Pub Date : 2025-02-01DOI: 10.1016/j.jag.2025.104381
Shangshu Cai , Yong Pang
Dividing a forest point cloud dataset into tiles is a common practice in point cloud processing (e.g., individual tree segmentation), aimed at addressing memory constraints and optimizing processing efficiency. Existing methods typically utilize automatic regular clipping (e.g., rectangular clipping), which tends to result in splitting tree crowns along the cutting lines. To preserve the completeness of tree crowns within predefined clipping boundaries (e.g., rectangles), we develop a tree crown edge-aware (E-A) point cloud clipping algorithm, named E-A algorithm. Firstly, the crown edge and distance features are enhanced and quantified using mathematical morphology and nearest neighbor pixel methods. Then, these two features are linearly weighted and integrated for cutting line detection. Finally, the optimal cutting lines are detected by exploring a set of edges with the minimum sum of integrated feature values. E-A algorithm was tested with airborne LiDAR point clouds collected from China’s Saihanba Forest Farm, comparing it against regular clipping methods. The results indicate that E-A algorithm can automatically and effectively emphasize preserving tree crown completeness within predefined clipping boundaries. It reduces crown fragmentation errors by 73.29% on average and maintains an average area difference of 6.42% compared to regular clippings. This algorithm provides a crucial tool for forest point cloud applications.
{"title":"A tree crown edge-aware clipping algorithm for airborne LiDAR point clouds","authors":"Shangshu Cai , Yong Pang","doi":"10.1016/j.jag.2025.104381","DOIUrl":"10.1016/j.jag.2025.104381","url":null,"abstract":"<div><div>Dividing a forest point cloud dataset into tiles is a common practice in point cloud processing (e.g., individual tree segmentation), aimed at addressing memory constraints and optimizing processing efficiency. Existing methods typically utilize automatic regular clipping (e.g., rectangular clipping), which tends to result in splitting tree crowns along the cutting lines. To preserve the completeness of tree crowns within predefined clipping boundaries (e.g., rectangles), we develop a tree crown edge-aware (E-A) point cloud clipping algorithm, named E-A algorithm. Firstly, the crown edge and distance features are enhanced and quantified using mathematical morphology and nearest neighbor pixel methods. Then, these two features are linearly weighted and integrated for cutting line detection. Finally, the optimal cutting lines are detected by exploring a set of edges with the minimum sum of integrated feature values. E-A algorithm was tested with airborne LiDAR point clouds collected from China’s Saihanba Forest Farm, comparing it against regular clipping methods. The results indicate that E-A algorithm can automatically and effectively emphasize preserving tree crown completeness within predefined clipping boundaries. It reduces crown fragmentation errors by 73.29% on average and maintains an average area difference of 6.42% compared to regular clippings. This algorithm provides a crucial tool for forest point cloud applications.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104381"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143050024","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2025.104374
Qi Li , Xingyuan Zu , Ming Zhang , Jinghua Li , Yan Feng
Underwater hyperspectral target detection (HTD) technology holds pivotal value in enhancing maritime military power. However, the absorption and scattering properties of the water bodies result in the inevitable issue of mixed pixels in underwater hyperspectral images (HSIs). To address the issue, a joint hyperspectral unmixing and target detection network for underwater HSI is proposed, denoted as HUTDNet, which utilizes the material type and abundance information for downstream semantic tasks. Specifically, a nonlinear underwater unmixing network is designed to extract pure underwater endmembers and their associated abundance information, which is essential in assisting the subsequent target detection task. The network also extracts underwater virtual endmembers and their abundance values to reconstruct a more realistic underwater HSI. Then, the abundance weighting module determines the abundance weighting factor by calculating the spectral distance between a priori target spectra and the estimated underwater pure endmembers, generating a weighted abundance map. Finally, due to the inherent limitations in the characterization capabilities of abundance maps and endmembers, the detection network extracts key spectral feature maps from the input underwater HSI. These feature maps serve as complementary terms, fused with the original and weighted abundance maps. Subsequently, convolutional and fully connected layers are employed to extract deeper features and generate the target detection maps. Experiments on both real and synthetic datasets demonstrate the superior performance and efficiency of the proposed method in this paper compared to other state-of-the-art methods.
