International journal of applied earth observation and geoinformation : ITC journal最新文献

{"title":"Early Deforestation Detection in the Tropics using L-band SAR and Optical multi-sensor data and Bayesian Statistics","authors":"Africa I. Flores-Anderson ,&nbsp;Jeffrey A. Cardille ,&nbsp;Josef Kellndorfer ,&nbsp;Franz J. Meyer ,&nbsp;Pontus Olofsson","doi":"10.1016/j.jag.2025.104831","DOIUrl":"10.1016/j.jag.2025.104831","url":null,"abstract":"<div><div>The growing availability of medium-resolution optical and radar satellite observations has prompted the development of synergistic change detection methodologies. Timely forest change detection, particularly early deforestation, is crucial for preventing illegal activities. This study proposes and evaluates an innovative model that integrates ALOS-2 PALSAR-2 L-band data with optical data from Landsat and Sentinel-2 to detect early deforestation, defined as the initial transition from stable to logged forest. Our model employs a 2-tier approach, combining harmonic curve fitting and z-scores to calculate differences between the time series. Bayesian updating statistics are then used to derive change probabilities. We comprehensively assessed the spatial and temporal detection accuracy of early deforestation maps generated by each sensor type, both individually and in combination. The integrated L-band Synthetic Aperture Radar (SAR) and optical method demonstrated the best performance, achieving a user’s accuracy of 99.19 ± 0.0081% (<span><math><mo>±</mo></math></span> 95 confidence interval) and a mean detection time lag of just 16 days. For comparison, L-band SAR data alone yielded a user’s accuracy of 93.70% (<span><math><mo>±</mo></math></span> 0.0333) with a mean time lag of 67 days, primarily due to ALOS-2’s lower repeat frequency. Optical-derived detections achieved a user’s accuracy of 98.39% (<span><math><mo>±</mo></math></span> 0.0113) and a mean time lag of 20 days. These findings confirm that combining radar and optical datasets significantly improves both detection accuracy and timeliness. Furthermore, detections were consistently captured shortly after logging activities, well before subsequent forest disturbances, underscoring true early deforestation. The high detection accuracies validate that both individual and combined L-band SAR and optical data can reliably detect early deforestation in this tropical region. We anticipate that the longer detection time lags observed with ALOS-2 PALSAR-2 will substantially improve with upcoming L-band SAR missions, such as NISAR and ALOS-4 PALSAR-3, which promise significantly enhanced global temporal sampling.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"143 ","pages":"Article 104831"},"PeriodicalIF":8.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145048969","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":"Insights into spatiotemporal dynamics and driving mechanisms of vegetation net primary productivity in African terrestrial ecosystems","authors":"Liang Liang,&nbsp;Meng Li,&nbsp;Ziru Huang,&nbsp;Qianjie Wang,&nbsp;Zhen Yang,&nbsp;Shuguo Wang","doi":"10.1016/j.jag.2025.104824","DOIUrl":"10.1016/j.jag.2025.104824","url":null,"abstract":"<div><div>Net Primary Productivity (NPP) is a critical measure of ecosystem vitality. This paper examines the spatiotemporal variation in NPP across Africa during 1981–2018 using Theil-Sen slope estimation and wavelet analysis. Sustainable change characteristics in different regions are analyzed using the Hurst exponent, and the influence of driving factors on African NPP are quantified through a structural equation model (SEM). The analysis revealed that: (1) The annual variation curve of African NPP demonstrated a fluctuating upward trajectory (p = 0.001) throughout the study period. Wavelet analysis revealed a cyclical pattern with a primary period of about 20 years, characterized by two upward and downward transitions during 1981–2018. (2) Spatial analysis indicates the distribution of NPP across Africa is centered around the equator and gradually decreases towards higher latitudes, in which the NPP of tropical rainforest and its adjacent areas increases significantly, covering 40.2 % of Africa’s area. However, Hurst exponent analysis reveals that NPP in Africa generally exhibits anti-sustainability changes, with 52.8 % of the total area potentially shifting from growth to decline in the future. (3) SEM analysis shows that NPP in Africa is mainly regulated by natural factors, particularly cumulative precipitation and temperature extremes, which exhibit the highest impact coefficient of 0.89. While topographic factors also have a substantial overall effect, their impact is primarily indirect through climate, with minimal direct influence. These findings offer a scientific foundation and policy support for sustainable development of environmental and socio-economic systems in Africa.