International journal of applied earth observation and geoinformation : ITC journal最新文献_第2页

ForResANeXt: Forest/non-forest segmentation with aggregated residual attention network in satellite imagery ForResANeXt：卫星图像中基于聚集残差注意力网络的森林/非森林分割

IF 8.6 Q1 REMOTE SENSING

International journal of applied earth observation and geoinformation : ITC journal

Pub Date : 2026-01-14 DOI: 10.1016/j.jag.2026.105105

Qianhuizi Guo , Liangzhi Li , Ling Han

Accurate mapping of forest (F) and non-forest (NF) areas is essential for ecological assessment, resource management, and deforestation monitoring. However, complex backgrounds, severe class imbalance and redundant features continue to limit the accuracy and efficiency of network segmentation. To overcome these issues, we present ForResANeXt, a novel semantic segmentation network that uses Sentinel-2 multispectral imagery for forest/non-forest mapping. The model incorporates an AResCAB to enrich contextual feature representations while reducing redundancy and a lightweight embedded attention module to improve positional awareness. Furthermore, attention-gated skip connections suppress background noise and emphasize key spatial information, and a Focal Dice Loss function mitigates the impact of severe class imbalance. Experimental results demonstrate that ForResANeXt achieves a mIoU of 95.31%, surpassing U-Net and mainstream CNN variants in recall and F1 score for the minority non-forest class. It also outperforms several representative advanced CNN architectures and Transformer-based models in terms of Boundary IoU and Small Object Recall. Qualitative comparisons further confirm its superior capability in preserving structural details and delineating complex boundaries with reduced misclassification. Cross-regional transfer experiments validate the model’s robustness and generalization capability across diverse geographical and temporal conditions, and ablation studies confirm the effectiveness of each proposed component. Overall, ForResANeXt shows great promise for efficient and accurate forest cover mapping using multispectral satellite data.

准确绘制森林和非森林区域的地图对于生态评估、资源管理和森林砍伐监测至关重要。然而，复杂的背景、严重的类不平衡和冗余的特征继续限制着网络分割的准确性和效率。为了克服这些问题，我们提出了一种新的语义分割网络ForResANeXt，该网络使用Sentinel-2多光谱图像进行森林/非森林制图。该模型结合了一个AResCAB来丰富上下文特征表示，同时减少冗余，并结合了一个轻量级的嵌入式注意模块来提高位置感知。此外，注意门控跳跃连接抑制背景噪声并强调关键空间信息，焦骰子损失函数减轻了严重的类别不平衡的影响。实验结果表明，ForResANeXt的mIoU达到95.31%，在召回率和少数非森林类的F1分数上超过了U-Net和主流CNN变体。在边界IoU和小对象召回方面，它也优于几种具有代表性的高级CNN架构和基于transformer的模型。定性比较进一步证实了其在保留结构细节和描绘复杂边界方面的优越能力，并减少了错误分类。跨区域转移实验验证了该模型在不同地理和时间条件下的稳健性和泛化能力，而消融研究证实了各组成部分的有效性。总的来说，ForResANeXt显示了利用多光谱卫星数据进行高效、准确的森林覆盖制图的巨大希望。

{"title":"ForResANeXt: Forest/non-forest segmentation with aggregated residual attention network in satellite imagery","authors":"Qianhuizi Guo , Liangzhi Li , Ling Han","doi":"10.1016/j.jag.2026.105105","DOIUrl":"10.1016/j.jag.2026.105105","url":null,"abstract":"<div><div>Accurate mapping of forest (F) and non-forest (NF) areas is essential for ecological assessment, resource management, and deforestation monitoring. However, complex backgrounds, severe class imbalance and redundant features continue to limit the accuracy and efficiency of network segmentation. To overcome these issues, we present ForResANeXt, a novel semantic segmentation network that uses Sentinel-2 multispectral imagery for forest/non-forest mapping. The model incorporates an AResCAB to enrich contextual feature representations while reducing redundancy and a lightweight embedded attention module to improve positional awareness. Furthermore, attention-gated skip connections suppress background noise and emphasize key spatial information, and a Focal Dice Loss function mitigates the impact of severe class imbalance. Experimental results demonstrate that ForResANeXt achieves a mIoU of 95.31%, surpassing U-Net and mainstream CNN variants in recall and F1 score for the minority non-forest class. It also outperforms several representative advanced CNN architectures and Transformer-based models in terms of Boundary IoU and Small Object Recall. Qualitative comparisons further confirm its superior capability in preserving structural details and delineating complex boundaries with reduced misclassification. Cross-regional transfer experiments validate the model’s robustness and generalization capability across diverse geographical and temporal conditions, and ablation studies confirm the effectiveness of each proposed component. Overall, ForResANeXt shows great promise for efficient and accurate forest cover mapping using multispectral satellite data.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105105"},"PeriodicalIF":8.6,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977788","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}

引用次数: 0

Global daily seamless XCO2 Mapping (2016–2020): Spatio-temporal trends and variations during wildfire events 全球每日无缝XCO2制图（2016-2020）：野火事件的时空趋势与变化

IF 8.6 Q1 REMOTE SENSING

International journal of applied earth observation and geoinformation : ITC journal

Pub Date : 2026-01-13 DOI: 10.1016/j.jag.2026.105092

Jie Li , Ziyi Zhang , Tongwen Li , Qiangqiang Yuan , Liangpei Zhang

Carbon dioxide (CO₂) is a dominant greenhouse gas and has a considerable effect on climate change. Satellite remote sensing is commonly used to acquire atmospheric CO₂ concentrations. However, the limited spatial coverage of a single satellite makes the obtainment of full-coverage CO₂ data difficult. In this study, a daily dataset of global seamless column-averaged dry-air mole fractions of CO₂ (XCO₂) was generated with a high spatial resolution of 0.1° from 2016 to 2020, by using a stacking machine learning method. The proposed XCO₂ dataset shows a satisfactory performance, with a root mean square error (RMSE) of 0.9697 ppm and correlation coefficient (R) of 0.9868 in the 10-fold cross validation. The spatial validation reveals good generalization ability, with continent-by-continent validation results showing an R greater than 0.93. The proposed dataset reports high consistency and accuracy in the ground-based validation, with an RMSE of 1.0855 ppm. Out of 24 stations, 22 demonstrate a precision of R greater than 0.95. In comparison with two XCO₂ model simulations, our reconstructions show a better consistency with ground observations. Spatial analyses at continent, national, and Chinese provincial levels, and temporal trends at daily, monthly, seasonal, and annual scales, are provided. Furthermore, benefitting from the daily temporal resolution, two typical examples of wildfire events, namely the Fort McMurray wildfire and the Blue Cut Fire, are evaluated. Our dataset can effectively capture fine-scale XCO₂ variations and has the potential to characterize carbon sources and sinks. The dataset can be obtained freely at https://zenodo.org/records/15191247.