{"title":"HUTDNet: A joint unmixing and target detection network for underwater hyperspectral imagery","authors":"Qi Li , Xingyuan Zu , Ming Zhang , Jinghua Li , Yan Feng","doi":"10.1016/j.jag.2025.104374","DOIUrl":"10.1016/j.jag.2025.104374","url":null,"abstract":"<div><div>Underwater hyperspectral target detection (HTD) technology holds pivotal value in enhancing maritime military power. However, the absorption and scattering properties of the water bodies result in the inevitable issue of mixed pixels in underwater hyperspectral images (HSIs). To address the issue, a joint hyperspectral unmixing and target detection network for underwater HSI is proposed, denoted as HUTDNet, which utilizes the material type and abundance information for downstream semantic tasks. Specifically, a nonlinear underwater unmixing network is designed to extract pure underwater endmembers and their associated abundance information, which is essential in assisting the subsequent target detection task. The network also extracts underwater virtual endmembers and their abundance values to reconstruct a more realistic underwater HSI. Then, the abundance weighting module determines the abundance weighting factor by calculating the spectral distance between a priori target spectra and the estimated underwater pure endmembers, generating a weighted abundance map. Finally, due to the inherent limitations in the characterization capabilities of abundance maps and endmembers, the detection network extracts key spectral feature maps from the input underwater HSI. These feature maps serve as complementary terms, fused with the original and weighted abundance maps. Subsequently, convolutional and fully connected layers are employed to extract deeper features and generate the target detection maps. Experiments on both real and synthetic datasets demonstrate the superior performance and efficiency of the proposed method in this paper compared to other state-of-the-art methods.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104374"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142990287","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2025.104396
Jianfeng Gao , Qingyan Meng , Linlin Zhang , Xinli Hu , Die Hu , Jiangkang Qian
Influenced by lockdown policies and anomalies in human activities, emergencies such as pandemic significantly altered the urban thermal environment. However, the spatiotemporal heat anomaly changes across and within cities during emergencies and their drivers have not been fully investigated. This study quantified the changes in the urban thermal environment in China before and during the COVID-19 pandemic. Based on z-scores and multiscale geographically weighted regression models, heat anomaly changes and transfer patterns of different land uses in cities with varying degrees of pandemic impact and drivers were estimated. During the entire year, we found that although the pandemic significantly reduced surface urban heat island intensity during 5 % to 35 % of days, it did not change significantly throughout 2020. During the first-level public health emergency response, the land surface temperatures of residential and commercial lands notably affected by the pandemic decreased by −0.195°C and −0.371°C, and the shifting of strong heat anomaly zones in industrial lands increased heat anomaly and no heat anomaly zones by 6.1 % and 1.4 %, respectively. Furthermore, thermal anomalies were highly correlated with changes in biophysical parameters during the pandemic. These findings provide insights and mitigation strategies for the fluctuations in the urban thermal environment caused by emergencies.
{"title":"Modeling the impact of pandemic on the urban thermal environment over megacities in China: Spatiotemporal analysis from the perspective of heat anomaly variations","authors":"Jianfeng Gao , Qingyan Meng , Linlin Zhang , Xinli Hu , Die Hu , Jiangkang Qian","doi":"10.1016/j.jag.2025.104396","DOIUrl":"10.1016/j.jag.2025.104396","url":null,"abstract":"<div><div>Influenced by lockdown policies and anomalies in human activities, emergencies such as pandemic significantly altered the urban thermal environment. However, the spatiotemporal heat anomaly changes across and within cities during emergencies and their drivers have not been fully investigated. This study quantified the changes in the urban thermal environment in China before and during the COVID-19 pandemic. Based on z-scores and multiscale geographically weighted regression models, heat anomaly changes and transfer patterns of different land uses in cities with varying degrees of pandemic impact and drivers were estimated. During the entire year, we found that although the pandemic significantly reduced surface urban heat island intensity during 5 % to 35 % of days, it did not change significantly throughout 2020. During the first-level public health emergency response, the land surface temperatures of residential and commercial lands notably affected by the pandemic decreased by −0.195°C and −0.371°C, and the shifting of strong heat anomaly zones in industrial lands increased heat anomaly and no heat anomaly zones by 6.1 % and 1.4 %, respectively. Furthermore, thermal anomalies were highly correlated with changes in biophysical parameters during the pandemic. These findings provide insights and mitigation strategies for the fluctuations in the urban thermal environment caused by emergencies.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104396"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143083299","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2024.104334
Muhammad Fahad Baqa , Linlin Lu , Huadong Guo , Xiaoning Song , Seyed Kazem Alavipanah , Syed Nawaz-ul-Huda , Qingting Li , Fang Chen
Due to the compounding impacts of urbanization and climate change-induced warming, urban inhabitants face increasing risks of thermal health issues. The use of high-resolution maps that categorize intra-urban thermal environment and Local Climate Zones (LCZ) could enhance the understanding of the correlation between heat-related health risks and microclimates. In this study, a fine-scale heat risk assessment framework was applied in an arid megacity, Karachi, Pakistan. Following Crichton’s Risk Triangle framework, heat health risks were mapped by considering hazard-exposure-vulnerability components at the census ward level. The heat hazard was mapped using SDGSAT-1 thermal infrared data at a 30 m spatial resolution during summer season. Factors contributing most to heat vulnerability were identified as the availability of electricity facilities, bathroom facilities, and housing density, with contribution rates of 47.51 %, 21.86 %, and 8.07 %, respectively. Heat risks were considerably higher for built types (0.16) compared to natural LCZ types (0.07), with 65 % of LCZ 2, 3, 6, and 7 (compact mid-rise, compact low-rise, open low-rise, and lightweight low-rise areas) identified as high-risk areas. To mitigate heat risks, green space should be planned in LCZ2 and LCZ3 characterized by dense population and compact buildings arrangement, and public cooling facilities and infrastructure should be improved in LCZ7 featured with squatter and slum settlements. Urban planners may consider restricting the growth of these areas in newly-developed regions, including encroachments and unplanned settlements, to prevent further exacerbation of heat stress. This study offers a valuable guide for assessing and alleviating heat risks at the community level, thereby promoting the development of heat resilient urban areas.
