Remote Sensing Applications-Society and Environment最新文献

{"title":"Investigation of evolutionary fuzzy MPCM machine learning and probabilistic SVM models for Butea monosperma species mapping","authors":"Payel Mani ,&nbsp;Dipanwita Dutta ,&nbsp;Anil Kumar","doi":"10.1016/j.rsase.2025.101667","DOIUrl":"10.1016/j.rsase.2025.101667","url":null,"abstract":"<div><div>Accurate identification of plant species at an optimal level of precision remains a major challenge in ecological observations when using conventional classification methods. This study explores the potentiality of multi-temporal datasets with machine learning classifiers for the identification and distribution of Butea monosperma tree species, a native floral species grown in many countries of South and Southeast Asia. For identifying optimum combinations of temporal images the Euclidian distance-based separability analysis was employed on the multi-temporal GCI, MSAVI2 indices database (24 toatal temporal dates). This study uses the fuzzy Modified Possibilistic <em>c-</em>Mean (MPCM) classification method combined with the green chlorophyll (GCl), MSAVI2 temporal index to handle the complexity and uncertainty inherent in the phenological data. Owing to its lesser variance on the testing target species over the other classes, the 21 optimum temporal combinations of GCl images were chosen as a benchmark for comparison of the output with the Probabilistic Support Vector Machine (PSVM) with Radial Basis Function (RBF) kernel approach machine learning classifier which is well known for its ability to handle probabilistic information and high-dimensional data. In this study, a diverse dataset of tree species phenological observations has been employed to evaluate the performance of both classifiers. Key metrices such as overall accuracy and F1-score were utilized for the comparison of different models. The MPCM classifier achieved notable performance, with 92 % overall accuracy and an F1-score of 0.93 when utilizing the 21-temporal GCI database. In contrast, a single-date output resulted in only 65% overall accuracy and an F1-score of 0.74. When compared to PSVM model, which exhibits an F-score of 0.88 and an overall accuracy of 82 %, the utilization of MPCM with combined 21 temporal GCI indices demonstrated superior classification performance. Additionally, this research provides insights into how various evolutionary strategies and algorithms can enhance the classifiers’ adaptability to changing data distributions.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"39 ","pages":"Article 101667"},"PeriodicalIF":4.5,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144750405","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":"Decoding urban complexity: Deep learning-based terrain-specific building segmentation for Indian cities","authors":"Akshit Koduru ,&nbsp;Reedhi Shukla","doi":"10.1016/j.rsase.2025.101673","DOIUrl":"10.1016/j.rsase.2025.101673","url":null,"abstract":"<div><div>Accurate building segmentation from satellite imagery is essential for urban planning, disaster management, and environmental monitoring. This paper presents a novel approach utilizing the UNET architecture for deep learning-based building segmentation, focusing on diverse terrains in Indian cities. Indian cities are uniquely complex in their urban complexity because of a highly densely packed urban landscape and patterns of buildings, thus making segmentation a challenging task. Our method includes meticulously performed data preprocessing and exhaustive validation to achieve high accuracy and adaptability in our trained terrain-based model. Very high-resolution satellite imagery with a 0.5-m spatial resolution was utilized for model training. Specialized models were developed for different terrain types—urban, coastal, and hilly—resulting in significant improvements in segmentation performance compared to generalist models. We reduce human effort and increase efficiency as the proposed system automates segmentation. Such research will, therefore, scale the solution very well in building segmentation. Its application will be practical to aspects of urban planning and disaster response while developing the smart city, and further work will be oriented towards expanding the dataset and generalizing and further developing the model.