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"143 ","pages":"Article 104824"},"PeriodicalIF":8.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145048971","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":"Utilization of multi-sensor remote sensing technologies for open pit mine monitoring","authors":"Francesco Falabella ,&nbsp;Antonio Pepe ,&nbsp;Krištof Oštir ,&nbsp;Rushaniia Gubaidullina ,&nbsp;Klemen Kozmus Trajkovski ,&nbsp;Dejan Grigillo ,&nbsp;Veronika Grabrovec ,&nbsp;Veton Hamza ,&nbsp;Polona Pavlovčič Prešeren ,&nbsp;Hannes Blaha ,&nbsp;Ana Cláudia Teodoro ,&nbsp;Fabiana Calò","doi":"10.1016/j.jag.2025.104834","DOIUrl":"10.1016/j.jag.2025.104834","url":null,"abstract":"<div><div>Nowadays, Earth Observation (EO) sensors with different technical characteristics installed on various platforms can provide data for multiple applications. Integrating, combining and processing data for various mining-related applications is required to assist decision-making and process adaptation procedures. Moreover, jointly using different data sets or products created from various sources allows for increasing precision and overcoming inherent measurement uncertainties, thereby enhancing the reliability of the results. Our research shows the potential of an integrated multi-sensor/multi-wavelength SAR data monitoring system that implements an innovative model-aided Phase Unwrapping (PhU) approach for generating ground displacement maps and time series in critical open pit mining areas. This allows us to mitigate the risk associated with rock falls and instabilities that can lead to severe damage to workers and neighbourhoods. Furthermore, the study also points out that jointly using Unmanned Aerial Vehicle (UAV) and spaceborne optical data is valuable to remotely estimate stockpile volume changes with enhanced accuracy and precision, supporting the mining companies’ management.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"143 ","pages":"Article 104834"},"PeriodicalIF":8.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145049436","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":"Evaluating the effectiveness of forest type stratification for aboveground biomass inference","authors":"Ziqiang Wu ,&nbsp;Xin Liu ,&nbsp;Shoumin Cheng ,&nbsp;Chenhui Yang ,&nbsp;Zongquan Wang ,&nbsp;Yongshuai Liu ,&nbsp;Lihu Dong ,&nbsp;Fengri Li ,&nbsp;Yuanshuo Hao","doi":"10.1016/j.jag.2025.104829","DOIUrl":"10.1016/j.jag.2025.104829","url":null,"abstract":"<div><div>Accurate quantification of aboveground biomass (AGB) in heterogeneous forest ecosystems is critical for reliable carbon cycle modeling and the effective climate policy development. Although remote sensing-assisted methods have significantly enhanced estimation efficiency, the impact of forest type stratification on estimation accuracy remains insufficiently investigated, especially when classified forest types from remote sensing data are used. In this study, we conducted a comprehensive comparison between model-assisted (MA) and model-based (MB) estimators and conventional simple random sampling (SRS) estimators under three different stratified or nonstratified scenarios: (A) a nonstratified estimation framework; (B) stratified estimation employing error-free forest type maps; and (C) stratified estimation predicated on classification results from remote sensing. Additionally, we assessed the effect of model specification—whether using a general model or strata-specific models—on estimation accuracy within stratified frameworks. The results showed that both the MA and MB estimators outperformed the SRS estimator. Stratification with ground truth reference maps significantly enhanced estimation accuracy, especially for the variance of the MB estimator employing strata-specific models is reduced from 13.65 t/ha to 10.42 t/ha, with the highest relative efficiency (RE = 2.95) achieved by the error-free stratified MA estimator using a general model. However, classification errors in remote sensing-derived maps substantially reduced these benefits, often leading to estimation variances exceeding those of the unstratified approach. Specifically, the variances of estimators MA and MB have increased from 8.89 t/ha to 24.17 t/ha, and from 10.42 t/ha to 23.65 t/ha, respectively. The predominant source of error was model misassignment due to misclassified forest types. This study provides a practical framework for estimating regional forest AGB using remote sensing data and offers decision support for the scientific formulation of forest management and sustainable utilization plans.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"143 ","pages":"Article 104829"},"PeriodicalIF":8.