Remote Sensing Applications-Society and Environment最新文献

{"title":"What's in a coral pixel? Spatial variability in co-occurrence patterns for coral reef benthic communities","authors":"Dylan Cowley,&nbsp;Faye F. Diederiks,&nbsp;Kirsten M. Golding,&nbsp;David E. Carrasco-Rivera,&nbsp;Chris M. Roelfsema","doi":"10.1016/j.rsase.2026.102210","DOIUrl":"10.1016/j.rsase.2026.102210","url":null,"abstract":"<div><div>The structural complexity and biodiversity of coral reefs provide vital ecosystem services to coastal communities, dependent on benthic community distribution. Earth Observation presents well developed tools that have been used for mapping and monitoring coral reef systems for decades, but the level of detail achieved is highly dependent on the spatial distribution of the benthic features of interest. The aim of this study was to better understand the spatial variability of reef habitats and the co-occurrence of key benthic features using Earth Observation imagery across broad spatial scales. We utilised over 100,000 geolocated benthic photoquadrats collected across multiple decades on Heron Reef, southern Great Barrier Reef, Australia. From these, high spatial resolution (&lt;1 m) in situ benthic cover data was derived to identify distinct assemblages of co-occurring benthic features across geomorphic zones. Benthic feature co-occurrence was assessed across spatial scales derived from satellite sensors used for coral reef habitat mapping, with pixel sizes between 1 and 300 m, to identify which features typically co-occur within satellite-derived pixels. We identified an optimal pixel size of &lt;20 m, where larger pixel sizes preclude separation of key benthic features defining regional-scale thematic classes (e.g., live coral, sand). Across all pixel sizes, co-occurrence with sand is lowest on the reef slope, as expected, while hard coral and turf/encrusting algae co-occur most frequently across all zones. Using Heron Reef as a case study, we highlight the variability expected within satellite derived image pixels, supporting both reef composition knowledge and informing Earth Observation imagery detection capabilities.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102210"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148770897","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"HF-YOLO: An improved object detection algorithm for typical targets in river and lake shoreline areas","authors":"Juncheng Bao,&nbsp;Zhen Zhang,&nbsp;George P. Petropoulos,&nbsp;Zihan Fang,&nbsp;Jiawang Shen,&nbsp;Li Jing","doi":"10.1016/j.rsase.2026.102168","DOIUrl":"10.1016/j.rsase.2026.102168","url":null,"abstract":"<div><div>Effective regulation of river and lake shoreline areas is vital for ecological and economic balance. However, conventional monitoring methods remain constrained by weather and efficiency limitations. Object detection technology enables automated, all-weather target identification, but models specifically designed for shoreline scenarios are still lacking. To address this gap, this paper proposes HF-YOLO, a specialized object detection model for river and lake shoreline scenarios. Built upon the YOLO11 framework, HF-YOLO integrates a Hybrid Attention Transformer (HAT) module for high-frequency enhancement of small targets, and a Frequency-Aware Linear Attention (FALA) module that mitigates the high-frequency suppression inherent in linear attention through focused kernel mapping, wavelet-based local enhancement, and multi-frequency position encoding, thereby preserving the fine-grained edge and texture features critical for small target detection. Furthermore, an Oriented Bounding Box (OBB) detection head is employed to improve localization for multi-oriented targets. This research also involves the construction of the RSI dataset, a UAV remote-sensing dataset tailored for river and lake shoreline regulation. Experimental results on the RSI dataset demonstrate that the HF-YOLO model achieves an mAP50 of 85.7%, representing a 12.5% improvement over the YOLO11s baseline and a 4.5% increase over the YOLO11s-OBB model. This research provides a technical solution for the object detection challenges in the management of shoreline areas.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102168"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148635665","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"GNSS-supervised machine learning framework for quality assessment of InSAR ground motion via spatial residual decomposition","authors":"Thalosang Tshireletso,&nbsp;Meghdad Bagheri,&nbsp;Seyed Ali Ghorashi","doi":"10.1016/j.rsase.2026.102199","DOIUrl":"10.1016/j.rsase.2026.102199","url":null,"abstract":"<div><div>Continental-scale InSAR products such as the European Ground Motion Service (EGMS) provide unprecedented spatial coverage of ground deformation, yet measurement reliability varies substantially across terrain types, coherence regimes, and scatterer densities. Distinguishing reliable from unstable measurements