results with the official MODIS cloud mask product, MOD35. The results indicate that our method achieves lower false detection rates on mixed surfaces of snow and bare land.

Accelerated climate warming in the 21st century has intensified glacier retreat relative to the 20th century, establishing glacier shrinkage as a key indicator of climate change. This has led to the rapid expansion of glacial lakes, particularly in high-mountain regions. These lakes accelerate glacier mass loss by enhancing frontal ablation and promoting ice destabilization, increasing the risk of glacial lake outburst floods (GLOFs). To elucidate the dynamics of the proglacial lake (present at the terminus of the Padam Glacier) and its impacts on glacier behaviour, this study examines the Padam Glacier (lake-terminating) situated in the Padam Valley, Zanskar, Western Himalaya, and compares it with the neighbouring Nateo Nala (land-terminating) Glacier. In this study, satellite images of varying resolutions from the Landsat archive and Sentinel-2B, as well as digital elevation models (DEMs) from the Shuttle Radar Topographic Mission (SRTM) and the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), covering the period from 1993 to 2022, were used. The lake-terminating glacier decreased in area by 11.6%, with a total retreat of 784 ± 51 m during 1993–2022. In comparison, the land-terminating glacier decreased in area by 5.3% and retreated to 324 ± 173 m during the same period. The proglacial lake increased in area from 0.35 to 0.57 km 2 (∼69%) between 1993 and 2022. Furthermore, the estimated surface ice velocity (SIV) of the glaciers has shown a significant reduction. The SIV of the lake-terminating glacier decelerated by 66%, ranging from 48 ± 6 to 16 ± 2 m a −1 between 2001 and 2022. Similarly, the SIV of land-terminating glacier declined by 47%, ranging from 23 ± 6 to 12 ± 2 m a −1 during the same period. In addition, the mass balance results indicate a negative mass balance of −0.22 ± 0.07 and −0.52 ± 0.07 m w.e.a −1 for lake- and land-terminating glaciers, respectively. These findings highlight a distinct pattern in glacier dynamics and mass balance between lake- and land-terminating glaciers in the western Himalaya.

Beach erosion represents a major hazard to coastal regions worldwide, yet the role of fluvial floods remains an understudied component of beach dynamics. Accounting for fluvial flooding is essential, as its neglect introduces biases and uncertainties in coastal zone management, particularly in complex, anthropized environments. We propose an integrated approach to investigate the relationship between fluvial floods and beach erosion along an anthropogenically impacted beach of the Atacama Desert in northern Chile. Bathymetric changes induced by flood events (2015 and 2017) were delimited on the beach formed by mining-related tailings accumulation (1938–1975). Interannual variations in beach erosion rates were quantified using satellite imagery by comparing post-flood conditions with the pre-flood period. In addition, a numerical model was implemented to assess the spatial variability of wave-driven erosion and to identify the main factors controlling the post-flood increase in erosion. Simulated erosion rates based on post-flood bathymetry are consistent with shoreline change rates derived from satellite data. We conclude that anthropogenic beach modification, combined with flood-triggered processes, generated a positive feedback cascade that led to a prolonged erosive phase lasting several years after the flood events.

Populus euphratica is a critical constructive species in arid desert regions, serving as a “natural barrier” for oasis protection. The sustainable management of Populus euphratica forests is directly related to regional ecological security, and the fine identification of sparse Populus euphratica forests is essential for the conservation of natural Populus euphratica forests. Currently, most mapping studies on Populus euphratica distribution focus on the extraction of dense, contiguous Populus euphratica forests, with insufficient attention paid to the identification of sparse Populus euphratica forests. This study utilizes Gaofen-2 (GF-2) satellite imagery as the data source and takes a typical sparse Populus euphratica forests distribution area in the Tarim River Basin as the study site. It systematically evaluates the performance of nine mainstream deep learning models, including U-Net, DeepLabV3+, and SegFormer, in the task of sparse Populus euphratica forests identification. The results indicate that: (1) The false-color sample set, synthesized from near-infrared, red, and green bands, contributes to improved model accuracy. Compared to the true-color (red, green, blue bands) dataset, the average Intersection over Union (IoU) of the nine models shows a relative improvement of approximately 20%. (2) For the sparse Populus euphratica forests identification task based on the false-color dataset, four models—U-Net, U-Net++, MA-Net, and DeepLabV3+—exhibited excellent performance, with IoU exceeding 75%. (3) Using U-Net as the baseline model, this study integrated the max-pooling indices mechanism, atrous spatial pyramid pooling, and residual connection modules to construct a semantic segmentation network tailored for sparse Populus euphratica forests, named Sparse Populus euphratica Segmentation Network (SPS-Net). This model achieved an IoU of 80%, a relative improvement of approximately 6.3% over the baseline model, and demonstrated good stability in large-scale classification tests. The identification scheme for sparse Populus euphratica forests constructed using GF-2 imagery and deep learning models proposed in this study can provide effective technical support for the refined monitoring and protection of natural Populus euphratica forests.