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"39 ","pages":"Article 101673"},"PeriodicalIF":4.5,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144841965","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":"A machine learning algorithm to retrieve the red peak of phytoplankton absorption spectra from ocean-colour remote sensing","authors":"Mohammad Ashphaq,&nbsp;Shovonlal Roy","doi":"10.1016/j.rsase.2025.101702","DOIUrl":"10.1016/j.rsase.2025.101702","url":null,"abstract":"<div><div>Light absorption by microscopic phytoplankton in marine ecosystems is a crucial process underpinning biological production and global biogeochemical cycles. Accurate estimation of phytoplankton absorption coefficients, an inherent optical property of ocean water, can improve remote sensing applications including spectral photosynthesis models and assessments of ocean health, biodiversity, and climate change impacts. However, considerable uncertainty exists in current satellite retrievals of phytoplankton absorption coefficients, particularly for <em>ɑ</em>_<em>ph</em>(676) - the phytoplankton absorption peak at red wavelengths near 676 nm - which is an input to several novel and advanced satellite algorithms. This uncertainty hinders operational use of algorithms for assessing phytoplankton physiology, size structure and oceanic carbon pools from space. We aimed to improve satellite-based estimation of <em>ɑ</em>_<em>ph</em> (676) using advanced machine learning (ML) techniques. We compiled a comprehensive <em>in situ</em> dataset (n = 1576) of <em>ɑ</em>_<em>ph</em>(676) from published databases and matched with remote-sensing reflectance <em>Rrs</em> at six wavelengths (412, 443, 490, 510, 560, and 665 nm) from the Ocean Colour Climate Change Initiative. We extensively evaluated multiple base ML algorithms: Random Forest (RF), Gradient Boosting Machines, and Linear Regression; and implemented ensemble ML models: RF with Grid Search Cross-Validation, eXtreme Gradient Boosting Ensembled Model, Ensemble Forecast, Stacked Voting, Optimised Ensemble and Meta Stacking, integrating the base models through cross-validated hyperparameter tuning. Meta Stacking outperformed individual ML models in predictive accuracy across temporal resolutions, showing best results with daily composites. Our study addresses key limitations of previous models, including small training datasets, inconsistent performances, and lack of ensemble comparisons. We present a robust, extensively trained and validated ensemble ML model that significantly improves <em>ɑ</em>_<em>ph</em>(676) estimation and opens the possibility of routinely using in ocean colour remote sensing.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"39 ","pages":"Article 101702"},"PeriodicalIF":4.5,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144988697","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 automated approach to generate reliable training samples without field survey labels for large-scale crop mapping","authors":"Shuo Zhang ,&nbsp;Xueqian Hu ,&nbsp;Li Li ,&nbsp;Xiaochuang Yao ,&nbsp;Yujiao Hong ,&nbsp;Bingqi Qiu ,&nbsp;Jianxi Huang","doi":"10.1016/j.rsase.2025.101693","DOIUrl":"10.1016/j.rsase.2025.101693","url":null,"abstract":"<div><div>Accurate and timely data on crop spatial distribution are essential for predicting grain production and assessing agricultural sustainability. However, large-scale crop mapping accuracy is often hindered by insufficient high-quality training samples. Particularly in regions with limited prior knowledge, the automated and repeatable generation of sufficient and reliable training samples remains challenging. To address this, we propose a novel method for generating abundant, well-distributed, representative, and reliable crop training samples directly from remote sensing imagery without needing field survey labels. Firstly, a multi-stage spatially uniform sampling model is designed to generate multi-class crop candidate samples from time-series imagery, leveraging phenological differences of land cover within local regions and spatially constrained uniform sampling strategies. Subsequently, a progressive sample purification framework is introduced to reduce the noise through intra-class neighborhood and inter-class spectral characteristics for each candidate sample, resulting in training samples with stable positions and labels. Finally, the quality of these samples and the performance of the proposed method are assessed and discussed using visual interpretation and machine learning methods. Validation in Northeast China showed the method's effectiveness in producing sufficient and high-quality crop samples from Sentinel-2 imagery, achieving a high visual similarity (&gt;93 %) to Google Earth high-resolution imagery. Applying generated samples to mainstream machine learning classifiers achieved overall accuracy and Kappa coefficients above 93.81 % and 0.9083, respectively. These results demonstrate the proposed method's effectiveness for accurate crop mapping, providing an automated, flexible, and valuable solution for large-scale crop monitoring across diverse regions.