{"title":"Investigating heat-related health risks related to local climate zones using SDGSAT-1 high-resolution thermal infrared imagery in an arid megacity","authors":"Muhammad Fahad Baqa , Linlin Lu , Huadong Guo , Xiaoning Song , Seyed Kazem Alavipanah , Syed Nawaz-ul-Huda , Qingting Li , Fang Chen","doi":"10.1016/j.jag.2024.104334","DOIUrl":"10.1016/j.jag.2024.104334","url":null,"abstract":"<div><div>Due to the compounding impacts of urbanization and climate change-induced warming, urban inhabitants face increasing risks of thermal health issues. The use of high-resolution maps that categorize intra-urban thermal environment and Local Climate Zones (LCZ) could enhance the understanding of the correlation between heat-related health risks and microclimates. In this study, a fine-scale heat risk assessment framework was applied in an arid megacity, Karachi, Pakistan. Following Crichton’s Risk Triangle framework, heat health risks were mapped by considering hazard-exposure-vulnerability components at the census ward level. The heat hazard was mapped using SDGSAT-1 thermal infrared data at a 30 m spatial resolution during summer season. Factors contributing most to heat vulnerability were identified as the availability of electricity facilities, bathroom facilities, and housing density, with contribution rates of 47.51 %, 21.86 %, and 8.07 %, respectively. Heat risks were considerably higher for built types (0.16) compared to natural LCZ types (0.07), with 65 % of LCZ 2, 3, 6, and 7 (compact mid-rise, compact low-rise, open low-rise, and lightweight low-rise areas) identified as high-risk areas. To mitigate heat risks, green space should be planned in LCZ2 and LCZ3 characterized by dense population and compact buildings arrangement, and public cooling facilities and infrastructure should be improved in LCZ7 featured with squatter and slum settlements. Urban planners may consider restricting the growth of these areas in newly-developed regions, including encroachments and unplanned settlements, to prevent further exacerbation of heat stress. This study offers a valuable guide for assessing and alleviating heat risks at the community level, thereby promoting the development of heat resilient urban areas.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104334"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142874794","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper introduces a method for estimating forest above-ground biomass (AGB) using the Interferometric SAR (InSAR)-based Phase Histogram (PH) technique. This novel technique allows for the extraction of 3D vertical forest structure using only a single interferometric pair to acquire a coarse-resolution backscatter intensity distribution in the height direction. Through 3D backscatter distribution, we can extract forest height, the intensity at predefined height bins and introduce the volume-to-ground intensity ratio (VGR) factor to investigate their sensitivities to forest AGB. To validate the method, we use the airborne fully polarized TomoSense dataset, flight-tested by European Space Agency (ESA) in Kermeter area at Eifel National Park, Germany, in 2020. We adopt both multivariate linear stepwise regression (MLSR) and random forest (RF) models to verify the feasibility of the PH technique in forest AGB estimation. Experimental results show that the PH technique effectively captures the vertical structure of the forest at a certain resolution. The forest height, the PH-derived backscatter intensity at a fixed height and VGR have good positive correlation with AGB. Notably, combining forest height, the intensity at fixed height layers and VGR significantly improves the inversion precision of forest AGB. Specifically, compared with LiDAR-derived AGB, the average root-mean-square error (RMSE) of MLSR and RF models estimates combining P- and L-band 2D + 3D observables are 57.92 ton/ha and 55.11 ton/ha, with Pearson correlation coefficient (PCC) of 0.75 and 0.77, respectively. This study presents a promising alternative approach for current and future SAR Earth observation missions aimed at forest vertical structure construction and AGB estimation when only a few of single-polarization SAR images are available.