The secure transmission of satellite imagery is essential for contemporary remote sensing, surveillance, and defense applications. Exchanging large, high-resolution datasets over insecure channels presents significant challenges. Conventional cryptographic algorithms are frequently inefficient for image data because of inherent redundancy and pixel correlations. This study introduces an enhanced encryption framework that combines adaptive hash-driven key generation, multi-chaotic synchronization, and reversible Fredkin logic to achieve both high security and computational efficiency. Dynamic encryption keys are generated using the SHA-256 hash of the input image. Unlike some existing encryption methods that prioritize pixels based on spatial frequency and local contrast features, the proposed framework generates dynamic image-dependent keys using a tri-chaotic system combined with SHA-256 hashing. This approach enhances randomness and key sensitivity while improving resistance to statistical and differential attacks in satellite image encryption. . Fredkin reversible logic gates facilitate bit-level swapping, ensuring complete reversibility of the encryption process. Reed–Solomon error correction, a coding technique for detecting and correcting data errors, is incorporated into the pipeline to enhance robustness in satellite communication and enable recovery from transmission errors. Experimental results on grayscale images demonstrate that the proposed scheme approaches ideal entropy (7.999), resists differential attacks (with NPCR ≈ 99.6% indicating the percentage of pixels changing between encrypted images after a single pixel change in the original, and UACI ≈ 33.2% measuring the average intensity change), achieves low pixel correlation, and passes all NIST SP 800-22 randomness tests. The framework offers a large key space exceeding 2²⁵⁶ and achieves real-time encryption at 0.19 seconds per frame using GPU-based parallel processing. These results confirm the scheme's security, efficiency, and robustness for real-time satellite image transmission.

Abstract Fundamental questions remain about where and when convection will occur within African easterly waves. In this study, we aim to better understand the dynamical processes that govern moist convective organization at the meso‐alpha scale in tropical easterly waves using NASA airborne field campaigns and satellite observations. We employ the SAMURAI 3D variational analysis technique in a vortex‐centric approach, integrating the fifth generation ECMWF (ERA5) reanalysis and research aircraft observations from 20 African easterly wave (AEW) cases sampled during the NASA African Monsoon Multidisciplinary Analyses (NAMMA) in 2006 and Convective Processes Experiment‐Cabo Verde (CPEX‐CV) in 2022. Infrared satellite imagery is used to obtain the frequency of occurrence within the wave of clear air, shallow/moderate convection, and deep convection relative to a potential vorticity (PV) centroid location. We identified four clusters of organized deep convection denoted as minimal deep, southern, southwestern, and widespread. Composites of the dynamical fields from the variational analyses show that high PV and relative humidity (RH) at mid‐levels were approximately co‐located with regions of low‐level convergence and more frequent deep convection, particularly for the southwestern and widespread clusters, which had the highest frequencies of deep convection. Waves with a higher frequency of deep convection are characterized by stronger PV and higher RH at mid‐levels compared to waves with a lower frequency of deep convection. The results suggest that improved understanding of the causal relationships between PV and RH and deep convection in easterly waves can lead to future forecast improvements of convective organization and tropical cyclogenesis.