at the point level therefore remains a central challenge for operational deformation monitoring. This study introduces a GNSS-supervised machine learning framework, Spatial Confidence Residual Deformation Analysis with Machine Learning (SCRDA-ML), designed to quantify and classify InSAR measurement quality through physically interpretable spatial residual modelling. Local deformation behaviour is characterised within 10 km moving windows using planar, quadratic, and cubic polynomial surface fitting. Residual magnitude, local variance, and normalised confidence metrics are derived for each scatterer, forming a nine-dimensional spatial feature set. At tile scale (100 km <span><math><mo>×</mo></math></span> 100 km), density-based clustering (HDBSCAN) is applied to characterise deformation regime structure prior to supervised classification. A total of 1005 GNSS–InSAR matched samples from tile E33N34 were obtained using a 10 km co-location threshold and partitioned, via a stratified 80–10–10 split (804 training, 100 validation, 101 testing). GNSS observations were used exclusively for bias-based quality labelling rather than regression modelling; because only one GNSS station satisfied the co-location constraint, velocity correlation is inherently limited despite clear separation of bias classes. Three classifiers, XGBoost, multilayer perceptron (MLP), and a Bayesian neural network (SCRDA-Bayes), were evaluated. On the independent test subset, XGBoost achieved perfect class separation (macro F1 = 1.00; ROC-AUC <span><math><mo>&gt;</mo></math></span> 0.98), reflecting strong discriminative power of the SCRDA-derived spatial features. During cross-site inference on two unseen tiles, SCRDA-Bayes demonstrated superior spatial robustness and smoother classification boundaries under low-coherence conditions. Feature importance analysis identified quadratic-model confidence and planar residual metrics as the dominant predictors of scatterer reliability.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102199"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148731706","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Flood susceptibility and mitigation in tropical cities using XAI and ‘Knowledge-infused transformer-based neural network’ model","authors":"Malay Pramanik,&nbsp;Swapan Talukdar,&nbsp;Harekrishna Manna,&nbsp;Rizwan Ahamed,&nbsp;Alok Das","doi":"10.1016/j.rsase.2026.102180","DOIUrl":"10.1016/j.rsase.2026.102180","url":null,"abstract":"<div><div>Floods increasingly threaten urban deltas like Bangkok due to rapid urbanization, low elevation, and climate variability, yet traditional flood susceptibility mapping (FSM) methods often struggle with scalability, interpretability, and robustness in complex urban environments. To address this limitation, the study introduced a novel Knowledge-Infused Transformer-based Neural Network (KITNN) integrated with Explainable AI (XAI) techniques. Fifteen flood-conditioning factors, including topography, Land use land cover (LULC), rainfall, soil, vegetation indices etc. Were used to calibrate and validate the model. KITNN was compared with Random Forest (RF), Quadratic Discriminant Analysis (QDA), Naive Bayes (NB), and Histogram-based Gradient Boosting (HistGB). The performance of the proposed model was evaluated using spatial block cross-validation and comparative analysis with existing methods. Results indicate that the KITNN model achieved a mean ROC-AUC of 0.94 under cross-validation, along with a PR-AUC exceeding 0.95, demonstrating strong and consistent predictive capability. A best single-trial validation AUC of 0.9725 was obtained during hyperparameter search; this value reflects performance under an optimal configuration on a single validation split and is reported separately from the cross-validated metrics, which provide the basis for all performance claims. Spatial analysis indicated that 43.71% of Bangkok was classified as high or very high flood susceptibility, consistent with historical flood-prone zones in Min Buri, Nong Chok, and Khlong Sam Wa. SHAP analysis identified LULC as the most influential predictor (mean absolute SHAP &gt;0.11), followed by soil type and NDVI, with curvature showing the highest gradient sensitivity. The model demonstrated minimal overgeneralization, with only 11.42% of the area classified as Very Low risk. These results underscore KITNN's strength in modeling nonlinear geospatial interactions while providing interpretable, stakeholder-relevant insights. The proposed FSM framework is scalable and transferable to other flood-prone megacities, supporting climate-resilient urban planning.