本文介绍了一种基于干涉SAR (InSAR)的相位直方图(PH)技术估算森林地上生物量(AGB)的方法。这种新技术允许仅使用单个干涉对提取三维垂直森林结构,以获得在高度方向上的粗分辨率后向散射强度分布。通过三维后向散射分布,提取森林高度、预定义高度仓的强度,并引入体地强度比(VGR)因子,考察其对森林AGB的敏感性。为了验证该方法,我们使用了机载全极化TomoSense数据集,该数据集于2020年由欧洲航天局(ESA)在德国艾菲尔国家公园的Kermeter地区进行了飞行测试。采用多元线性逐步回归(MLSR)和随机森林(RF)模型验证PH技术在森林AGB估计中的可行性。实验结果表明,PH技术在一定分辨率下能有效地捕获森林的垂直结构。森林高度、固定高度ph反演后向散射强度和VGR与AGB呈良好的正相关关系。值得注意的是,结合森林高度、固定高程层强度和VGR显著提高了森林AGB的反演精度。与激光雷达AGB相比,结合P波段和l波段2D + 3D观测数据的MLSR和RF模型估计的平均均方根误差(RMSE)分别为57.92和55.11 t /ha, Pearson相关系数(PCC)分别为0.75和0.77。该研究为当前和未来针对森林垂直结构构建和AGB估算的SAR对地观测任务提供了一种有希望的替代方法。
{"title":"InSAR-based estimation of forest above-ground biomass using phase histogram technique","authors":"Chuanjun Wu , Peng Shen , Stefano Tebaldini , Mingsheng Liao , Lu Zhang","doi":"10.1016/j.jag.2024.104350","DOIUrl":"10.1016/j.jag.2024.104350","url":null,"abstract":"<div><div>This paper introduces a method for estimating forest above-ground biomass (AGB) using the Interferometric SAR (InSAR)-based Phase Histogram (PH) technique. This novel technique allows for the extraction of 3D vertical forest structure using only a single interferometric pair to acquire a coarse-resolution backscatter intensity distribution in the height direction. Through 3D backscatter distribution, we can extract forest height, the intensity at predefined height bins and introduce the volume-to-ground intensity ratio (VGR) factor to investigate their sensitivities to forest AGB. To validate the method, we use the airborne fully polarized TomoSense dataset, flight-tested by European Space Agency (ESA) in Kermeter area at Eifel National Park, Germany, in 2020. We adopt both multivariate linear stepwise regression (MLSR) and random forest (RF) models to verify the feasibility of the PH technique in forest AGB estimation. Experimental results show that the PH technique effectively captures the vertical structure of the forest at a certain resolution. The forest height, the PH-derived backscatter intensity at a fixed height and VGR have good positive correlation with AGB. Notably, combining forest height, the intensity at fixed height layers and VGR significantly improves the inversion precision of forest AGB. Specifically, compared with LiDAR-derived AGB, the average root-mean-square error (RMSE) of MLSR and RF models estimates combining P- and L-band 2D + 3D observables are 57.92 ton/ha and 55.11 ton/ha, with Pearson correlation coefficient (PCC) of 0.75 and 0.77, respectively. This study presents a promising alternative approach for current and future SAR Earth observation missions aimed at forest vertical structure construction and AGB estimation when only a few of single-polarization SAR images are available.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104350"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142929501","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2025.104367
Keltoum Khechba , Mariana Belgiu , Ahmed Laamrani , Alfred Stein , Abdelhakim Amazirh , Abdelghani Chehbouni
<div><div>Climate change poses significant challenges to food security, especially in semi-arid agriculture areas. Effective monitoring of crop yield is important for establishing food emergency responses and developing long-term sustainable strategies. In Morocco, where cereals are the predominant crops, yield forecasting is important for addressing the yield gap as it enables farmers to take preventive actions before the harvesting period. This study aims to assess the impact of spatial and temporal heterogeneity of environmental conditions on wheat yield forecasting using machine learning models. It compares the 2019–2020 and 2020–2021 agricultural seasons using three sets of variables: (1) spectral indices; (2) weather data; and (3) a combination of both spectral indices and weather data. Weather data, including cumulative monthly precipitation from ERA5 data and average monthly temperature from PERSIANN data, were extracted for the wheat growing season (November to June). Spectral indices including the Normalized Difference Vegetation Index, Moisture Stress Index, and Terrestrial Chlorophyll Index were calculated from Sentinel-2 imagery for the same period and processed using Google Earth Engine. The study area was divided into homogeneous zones based on an existing landform classification, and XGBoost and Random Forest (RF) models were used for yield forecasting in each zone separately. The two models performed equally well across both the zones and the whole study area (SA) when using weather data as the input variable. For instance, across SA, they achieved average R<sup>2</sup> values of 0.60 and 0.81 for all months during the 2019–2020 and 2020–2021 agricultural seasons, respectively. However, when using spectral indices or combining these indices with weather data, RF consistently outperformed XGBoost. For example, in SA during the 2019–2020 season, RF achieved an average R<sup>2</sup> of 0.48 across the growing season, compared to XGBoost’s R<sup>2</sup> of 0.43. Similarly, in the 2020–2021 season, RF achieved an R<sup>2</sup> of 0.35 and an RMSE of 1083 kg ha<sup>-1</sup>, while XGBoost performed slightly lower, with an R<sup>2</sup> of 0.29 and an RMSE of 1137 kg ha<sup>-1</sup>. Comparing the prediction accuracy between the seasons for each set of variables, the RF model performs better when using spectral indices during the relatively dry 2019–2020 season as compared to the wet 2020–2021 season. Incorporating weather data, the model improved its performance for the 2020–2021 season. April showed the highest prediction performance overall, with R<sup>2</sup> values of 0.6 for SA using weather data alone in the 2019–2020 season, and 0.8 for SA using a combination of weather data and spectral indices in the 2020–2021 season. The 2019–2020 season showed strong fluctuations in accuracy throughout the growing season, whereas the 2020–2021 season had a consistent improvement in accuracy over time. These variations in accuracy are due to
