Remote Sensing Applications-Society and Environment最新文献

{"title":"Fine-scale classification of sand-fixing vegetation using UAV multispectral imagery and deep learning models","authors":"Yaru Zhang,&nbsp;Junfang Wang,&nbsp;Jie Meng,&nbsp;Haodong Ding,&nbsp;Wei Deng,&nbsp;Duanyang Xu","doi":"10.1016/j.rsase.2026.102212","DOIUrl":"10.1016/j.rsase.2026.102212","url":null,"abstract":"<div><div>Precise classification of sand-fixing vegetation is essential for desertification control and ecological monitoring. However, fine-scale vegetation classification in desert environments remains challenging due to the high spectral similarity among species, complex sandy backgrounds, and substantial variations in canopy morphology. In this study, typical sand-fixing vegetation in the eastern sandy area of Ongniud Banner, Inner Mongolia, was investigated using centimeter-level UAV imagery. A high-quality pixel-level labeled dataset was constructed through manual delineation of vegetation canopies. A total of 2181 spatially independent paired UAV image patches and their corresponding pixel-wise label patches were generated, and data augmentation was applied to improve dataset diversity for model training and evaluation. Based on this dataset, a lightweight semantic segmentation network, RPB-UNet, was proposed for fine-scale classification of seven land-cover categories. The model incorporates an RPBlock with an embedded SE attention mechanism to enhance multi-scale feature representation, improve vegetation feature localization, and suppress background interference. Extensive experiments, including ablation studies and comparisons with representative semantic segmentation models, were conducted to evaluate the effectiveness of the proposed framework. The results demonstrated that RPB-UNet achieves competitive classification performance, with an OA of 88.61%, an mIoU of 73.49%, and a weighted F1 score of 88.86%, while requiring only 1.06 M parameters. The proposed model effectively captures fine canopy structures and multi-scale spatial–spectral information, improving the discrimination of sand-fixing vegetation species with similar spectral responses and spatial characteristics while maintaining computational efficiency. Overall, RPB-UNet provides an accurate and computationally efficient solution for UAV-based fine-scale sand-fixing vegetation mapping, with potential applications in desert ecosystem monitoring and restoration assessment.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102212"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148770895","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":"Medicinal plant discrimination in Paramos and cloud forests of northern Peru: A study of deep learning techniques and a novel approach for data augmentation","authors":"W. Castro,&nbsp;M. Juarez,&nbsp;D. Gálvez-Paucar,&nbsp;V. Tirado,&nbsp;J. Gonzáles,&nbsp;F. Torres-Guevara","doi":"10.1016/j.rsase.2026.102204","DOIUrl":"10.1016/j.rsase.2026.102204","url":null,"abstract":"<div><h3>Introduction:</h3><div>Medicinal plants in remote mountain ecosystems are threatened, requiring accurate identification for conservation and sustainable management. However, traditional methods require specialized expertise, and ground surveys in Peru’s Paramos and cloud forests are logistically challenging. This study develops and validates a methodology for discriminating six medicinal plant species using RGB images acquired from UAV and deep learning.</div></div><div><h3>Methodology:</h3><div>Individual plants from six species of medicinal plants were identified and georeferenced; from these, aerial RGB images were acquired at 3 to 5 m altitude. the rectangular ROIs of each image were extracted using the Segment Anything model (SAM), which constituted our image database. Then the image database was augmented through two methods (a) Geometric transformation and (b) a proposal called “Cartesian subsampling” Finally, five CNN architectures for classifying tasks were trained, validated, and their results statistically compared.</div></div><div><h3>Results:</h3><div>Geometric transformation yielded higher overall accuracy, with MobileNetV2 achieving 97.86% and ResNet-50 97.66%, while models trained from scratch performed lower (AlexNet-Deep: 90.14%, PhytoNet: 81.11%). Cartesian subsampling showed advantages when geometric invariance was critical for transfer-learning models. Confusion persisted between species of Myrcianthes that were morphologically similar, with F1 scores ranging from 0.76 to 0.97.</div></div><div><h3>Conclusion:</h3><div>classification precisions from 81% to 97% were achieved, with geometric transformation outperforming Cartesian methods in all models. ResNet-50 offers the best balance of accuracy, computational efficiency, and training stability for practical conservation applications in remote mountain ecosystems.