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144996793","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":"Mapping forest fine-grained soil particle size distributions: a holistic GeoAI approach via graph neural networks, LiDAR, and Sentinel-2","authors":"Omid Abdi,&nbsp;Ville Laamanen,&nbsp;Jori Uusitalo","doi":"10.1016/j.jag.2025.104807","DOIUrl":"10.1016/j.jag.2025.104807","url":null,"abstract":"<div><div>Fine-grained soils are crucial for assessing forest diversity and soil disturbances. Existing models for predicting particle size distributions (PSDs) often rely heavily on soil samples or lack necessary spatial dependencies, scalability and flexibility. This study introduces a holistic GeoAI model using graph neural networks (GNNs), LiDAR, and Sentinel-2 data to address these limitations. We collected 330 soil samples from 47 forest stands with a random-stratified method in southwestern Finland. The samples were pre-processed and analyzed for PSDs using a laser diffraction method, and classified into four groups: &lt;2 µm, 2–6 µm, 6–20 µm, and 20–60 µm. To increase the number of annotations, we predicted soil PSDs at unmeasured locations using CoKriging within stands. The forests were segmented into small homogeneous polygons to construct the graph layer. We mapped 61 covariates using LiDAR and Sentinel-2 based on <em>scorpan</em> model, which were then summarized into the graph layer. Subsequently, we established the pipelines of five GNN models regarding the top covariates. The results indicate that geomorphometry and organisms covariates accounted for the majority of importance. The graph attention network (GAT) recorded high stability during training and remarkable prediction accuracy after testing with R² values above 0.98 in predicting fine-grained soil PSDs across all four soil groups. Conversely, the relational graph convolutional networks (RGCN) also achieved R² values above 0.97, but with lower stability and longer training times. However, the high accuracy of the predictive models is partly due to the large number of annotations derived from CoKriging, which may introduce uncertainties. Our GAT model demonstrated strong transferability when applied to an independent test stand using CoKriging-derived data (R²: 0.98–0.99) and showed robust performance when evaluated against real ground-truth samples (R²: 0.88–0.95). The observed prediction errors (RMSE: 0.68–2.82) reflect a combination of uncertainties originating from the CoKriging training data (RMSE: 0.34–2.46) and model-induced errors during training (RMSE: 0.37–1.46). Nevertheless, the consistently high R² values indicate a strong agreement between predicted and measured soil PSDs. Future studies should focus on training the model with a larger number of ground-truth soil samples and evaluating its transferability across diverse boreal forest landscapes.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"143 ","pages":"Article 104807"},"PeriodicalIF":8.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144925318","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":"Predicting Household Income with Sentinel and Street View Imagery: A Comparative Study across Amsterdam, Sydney, and New York","authors":"Oleksandr Karasov ,&nbsp;Evelyn Uuemaa ,&nbsp;Olle Järv ,&nbsp;Monika Kuffer ,&nbsp;Tiit Tammaru","doi":"10.1016/j.jag.2025.104828","DOIUrl":"10.1016/j.jag.2025.104828","url":null,"abstract":"<div><div>In the context of urbanisation and growing disparities, timely and detailed spatial data on income inequality in cities is essential. We combined satellite imagery with streetlevel photographs provided by Google Street View to reveal the spatial distribution of household income. For this, we suggest a harmonised framework for median household income modelling based on deconstructing landscape patterns using a machine-learning approach, applied across three ’global cities’: Amsterdam, New York, and Sydney. First, we classified Sentinel-1 and Sentinel-2 mosaics and Google Street View scenes to detect functional elements of the built environment. Second, we calculated spatial indices for Sentinel imagery and visual indices for Google Street View scenes to characterise the urban landscape. Third, by combining various indicators, we trained city-specific income prediction models according to ground truth census data. The correlation between actual and predicted income in New York and Sydney reached 0.76 and 0.78, respectively. The accuracy of income prediction in Amsterdam reached 51.13%. We revealed relationships between spatial indicators of landscape patterns and spatial income distribution and recommend using Sentinel-1 and Sentinel-2 imagery as the primary data choice for income modelling in datarestricted regions. Google Street View data can be used complementarily when available.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"143 ","pages":"Article 104828"},"PeriodicalIF":8.