二氧化碳（CO2）是主要的温室气体，对气候变化有相当大的影响。卫星遥感通常用于获取大气中的二氧化碳浓度。然而，由于单个卫星的空间覆盖有限，很难获得全覆盖的CO2数据。本研究采用堆叠机器学习方法，生成了2016 - 2020年全球无缝柱平均干空气摩尔分数（XCO2）的日数据集，空间分辨率为0.1°。在10倍交叉验证中，XCO2数据集的均方根误差（RMSE）为0.9697 ppm，相关系数(R)为0.9868。空间验证显示出较好的泛化能力，各大洲验证结果的R均大于0.93。该数据集在地面验证中具有较高的一致性和准确性，RMSE为1.0855 ppm。在24个站点中，22个站点的R精度大于0.95。与两个XCO2模式的模拟结果相比，我们的重建结果与地面观测结果具有更好的一致性。提供了大陆、国家和中国省级的空间分析，以及日、月、季、年尺度的时间趋势。此外，利用日时间分辨率，对两个典型的野火事件，即Fort McMurray野火和Blue Cut Fire进行了评估。我们的数据集可以有效地捕获精细尺度的XCO2变化，并具有表征碳源和汇的潜力。该数据集可以在https://zenodo.org/records/15191247上免费获得。

{"title":"Global daily seamless XCO2 Mapping (2016–2020): Spatio-temporal trends and variations during wildfire events","authors":"Jie Li , Ziyi Zhang , Tongwen Li , Qiangqiang Yuan , Liangpei Zhang","doi":"10.1016/j.jag.2026.105092","DOIUrl":"10.1016/j.jag.2026.105092","url":null,"abstract":"<div><div>Carbon dioxide (CO<sub>2</sub>) is a dominant greenhouse gas and has a considerable effect on climate change. Satellite remote sensing is commonly used to acquire atmospheric CO<sub>2</sub> concentrations. However, the limited spatial coverage of a single satellite makes the obtainment of full-coverage CO<sub>2</sub> data difficult. In this study, a daily dataset of global seamless column-averaged dry-air mole fractions of CO<sub>2</sub> (XCO<sub>2</sub>) was generated with a high spatial resolution of 0.1° from 2016 to 2020, by using a stacking machine learning method. The proposed XCO<sub>2</sub> dataset shows a satisfactory performance, with a root mean square error (RMSE) of 0.9697 ppm and correlation coefficient (R) of 0.9868 in the 10-fold cross validation. The spatial validation reveals good generalization ability, with continent-by-continent validation results showing an R greater than 0.93. The proposed dataset reports high consistency and accuracy in the ground-based validation, with an RMSE of 1.0855 ppm. Out of 24 stations, 22 demonstrate a precision of R greater than 0.95. In comparison with two XCO<sub>2</sub> model simulations, our reconstructions show a better consistency with ground observations. Spatial analyses at continent, national, and Chinese provincial levels, and temporal trends at daily, monthly, seasonal, and annual scales, are provided. Furthermore, benefitting from the daily temporal resolution, two typical examples of wildfire events, namely the Fort McMurray wildfire and the Blue Cut Fire, are evaluated. Our dataset can effectively capture fine-scale XCO<sub>2</sub> variations and has the potential to characterize carbon sources and sinks. The dataset can be obtained freely at <span><span>https://zenodo.org/records/15191247</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105092"},"PeriodicalIF":8.6,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957311","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}

引用次数: 0

Urban vegetation semantics in CityGML: Key stakeholder survey findings and vegetation ADE development CityGML中的城市植被语义：关键利益相关者调查结果和植被ADE发展

IF 8.6 Q1 REMOTE SENSING

International journal of applied earth observation and geoinformation : ITC journal

Pub Date : 2026-01-13 DOI: 10.1016/j.jag.2025.105043

Laura Mrosla , Dessislava Petrova-Antonova , Simeon Malinov , Henna Fabritius

Urban vegetation, providing critical ecosystem services and supporting biodiversity, is essential for sustainable and resilient cities. Yet its semantic representation in urban digital twins and 3D city models remains inadequate for advanced modeling. Open data standards facilitate interoperable urban modeling. However, the CityGML standard, despite its widespread adoption, provides limited semantic depth for vegetation in its current Vegetation module, constraining dynamic and interdisciplinary applications.

To address this, we present the international stakeholder survey evaluation alongside the updated version of the conceptual model of the CityGML 3.0 Vegetation Application Domain Extension (Vegetation ADE v1.1). The ADE enhances semantic richness by introducing extended attributes, feature types, enumerations, and code lists for both SolitaryVegetationObject and PlantCover, including structural components (crown, trunk, root) and dynamic properties via the CityGML Dynamizer module. Additionally, a vegetation management class is established.

Refined through feedback from an international stakeholder survey, the model reflects practical requirements from domains including urban studies, ecology, and public administration. The core data modeling structure of ADE v1.0, unanimously supported by survey participants, was retained for ADE v1.1. The new version introduces substantial improvements: 15 attributes were modified, 12 added, and one removed. Nine code lists were revised and one added. Two enumerations were updated. These enhancements ensure ADE v1.1 achieves improved semantic clarity and usability, supporting diverse applications such as vegetation monitoring and adaptive green infrastructure management. This work demonstrates a replicable methodology for participatory standard development and advances the integration of vegetation into data-driven urban digital twins and 3D city models.