气候变化对粮食安全构成重大挑战,特别是在半干旱农业区。有效监测作物产量对于制定粮食应急措施和制定长期可持续战略至关重要。在以谷物为主要作物的摩洛哥,产量预测对于解决产量差距非常重要,因为它使农民能够在收获期之前采取预防行动。本研究旨在利用机器学习模型评估环境条件时空异质性对小麦产量预测的影响。利用三组变量对2019-2020年和2020-2021年的农业季节进行比较:(1)光谱指数;(2)气象资料;(3)光谱指数与气象资料相结合。提取了小麦生长季节(11月至6月)的气象数据,包括ERA5数据的月累积降水量和persann数据的月平均气温。利用同一时期的Sentinel-2遥感影像计算归一化植被指数、水分胁迫指数和陆地叶绿素指数,并使用谷歌Earth Engine进行处理。在现有地貌分类的基础上,将研究区划分为均匀带,分别使用XGBoost和Random Forest (RF)模型对每个带进行产量预测。当使用天气数据作为输入变量时,这两个模型在两个区域和整个研究区域(SA)上都表现得同样好。例如,在整个SA中,2019-2020年和2020-2021年农业季节所有月份的平均R2分别为0.60和0.81。然而,当使用光谱指数或将这些指数与天气数据相结合时,RF的表现始终优于XGBoost。例如,在2019-2020季节,在SA中,RF在整个生长季节的平均R2为0.48,而XGBoost的R2为0.43。同样,在2020-2021赛季,RF的R2为0.35,RMSE为1083 kg ha-1,而XGBoost的R2略低,为0.29,RMSE为1137 kg ha-1。对比各变量季节间的预测精度,RF模型在相对干旱的2019-2020季节比湿润的2020-2021季节表现更好。结合天气数据,该模型提高了其在2020-2021赛季的表现。总体而言,4月份的预测性能最高,2019-2020年季节仅使用天气数据的SA R2值为0.6,2020-2021年季节使用天气数据和光谱指数组合的SA R2值为0.8。2019-2020赛季在整个生长季中准确性波动较大,而2020-2021赛季随着时间的推移准确性持续提高。这些准确性的差异是由于不同的环境条件造成的,为了做出更好和更可靠的产量预测,应将这些环境条件考虑在内。
{"title":"The impact of spatiotemporal variability of environmental conditions on wheat yield forecasting using remote sensing data and machine learning","authors":"Keltoum Khechba , Mariana Belgiu , Ahmed Laamrani , Alfred Stein , Abdelhakim Amazirh , Abdelghani Chehbouni","doi":"10.1016/j.jag.2025.104367","DOIUrl":"10.1016/j.jag.2025.104367","url":null,"abstract":"<div><div>Climate change poses significant challenges to food security, especially in semi-arid agriculture areas. Effective monitoring of crop yield is important for establishing food emergency responses and developing long-term sustainable strategies. In Morocco, where cereals are the predominant crops, yield forecasting is important for addressing the yield gap as it enables farmers to take preventive actions before the harvesting period. This study aims to assess the impact of spatial and temporal heterogeneity of environmental conditions on wheat yield forecasting using machine learning models. It compares the 2019–2020 and 2020–2021 agricultural seasons using three sets of variables: (1) spectral indices; (2) weather data; and (3) a combination of both spectral indices and weather data. Weather data, including cumulative monthly precipitation from ERA5 data and average monthly temperature from PERSIANN data, were extracted for the wheat growing season (November to June). Spectral indices including the Normalized Difference Vegetation Index, Moisture Stress Index, and Terrestrial Chlorophyll Index were calculated from Sentinel-2 imagery for the same period and processed using Google Earth Engine. The study area was divided into homogeneous zones based on an existing landform classification, and XGBoost and Random Forest (RF) models were used for yield forecasting in each zone separately. The two models performed equally well across both the zones and the whole study area (SA) when using weather data as the input variable. For instance, across SA, they achieved average R<sup>2</sup> values of 0.60 and 0.81 for all months during the 2019–2020 and 2020–2021 agricultural seasons, respectively. However, when using spectral indices or combining these indices with weather data, RF consistently outperformed XGBoost. For example, in SA during the 2019–2020 season, RF achieved an average R<sup>2</sup> of 0.48 across the growing season, compared to XGBoost’s R<sup>2</sup> of 0.43. Similarly, in the 2020–2021 season, RF achieved an R<sup>2</sup> of 0.35 and an RMSE of 1083 kg ha<sup>-1</sup>, while XGBoost performed slightly lower, with an R<sup>2</sup> of 0.29 and an RMSE of 1137 kg ha<sup>-1</sup>. Comparing the prediction accuracy between the seasons for each set of variables, the RF model performs better when using spectral indices during the relatively dry 2019–2020 season as compared to the wet 2020–2021 season. Incorporating weather data, the model improved its performance for the 2020–2021 season. April showed the highest prediction performance overall, with R<sup>2</sup> values of 0.6 for SA using weather data alone in the 2019–2020 season, and 0.8 for SA using a combination of weather data and spectral indices in the 2020–2021 season. The 2019–2020 season showed strong fluctuations in accuracy throughout the growing season, whereas the 2020–2021 season had a consistent improvement in accuracy over time. These variations in accuracy are due to ","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104367"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142975200","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2025.104370
Mengyuan Xu , Haoxuan Yang , Annan Hu , Lee Heng , Linyi Li , Ning Yao , Gang Liu
Deep learning (DL) based methods have recently made remarkable progress in remote sensing (RS) soil moisture (SM) retrieval applications. However, their purely “black box” algorithms suffer from a lack of interpretability, whereas methods based solely on physical mechanisms often underperform in complex scenarios. In this study, we attempt to use an SM downscaling approach that integrates thermal inertia (TI) theory with the DenseNet deep network algorithm. This approach provides partial interpretability of the physical mechanisms while utilizing DenseNet’s superior nonlinear learning ability and feature reuse capability for downscaling the Soil Moisture Active Passive (SMAP) satellite product, generating daily 1 km × 1 km SM. A comprehensive assessment of the downscaled results using in-situ SM acquired from 264 International Soil Moisture Network (ISMN) sites densely distributed across the continental U.S. indicated that this downscaling approach had an overall high accuracy, with an average unbiased root mean square error (ubRMSE) of 0.048 m3/m3. In addition, the downscaled SM exhibited marked improvement in spatial details over the original SMAP SM maps, providing clearer land surface features. The proposed SM downscaling approach is a valuable attempt to adopt DL methods with more practical physical meaning and more interpretability in the current RS Big Data era.