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102204"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148770896","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":"Precision soil property mapping through SAR–optical data synergy and random forest regression kriging","authors":"Partha Deb Roy,&nbsp;Somsubhra Chakraborty,&nbsp;David C. Weindorf,&nbsp;Subhadip Dey","doi":"10.1016/j.rsase.2026.102184","DOIUrl":"10.1016/j.rsase.2026.102184","url":null,"abstract":"<div><div>Accurate mapping of soil moisture (Mv), texture, and soil organic carbon (SOC) is vital for sustainable land management, yet simultaneous prediction of these properties remains limited. In this study, a hybrid regression kriging (RK) modellingapproach combining random forest (RF) for the deterministic component and ordinary kriging (OK) for spatially correlated residuals was applied to predict Mv, sand, silt, and SOC concurrently. A total of 246 soil samples, 128 from fallow and 118 from crop-covered fields, were collected from Paschim Medinipur and South 24 Parganas districts in West Bengal, India. Twenty-two predictor variables, including Sentinel-1 and Sentinel-2 data, synthetic aperture radar (SAR) indices, vegetation descriptors, terrain parameters, and weather variables (maximum and minimum temperature, maximum relative humidity, wind speed, sunshine hours, and evaporation), were used. The relatively small sample size (246 samples with 22 predictors) may lead to overfitting and limit the generalizability of the models. Therefore, separate models developed for fallow, crop-covered, and combined datasets were evaluated using an 80:20 train-test split along with an independent blind validation set. The hybrid (RF + OK)-RK model consistently outperformed standalone RF, reducing root mean squared error (RMSE) by 29% for sand and silt, and by 14% and 12% for Mv and SOC, respectively, in crop-covered areas; by 1.5%, 13%, 8%, and 9% in fallow lands; and by 16%, 16%, 11%, and 14% across the overall dataset. In this study, models trained on 246 samples with 22 predictors may be prone to overfitting and limited generalizability; therefore, model performance was evaluated using an independent blind validation set. The results demonstrated that this hybrid regression kriging approach effectively integrated multisensor, terrain, and weather variables for the concurrent estimation of key soil attributes across diverse land-use conditions.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102184"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148730277","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":"National-scale detection of agricultural landscape elements from sentinel-2 imagery with vision transformers","authors":"Manuel Reese,&nbsp;Björn Waske","doi":"10.1016/j.rsase.2026.102190","DOIUrl":"10.1016/j.rsase.2026.102190","url":null,"abstract":"<div><div>The agricultural landscapes of Germany have undergone significant intensification and land consolidation. Both contribute to the decline of non-productive landscape elements, such as hedgerows, tree rows, field margins, and ditches. Mapping these elements is essential for monitoring policy targets, yet comprehensive cadastral data is lacking. This study investigates whether landscape elements in agricultural areas in Germany can be mapped using freely available multispectral Sentinel-2 imagery. First, we generate a nationwide, least-vegetated image composite for the 2023 and 2024 vegetation periods. For each pixel, we select the observation with the maximum bare soil index from available Sentinel-2 scenes. This maximizes the contrast between permanently vegetated landscape elements and the surrounding arable land. We fine-tuned a transformer-based semantic segmentation model (SegFormer MiT-B4) on this data and compared it with a proposed added refinement head designed to promote the spatial continuity of linear structures. We validate the model on independent data, covering various terrestrial landscapes in Germany. The refinement head improves the model’s overall performance (macro IoU: 0.44 to 0.53), with the largest gains observed for woody elements (IoU: 0.46 to 0.56). However, improvements for herbaceous elements are marginal (IoU: 0.42 to 0.49). Qualitative maps and a 1 km<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span> hexagon aggregation demonstrate the model’s applicability at the national scale and reveal pronounced regional patterns in landscape element density and composition.