城市植被提供关键的生态系统服务和支持生物多样性，对可持续和有复原力的城市至关重要。然而，它在城市数字孪生和3D城市模型中的语义表示仍然不足以进行高级建模。开放数据标准促进了可互操作的城市建模。然而，尽管CityGML标准被广泛采用，但在其当前的vegetation模块中，对植被的语义深度有限，限制了动态和跨学科的应用。

{"title":"Urban vegetation semantics in CityGML: Key stakeholder survey findings and vegetation ADE development","authors":"Laura Mrosla , Dessislava Petrova-Antonova , Simeon Malinov , Henna Fabritius","doi":"10.1016/j.jag.2025.105043","DOIUrl":"10.1016/j.jag.2025.105043","url":null,"abstract":"<div><div>Urban vegetation, providing critical ecosystem services and supporting biodiversity, is essential for sustainable and resilient cities. Yet its semantic representation in urban digital twins and 3D city models remains inadequate for advanced modeling. Open data standards facilitate interoperable urban modeling. However, the CityGML standard, despite its widespread adoption, provides limited semantic depth for vegetation in its current Vegetation module, constraining dynamic and interdisciplinary applications.</div><div>To address this, we present the international stakeholder survey evaluation alongside the updated version of the conceptual model of the CityGML 3.0 Vegetation Application Domain Extension (Vegetation ADE v1.1). The ADE enhances semantic richness by introducing extended attributes, feature types, enumerations, and code lists for both <em>SolitaryVegetationObject</em> and <em>PlantCover</em>, including structural components (crown, trunk, root) and dynamic properties via the CityGML <em>Dynamizer</em> module. Additionally, a vegetation management class is established.</div><div>Refined through feedback from an international stakeholder survey, the model reflects practical requirements from domains including urban studies, ecology, and public administration. The core data modeling structure of ADE v1.0, unanimously supported by survey participants, was retained for ADE v1.1. The new version introduces substantial improvements: 15 attributes were modified, 12 added, and one removed. Nine code lists were revised and one added. Two enumerations were updated. These enhancements ensure ADE v1.1 achieves improved semantic clarity and usability, supporting diverse applications such as vegetation monitoring and adaptive green infrastructure management. This work demonstrates a replicable methodology for participatory standard development and advances the integration of vegetation into data-driven urban digital twins and 3D city models.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105043"},"PeriodicalIF":8.6,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957312","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}

引用次数: 0

A semi-supervised framework for UAV-based individual tree crown segmentation in structurally heterogeneous planted forests 结构异质人工林中基于无人机的树冠分割半监督框架

IF 8.6 Q1 REMOTE SENSING

International journal of applied earth observation and geoinformation : ITC journal

Pub Date : 2026-01-13 DOI: 10.1016/j.jag.2025.105078

Qing Wang , Yihui Zhao , Yingpu Che , Han Shen , Yongbin Qiu , Yixiang Wang

Accurate individual tree crown (ITC) segmentation is essential for quantifying forest fine carbon stocks. However, deep learning segmentation methods face prohibitive annotation costs while unsupervised algorithms struggle in structurally complex forests. This study proposes a semi-supervised framework integrating unsupervised pseudo-label generation, deep instance segmentation, and staged fine-tuning to overcome these limitations. The framework transforms outputs from marker-controlled watershed (MCWS) and region-growing (itcSegment) algorithms into initial training targets to obtain generalizable instance segmentation representations, then refines Mask R-CNN models with minimal manual annotations to enhance structural awareness of instance boundaries and semantic context understanding, significantly reducing labeling dependency while improving robustness. Validated across three structurally heterogeneous Chinese fir stands spanning age (11–67 years), density (450–2500 stems·ha⁻¹), and elevation (192–1047 m) gradients, our UAV RGB-based framework achieved consistent superiority over LiDAR and fused inputs, with spectral-textural features proving dominant for boundary delineation. It attained F1-scores of 0.826 (young, undulating terrain), 0.836 (mature, high-density), and 0.711 (over-mature, occluded) using only 40 % expert annotations (0.6 personnel-hours), representing a 0.42 average improvement over MCWS/itcSegment baselines. The designed staged fine-tuning strategy effectively mitigated pseudo-label error propagation, while expanding this annotation effort to 70–100 % yielded marginal accuracy gains, demonstrating exceptional efficiency in leveraging minimal supervision. Based on the segmented crowns, structural parameters were extracted with high fidelity: crown diameter (R² ≥ 0.76, rRMSE ≤ 11.3 %) and crown area (R² ≥ 0.88, rRMSE ≤ 16.4 %). This approach reduces annotation demands while maintaining robustness across forest heterogeneity, enabling operationally scalable solutions for precision forestry.

准确的单株树冠分割是森林细碳储量量化的关键。然而，深度学习分割方法面临着令人望而却步的标注成本，而非监督算法在结构复杂的森林中挣扎。本研究提出了一种半监督框架，集成了无监督伪标签生成、深度实例分割和阶段微调来克服这些限制。该框架将标记控制分水岭（MCWS）和区域增长（itcSegment）算法的输出转换为初始训练目标，以获得可推广的实例分割表示，然后使用最少的手动注释来改进Mask R-CNN模型，以增强对实例边界的结构感知和语义上下文理解，显著降低标记依赖，同时提高鲁棒性。在三种不同结构杉木林分的年龄（11-67年）、密度（450-2500株·ha - 1）和海拔（192-1047 m）梯度上进行验证，我们基于rgb的无人机框架比激光雷达和融合输入具有一致的优势，光谱纹理特征在边界划分方面占主导地位。仅使用40%的专家注释（0.6人-小时），它就获得了f1分数0.826（年轻，起伏地形），0.836（成熟，高密度）和0.711（过度成熟，闭塞），比MCWS/itcSegment基线平均提高了0.42。设计的分阶段微调策略有效地减轻了伪标签错误传播，同时将注释工作扩展到70 - 100%，产生了边际精度增益，在利用最小监督方面展示了卓越的效率。基于分割的冠提取结构参数：冠直径（R2≥0.76,rRMSE≤11.3%）和冠面积（R2≥0.88,rRMSE≤16.4%）。这种方法减少了注释需求，同时保持了森林异质性的健壮性，为精准林业提供了可操作的可扩展解决方案。