基于深度学习(DL)的方法最近在遥感(RS)土壤湿度(SM)检索应用中取得了显著进展。然而,这些纯粹的 "黑箱 "算法缺乏可解释性,而仅基于物理机制的方法在复杂场景中往往表现不佳。在本研究中,我们尝试使用一种将热惯性(TI)理论与 DenseNet 深度网络算法相结合的 SM 降尺度方法。这种方法提供了物理机制的部分可解释性,同时利用 DenseNet 优越的非线性学习能力和特征重用能力对土壤水分主动被动(SMAP)卫星产品进行降尺度,生成每日 1 km × 1 km 的土壤水分。利用密集分布在美国大陆的 264 个国际土壤水分网络(ISMN)站点获取的原位土壤水分,对降级结果进行了综合评估,结果表明这种降级方法总体精度较高,平均无偏均方根误差(ubRMSE)为 0.048 m3/m3。此外,缩小尺度后的 SM 与原始的 SMAP SM 地图相比,在空间细节方面有明显改善,提供了更清晰的地表特征。所提出的SM降尺度方法是在当前RS大数据时代采用更具实际物理意义和可解释性的DL方法的有益尝试。
{"title":"A deep learning approach for SMAP soil moisture downscaling informed by thermal inertia theory","authors":"Mengyuan Xu , Haoxuan Yang , Annan Hu , Lee Heng , Linyi Li , Ning Yao , Gang Liu","doi":"10.1016/j.jag.2025.104370","DOIUrl":"10.1016/j.jag.2025.104370","url":null,"abstract":"<div><div>Deep learning (DL) based methods have recently made remarkable progress in remote sensing (RS) soil moisture (SM) retrieval applications. However, their purely “black box” algorithms suffer from a lack of interpretability, whereas methods based solely on physical mechanisms often underperform in complex scenarios. In this study, we attempt to use an SM downscaling approach that integrates thermal inertia (TI) theory with the DenseNet deep network algorithm. This approach provides partial interpretability of the physical mechanisms while utilizing DenseNet’s superior nonlinear learning ability and feature reuse capability for downscaling the Soil Moisture Active Passive (SMAP) satellite product, generating daily 1 km × 1 km SM. A comprehensive assessment of the downscaled results using <em>in-situ</em> SM acquired from 264 International Soil Moisture Network (ISMN) sites densely distributed across the continental U.S. indicated that this downscaling approach had an overall high accuracy, with an average unbiased root mean square error (ubRMSE) of 0.048 m<sup>3</sup>/m<sup>3</sup>. In addition, the downscaled SM exhibited marked improvement in spatial details over the original SMAP SM maps, providing clearer land surface features. The proposed SM downscaling approach is a valuable attempt to adopt DL methods with more practical physical meaning and more interpretability in the current RS Big Data era.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104370"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142990330","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2025.104357
Seyed Vahid Razavi-Termeh , Ali Pourzangbar , Abolghasem Sadeghi-Niaraki , Mário J. Franca , Soo-Mi Choi
Managing and controlling costly natural hazards such as floods has been a fundamental and essential issue for decision-makers and planners from the past to the present. Artificial intelligence (AI) has recently proven promising to improve disaster management. There is growing interest in using AI to predict and identify flood-prone areas. However, creating accurate flood susceptibility maps with AI remains a significant challenge. Therefore, the present work endeavors to cope with this gap and produce the most efficient flood susceptibility maps employing Categorical Boosting (CatBoost) algorithms and three system-based metaheuristic methods, including Augmented Artificial Ecosystem Optimization (AAEO), Germinal Center Optimization (GCO), and Water Circle Algorithm (WCA). We selected Jahrom County, Iran, to develop machine learning-based models as our case study. We used 13 flood conditioning geophysical factors as input parameters and flood occurrence (binary classification), derived from satellite imagery, as the output. Our results show that CatBoost-AAEO performs better in flood susceptibility mapping than the other combined models, CatBoost-WCA, CatBoost-GCO, and the basic CatBoost model, which are mentioned in descending order of performance. The partial Dependence Plots (PDP) approach is used to interpret the results of the developed algorithms, highlighting that low slope, low elevation, minimal vegetation cover, flat curvature, and proximity to rivers significantly impact the performance of ML models to predict flood occurrence. The findings of this research can help planners manage and prevent floods and avoid development in sensitive areas to reduce financial losses caused by floods.