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102190"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148730281","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":"Mapping of coffee cultivation based on land use/land cover in the tropics of Cajamarca (Peru): An integration of Sentinel data, machine learning, and Google Earth Engine","authors":"Elgar Barboza,&nbsp;Amanda Aragon,&nbsp;Samuel Pizarro,&nbsp;Alcides Roman,&nbsp;Brandon Adams,&nbsp;Ellen Delgado,&nbsp;Alexander Cotrina-Sanchez,&nbsp;Aqil Tariq,&nbsp;Angel J. Medina-Medina,&nbsp;Katerin M. Tuesta-Trauco,&nbsp;Abner S. Rivera-Fernandez,&nbsp;Jhon A. Zabaleta-Santisteban,&nbsp;Candy Ocaña,&nbsp;Marguerite Madden","doi":"10.1016/j.rsase.2026.102200","DOIUrl":"10.1016/j.rsase.2026.102200","url":null,"abstract":"<div><div>Coffee cultivation is a strategic economic activity in the Andean-Amazonian region of Peru, characterized by its diversity of production systems and complex landscape. In this study, a methodological approach based on multitemporal mapping of land cover and land use (LULC) was developed to identify coffee crops by integrating optical satellite imagery (Sentinel-2) and radar data (Sentinel-1), as well as topographic variables, processed on the Google Earth Engine (GEE) platform, with supervised classification based on the Random Forest (RF) algorithm. The study area covered the provinces of Jaen and San Ignacio (Cajamarca), the main coffee-growing areas in the country's north. A time series of satellite data from 2019 to 2024 was used, considering two climatic seasons (dry and wet). The 69 spectral, textural, and topographic variables were evaluated but subsequently reduced to 22 using the Variance Inflation Factor (VIF). The classification distinguished 11 LULC classes, including shade-grown and non-shade-grown coffee, achieving an overall accuracy of over 84% and a Kappa index of 0.90. The binary coffee maps made it possible to quantify and analyze the spatial dynamics of the crop in both provinces, showing a growing trend toward shade-grown coffee, especially in San Ignacio. The proposed approach proved to be operational for mountainous tropical areas with high cloud cover and represents a replicable methodological contribution to agricultural monitoring and sustainable rural landscape management in similar contexts.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102200"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148731704","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":"Interferometric pattern recognition of mangrove dynamics through deep learning multi-source fusion","authors":"Anindita Das Bhattacharjee,&nbsp;Rahul De,&nbsp;Debangshu Pandey,&nbsp;Nilanjan Dey","doi":"10.1016/j.rsase.2026.102181","DOIUrl":"10.1016/j.rsase.2026.102181","url":null,"abstract":"<div><div>A cyclone causes structural damage by breaking the mangrove canopy, uprooting trees, increasing soil moisture, and reducing chlorophyll. No single sensor-driven datum can capture simultaneous structural disruption, hydrological change, and stress on mangroves. In this study, a synthetic aperture radar (SAR)–optical fusion U-Net-based convolutional (U-Net-CNN) fusion that explicitly incorporates interferometric coherence (γ) is used to address these requirements. A U-Net-CNN fusion is applied to synthetic aperture radar (SAR) and Optical data of Henry Island, Sundarbans, India, across the pre- (2016–2018), during (2019–9 2020), and post-cyclone (2021–2023) phases. Two metrics are introduced: the Interferometric Damage Score and the Moisture–Vegetation Damage Score, representing rapid structural change derived from SAR signals and delayed ecological and hydrological responses from optical features, respectively. Model performance is evaluated using Temporal trajectory analysis, half-life estimation for recovery analysis, hotspot agreement (IoU), monotonic consistency, spatial accuracy, and model explainability assessment via the Saliency–Mask Overlap. Pixel-wise fusion, along with DeepLab variants (NEURALPRISM) and transformer-based models (SPECTRAFORMER), is tested for multi-scale feature extraction for heterogeneous mangrove damage pattern detection. Results demonstrate that the proposed approach improves spatial consistency and damage localization, preserves the cyclone track distance–damage relationship under cyclone exposure, captures both instantaneous structural disruption and lagged ecological recovery, and achieves higher spatial accuracy, stronger SAR signal representation, lower cross-ROI variance, and zero collapse (models do not fail to detect structural loss). Under heterogeneous conditions, NEURALPRISM and SPECTRAFORMER exhibit partial collapse in ROI 2 and multiple collapses in mapping nonlinear cyclone disturbances and resilience, respectively. The γ improves the detection of cyclone-driven transitions in mangrove ecosystems. Quantification of spatial accuracy, structural–ecological consistency, and physical coherence of cyclonic damage is validated by measuring detectability (the movement of the mangrove patch centroid during and after the cyclone).