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144996792","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":"Frequently updating DEMs based on multi-track repeat-pass InSAR observations using robust variance component estimation","authors":"Zhanpeng Cao ,&nbsp;Zefa Yang ,&nbsp;Cui Zhou ,&nbsp;Zhiwei Li","doi":"10.1016/j.jag.2025.104821","DOIUrl":"10.1016/j.jag.2025.104821","url":null,"abstract":"<div><div>Space-borne interferometric synthetic aperture radar (InSAR) is a useful technique to generate or update digital elevation models (DEMs) over large regions. Specifical InSAR missions for DEM generation/update currently work in bistatic mode. The bistatic InSAR satellites have a low temporal coverage, causing the difficulty to keep DEM products up to date. InSAR satellites working in a repeat-pass mode can offer numerous data sources with a short temporal coverage, offering a great potential to frequently update DEMs to keep DEM valid with time. However, the accuracy of repeat-pass InSAR DEMs using the existing algorithms is too low for practical uses currently. To circumvent this, we proposed a new method to frequently update DEMs from repeat-pass InSAR datasets, in order to improve update accuracy. Firstly, multi-track repeat-pass InSAR datasets were utilized to offer more redundant observations to mitigate InSAR noises. A new quantitative model was then developed to scientifically guide the exclusion of multi-track interferograms with very short spatial baselines, in order to further reduce the propagation of InSAR errors into DEM products. Thirdly, a robust variance component estimation (RVCE) algorithm, which can adaptively weight multi-track InSAR observations and automatically exclude outliers, was used to dynamically update the DEMs. The proposed method was tested over the Hambach open-pit mine in Germany. The results show that the mean accuracy of the updated DEMs is about 8.7 m, demonstrating a 60 % improvement over classical single-track repeat-pass InSAR techniques. The proposed method offers a new option to frequently update DEMs, especially over areas with changes of surface terrain.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"143 ","pages":"Article 104821"},"PeriodicalIF":8.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144925317","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":"Geological hazard susceptibility assessment and forecasting analysis based on InSAR and C-L-A model","authors":"Jie Hu,&nbsp;Zhihua Zhang,&nbsp;Xinyu Zhu,&nbsp;Xinxiu Zhang,&nbsp;Shuwen Yang,&nbsp;Chunlin Huang,&nbsp;Wei Wang,&nbsp;Xuhui Li,&nbsp;Li Hou,&nbsp;Lujia Zhao","doi":"10.1016/j.jag.2025.104840","DOIUrl":"10.1016/j.jag.2025.104840","url":null,"abstract":"<div><div>Subsidence along expressways and railways poses significant risks to transportation infrastructure safety and environmental stability. Predicting ground settlement enables enhanced understanding of deformation characteristics along transportation corridors and facilitates timely warnings for high risk areas. This study focuses on the area within 50 km of the Jishishan earthquake epicenter, employing InSAR technology to obtain preseismic and postseismic surface deformation maps. Utilizing 35 preseismic and 21 postseismic Sentinel-1A descending orbit images from 2021 to 2024, we derived surface deformation rates through PS-InSAR and SBAS-InSAR time series analysis. The preseismic annual deformation rate was incorporated with 12 influencing factors including PGA for coseismic geological hazard susceptibility evaluation. A novel Convolutional Neural Network and Long Short-Term Memory model with attention mechanism (C-L-A) was developed for settlement forecasting by integrating deformation rate data. The results show, integrated analysis incorporating InSAR derived deformation velocities and seismic dynamic factors such as PGA significantly enhances geological hazard susceptibility assessment precision. Compared to conventional static evaluation models, the novel methodology achieves a 21 % reduction in the spatial extent of very high susceptibility zones while elevating the hazard occurrence frequency ratio by 33 %, effectively mitigating false alarm risks. This approach particularly highlight extreme hazard vulnerability in areas exhibiting annual deformation rates exceeding 30 mm. Time series InSAR monitoring unequivocally delineates regional deformation patterns: significant preseismic subsidence (reaching 118 mm/year) prevailed across the study area, while the coseismic deformation field (maximum uplift: 7.85 cm) confirms a thrust type earthquake with strike slip components, tectonically linked to the South Margin Fault of the Lajishan Mountains. Persistent postseismic uplift within a 25-km radius of the epicenter reflects ongoing stress adjustment processes. The proposed C-L-A land subsidence forecasting model demonstrates superior performance across critical metrics, including Δx(MAX) = 2.94 mm, MAE = 1.74 mm, MSE = 3.39 mm², and RMSE = 1.84 mm, outperforming benchmark models (RF, CNN, LSTM, CNN-LSTM). This architecture effectively captures spatiotemporal deformation characteristics along transportation corridors, with its high accuracy short term forecasts (about 5 months) providing reliable foundations for infrastructure risk early warning systems and disaster mitigation decision support.