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"39 ","pages":"Article 101693"},"PeriodicalIF":4.5,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144879588","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":"Estimate of sugarcane productivity using machine learning algorithm from time series of WFI/CBERS-4 and WPM/CBERS-4A time series","authors":"João Pedro de Sousa Costa ,&nbsp;Ieda Del’Arco Sanches ,&nbsp;Luiz Gabriel da Silva ,&nbsp;Max Well de Oliveira Rabelo ,&nbsp;Ana Cláudia dos Santos Luciano","doi":"10.1016/j.rsase.2025.101689","DOIUrl":"10.1016/j.rsase.2025.101689","url":null,"abstract":"<div><div>Brazil is the world's largest sugarcane producer, meaning productivity monitoring is crucial to enable mills to plan for future seasons and make decisions during the harvest. This results in time, labor, and resource savings. This study aims to leverage machine learning models and spectral data from the Wide Swath Multispectral and Panchromatic Camera (WPM/CBERS-4A) and the Wide Field Imager (WFI/CBERS-4) to estimate sugarcane productivity at various stages of crop development. The study focuses on 6229 sugarcane fields in the Araçatuba region of São Paulo. Productivity data for these fields from the 2020 to 2022 harvest, were provided by a partnering sugarcane mill. A time series of spectral data (bands and the vegetation indices) from the two sensors were used, combined with meteorological, water balance and agronomic data. From these variables, two distinct datasets were created, one for each sensor (WPM and WFI).</div><div>Sixteen empirical models were developed for each dataset, each representing different stages of sugarcane development, both for plant cane (PC) and ratoon cane (RC), giving 32 models. The models were created iteratively, where each model was developed from the input data set of the previous and current month, allowing estimates throughout crop development along with choosing the most important variables. Data processing included sensor cross-calibration, cloud removal, vegetation index calculation, and integration of meteorological data. The models were trained for different phenological stages of the crops, considering variables such as precipitation, solar radiation, and the number of cuts. The Random Forest model was chosen for its robustness in dealing with large volumes of data, ability to capture complex relationships between variables, and resistance to overfitting.</div><div>Additionally, the Nemenyi post-hoc test for mean comparison was applied. Upon analyzing the results, it was observed that for both the WFI and WPM models, the optimal stage to begin productivity estimations is from the 6th/7th month for plant cane and the 3.5th/4th month for ratoon cane. Before these time points, the predicted data statistically differed from those observed. Models using data from the complete crop development cycle yielded the best results, which achieved a Coefficient of determination (R²) of 0.63, a modified Willmott index (d_mod) of 0.70, and a root mean square error (RMSE) of 10.34 t ha⁻¹. Thus, data from the WPM/CBERS-4A and WFI/CBERS-4 sensors integrated with additional data demonstrated a promising potential for predicting sugarcane productivity.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"39 ","pages":"Article 101689"},"PeriodicalIF":4.5,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144886747","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":"Spatio-temporal net primary productivity prediction in western Sanjiangyuan region using deep learning with spatial aggregation model and non-local attention mechanism","authors":"Fujiang Liu ,&nbsp;Zhe Zhu ,&nbsp;Weihua Lin ,&nbsp;Yan Guo ,&nbsp;Bo Li ,&nbsp;Kurelijiang Aymer ,&nbsp;Ziheng Cao ,&nbsp;Jun Zheng","doi":"10.1016/j.rsase.2025.101714","DOIUrl":"10.1016/j.rsase.2025.101714","url":null,"abstract":"<div><div>As a critical component of the global carbon cycle and climate regulation, terrestrial ecosystems' net primary productivity (NPP) necessitates accurate monthly prediction. However, most existing studies neglect the index's spatial aggregation and dependence, thereby limiting prediction accuracy. This study proposes GOG-PredRNN-NL, a novel spatio-temporal deep learning model, using monthly NPP imagery data from the western Sanjiangyuan region (2000–2020). The model integrates the PredRNN architecture with a nonlocal attention block (NL) and spatial agglomeration modeling. It employs the Getis-Ord Gi∗ statistic to capture NPP's spatial aggregation features and the NL block to establish long-range dependencies. Compared with six baseline models (SVM, RF, MLP, GCN, LSTM, and GRU), GOG-PredRNN-NL significantly outperforms them, reducing MAE by 21 %–29 % and RMSE by 16 %–46 %, while increasing R² by 11 %–32 %. Ablation experiments further validate the effectiveness of input features. Collectively, this model provides a high-precision NPP prediction framework and offers an innovative approach for future spatio-temporal model optimization.