{"title":"Metaheuristic-driven enhancement of categorical boosting algorithm for flood-prone areas mapping","authors":"Seyed Vahid Razavi-Termeh , Ali Pourzangbar , Abolghasem Sadeghi-Niaraki , Mário J. Franca , Soo-Mi Choi","doi":"10.1016/j.jag.2025.104357","DOIUrl":"10.1016/j.jag.2025.104357","url":null,"abstract":"<div><div>Managing and controlling costly natural hazards such as floods has been a fundamental and essential issue for decision-makers and planners from the past to the present. Artificial intelligence (AI) has recently proven promising to improve disaster management. There is growing interest in using AI to predict and identify flood-prone areas. However, creating accurate flood susceptibility maps with AI remains a significant challenge. Therefore, the present work endeavors to cope with this gap and produce the most efficient flood susceptibility maps employing Categorical Boosting (CatBoost) algorithms and three system-based metaheuristic methods, including Augmented Artificial Ecosystem Optimization (AAEO), Germinal Center Optimization (GCO), and Water Circle Algorithm (WCA). We selected Jahrom County, Iran, to develop machine learning-based models as our case study. We used 13 flood conditioning geophysical factors as input parameters and flood occurrence (binary classification), derived from satellite imagery, as the output. Our results show that CatBoost-AAEO performs better in flood susceptibility mapping than the other combined models, CatBoost-WCA, CatBoost-GCO, and the basic CatBoost model, which are mentioned in descending order of performance. The partial Dependence Plots (PDP) approach is used to interpret the results of the developed algorithms, highlighting that low slope, low elevation, minimal vegetation cover, flat curvature, and proximity to rivers significantly impact the performance of ML models to predict flood occurrence. The findings of this research can help planners manage and prevent floods and avoid development in sensitive areas to reduce financial losses caused by floods.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104357"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142990331","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2024.104259
Bo Guo , Zhihai Huang , Haitao Luo , Perpetual Hope Akwensi , Ruisheng Wang , Bo Huang , Tsz Nam Chan
The tunnel environment is characterized by insufficient ambient light, obstructed view, and complex inner lining construction conditions. These factors frequently result in limited anti-interference capability, reduced recognition accuracy, and suboptimal segmentation results for defect extraction. We propose a deep network model utilizing an encoder–decoder framework that integrates Transformer and convolution for comprehensive defect extraction. The proposed model utilizes an encoder that integrates a hierarchical Transformer backbone with an efficient attention mechanism to fully explore complete information at multi-scale granularities. In the decoder, multi-scale information is initially aggregated using a Multi-Layer Perceptron (MLP) module. Additionally, the Stacking Filters with Atrous Convolutions (SFAC) module are implemented to enhance the perception of the complete defect scope. Furthermore, a Boundary-aware Attention Module (BAM) is implemented to enhance edge information to improve the detection of defects. With this well-designed decoder, the multi-scale information from the encoder can be fully aggregated and exploited for complete defect detection. Experimental findings illustrate the effectiveness of our proposed approach in addressing tunnel lining defects within the image dataset. The outcomes reveal that our proposed network achieves an accuracy (Acc) of 94.4% and a mean intersection over union (mIoU) of 78.14%. Compared to state-of-the-art segmentation networks, our model improves the accuracy of tunnel lining defect extraction, showcasing enhanced extraction effectiveness and anti-interference capability, thus meeting the engineering requirements for defect detection in complex environments of tunnels.