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102180"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148731803","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The new paradigm of hydrogeology: A critical review of remote sensing and AI approaches for groundwater assessment and monitoring in Sub-Saharan Africa","authors":"Boitumelo L. Maxala,&nbsp;Timothy Dube","doi":"10.1016/j.rsase.2026.102226","DOIUrl":"10.1016/j.rsase.2026.102226","url":null,"abstract":"<div><div>Faced with growing water scarcity, the delineation of groundwater potential zones (GWPZ) using remote sensing and machine learning techniques has become an indispensable tool. This article critically reviews the evolution and application of these techniques between 2000 and 2025. We synthesize prevailing approaches, document progress, address critical challenges, and identify urgent research opportunities to guide future sustainable groundwater exploration and management. Our review demonstrates that the advent and continuous advancements of remote sensing, providing critical long-term data, have significantly propelled the field of GWPZ. The availability of various multi-spectral sensors and sophisticated models has been instrumental in this progress. Additionally, emerging trends such as machine learning techniques (e.g., Random Forest and Support Vector Machine) and cloud computing platforms enabled effective multi-source data fusion, enhancing predictive accuracy and overcoming limitations of individual sensors. However, significant challenges persist. The scarcity of ground-based data impedes robust model validation, while a lack of high-resolution spatial and temporal data hinders precise and small-scale delineation. This is particularly acute in Sub-Saharan Africa, which remains critically understudied. To address these gaps, we recommend that future research work focus on using spatial cross-validation to achieve accurate results, assess data-scarcity solutions including data augmentation, use advanced approaches such as explainable artificial intelligence, hybrid and ensemble machine learning models to improve delineation precision and integrate these approaches into broader water resource resilience frameworks.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102226"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148995231","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Bridging satellite and UAS scales for surface soil moisture mapping: A two-step random forest downscaling framework","authors":"Ruodan Zhuang,&nbsp;Salvatore Manfreda,&nbsp;Yijian Zeng,&nbsp;Lijie Zhang,&nbsp;Brigitta Szabó,&nbsp;Paolo Nasta,&nbsp;Nunzio Romano,&nbsp;Zhongbo Su","doi":"10.1016/j.rsase.2026.102178","DOIUrl":"10.1016/j.rsase.2026.102178","url":null,"abstract":"<div><div>Quantifying surface soil moisture (SSM) is essential for understanding hydrological cycles, land-atmosphere exchanges, and field-scale water management. However, satellite-based and land surface model products, typically available at kilometer-scale resolution, cannot resolve the sub-field heterogeneity required for small-catchment or precision agriculture applications. This study develops and evaluates a two-step Random Forest (RF) downscaling framework that links a regional 1 km SSM product to a 30 m product and then tests a UAS-supported 16 cm super-resolution mapping experiment.</div><div>The framework was implemented at the Monteforte Cilento experimental sub-catchment in southern Italy. <em>Model I</em> used 1 km SSM and land surface predictors to generate a five-year 30 m SSM product (SSM_30m). <em>Model II</em> used the 30 m product and scale-consistent predictors to test whether Uncrewed Aerial Systems (UAS)-derived 16 cm features, including land surface temperature, diurnal temperature difference, vegetation index, and micro-topographic variables, can further disaggregate the 30 m patterns for a single UAS campaign day.</div><div>Internal model testing yielded R² values of 0.805 for Model I and 0.837 for Model II. For independent field evaluation: Dynamics of SSM_30m closely tracked cosmic-ray neutron sensor references, whereas the single-day 16 cm resolution surface soil moisture map (SSM_16cm) achieved a Pearson correlation coefficient of 0.71 and an unbiased root mean squared error (ubRMSE) of 0.0473 cm³ cm⁻³ against 20 time-domain reflectometry (TDR) measurements. From <em>Model I</em> to <em>Model II</em>, predictor importance shifted from regional LST at coarser scales to diurnal thermal contrast and micro-topography at sub-meter scale. Furthermore, an analyse of spatial variance revealed strong spatial coherence across all resolutions, from 1 km to 16 cm. This research demonstrates a scalable framework from regional monitoring to “single plant” scale water management, providing a vital tool for sustainable environmental and agricultural decision-making.