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"143 ","pages":"Article 104840"},"PeriodicalIF":8.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145048975","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":"MSFDmap: A novel scheme to map monthly soil freeze depth in the pan-Arctic considering spatiotemporal heterogeneity in heat transfer capability","authors":"Liyuan Chen ,&nbsp;Wenquan Zhu ,&nbsp;Cunde Xiao ,&nbsp;Cenliang Zhao ,&nbsp;Hongxiang Guo","doi":"10.1016/j.jag.2025.104820","DOIUrl":"10.1016/j.jag.2025.104820","url":null,"abstract":"<div><div>Accurately characterizing the spatiotemporal dynamics of soil freeze depth (SFD) is critical for understanding the response of frozen soils to climate change. Existing SFD mapping schemes mainly focus on annual maximum values, rarely address monthly variations, and fail to capture both spatiotemporal heterogeneity and physical constraints. We developed a monthly SFD mapping scheme (MSFDmap) that considers spatiotemporal heterogeneity in heat transfer capability. Based on the simplified Stefan equation, which is physically constrained by energy conservation, MSFDmap first predicts the spatial distribution of monthly heat transfer factor (HTF) using a random forest regression model driven by soil clay content, precipitation, soil bulk density, soil organic carbon content, soil water content, and leaf area index, and then maps monthly SFD. MSFDmap was implemented using 2123 site-month observations from 60 pan-Arctic sites over 20 years. Results show that MSFDmap achieves a root mean square error (RMSE) of 19.21 cm and an R² of 0.91 for monthly SFD estimates, reducing RMSE by 24–55 % and improving R² by 8–65 % over existing schemes. For monthly SFD averaged across sites, estimates exhibit strong temporal agreement with quasi-true SFD series (Pearson correlation coefficient <em>r</em> = 0.99, RMSE = 9.13 cm). The MSFDmap-derived SFD distribution exhibits expected latitudinal and altitudinal gradients, with <em>r</em> = 0.60 relative to an ERA5-Land-based reference distribution. These results demonstrate that MSFDmap effectively characterizes the spatiotemporal dynamics of monthly SFD and outperforms existing schemes. It is attributed to the capture of heterogeneous HTF, which enables the representation of SFD heterogeneity under physical constraints.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"143 ","pages":"Article 104820"},"PeriodicalIF":8.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144920027","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":"In the search for optimal multi-view learning models for crop classification with global remote sensing data","authors":"Francisco Mena ,&nbsp;Diego Arenas ,&nbsp;Andreas Dengel","doi":"10.1016/j.jag.2025.104823","DOIUrl":"10.1016/j.jag.2025.104823","url":null,"abstract":"<div><div>Studying and analyzing cropland is a difficult task due to its dynamic and heterogeneous growth behavior. Usually, diverse data sources can be collected for its estimation. Although deep learning models have proven to excel in the crop classification task, they face substantial challenges when dealing with multiple inputs, named multi-modal or Multi-View Learning (MVL). The methods used in the MVL scenario can be structured based on the encoder architecture, the fusion strategy, and the optimization technique. Here, the literature has primarily focused on using specific encoder architectures for local regions, lacking a deeper exploration of other components in the MVL methodology. In contrast, we investigate the simultaneous selection of the fusion strategy and encoder architecture, assessing global-scale cropland and crop-type classifications. We use a range of five fusion strategies (Input, Feature, Decision, Ensemble, Hybrid) and five temporal encoders (LSTM, GRU, TempCNN, TAE, L-TAE) as possible configurations in the MVL method. We use the CropHarvest dataset for validation, which provides optical, radar, weather time series, and topographic information as input data. We found that in scenarios with a limited number of labeled samples, a unique configuration is insufficient for all cases. Instead, a specific combination should be meticulously sought, including an encoder and fusion strategy. To streamline this search process, we suggest identifying the optimal encoder architecture tailored for a particular fusion strategy and then determining the most suitable fusion strategy for the classification task. We provide a standardized model schema for the exploration of crop classification through an MVL methodology. jn</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"143 ","pages":"Article 104823"},"PeriodicalIF":8.6,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145019253","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}