Origin-Destination (OD) flow matrices are critical for urban mobility analysis, supporting traffic forecasting, infrastructure planning, and policy design. Existing methods face two key limitations: (1) reliance on costly auxiliary features (e.g., Points of Interest, socioeconomic statistics) with limited spatial coverage, and (2) fragility to spatial topology changes, where reordering urban regions disrupts the structural coherence of generated flows. We propose Sat2Flow, a structure-aware diffusion framework that generates structurally coherent OD flows using only satellite imagery. Our approach employs a multi-kernel encoder to capture diverse regional interactions and a permutation-aware diffusion process that maintains consistency across regional orderings. Through joint contrastive training linking satellite features with OD patterns and equivariant diffusion training enforcing structural invariance, Sat2Flow ensures topological robustness under arbitrary regional reindexing. Experiments on real-world datasets show that Sat2Flow outperforms physics-based and data-driven baselines in accuracy while preserving flow distributions and spatial structures under index permutations. Sat2Flow offers a globally scalable solution for OD flow generation in data-scarce environments, eliminating region-specific auxiliary data dependencies while maintaining structural robustness for reliable mobility modeling.

Three-dimensional scene reconstruction from sparse-view satellite images is a long-standing and challenging task. While 3D Gaussian Splatting (3DGS) and its variants have recently attracted attention for its high efficiency, existing methods remain unsuitable for satellite images due to incompatibility with rational polynomial coefficient (RPC) models and limited generalization capability. Recent advances in generalizable 3DGS approaches show potential, but they perform poorly on multi-temporal sparse satellite images due to limited geometric constraints, transient objects, and radiometric inconsistencies. To address these limitations, we propose SkySplat, a novel self-supervised framework that integrates the RPC model into the generalizable 3DGS pipeline, enabling more effective use of sparse geometric cues for improved reconstruction. SkySplat relies only on RGB images and radiometric-robust relative height supervision, thereby eliminating the need for ground-truth height maps. Key components include a Cross-Self Consistency Module (CSCM), which mitigates transient object interference via consistency-based masking, and a multi-view consistency aggregation strategy that refines reconstruction results. Compared to per-scene optimization methods, SkySplat achieves an 86 times speedup over EOGS with higher accuracy. It also outperforms generalizable 3DGS baselines, reducing MAE from 13.18 m to 1.80 m on the DFC19 dataset significantly, and demonstrates strong cross-dataset generalization on the MVS3D benchmark.

ABSTRACT Assessing tree leaf chlorophyll (Chl) levels at the woodland scale is useful for indicating ecosystem productivity. Chl-sensitive vegetation indices (VIs) derived from the synoptic, repetitive satellite imagery can be a beneficial supplement to field measurements. The sparse tree distribution among grass in savannah woodlands makes the application of the VIs challenging. Therefore, optimising their application in savannah woodlands is beneficial. In this study, relationships between field-measured Chl concentrations in savannah woodland stands and a selection of Chl-sensitive VIs derived from a 30 m resolution PRISMA (PRecursore IperSpettrale della Missione Applicativa) hyperspectral image were examined, to identify suitable VIs. To reduce the error from the tree canopy background Chl content, an autumn image of the study area was used, when the grass was senescent but the tree leaves still had some Chl content. Within a week prior to the image date, tree leaf Chl concentrations were measured in-situ using a Chl meter, in 30 m × 30 m sampling sites at widely distributed stands. Due to the numerous Chl-sensitive VIs that have been proposed, only 10 were selected for the analysis. The 10 represented VIs employing visible (Vis), red edge (RE), and near-infrared (NIR) spectral ranges, which are the most common in Chl-sensitive VIs. At the sampling sites, the values of the respective VIs were correlated with the measured tree leaf Chl concentrations. VIs employing the reflectance contrast between the NIR and a Vis Chl absorption or reflectance feature yielded higher, statistically significant correlations (highest: Green Normalized Difference Vegetation Index, r = 0.642, p < 0.001). RE-employing VIs were only second best (Vogelmann red edge index 1: r = 0.556, p < 0.01). It is deduced that, due to the sparse distribution of savannah trees, NIR-Chl feature-contrasting VIs are more suitable for studying tree leaf Chl levels.