{"title":"A semi-supervised framework for UAV-based individual tree crown segmentation in structurally heterogeneous planted forests","authors":"Qing Wang , Yihui Zhao , Yingpu Che , Han Shen , Yongbin Qiu , Yixiang Wang","doi":"10.1016/j.jag.2025.105078","DOIUrl":"10.1016/j.jag.2025.105078","url":null,"abstract":"<div><div>Accurate individual tree crown (ITC) segmentation is essential for quantifying forest fine carbon stocks. However, deep learning segmentation methods face prohibitive annotation costs while unsupervised algorithms struggle in structurally complex forests. This study proposes a semi-supervised framework integrating unsupervised pseudo-label generation, deep instance segmentation, and staged fine-tuning to overcome these limitations. The framework transforms outputs from marker-controlled watershed (MCWS) and region-growing (itcSegment) algorithms into initial training targets to obtain generalizable instance segmentation representations, then refines Mask R-CNN models with minimal manual annotations to enhance structural awareness of instance boundaries and semantic context understanding, significantly reducing labeling dependency while improving robustness. Validated across three structurally heterogeneous Chinese fir stands spanning age (11–67 years), density (450–2500 stems·ha<sup>−1</sup>), and elevation (192–1047 m) gradients, our UAV RGB-based framework achieved consistent superiority over LiDAR and fused inputs, with spectral-textural features proving dominant for boundary delineation. It attained F1-scores of 0.826 (young, undulating terrain), 0.836 (mature, high-density), and 0.711 (over-mature, occluded) using only 40 % expert annotations (0.6 personnel-hours), representing a 0.42 average improvement over MCWS/itcSegment baselines. The designed staged fine-tuning strategy effectively mitigated pseudo-label error propagation, while expanding this annotation effort to 70–100 % yielded marginal accuracy gains, demonstrating exceptional efficiency in leveraging minimal supervision. Based on the segmented crowns, structural parameters were extracted with high fidelity: crown diameter (<em>R</em><sup>2</sup> ≥ 0.76, rRMSE ≤ 11.3 %) and crown area (<em>R</em><sup>2</sup> ≥ 0.88, rRMSE ≤ 16.4 %). This approach reduces annotation demands while maintaining robustness across forest heterogeneity, enabling operationally scalable solutions for precision forestry.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105078"},"PeriodicalIF":8.6,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977719","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}

引用次数: 0

Few-Shot change detection in optical and SAR remote sensing images for disaster response 面向灾害响应的光学和SAR遥感影像少镜头变化检测

IF 8.6 Q1 REMOTE SENSING

International journal of applied earth observation and geoinformation : ITC journal

Pub Date : 2026-01-12 DOI: 10.1016/j.jag.2026.105100

Di Wang , Guorui Ma , Xiao Wang , Ronghao Yang , Yongxian Zhang

Few-shot change detection in optical and Synthetic Aperture Radar images is a critical task for disaster monitoring. offering significant application value in complex scenarios with extremely limited labeled samples. However, the randomness of disasters causes a notable data distribution shift between public datasets and real disaster scenarios. With only a few annotated image pairs, existing methods struggle to effectively fuse features from heterogeneous images, leading to severe performance degradation. To address this challenge, we propose a Dual-Stage Training framework for Change Detection (DSTCD), specifically designed for few-shot scenarios involving fewer than 20 labeled image pairs. DSTCD first undergoes source task pre-training on a heterogeneous image registration dataset. Subsequently, in the target task stage, it leverages task guided feature transfer module to transfer the structural and semantic features of image registration to the change detection task. This mechanism significantly enriches the feature representations under few-shot conditions, enabling accurate identification of affected areas. To validate its performance, we conducted comparative and ablation studies against eleven state-of-the-art methods on four public datasets covering both urban expansion and water expansion scenarios. Experimental results demonstrate that DSTCD achieves a significant performance lead. Its average F1-score surpasses the second-best method by 6.98% in urban expansion scenarios and by 13.09% in water expansion scenarios, proving its superior performance and strong multi-scenario adaptability. Furthermore, robustness analysis of varying training sample sizes and real-world disaster application validation further confirm the method’s practicality and robustness for data-scarce disaster monitoring tasks. The code of the proposed method will be made available at https://github.com/Lucky-DW/DSTCD.

光学和合成孔径雷达图像的少镜头变化检测是灾害监测的关键任务。在极其有限的标记样本的复杂场景中提供重要的应用价值。然而，灾害的随机性导致公共数据集和真实灾害场景之间的数据分布发生了明显的变化。由于只有少数带注释的图像对，现有方法难以有效地融合异构图像的特征，导致性能严重下降。为了解决这一挑战，我们提出了一个双阶段变化检测训练框架（DSTCD），专门为涉及少于20个标记图像对的少量场景设计。DSTCD首先在异构图像配准数据集上进行源任务预训练。随后，在目标任务阶段，利用任务导向特征转移模块将图像配准的结构特征和语义特征转移到变化检测任务中。这一机制大大丰富了少镜头条件下的特征表示，使受影响区域的准确识别成为可能。为了验证其性能，我们在四个公共数据集上对11种最先进的方法进行了比较和消融研究，包括城市扩张和水扩张情景。实验结果表明，DSTCD具有明显的性能领先优势。该方法在城市扩张情景下的平均f1得分比次优方法高6.98%，在水域扩张情景下的平均f1得分比次优方法高13.09%，证明了其优越的性能和较强的多情景适应性。此外，对不同训练样本大小的鲁棒性分析和实际灾难应用验证进一步证实了该方法对于数据稀缺的灾害监测任务的实用性和鲁棒性。建议方法的代码将在https://github.com/Lucky-DW/DSTCD上提供。

{"title":"Few-Shot change detection in optical and SAR remote sensing images for disaster response","authors":"Di Wang , Guorui Ma , Xiao Wang , Ronghao Yang , Yongxian Zhang","doi":"10.1016/j.jag.2026.105100","DOIUrl":"10.1016/j.jag.2026.105100","url":null,"abstract":"<div><div>Few-shot change detection in optical and Synthetic Aperture Radar images is a critical task for disaster monitoring. offering significant application value in complex scenarios with extremely limited labeled samples. However, the randomness of disasters causes a notable data distribution shift between public datasets and real disaster scenarios. With only a few annotated image pairs, existing methods struggle to effectively fuse features from heterogeneous images, leading to severe performance degradation. To address this challenge, we propose a Dual-Stage Training framework for Change Detection (DSTCD), specifically designed for few-shot scenarios involving fewer than 20 labeled image pairs. DSTCD first undergoes source task pre-training on a heterogeneous image registration dataset. Subsequently, in the target task stage, it leverages task guided feature transfer module to transfer the structural and semantic features of image registration to the change detection task. This mechanism significantly enriches the feature representations under few-shot conditions, enabling accurate identification of affected areas. To validate its performance, we conducted comparative and ablation studies against eleven state-of-the-art methods on four public datasets covering both urban expansion and water expansion scenarios. Experimental results demonstrate that DSTCD achieves a significant performance lead. Its average F1-score surpasses the second-best method by 6.98% in urban expansion scenarios and by 13.09% in water expansion scenarios, proving its superior performance and strong multi-scenario adaptability. Furthermore, robustness analysis of varying training sample sizes and real-world disaster application validation further confirm the method’s practicality and robustness for data-scarce disaster monitoring tasks. The code of the proposed method will be made available at <span><span>https://github.com/Lucky-DW/DSTCD</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105100"},"PeriodicalIF":8.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957348","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}

