Science of Remote Sensing最新文献_第3页

The impacts of training data spatial resolution on deep learning in remote sensing

IF 5.7 Q1 ENVIRONMENTAL SCIENCES

Science of Remote Sensing

Pub Date : 2024-12-12 DOI: 10.1016/j.srs.2024.100185

Christopher Ardohain, Songlin Fei

Deep learning (DL) is ubiquitous in remote sensing analysis with continued evolution in model architectures and advancement of model types. However, DL is still constrained by the need for extensive training datasets, which are costly and time-consuming to produce. One potential solution is adapting training data annotations from different spatial resolutions, though the feasibility of such an application has yet to be tested. In this study, we explore the effects of using forest boundary training data derived from the 3D Elevation Program (3DEP) at 1.5m resolution and the National Land Cover Database (NLCD) at 30m to compare the effects on DL model performance. Our research covers diverse landscapes across 11 counties in Indiana (∼11,636 km²), developing 36 DL models to assess the impact of spatial resolution, model architectures, land cover, and training chip sizes. Our results show that higher-resolution training data yield more accurate models, regardless of imagery resolution, though the performance gap (F1 score) was limited to ∼2.7% even at its most extreme. We also found significant variation in performance based on land cover, with average F1 scores of 0.923 in homogeneous forested areas compared to 0.684 in complex urban settings. Despite similar training times between data sources, chipping 3DEP data took roughly five times longer. We expect that the findings from this study will assist future research in optimizing the development of DL training datasets, selection of source imagery at the proper resolution given training data availability, and application of appropriate model tuning depending on landscape complexity.

{"title":"The impacts of training data spatial resolution on deep learning in remote sensing","authors":"Christopher Ardohain, Songlin Fei","doi":"10.1016/j.srs.2024.100185","DOIUrl":"10.1016/j.srs.2024.100185","url":null,"abstract":"<div><div>Deep learning (DL) is ubiquitous in remote sensing analysis with continued evolution in model architectures and advancement of model types. However, DL is still constrained by the need for extensive training datasets, which are costly and time-consuming to produce. One potential solution is adapting training data annotations from different spatial resolutions, though the feasibility of such an application has yet to be tested. In this study, we explore the effects of using forest boundary training data derived from the 3D Elevation Program (3DEP) at 1.5m resolution and the National Land Cover Database (NLCD) at 30m to compare the effects on DL model performance. Our research covers diverse landscapes across 11 counties in Indiana (∼11,636 km<sup>2</sup>), developing 36 DL models to assess the impact of spatial resolution, model architectures, land cover, and training chip sizes. Our results show that higher-resolution training data yield more accurate models, regardless of imagery resolution, though the performance gap (F1 score) was limited to ∼2.7% even at its most extreme. We also found significant variation in performance based on land cover, with average F1 scores of 0.923 in homogeneous forested areas compared to 0.684 in complex urban settings. Despite similar training times between data sources, chipping 3DEP data took roughly five times longer. We expect that the findings from this study will assist future research in optimizing the development of DL training datasets, selection of source imagery at the proper resolution given training data availability, and application of appropriate model tuning depending on landscape complexity.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100185"},"PeriodicalIF":5.7,"publicationDate":"2024-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143094976","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}

引用次数: 0

Salinity Indian Ocean Dipole: Another facet of the Indian Ocean Dipole phenomenon from satellite remote sensing

IF 5.7 Q1 ENVIRONMENTAL SCIENCES

Science of Remote Sensing

Pub Date : 2024-12-10 DOI: 10.1016/j.srs.2024.100184

Wei Shi , Menghua Wang

Using satellite-measured sea surface salinity (SSS) from the Aquarius and Soil Moisture Active Passive (SMAP) missions since 2011, we show that SSS in the Equatorial Indian Ocean (EIO) experienced dipolar changes in the well-defined east EIO and west EIO regions during the Indian Ocean Dipole (IOD) events. Similar to the concepts of dipole mode Index (DMI) and biological dipole mode index (BDMI), a salinity dipole mode index (SDMI) is proposed using the same definition for the east and west IOD zones. The results show that the salinity IOD in this study is in general co-located and co-incidental with the sea surface temperature (SST) IOD and biological IOD in previous studies. In the positive IOD event in 2019, the SSS anomaly was >1 psu for most of the east IOD zone, while the average SSS in the west IOD zone was ∼0.2–0.3 psu lower than the climatology monthly SSS. The reversed SSS dipolar variability in the EIO was also found during the 2022 negative IOD event. The SSS anomaly difference between the east IOD zone and west IOD zone shows the same variation as the SST-based DMI and chlorophyll-a (Chl-a)-based BDMI. The in situ measurements show that, in the 2019 positive IOD event, the significant IOD-driven salinity change reached water depths at ∼70–80 m and ∼50 m in the east and the west IOD zones, respectively. Results also reveal that the salinity IOD is not only driven by the various ocean processes (e.g., upwelling, downwelling, propagation of the planetary waves, etc.), which are also the main driving forcing for the SST IOD and biological IOD, but also the precipitation and evaporation in the two IOD zones, especially in the west IOD zone. In addition to the traditional SST IOD and recently proposed biological IOD, the salinity IOD indeed features another facet of the entire IOD phenomenon.