Land degradation poses a significant threat to environmental sustainability, agricultural productivity, and socio-economic stability, particularly in cities and villages where mining activities are very rampant. This study aims to assess land degradation vulnerability in Atakunmosa Area, Osun State using Geospatial techniques. By leveraging multi-temporal satellite imagery and spatial analysis techniques, the study identifies and maps degradation-prone areas based on a combination of biophysical and anthropogenic indicators, including land use/land cover (LULC), slope, modified soil adjusted vegetation indices (MSAVI), topographic wetness index (TWI), geology, soil types, soil acidity, soil texture, soil depth and drainage density. The SRTM DEM, Landsat 8 OLI, soil map, and geology map were used to create these thematic maps. Each class of these parameters was given appropriate weighting factors. Using the analytical hierarchical process, weighting factors were assigned to the different themes according to their influence to land degradation. GIS tools were utilized to integrate these spatial layers using weighted overlay analysis to produce the Land Degradation Vulnerability Index (LDVI) for the study area. The results indicated that 36.6 % (297.19 km&amp;lt;sup&amp;gt;2&amp;lt;/sup&amp;gt;) of the total area was prone to high degradation risks, 17.1% (138.85 km&amp;lt;sup&amp;gt;2&amp;lt;/sup&amp;gt;) was prone to moderate risks, 14.2% (115.30 km&amp;lt;sup&amp;gt;2&amp;lt;/sup&amp;gt;) was prone to low risks, while 32.1% (260.65 km&amp;lt;sup&amp;gt;2&amp;lt;/sup&amp;gt;) was prone to very low risks. The consistency ratio (CR) for the study was less than 0.1, which is an indication of the acceptability of the pairwise comparism. The study highlights the critical role of geospatial technologies in environmental assessment and decision-making processes.

Accurate and early prediction of crop yield at the sub-field scale is essential for precision-agriculture and food-system planning. This study evaluates a phenology-based machine learning framework for winter wheat yield prediction using Sentinel-2 satellite imagery, climate reanalysis data, and field-level yield data. Phenological metrics derived from the normalised difference vegetation index (NDVI), the normalised difference water index (NDWI), and the normalised difference red-edge index (NDRE) were combined with accumulated seasonal rainfall and seasonal potential evapotranspiration, and multiple modelling strategies were assessed using a leave-one-field-out cross-validation (LOFO CV) scheme to ensure spatial generalisation. Among the evaluated models, the Random Forest (RF) algorithm achieved the highest overall performance, explaining up to 73% of the yield variability with a root mean square error (RMSE) of 0.88 t ha−1 at optimal prediction timing (day of year 160–175). Integrating phenological and climatic covariates consistently improved prediction accuracy compared to models based only on phenological variables, while the inclusion of soil properties provided limited additional benefit at the examined spatial scale. Phenological metrics based on red-edge data, particularly the maximum NDRE, were the most influential predictors, highlighting the added value of red-edge spectral information beyond traditional red–near-infrared indices. Uncertainty analysis revealed spatially heterogeneous prediction uncertainty, particularly near field boundaries and in areas of complex spatial patterns. Overall, the proposed framework enables robust, early, and interpretable yield prediction at the sub-field scale, supporting uncertainty-aware decision-making in precision agriculture and offering a scalable foundation for regional crop monitoring.

We investigated whether living in greener neighborhoods in midlife is associated with slower cognitive decline in later life. We used data on 2,881 participants from the population-based Multi-Ethnic Study of Atherosclerosis. Geocoded residential addresses (1980–2009) were used to derive midlife neighborhood greenness exposure defined as a 10-year mean of annual normalized difference vegetation index values (based on satellite imagery) during the midlife period (ages 45–54). Cognitive testing over ~ 10 years, when the participants were ≥ 55-year-olds, captured global cognition and processing speed. Multivariable linear mixed effects regression estimated associations between the 10-year midlife greenness measure and global cognition and processing speed z-scores in later life and whether greenness-cognition associations varied by age at first cognitive visit. Greater midlife greenness was associated with slower annual decline in processing speed in the overall sample. We found no differences in associations by age at first cognitive visit. In an ethnoracially and geographically diverse US cohort, living in greener neighborhoods in midlife was associated with slower cognitive decline (i.e., processing speed) in later life.