引用次数: 0

Non-negative matrix factorization-assisted independent component analysis for land subsidence feature decomposition and attribution in Lubei Plain, and further integration with deep learning for subsidence prediction 陆北平原非负矩阵分解辅助独立分量分析沉降特征分解与归因，并进一步与深度学习结合进行沉降预测

IF 8.6 Q1 REMOTE SENSING

International journal of applied earth observation and geoinformation : ITC journal

Pub Date : 2026-01-12 DOI: 10.1016/j.jag.2026.105097

Yongkang Wang , Huili Gong , Beibei Chen , Chaofan Zhou , Lin guo , Haotong Wang , Qin Wang , Shuqi Hao , Yabin Yang

Interferometric Synthetic Aperture Radar (InSAR) has become crucial for large-scale land subsidence monitoring, but its results represent complex superpositions of deformation signals from multiple compressible layers driven by various anthropogenic and natural factors. Decomposing subsidence patterns by different driving factors is vital for attribution and prediction. However, existing feature decomposition methods suffer from problems of incomplete decomposition and difficulties in direction determination. This study employs non-negative matrix factorization (NMF) assisted independent component analysis (ICA) to achieve feature decomposition and attribution of land subsidence in the Lubei Plain (LBP) from 2018 to 2023, and integration with deep learning to realize land subsidence prediction. Key contributions include: (1) Using NMF component directions as reference to determine three independent component directions for subsequent attribution and prediction analysis. (2) Groundwater wells in high-spatial-score regions were selected to analyze the relationship between ICs, deformation, and groundwater levels. IC1 showed a high correlation exceeding 0.8 with subsidence, reflecting near-linear patterns from deep confined aquifer over-exploitation and petroleum extraction. IC2 reached maximum correlations of 0.72 with subsidence and 0.56 with groundwater levels (GWL), reflecting inter-annual precipitation-driven variations. Similarly, IC3 exhibited maximum correlations of 0.63 and 0.71, respectively, characterizing seasonal cycles linked to agricultural irrigation and precipitation. (3) The proposed prediction model demonstrated excellent performance with RMSE of 0.06–0.69 and R² of 0.80–0.86 in test datasets. This study provides a new reference perspective for future driving factor analysis and regulation of land subsidence in the Lubei Plain, and offers a new method for land subsidence prediction.

干涉合成孔径雷达（InSAR）已成为大尺度地面沉降监测的重要手段，但其结果是受各种人为和自然因素驱动的多个可压缩层变形信号的复杂叠加。不同驱动因素对沉降模式的分解对沉降成因和预测具有重要意义。然而，现有的特征分解方法存在分解不完全、方向确定困难等问题。本研究采用非负矩阵分解（NMF）辅助独立分量分析（ICA）对鲁北平原2018 - 2023年地面沉降进行特征分解与归因，并结合深度学习实现地面沉降预测。主要贡献包括：(1)以NMF分量方向为参考，确定了3个独立分量方向，用于后续归因和预测分析。(2)选取高空间分值区域的地下水井，分析地表变形与地下水位的关系。IC1与沉降呈高度相关，相关性超过0.8，反映了深部承压含水层过度开采和石油开采的近线性模式。IC2与沉降和地下水位（GWL）的相关性最大，分别为0.72和0.56，反映了降水驱动的年际变化。同样，IC3的最大相关性分别为0.63和0.71，表征了与农业灌溉和降水有关的季节周期。(3)所提出的预测模型在测试数据集上RMSE为0.06 ~ 0.69，R2为0.80 ~ 0.86，具有良好的预测性能。该研究为今后鲁北平原地面沉降驱动因素分析和调控提供了新的参考视角，并为地面沉降预测提供了新的方法。

{"title":"Non-negative matrix factorization-assisted independent component analysis for land subsidence feature decomposition and attribution in Lubei Plain, and further integration with deep learning for subsidence prediction","authors":"Yongkang Wang , Huili Gong , Beibei Chen , Chaofan Zhou , Lin guo , Haotong Wang , Qin Wang , Shuqi Hao , Yabin Yang","doi":"10.1016/j.jag.2026.105097","DOIUrl":"10.1016/j.jag.2026.105097","url":null,"abstract":"<div><div>Interferometric Synthetic Aperture Radar (InSAR) has become crucial for large-scale land subsidence monitoring, but its results represent complex superpositions of deformation signals from multiple compressible layers driven by various anthropogenic and natural factors. Decomposing subsidence patterns by different driving factors is vital for attribution and prediction. However, existing feature decomposition methods suffer from problems of incomplete decomposition and difficulties in direction determination. This study employs non-negative matrix factorization (NMF) assisted independent component analysis (ICA) to achieve feature decomposition and attribution of land subsidence in the Lubei Plain (LBP) from 2018 to 2023, and integration with deep learning to realize land subsidence prediction. Key contributions include: (1) Using NMF component directions as reference to determine three independent component directions for subsequent attribution and prediction analysis. (2) Groundwater wells in high-spatial-score regions were selected to analyze the relationship between ICs, deformation, and groundwater levels. IC1 showed a high correlation exceeding 0.8 with subsidence, reflecting near-linear patterns from deep confined aquifer over-exploitation and petroleum extraction. IC2 reached maximum correlations of 0.72 with subsidence and 0.56 with groundwater levels (GWL), reflecting inter-annual precipitation-driven variations. Similarly, IC3 exhibited maximum correlations of 0.63 and 0.71, respectively, characterizing seasonal cycles linked to agricultural irrigation and precipitation. (3) The proposed prediction model demonstrated excellent performance with RMSE of 0.06–0.69 and R<sup>2</sup> of 0.80–0.86 in test datasets. This study provides a new reference perspective for future driving factor analysis and regulation of land subsidence in the Lubei Plain, and offers a new method for land subsidence prediction.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105097"},"PeriodicalIF":8.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957349","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}