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"39 ","pages":"Article 101714"},"PeriodicalIF":4.5,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145019183","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":"AttUNet+: Towards high-fidelity building footprints","authors":"Davood Saadati ,&nbsp;Hong Hao","doi":"10.1016/j.rsase.2025.101715","DOIUrl":"10.1016/j.rsase.2025.101715","url":null,"abstract":"<div><div>Automated extraction of building footprints from high-resolution aerial imagery is recognized as underpinning critical applications in urban planning, disaster response, and environmental monitoring. A lightweight encoder–decoder model, AttUNet+, has been developed by fusing attention gates, Atrous Spatial Pyramid Pooling (ASPP), and Squeeze-and-Excitation (SE) blocks into a unified architecture. When trained and evaluated on the AIRS dataset, 97.3 % overall accuracy, 0.880 IoU, 0.935 Dice, and 0.919 MCC were achieved, with 94.8 % precision and 92.4 % recall. Boundary fidelity was confirmed by a mean area-difference of 567 px and a surface-to-surface distance of 2.03 px, while an AUC-ROC of 0.955 indicated robust threshold discrimination. Compared to a state-of-the-art ViT baseline, improvements of +0.5 pp in IoU and recall were delivered, despite 60 % fewer parameters (38 M vs. 97 M), 72 % less GPU memory usage (161 MB vs. 581 MB), and 23 % faster inference (0.0907 s vs. 0.117 s per 256 × 256 tile). By combining best-in-class segmentation quality with low computational cost, AttUNet+ is positioned as a practical, scalable solution for large-scale urban mapping, real-time UAV deployment, and cloud-based disaster assessment pipelines.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"39 ","pages":"Article 101715"},"PeriodicalIF":4.5,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145007665","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":"Urban three-dimension green quantity Estimation: An approach utilizing UAV, satellite imagery, and machine learning","authors":"Saiqiang Li ,&nbsp;Weiqiang Li ,&nbsp;Minfen Yu ,&nbsp;Daosheng Chen ,&nbsp;Mingshan Xu ,&nbsp;Min Ren ,&nbsp;Xiaodong Yang","doi":"10.1016/j.rsase.2025.101691","DOIUrl":"10.1016/j.rsase.2025.101691","url":null,"abstract":"<div><div>As a critical indicator of urban vegetation structural complexity and ecological function, three-dimensional green quantity (3DGQ) provides a more spatially detailed and holistic assessment of urban ecosystem health compared than traditional two-dimensional green metrics. However, current methods for estimating 3DGQ are constrained by cost and data accuracy, which restricts their applicability in large-scale urban green volume assessments. This study presents an innovative multi-source data fusion method to calculate 3DGQ (dependent variable) using tree structural parameters extracted from high-resolution UAV imagery. It integrates 18 vegetation indices (independent variables) derived from the Sentinel-2 imagery, yielding 608 3DGQ units. Six machine learning models are employed for estimation, with the best-performing model subsequently used to estimate 3DGQ in Ningbo's built-up areas. The results showed that UAV remote sensing imagery was highly effective in extracting tree height and crown dimensions (<em>R</em>² ≥ 0.84, 1.07m &gt; <em>RMSE</em> &gt; 0.71m, <em>P</em> &lt; 0.001). The coefficient of determination (<em>R</em>²) for the six machine learning models ranged from 0.67 to 0.85, and the root mean square error (<em>RMSE</em>) varied from 74.02 m³/pixel to 109.91 m³/pixel, indicating overall robust estimation performance. However, the Random Forest model (RF) achieved the best performance in predicting 3DGQ (<em>R</em>² = 0.85, <em>RMSE</em> = 74.02 m³/pixel, <em>nRMSE</em> = 9.49 %), significantly outperforming other models. After feature importance analysis, nine key independent variables were selected, and the simplified RF maintained high predictive accuracy (<em>R</em>² = 0.84, <em>RMSE</em> = 75.23 m³/pixel, <em>nRMSE</em> = 9.64 %). The simplified model was then applied to estimate 3DGQ of Ningbo's urban built-up areas. The results showed that at a 10 m × 10 m scale, 3DGQ values ranged from 0 m³ to 770 m³, while at a 100 m × 100 m scale, they ranged from 0 m³/ha to 75,276 m³/ha. Our method establishes an efficient and scalable framework for large-scale urban 3DGQ estimation, providing valuable insights into urban ecosystem service assessments and green space management.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"39 ","pages":"Article 101691"},"PeriodicalIF":4.5,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144865775","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":"Contribution of topographical and morphological parameters in glacier response to change in climate in the Western Himalaya","authors":"Supratim Guha ,&nbsp;Anugrah Pratap ,&nbsp;Reet Kamal Tiwari ,&nbsp;Tanisha Ghosh ,&nbsp;Deepali Gaikwad","doi":"10.1016/j.rsase.2025.101643","DOIUrl":"10.1016/j.rsase.2025.101643","url":null,"abstract":"<div><div><span>Glacier thinning patterns are crucial in determining the availability and distribution of meltwater in Himalayan catchments, making precise assessments essential for predicting future water resources and formulating effective water management strategies. </span><strong>This study aims to</strong> <em>quantify the specific influence of key topographic and morphological parameters on glacier thinning variability</em> <strong>in the</strong> <span><em>Himachal </em><em>Himalaya</em><em> and Kashmir Himalaya</em></span> <strong>regions.