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102178"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148635657","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Machine learning-based inversion of water quality parameters and trophic state evaluation in plateau lakes: Case study of Lake Dianchi (China)","authors":"Hang Zhang,&nbsp;Wenying Hu,&nbsp;Jinliang Wang,&nbsp;Yuanmei Jiao","doi":"10.1016/j.rsase.2026.102177","DOIUrl":"10.1016/j.rsase.2026.102177","url":null,"abstract":"<div><div>Lake water quality is critical to human well-being and sustainable development. However, Lake Dianchi—the largest highland lake in southwestern China—has faced severe eutrophication in recent decades, threatening regional water sustainability. Accurate and high-precision water quality monitoring is therefore essential for its ecological restoration. To address the challenge of a small in-situ dataset (N = 30), this study employed three machine learning models—eXtreme Gradient Boosting (XGBoost), Random Forest-Adaptive Boosting (RF-AdaBoost), and Particle Swarm Optimization-Back Propagation (PSO-BP)—to construct retrieval models for water quality parameters in a typical highland lake, and systematically evaluated their performance. Using Landsat 8 Operational Land Imager (OLI) imagery from April 2013 to April 2023, we retrieved the spatiotemporal distributions of chlorophyll <em>a</em> (Chla), total phosphorus (TP), and total nitrogen (TN), and assessed eutrophication dynamics using the Trophic Level Index (TLI). The results indicate that XGBoost demonstrated superior robustness under the small-sample constraint, outperforming the other models in validation accuracy for Chla, TP, and TN. Leave-one-out cross-validation (LOOCV) validated the model's performance, yielding Coefficient of Determination (R²) values of 0.683, 0.441, and 0.451, alongside Root Mean Square Error (RMSE) values of 0.023 mg/L, 0.023 mg/L, and 0.405 mg/L, and Mean Absolute Percentage Error (MAPE) values of 78.00%, 22.33%, and 16.92% for Chla, TP, and TN, respectively, confirming its strong generalization stability under strict small-sample constraints. Spatially, water quality followed a clear north-to-south gradient, with higher TLI values in the northern urban-adjacent zone. Temporally, Lake Dianchi transitioned from severe eutrophication in 2013 to light eutrophication by 2023. Collectively, the findings confirm that the XGBoost model is both viable and scalable for small-sample water quality retrieval. This study underscores the viability and transferability of XGBoost under severe data constraints. The proposed approach offers a valuable methodological blueprint and technical pipeline for remote sensing-based water quality surveillance and trophic state diagnostics in highland lakes with limited availability of in-situ data and matching phenological baselines.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102177"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148635730","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dealing with and communicating data uncertainties in agricultural remote sensing: a comprehensive approach","authors":"Sebastian Goihl,&nbsp;Volker Hochschild,&nbsp;Peter Dietrich","doi":"10.1016/j.rsase.2026.102202","DOIUrl":"10.1016/j.rsase.2026.102202","url":null,"abstract":"<div><div>The European Space Agency's (ESA) Copernicus program has provided free access to satellite remote sensing data, making it an attractive option for public institutions. Government institutions are currently evaluating and identifying use cases for satellite imagery. Agricultural administration stands to benefit particularly, with the potential uses being substantial, especially in this area. This is primarily driven by two factors: (1) legal requirements within the framework of the Common Agricultural Policy (CAP), and (2) the substantial proportion of agricultural land within the total land cover in a mid-European study area. In the domain of agriculture and environmental discourse, there are a variety of applications for agricultural remote sensing in the context of providing a database for decision makers within the agricultural administration.</div><div>The administration's activities are connected with the careful handling of the created geodata products, depending on the data requirements and the uncertainties inherent in the products as a result of the creation process. Government institutions have high accuracy requirements for the products they use to make decisions. In this regard, this thesis examines the factors that influence the informative value of agricultural remote sensing products and proposes possible solutions in understanding and communicating data uncertainties. This applied approach aims to increase the social benefits of remote sensing. The objective of this study was to assess the potential and limitations of applications by remote sensing data from the point of view of the agricultural administration. To this end, a total of 53 uncertainty factors in agricultural remote sensing were compiled. A selection of practices for communicating and visualizing uncertainty was created to address these factors.