引用次数: 0

Satellite assessment of rooftop greenness dynamics and equity in Hong Kong 香港屋顶绿化动态及公平性的卫星评估

IF 8.6 Q1 REMOTE SENSING

International journal of applied earth observation and geoinformation : ITC journal

Pub Date : 2026-01-12 DOI: 10.1016/j.jag.2026.105096

Jing Ling , Shan Wei , Genyun Sun , Liqun Sun , Xiaojun Yang , Hongsheng Zhang

Rooftop greening is increasingly recognized as a pathway to urban sustainability and environmental equity, yet its fine-scale temporal dynamics and distributional fairness remain insufficiently examined. This study develops a building-level framework that constructs annual rooftop greenness time series from satellite-derived NDVI and introduces a Rooftop Greenness Inequality Index (RGII) based on the rooftop green fraction to evaluate distributive fairness. By integrating these measures with a multi-pattern trend classification, the framework captures the co-evolution of temporal dynamics and spatial inequality, and links them to socioeconomic conditions to identify key drivers and vulnerable communities. Using Hong Kong as a case study (2018–2024), we find overall greenness gains and inequality reductions, though unevenly distributed. Citywide, 22.1% of buildings showed significant increases, while 2.1% showed consistent decreases. All 452 communities saw inequality decrease, with the median RGII falling from 0.65 to 0.45. Greenness trajectories varied by land use and development stage: public housing showed a 38.9% rise compared with 21.1% in private housing, and late-stage new towns grew faster but with less stability than older districts. Socioeconomic disparities shaped outcomes: communities with high rent burden, large non-local populations, or many single-person households showed limited improvement, whereas those with more spatial capacity or stronger cultural cohesion performed better. Vulnerable areas were concentrated in older high-density districts where greenness remained low and inequality persisted. These findings demonstrate the framework’s capacity to integrate rooftop greenness change with social dimensions, providing transferable tools for equity-oriented greening strategies in high-density cities.

屋顶绿化越来越被认为是城市可持续发展和环境公平的途径，但其精细尺度的时间动态和分配公平性仍然没有得到充分的研究。本研究开发了一个基于卫星NDVI数据构建年度屋顶绿色度时间序列的建筑级框架，并引入了基于屋顶绿色度比例的屋顶绿色度不平等指数（RGII）来评估分配公平性。通过将这些措施与多模式趋势分类相结合，该框架捕捉了时间动态和空间不平等的共同演变，并将其与社会经济条件联系起来，以确定关键驱动因素和脆弱社区。以香港为例（2018-2024年），我们发现整体绿色增长和不平等减少，尽管分布不均。在全市范围内，22.1%的建筑物显著增加，而2.1%的建筑物持续下降。所有452个社区的不平等程度都有所下降，RGII中位数从0.65降至0.45。绿化轨迹因土地用途和发展阶段而异：公共住房增长38.9%，而私人住房增长21.1%，后期新市镇增长更快，但稳定性不如旧区。社会经济差异影响了结果：租金负担高、非本地人口多或单身家庭多的社区改善有限，而空间容量大或文化凝聚力强的社区表现更好。脆弱地区集中在旧的高密度地区，那里的绿化率仍然很低，不平等现象持续存在。这些发现证明了该框架将屋顶绿化变化与社会维度相结合的能力，为高密度城市的公平导向绿化战略提供了可转移的工具。

{"title":"Satellite assessment of rooftop greenness dynamics and equity in Hong Kong","authors":"Jing Ling , Shan Wei , Genyun Sun , Liqun Sun , Xiaojun Yang , Hongsheng Zhang","doi":"10.1016/j.jag.2026.105096","DOIUrl":"10.1016/j.jag.2026.105096","url":null,"abstract":"<div><div>Rooftop greening is increasingly recognized as a pathway to urban sustainability and environmental equity, yet its fine-scale temporal dynamics and distributional fairness remain insufficiently examined. This study develops a building-level framework that constructs annual rooftop greenness time series from satellite-derived NDVI and introduces a Rooftop Greenness Inequality Index (RGII) based on the rooftop green fraction to evaluate distributive fairness. By integrating these measures with a multi-pattern trend classification, the framework captures the co-evolution of temporal dynamics and spatial inequality, and links them to socioeconomic conditions to identify key drivers and vulnerable communities. Using Hong Kong as a case study (2018–2024), we find overall greenness gains and inequality reductions, though unevenly distributed. Citywide, 22.1% of buildings showed significant increases, while 2.1% showed consistent decreases. All 452 communities saw inequality decrease, with the median RGII falling from 0.65 to 0.45. Greenness trajectories varied by land use and development stage: public housing showed a 38.9% rise compared with 21.1% in private housing, and late-stage new towns grew faster but with less stability than older districts. Socioeconomic disparities shaped outcomes: communities with high rent burden, large non-local populations, or many single-person households showed limited improvement, whereas those with more spatial capacity or stronger cultural cohesion performed better. Vulnerable areas were concentrated in older high-density districts where greenness remained low and inequality persisted. These findings demonstrate the framework’s capacity to integrate rooftop greenness change with social dimensions, providing transferable tools for equity-oriented greening strategies in high-density cities.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105096"},"PeriodicalIF":8.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957350","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}

引用次数: 0

A dual-path network for semantic scene completion of single-frame LiDAR point clouds 单帧激光雷达点云语义场景补全的双路径网络

IF 8.6 Q1 REMOTE SENSING

International journal of applied earth observation and geoinformation : ITC journal

Pub Date : 2026-01-12 DOI: 10.1016/j.jag.2025.105020

Wei Liu , Ziwen Kang , Yongtao Yu , Zheng Gong , Yuchao Zheng , Xiaohui Huang , Haiyan Guan , Lingfei Ma , Dedong Zhang

Semantic Scene Completion (SSC) is a fundamental yet challenging task in 3D environment perception, as the sparsity and noise of LiDAR point clouds make it difficult to accurately recover both geometry and semantics. To address these challenges, we propose DPS2CNet, a novel Dual-Path SSC Network that integrates voxel-based and bird’s-eye view (BEV) representations to exploit their complementary strengths. Specifically, DPS2CNet employs a Cylinder3D-enhanced voxel branch to capture fine-grained 3D geometry and a UNet-based BEV branch to model large-scale contextual information. To further boost performance, we incorporate CARAFE for efficient feature upsampling and design a tailored loss function optimized for SSC. Extensive experiments on SemanticKITTI and SSCBench-KITTI-360 demonstrate that DPS2CNet achieves state-of-the-art results. In particular, it ranks first on the SemanticKITTI test set with an IoU of 62.6% among all open-source submissions¹ , highlighting its effectiveness in complex real-world driving scenarios.