</strong> Using <strong>Multiple Linear Regression,</strong> we systematically evaluate how <strong>maximum elevation, mean elevation, snout elevation, slope, aspect, glacier area, terminus type, and debris cover</strong> control variations in glacier thinning. <strong>To refine the model, a</strong> <em>backward stepwise subset selection technique</em> <strong>was employed to identify the most significant predictors, followed by least squares regression to quantify their contributions.</strong></div><div>The results reveal that in the <strong>Himachal Himalaya, glacier terminus type, mean elevation, and slope</strong> play dominant roles in influencing glacier thinning. <strong>Lake-terminating glaciers experience</strong> <em>0.23 ± 0.09 m/year more thinning</em> <strong>than land-terminating glaciers, independent of other factors.</strong> Glaciers located <strong>100 m higher in elevation exhibit</strong> <em>0.01 ± 0.004 m/year less thinning</em>, while a <strong>10 % increase in slope is associated with</strong> <em>0.19 ± 0.03 m/year less thinning</em>.</div><div>In the <strong>Kashmir Himalaya, debris cover extension, snout elevation, and glacier slope</strong> were identified as the primary controls on thinning heterogeneity. <strong>A</strong> <em>10 % increase in debris cover extension</em> <strong>correlates with a</strong> <em>0.08 ± 0.02 m/year reduction</em> <strong>in thinning, irrespective of other factors.</strong> Glaciers with <strong>snouts located 100 m higher in elevation experience</strong> <em>0.01 ± 0.003 m/year more thinning</em>, even when controlling for debris cover and slope. Additionally, <strong>a</strong> <em>10 % increase in slope</em> <strong>corresponds to</strong> <em>0.28 ± 0.04 m/year less thickness change</em> <strong>compared to lower-sloped glaciers.</strong></div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"39 ","pages":"Article 101643"},"PeriodicalIF":4.5,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145094819","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":"Attributing vegetation disturbance change agent from Landsat time series in the arid and semi-arid ecosystem of Qilian Mountains, China","authors":"Lipeng Jiao ,&nbsp;Randolph H. Wynne ,&nbsp;Liqin Han ,&nbsp;Pi Chen ,&nbsp;Yaonan Zhang ,&nbsp;Feng Yang","doi":"10.1016/j.rsase.2025.101720","DOIUrl":"10.1016/j.rsase.2025.101720","url":null,"abstract":"<div><div>Vegetation disturbances can fundamentally alter ecosystem structure and function, with profound implications for carbon emissions and sequestration. While disturbance attribution has been extensively studied in forest ecosystems, its application to other land cover types, particularly pasture-dominated systems, remains underexplored. Here, we introduce an approach—Shapelet-based Vegetation Change Detection with Random Forest (SVCD-RF)— that integrates time-series analysis with a Random Forest (RF) machine learning algorithm to attribute causal disturbance agents in mixed pasture and forest ecosystems. We apply the SVCD algorithm to Landsat time-series data in the arid and semi-arid Qilian Mountains of China, detecting vegetation disturbances and categorizing them into three phases: pre-disturbance, onset, and post-disturbance. For each disturbed pixel, we derive phase-specific spectral and temporal metrics, along with topographic variables, to construct feature sets. These are linked to 1800 purposively sampled and manually interpreted training labels, used to train separate RF classifiers for forest and pasture areas. Disturbance agents are classified into Development or Degradation (DD), Agricultural Expansion (AE), and Greenup. Classification accuracy is evaluated via stratified random sampling, generating 1137 and 1849 validation labels for pasture and forest, respectively. The resulting map achieves high accuracy, with overall accuracies of 89.1 % ± 1.8 % for pasture and 85.6 % ± 1.5 % for forest at the 95 % confidence level. User's and producer's accuracies for most classes exceed 80 %, except for AE in the forest, yielding lower values (56.3 % and 61.2 %, respectively). Stratified area estimates indicate that AE was the dominant disturbance agent before 2012, especially in pasture ecosystems, with an estimated area of 1107.90 ± 51.14 km², significantly surpassing DD (818.58 ± 24.71 km²) and Greenup (713.36 ± 32.10 km²). In contrast, Greenup emerges as the dominant disturbance process after 2012, particularly within forest ecosystems. We underscore the potential of the SVCD-RF approach for mapping vegetation disturbance agents in complex, mixed-use landscapes.</div></div>","PeriodicalId":53227,"journal":{"name":"Remote Sensing Applications-Society and Environment","volume":"39 ","pages":"Article 101720"},"PeriodicalIF":4.5,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145007666","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}