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102202"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148731705","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"S2GAIA: A seasonally aware multi-year dataset for deep learning-based land cover mapping using sentinel-2 imagery in Greece","authors":"Anastasios Temenos,&nbsp;Styliani Verykokou,&nbsp;Eugenia Papatheodorou,&nbsp;Sofia Soile,&nbsp;Charalabos Ioannidis,&nbsp;Anastasios Doulamis,&nbsp;Nikolaos Doulamis,&nbsp;Chryssy Potsiou,&nbsp;Christos Kontopoulos,&nbsp;Pol Kolokousis,&nbsp;Georgios Voskopoulos,&nbsp;Betty Charalabopoulou,&nbsp;Maria Lamprianidou,&nbsp;Konstantinos Karalidis,&nbsp;George Vergos,&nbsp;Christina Karakizi,&nbsp;Dimitris Bliziotis,&nbsp;Konstantinos Karantzalos,&nbsp;Sofia Oliveira,&nbsp;Ariane Muting,&nbsp;Svein-Jarle Lokas","doi":"10.1016/j.rsase.2026.102197","DOIUrl":"10.1016/j.rsase.2026.102197","url":null,"abstract":"<div><div>Accurate, up-to-date, and high-resolution land use/land cover (LULC) maps are essential for environmental monitoring, spatial planning, climate resilience, and sustainable land management. Despite the increasing availability of Sentinel-2 imagery, existing publicly available datasets often lack explicit seasonal representation, consistent multi-year annotations, or thematic detail suitable for heterogeneous Mediterranean landscapes. To address these limitations, this study introduces S2GAIA, a multi-year, seasonally aware Sentinel-2 dataset for pixel-wise LULC mapping in Greece. The dataset is generated by harmonizing multiple geospatial sources, including Copernicus products and national datasets, into a unified 22-class taxonomy. S2GAIA comprises 34,030 annotated Sentinel-2 image patches (256 × 256 pixels) at 10 m spatial resolution, spanning the period 2017–2024. Each year is represented by four cloud-free seasonal composites, enabling the exploitation of phenological variability and improving the discrimination of temporally dynamic land-cover classes. The dataset was evaluated using four semantic segmentation architectures, with ResUNet achieving the highest performance, obtaining a Cohen’s Kappa coefficient of 91.23% and a weighted F1-score of 92.23%. Additional class-wise, confusion, and seasonal analyses demonstrate the effectiveness of the proposed multi-season design while identifying the remaining challenges associated with fragmented and spectrally similar land-cover classes. The resulting national-scale LULC maps demonstrate the practical applicability of S2GAIA for environmental monitoring and provide a valuable resource for developing and evaluating deep learning models for Mediterranean land-cover mapping. The S2GAIA dataset and the associated source code are publicly available through Zenodo (<span><span>https://zenodo.org/records/21883467</span><svg><path></path></svg></span>) and GitHub (<span><span>https://github.com/eugeniapapathe/S2GAIA</span><svg><path></path></svg></span>), respectively.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102197"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148731804","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Within-season crop yield prediction in the Vojvodina Province (Serbia) using Sentinel-2 satellite vegetation indices","authors":"Branislav Živaljević,&nbsp;Samuel Jonson Sutanto,&nbsp;Spyridon Paparrizos,&nbsp;Minučer Mesaroš,&nbsp;Gordan Mimić","doi":"10.1016/j.rsase.2026.102192","DOIUrl":"10.1016/j.rsase.2026.102192","url":null,"abstract":"<div><div>The main objective of this study was to develop and evaluate machine learning-based models capable of providing within-season predictions of crop yield. To achieve this, we developed algorithms based on Sentinel-2 satellite vegetation indices of plant health and moisture content, at the field level in the Vojvodina Autonomous Province (Serbia) for maize, wheat, soybean and sunflower. Yield data collected during several growing seasons (2017–2024) were used to train Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost) algorithms and to develop regression models to predict yields at least two months prior to harvest. Model performance was assessed using monthly mean 3-month and 2-month datasets and statistical metrics R², RMSE, MAE, rRMSE and Bias. RF consistently achieved the best predictive performance across all crops, except for soybean, where the results of XGBoost matched RF’s performance (R²=0.90, MAE = 0.26 t/ha) while other models also showed strong performance (R² <span><math><mrow><mo>≥</mo><mn>0.86</mn></mrow></math></span>). For maize, RF and XGBoost performed similarly well (R² up to 0.82, MAE to 0.93 t/ha), whereas SVM showed lower accuracy. Wheat was predicted with moderate performance using RF (R²=0.73, RMSE = 0.4 t/ha) and sunflower exhibited the best predictability also with RF (R²=0.62). SHAP feature analysis showed that mid-season variables consistently provided the greatest contribution to yield prediction across all crops. Differences between the 2-month and 3-month windows are minor, conforming to the reliability of within-season predictions, with RF providing the most robust and stable results overall.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102192"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148731801","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}