语义场景补全（SSC）是3D环境感知中的一项基本但具有挑战性的任务，因为激光雷达点云的稀疏性和噪声使得难以准确恢复几何和语义。为了应对这些挑战，我们提出了一种新的双路径SSC网络DPS2CNet，它集成了基于体素和鸟瞰图（BEV）的表示，以利用它们的互补优势。具体来说，DPS2CNet使用了一个圆柱体3D增强的体素分支来捕获细粒度的3D几何图形，以及一个基于unet的BEV分支来建模大规模的上下文信息。为了进一步提高性能，我们结合CARAFE进行有效的特征上采样，并设计了针对SSC优化的定制损失函数。在SemanticKITTI和sschbench - kitti -360上进行的大量实验表明，DPS2CNet达到了最先进的效果。特别是，它在SemanticKITTI测试集上排名第一，在所有开源提交任务中，IoU为62.6% 11https://codalab.lisn.upsaclay.fr/competitions/7170#results。突出了其在复杂的现实驾驶场景中的有效性。

{"title":"A dual-path network for semantic scene completion of single-frame LiDAR point clouds","authors":"Wei Liu , Ziwen Kang , Yongtao Yu , Zheng Gong , Yuchao Zheng , Xiaohui Huang , Haiyan Guan , Lingfei Ma , Dedong Zhang","doi":"10.1016/j.jag.2025.105020","DOIUrl":"10.1016/j.jag.2025.105020","url":null,"abstract":"<div><div>Semantic Scene Completion (SSC) is a fundamental yet challenging task in 3D environment perception, as the sparsity and noise of LiDAR point clouds make it difficult to accurately recover both geometry and semantics. To address these challenges, we propose DPS2CNet, a novel Dual-Path SSC Network that integrates voxel-based and bird’s-eye view (BEV) representations to exploit their complementary strengths. Specifically, DPS2CNet employs a Cylinder3D-enhanced voxel branch to capture fine-grained 3D geometry and a UNet-based BEV branch to model large-scale contextual information. To further boost performance, we incorporate CARAFE for efficient feature upsampling and design a tailored loss function optimized for SSC. Extensive experiments on SemanticKITTI and SSCBench-KITTI-360 demonstrate that DPS2CNet achieves state-of-the-art results. In particular, it ranks first on the SemanticKITTI test set with an IoU of 62.6% among all open-source submissions<span><span><sup>1</sup></span></span> , highlighting its effectiveness in complex real-world driving scenarios.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105020"},"PeriodicalIF":8.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957352","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}

引用次数: 0

Federated earth-observation models for collaborative farm-scale soil mapping 协同农场尺度土壤制图的联合地球观测模型

IF 8.6 Q1 REMOTE SENSING

International journal of applied earth observation and geoinformation : ITC journal

Pub Date : 2026-01-12 DOI: 10.1016/j.jag.2025.105067

Giannis Gallios , José A.M. Demattê , Nikolaos Tsakiridis , Matheus Carraco Cardoso , Anastasia Kritharoula , Nikolaos Tziolas

Accurate, privacy-respecting soil information is essential for site-specific nutrient management and carbon accounting, yet the cost of laboratory analyses limits many farms to relatively sparse sampling grids. We propose a collaborative framework that couples a national Sentinel-2 bare-soil composite with FL to produce high-resolution clay and soil organic carbon (SOC) maps while keeping all local data on-premise. A one-dimensional convolutional neural network was first pre-trained on a 53,570-sample Brazilian archive and then fine-tuned across 50 farms through synchronous Federated Averaging. We benchmarked this hybrid configuration against (i) a purely centralized model trained on the same archive and (ii) a fully decentralized FL model initialized at random.

Across farm-level validation sets, pre-trained FL lowered median RMSE by 42% for clay and 31% for SOC relative to the centralized baseline, while increasing median RPIQ by 33% and 25%, respectively. On farms with

\leq 150

samples, the gains remained substantial, confirming that the approach remains effective when local datasets are modest compared with national archives. Error distributions differed significantly between scenarios (Friedman and Wilcoxon tests), and the pre-trained FL maps removed most spatial artefacts observed in the centralized outputs while preserving fine-scale gradients. Because only encrypted weight updates are exchanged, sensitive information never leaves the farm, satisfying GDPR/LGPD-style constraints and allowing late-joining clients to inherit an increasingly mature global model. Taken together, these results indicate that continental pre-training followed by federated fine-tuning reconciles global generality with local specificity and offers a scalable blueprint for privacy-preserving, high-resolution soil mapping in settings where sample densities are often substantially lower than in experimental setups, without compromising data sovereignty.

准确、尊重隐私的土壤信息对于特定地点的养分管理和碳核算至关重要，但实验室分析的成本限制了许多农场相对稀疏的采样网格。我们提出了一个协作框架，将国家Sentinel-2裸土复合材料与FL相结合，生成高分辨率粘土和土壤有机碳（SOC）地图，同时保留所有本地数据。一维卷积神经网络首先在53570个巴西档案样本上进行预训练，然后通过同步联邦平均对50个农场进行微调。我们将这种混合配置与(i)在同一存档上训练的纯集中式模型和（ii）随机初始化的完全分散的FL模型进行基准测试。