{"title":"An enhanced network for extracting tunnel lining defects using transformer encoder and aggregate decoder","authors":"Bo Guo , Zhihai Huang , Haitao Luo , Perpetual Hope Akwensi , Ruisheng Wang , Bo Huang , Tsz Nam Chan","doi":"10.1016/j.jag.2024.104259","DOIUrl":"10.1016/j.jag.2024.104259","url":null,"abstract":"<div><div>The tunnel environment is characterized by insufficient ambient light, obstructed view, and complex inner lining construction conditions. These factors frequently result in limited anti-interference capability, reduced recognition accuracy, and suboptimal segmentation results for defect extraction. We propose a deep network model utilizing an encoder–decoder framework that integrates Transformer and convolution for comprehensive defect extraction. The proposed model utilizes an encoder that integrates a hierarchical Transformer backbone with an efficient attention mechanism to fully explore complete information at multi-scale granularities. In the decoder, multi-scale information is initially aggregated using a Multi-Layer Perceptron (MLP) module. Additionally, the Stacking Filters with Atrous Convolutions (SFAC) module are implemented to enhance the perception of the complete defect scope. Furthermore, a Boundary-aware Attention Module (BAM) is implemented to enhance edge information to improve the detection of defects. With this well-designed decoder, the multi-scale information from the encoder can be fully aggregated and exploited for complete defect detection. Experimental findings illustrate the effectiveness of our proposed approach in addressing tunnel lining defects within the image dataset. The outcomes reveal that our proposed network achieves an accuracy (Acc) of 94.4% and a mean intersection over union (mIoU) of 78.14%. Compared to state-of-the-art segmentation networks, our model improves the accuracy of tunnel lining defect extraction, showcasing enhanced extraction effectiveness and anti-interference capability, thus meeting the engineering requirements for defect detection in complex environments of tunnels.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104259"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142874861","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.jag.2025.104361
Jinpeng Li , Yuan Li , Shuhang Zhang , Yiping Chen
Farmland extraction has been a pivotal research challenge for decades in remote sensing. Breakthrough progress has been made by relevant studies due to the advanced deep learning-based techniques. However, existing methods still pay little attention to the simultaneous instance-level farmland extraction and semantic-based 3D attribute analysis, which are essential for enabling more various agricultural applications. Additionally, most bimodal methods apply simple projection to convert high-dimensional features to low-dimensional space for feature interaction, which inevitably underutilizes the advantages of bimodal learning and leads to lamentable information loss. To address this issue, we propose a novel end-to-end bimodal network, named Image-Point Cloud Embedding Network (IPCE-Net), that innovatively employs a dual-stream branch architecture to concurrently perform image-based farmland instance segmentation and point cloud-based semantic segmentation. Furthermore, by leveraging the Heterogeneous Conversion Module (HCM), the IPCE-Net effectively reconciles the modality disparities between images and point clouds and achieves stage-by-stage feature interaction during the bimodal learning process, thus achieving higher performance than unimodal learning. Experiments on two datasets show that IPCE-Net achieves superior performance in both farmland instance extraction and point cloud semantic segmentation tasks. For farmland instance extraction, the instance-level mAP and pixel-level IoU metrics reach 74.9% and 79.6%, respectively, being considerably higher than other classical image-based instance segmentation methods. For the point cloud semantic segmentation, the OA and mIoU metrics are 93.8% and 66.1%, with a remarkable improvement of at least 1.3% and 8.2%, respectively, compared with the state-of-the-art semantic segmentation approaches. Moreover, intelligent analysis based on the interconnection of IPCE-Net and GPT-4 transforms the abstract categorical information into easy-to-understand measurable information, demonstrating its great potential for practical applications in precision and smart agriculture.
{"title":"Image-point cloud embedding network for simultaneous image-based farmland instance extraction and point cloud-based semantic segmentation","authors":"Jinpeng Li , Yuan Li , Shuhang Zhang , Yiping Chen","doi":"10.1016/j.jag.2025.104361","DOIUrl":"10.1016/j.jag.2025.104361","url":null,"abstract":"<div><div>Farmland extraction has been a pivotal research challenge for decades in remote sensing. Breakthrough progress has been made by relevant studies due to the advanced deep learning-based techniques. However, existing methods still pay little attention to the simultaneous instance-level farmland extraction and semantic-based 3D attribute analysis, which are essential for enabling more various agricultural applications. Additionally, most bimodal methods apply simple projection to convert high-dimensional features to low-dimensional space for feature interaction, which inevitably underutilizes the advantages of bimodal learning and leads to lamentable information loss. To address this issue, we propose a novel end-to-end bimodal network, named Image-Point Cloud Embedding Network (IPCE-Net), that innovatively employs a dual-stream branch architecture to concurrently perform image-based farmland instance segmentation and point cloud-based semantic segmentation. Furthermore, by leveraging the Heterogeneous Conversion Module (HCM), the IPCE-Net effectively reconciles the modality disparities between images and point clouds and achieves stage-by-stage feature interaction during the bimodal learning process, thus achieving higher performance than unimodal learning. Experiments on two datasets show that IPCE-Net achieves superior performance in both farmland instance extraction and point cloud semantic segmentation tasks. For farmland instance extraction, the instance-level mAP and pixel-level IoU metrics reach 74.9% and 79.6%, respectively, being considerably higher than other classical image-based instance segmentation methods. For the point cloud semantic segmentation, the OA and mIoU metrics are 93.8% and 66.1%, with a remarkable improvement of at least 1.3% and 8.2%, respectively, compared with the state-of-the-art semantic segmentation approaches. Moreover, intelligent analysis based on the interconnection of IPCE-Net and GPT-4 transforms the abstract categorical information into easy-to-understand measurable information, demonstrating its great potential for practical applications in precision and smart agriculture.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"136 ","pages":"Article 104361"},"PeriodicalIF":7.6,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142990325","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}