</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102181"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148731808","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":"Uncovering land subsidence recovery in the Jakarta metropolitan region, Indonesia, from satellite remote sensing","authors":"Muhammad Yunus,&nbsp;Ruey-Juin Rau,&nbsp;Sung-Ping Chang,&nbsp;Reyhan Azeriansyah,&nbsp;Chih-Heng Lu","doi":"10.1016/j.rsase.2026.102176","DOIUrl":"10.1016/j.rsase.2026.102176","url":null,"abstract":"<div><div>Groundwater regulations have intensified in many subsiding regions, yet their impacts on land deformation remain poorly understood. In the Jakarta Metropolitan Region (JMR), deformation patterns have evolved from persistent subsidence to a mixture of deceleration and localized uplift following groundwater-management interventions. Here, we integrated geodetic, groundwater-level, and extensometer observations to characterize this heterogeneous response. We identified two deformation episodes. During 2007-2010, both coastal and inland areas experienced nearly linear subsidence. Thereafter, coastal sites continued to subside at near-linear rates, whereas inland sites exhibited deceleration, acceleration, and localized uplift. Subsidence rates decreased from approximately −180 to −70 mm/yr. Analysis of groundwater and extensometer data indicates that persistent coastal subsidence sites are mainly driven by delayed consolidation of shallow sediments, reflecting ongoing inelastic deformation. In contrast, inland deformation is associated with groundwater recovery in Aquifers II and III, resulting in predominantly elastic responses and localized rebound. These findings reveal spatially contrasting responses to groundwater management and underscore the need for adaptive strategies that address both reversible and irreversible deformation processes.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102176"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148635662","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":"Crop yield prediction for monitoring within-field variability using MultiRocket and XGBoost on Sentinel-2 time series data","authors":"Maleeha Najam,&nbsp;Hasnat Khurshid,&nbsp;Faisal Akram","doi":"10.1016/j.rsase.2026.102167","DOIUrl":"10.1016/j.rsase.2026.102167","url":null,"abstract":"<div><div>Crop yield estimation supports a range of precision agriculture operations aimed at optimizing resource management. It also plays a crucial role in monitoring the within-field yield variability of crops, which can be stabilized using planning and site-specific management practices. Several machine learning and deep learning models have been developed and applied for crop yield prediction using remote sensing data. Deep learning models, especially convolution-based architectures, frequently struggle with learning kernel weights due to extensive training time on large multivariate satellite data. ROCKET (RandOm Convolutional KErnel Transform) has been applied in various domains for time-series classification tasks, which uses thousands of non-learnable kernels for feature extraction. In this work, we propose a framework based on a variant of ROCKET, referred to as MultiRocket, for field-level corn yield prediction using time-series Sentinel-2 imagery. Features are extracted from a combination of saturation-resistant vegetation indices using the MultiRocket method and are used to train the XGBoost model. The study was carried out on a 22-hectare (ha) corn field in Italy using satellite data collected during the 2017 and 2018 crop growing seasons. The proposed framework showed marginal improvement compared to the two baseline methods, achieving <span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>2</mn></mrow></msup></math></span> values of 0.566 and 0.721, RMSE values of 1.084 ton/ha and 0.972 ton/ha, and MAE values of 0.784 ton/ha and 0.684 ton/ha for the 2017 and 2018 datasets, respectively. Thus, this study presents a proof of concept for the proposed MultiRocket-based framework for crop yield estimation, demonstrating its potential while also indicating the need for further validation on diverse, multi-year agricultural datasets.