{"title":"Federated earth-observation models for collaborative farm-scale soil mapping","authors":"Giannis Gallios , José A.M. Demattê , Nikolaos Tsakiridis , Matheus Carraco Cardoso , Anastasia Kritharoula , Nikolaos Tziolas","doi":"10.1016/j.jag.2025.105067","DOIUrl":"10.1016/j.jag.2025.105067","url":null,"abstract":"<div><div>Accurate, privacy-respecting soil information is essential for site-specific nutrient management and carbon accounting, yet the cost of laboratory analyses limits many farms to relatively sparse sampling grids. We propose a collaborative framework that couples a national Sentinel-2 bare-soil composite with FL to produce high-resolution clay and soil organic carbon (SOC) maps while keeping all local data on-premise. A one-dimensional convolutional neural network was first pre-trained on a 53,570-sample Brazilian archive and then fine-tuned across 50 farms through synchronous Federated Averaging. We benchmarked this hybrid configuration against (i) a purely centralized model trained on the same archive and (ii) a fully decentralized FL model initialized at random.</div><div>Across farm-level validation sets, pre-trained FL lowered median RMSE by 42% for clay and 31% for SOC relative to the centralized baseline, while increasing median RPIQ by 33% and 25%, respectively. On farms with <span><math><mrow><mo>≤</mo><mn>150</mn></mrow></math></span> samples, the gains remained substantial, confirming that the approach remains effective when local datasets are modest compared with national archives. Error distributions differed significantly between scenarios (Friedman and Wilcoxon tests), and the pre-trained FL maps removed most spatial artefacts observed in the centralized outputs while preserving fine-scale gradients. Because only encrypted weight updates are exchanged, sensitive information never leaves the farm, satisfying GDPR/LGPD-style constraints and allowing late-joining clients to inherit an increasingly mature global model. Taken together, these results indicate that continental pre-training followed by federated fine-tuning reconciles global generality with local specificity and offers a scalable blueprint for privacy-preserving, high-resolution soil mapping in settings where sample densities are often substantially lower than in experimental setups, without compromising data sovereignty.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105067"},"PeriodicalIF":8.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957351","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}

引用次数: 0

Wildfire susceptibility mapping with multiple machine learning algorithms utilizing forest inventory and FIRMS data: A case study in Arsin, Trabzon, Türkiye 利用森林清查和FIRMS数据的多种机器学习算法绘制野火易感性图：以Arsin， Trabzon, t<s:1> rkiye为例

IF 8.6 Q1 REMOTE SENSING

International journal of applied earth observation and geoinformation : ITC journal

Pub Date : 2026-01-10 DOI: 10.1016/j.jag.2026.105091

Sude Gül Yel , Derya Mumcu Küçüker , Esra Tunç Görmüş

Forest fires pose a significant threat to ecosystems, biodiversity, and human settlements. This study focuses on the Arsin Forest Sub-district Directorate, located in Trabzon, Türkiye, with the aim of developing wildfire susceptibility maps using machine learning techniques. To improve the completeness of the wildfire inventory dataset, official fire records from 2013 to 2022 were integrated with active fire pixel data obtained from the Fire Information for Resource Management System (FIRMS) for the period 2001–2012. Five machine learning models Random Forest (RF), Artificial Neural Network (ANN), Gradient Boosting Machine (GBM), Extreme Gradient Boosting (XGBoost), and Deep Neural Network (DNN) were employed to generate susceptibility maps. Model performance was evaluated using accuracy, precision, recall, F1-score, ROC curves, and AUC-ROC metrics. Seventeen wildfire conditioning factors, categorized into four groups (topographic, meteorological, stand-related, and anthropogenic), were used to assess fire risk. Unlike previous studies, this research incorporates a region-specific anthropogenic variable: proximity to hazelnut cultivation. Feature importance scores were computed to determine the influence of each factor on fire occurrence. Additionally, SHapley Additive exPlanations (SHAP), accompanied by graphical analysis, were used to interpret the relationships between predictors and fire events. Areas of very high fire susceptibility covering 4.62% and 4.86% of the study area were identified by the RF and GBM models, both achieving an accuracy of 0.98.

森林火灾对生态系统、生物多样性和人类住区构成重大威胁。本研究的重点是位于 rkiye省Trabzon的Arsin森林分区局，目的是利用机器学习技术开发野火易感性地图。为了提高野火清单数据集的完整性，将2013年至2022年的官方火灾记录与2001年至2012年期间从火灾信息资源管理系统（FIRMS）获得的活火像素数据进行了整合。采用随机森林（Random Forest， RF）、人工神经网络（Artificial Neural Network， ANN）、梯度增强机（Gradient Boosting machine， GBM）、极端梯度增强机（Extreme Gradient Boosting, XGBoost）和深度神经网络（Deep Neural Network， DNN） 5种机器学习模型生成敏感性图。采用准确度、精密度、召回率、f1评分、ROC曲线和AUC-ROC指标评价模型的性能。17个野火调节因子被分为四组（地形、气象、森林相关和人为），用于评估火灾风险。与以前的研究不同，这项研究纳入了一个区域特定的人为变量：接近榛子种植。计算特征重要性分数，以确定每个因素对火灾发生的影响。此外，采用SHapley加性解释（SHAP）和图形分析来解释预测因子与火灾事件之间的关系。RF和GBM模型分别确定了4.62%和4.86%的高火易感性区域，两者的准确率均为0.98。

{"title":"Wildfire susceptibility mapping with multiple machine learning algorithms utilizing forest inventory and FIRMS data: A case study in Arsin, Trabzon, Türkiye","authors":"Sude Gül Yel , Derya Mumcu Küçüker , Esra Tunç Görmüş","doi":"10.1016/j.jag.2026.105091","DOIUrl":"10.1016/j.jag.2026.105091","url":null,"abstract":"<div><div>Forest fires pose a significant threat to ecosystems, biodiversity, and human settlements. This study focuses on the Arsin Forest Sub-district Directorate, located in Trabzon, Türkiye, with the aim of developing wildfire susceptibility maps using machine learning techniques. To improve the completeness of the wildfire inventory dataset, official fire records from 2013 to 2022 were integrated with active fire pixel data obtained from the Fire Information for Resource Management System (FIRMS) for the period 2001–2012. Five machine learning models Random Forest (RF), Artificial Neural Network (ANN), Gradient Boosting Machine (GBM), Extreme Gradient Boosting (XGBoost), and Deep Neural Network (DNN) were employed to generate susceptibility maps. Model performance was evaluated using accuracy, precision, recall, F1-score, ROC curves, and AUC-ROC metrics. Seventeen wildfire conditioning factors, categorized into four groups (topographic, meteorological, stand-related, and anthropogenic), were used to assess fire risk. Unlike previous studies, this research incorporates a region-specific anthropogenic variable: proximity to hazelnut cultivation. Feature importance scores were computed to determine the influence of each factor on fire occurrence. Additionally, SHapley Additive exPlanations (SHAP), accompanied by graphical analysis, were used to interpret the relationships between predictors and fire events. Areas of very high fire susceptibility covering 4.62% and 4.86% of the study area were identified by the RF and GBM models, both achieving an accuracy of 0.98.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105091"},"PeriodicalIF":8.6,"publicationDate":"2026-01-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145925603","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}

引用次数: 0