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102167"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148635664","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":"An optimized deep learning based segmentation framework for land-use mapping from Cartosat-2S satellite images","authors":"Sravya Nedungatt,&nbsp;Shyam Lal,&nbsp;J. Nalini,&nbsp;Chintala Sudhakar Reddy","doi":"10.1016/j.rsase.2026.102182","DOIUrl":"10.1016/j.rsase.2026.102182","url":null,"abstract":"<div><div>As the world’s second-largest road network connecting rural and urban areas and a high-density, heterogeneous building infrastructure, extracting roads and buildings from the Indian region is a challenging task. Although there is extensive research on deep learning (DL)-based object segmentation of satellite images, model performance deteriorates in such complex scenarios. The models trained on datasets that include topographic regions similar to those in India improve building and road extraction results for the Indian region. This paper presents a new land-use mapping dataset, namely Cartosat Land-Use (CartoLU), created from Cartosat-2S satellite imagery of the Indian regions of Bangalore, Bhopal, and Imphal, which represent complex structural and spectral variations, along with their ground-truth data. Moreover, to overcome the limitations of segmentation models in patch-based training and to extract enhanced features by considering the various-sized objects and relevance, a Vision Transformer (ViT) and Convolutional Neural Network (CNN) based model, namely Optimized Transidual Transformer based Segmentation Network (OTTSegNet), is proposed for land-use mapping from Cartosat-2S satellite images. The OTTSegNet mainly consists of an Optimized Transidual block (OTB) for avoiding ambiguities at patch borders in patch-based training and a Lightweight Atrous Convolution ViT (LAC ViT) for incorporating multiscale features along with the extraction of broader dependencies with a smaller number of parameters. The proposed OTTSegNet achieves better results than state-of-the-art (SOTA) models, with mean IoUs of 0.647, 0.656, and 0.281 on the introduced CartoLU and existing BLU and UOS2 datasets, respectively, while maintaining low computational complexity. Thus, this model can be a valid solution for satellite image segmentation and can accelerate sustainable urbanization. This research can be beneficial to India and similar regions for smart city infrastructure planning with the introduction of the CartoLU dataset.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102182"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148730279","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":"Investigating the phenology of two invasive macrophytes across environmental gradients using satellite time series","authors":"Alessandro Q. Scotti,&nbsp;Mariano Bresciani,&nbsp;Federico Filipponi,&nbsp;Claudia Giardino,&nbsp;Monica Pinardi,&nbsp;Paolo Villa","doi":"10.1016/j.rsase.2026.102183","DOIUrl":"10.1016/j.rsase.2026.102183","url":null,"abstract":"<div><div>Remote sensing data enables the accurate tracking of short- and long-term vegetation changes and provides valuable information for assessing plant phenology at different scales. However, investigating plant phenology using satellites and relating it to the environmental context still presents challenges in natural, complex ecosystems such as wetlands. Within these systems, both anthropogenic disturbance and climate change pose a threat to habitat equilibrium, for example by facilitating the establishment of invasive alien species.</div><div>To characterise the phenology of two invasive aquatic plant species (<em>Ludwigia hexapetala</em> and <em>Nelumbo nucifera</em>), which have a notable impact on wetlands across Europe and North America, we utilised a medium-term Sentinel-2 time series collected across seven sites, representing diverse wetland types and biogeographical regions. We then investigated the environmental drivers (meteo-climatic factors and water quality parameters) influencing different phenometrics linked to the invasion success of these plants across sites and growing seasons, using non-linear parametric models.</div><div>The models demonstrated different strengths of linkage between environmental variables and selected phenological metrics. The phenometrics that were modelled most efficiently were the growth and senescence rates, and peak biomass for <em>L. hexapetala</em>, and the green-up and senescence start timing for <em>N. nucifera</em>. Assessing the most important inputs for the phenometrics models showed that the two species respond to environmental forcing at different spatial scales, reflecting their contrasting ecological strategies. Key phenology drivers are linked to climatic constraints, mainly temperature and light resources, for <em>N. nucifera</em>, or to local habitat features, such as residence time and wind disturbance, for <em>L. hexapetala</em>.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"43 ","pages":"Article 102183"},"PeriodicalIF":4.9,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148730280","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}