{"title":"Salinity Indian Ocean Dipole: Another facet of the Indian Ocean Dipole phenomenon from satellite remote sensing","authors":"Wei Shi , Menghua Wang","doi":"10.1016/j.srs.2024.100184","DOIUrl":"10.1016/j.srs.2024.100184","url":null,"abstract":"<div><div>Using satellite-measured sea surface salinity (SSS) from the Aquarius and Soil Moisture Active Passive (SMAP) missions since 2011, we show that SSS in the Equatorial Indian Ocean (EIO) experienced dipolar changes in the well-defined east EIO and west EIO regions during the Indian Ocean Dipole (IOD) events. Similar to the concepts of dipole mode Index (DMI) and biological dipole mode index (BDMI), a salinity dipole mode index (SDMI) is proposed using the same definition for the east and west IOD zones. The results show that the salinity IOD in this study is in general co-located and co-incidental with the sea surface temperature (SST) IOD and biological IOD in previous studies. In the positive IOD event in 2019, the SSS anomaly was >1 psu for most of the east IOD zone, while the average SSS in the west IOD zone was ∼0.2–0.3 psu lower than the climatology monthly SSS. The reversed SSS dipolar variability in the EIO was also found during the 2022 negative IOD event. The SSS anomaly difference between the east IOD zone and west IOD zone shows the same variation as the SST-based DMI and chlorophyll-a (Chl-a)-based BDMI. The in situ measurements show that, in the 2019 positive IOD event, the significant IOD-driven salinity change reached water depths at ∼70–80 m and ∼50 m in the east and the west IOD zones, respectively. Results also reveal that the salinity IOD is not only driven by the various ocean processes (e.g., upwelling, downwelling, propagation of the planetary waves, etc.), which are also the main driving forcing for the SST IOD and biological IOD, but also the precipitation and evaporation in the two IOD zones, especially in the west IOD zone. In addition to the traditional SST IOD and recently proposed biological IOD, the salinity IOD indeed features another facet of the entire IOD phenomenon.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100184"},"PeriodicalIF":5.7,"publicationDate":"2024-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099661","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}

引用次数: 0

Developing a forest description from remote sensing: Insights from New Zealand

IF 5.7 Q1 ENVIRONMENTAL SCIENCES

Science of Remote Sensing

Pub Date : 2024-12-10 DOI: 10.1016/j.srs.2024.100183

Grant D. Pearse , Sadeepa Jayathunga , Nicolò Camarretta , Melanie E. Palmer , Benjamin S.C. Steer , Michael S. Watt , Pete Watt , Andrew Holdaway

Remote sensing is increasingly being used to create large-scale forest descriptions. In New Zealand, where radiata pine (Pinus radiata) plantations dominate the forestry sector, the current national forest description lacks spatially explicit information and struggles to capture data on small-scale forests. This is important as these forests are expected to contribute significantly to future wood supply and carbon sequestration. This study demonstrates the development of a spatially explicit, remote sensing-based forest description for the Gisborne region, a major forest growing area. We combined deep learning-based forest mapping using high-resolution aerial imagery with regional airborne laser scanning (ALS) data to map all planted forest and estimate key attributes. The deep learning model accurately delineated planted forests, including large estates, small woodlots, and newly established stands as young as 3-years post planting. It achieved an intersection over union of 0.94, precision of 0.96, and recall of 0.98 on a withheld dataset. ALS-derived models for estimating mean top height, total stem volume, and stand age showed good performance (R² = 0.94, 0.82, and 0.94 respectively). The resulting spatially explicit forest description provides wall-to-wall information on forest extent, age, and volume for all sizes of forest. This enables stratification by key variables for wood supply forecasting, harvest planning, and infrastructure investment decisions. We propose satellite-based harvest detection and digital photogrammetry to continuously update the initial forest description. This methodology enables near real-time monitoring of planted forests at all scales and is adaptable to other regions with similar data availability.

{"title":"Developing a forest description from remote sensing: Insights from New Zealand","authors":"Grant D. Pearse , Sadeepa Jayathunga , Nicolò Camarretta , Melanie E. Palmer , Benjamin S.C. Steer , Michael S. Watt , Pete Watt , Andrew Holdaway","doi":"10.1016/j.srs.2024.100183","DOIUrl":"10.1016/j.srs.2024.100183","url":null,"abstract":"<div><div>Remote sensing is increasingly being used to create large-scale forest descriptions. In New Zealand, where radiata pine (<em>Pinus radiata</em>) plantations dominate the forestry sector, the current national forest description lacks spatially explicit information and struggles to capture data on small-scale forests. This is important as these forests are expected to contribute significantly to future wood supply and carbon sequestration. This study demonstrates the development of a spatially explicit, remote sensing-based forest description for the Gisborne region, a major forest growing area. We combined deep learning-based forest mapping using high-resolution aerial imagery with regional airborne laser scanning (ALS) data to map all planted forest and estimate key attributes. The deep learning model accurately delineated planted forests, including large estates, small woodlots, and newly established stands as young as 3-years post planting. It achieved an intersection over union of 0.94, precision of 0.96, and recall of 0.98 on a withheld dataset. ALS-derived models for estimating mean top height, total stem volume, and stand age showed good performance (<em>R</em><sup>2</sup> = 0.94, 0.82, and 0.94 respectively). The resulting spatially explicit forest description provides wall-to-wall information on forest extent, age, and volume for all sizes of forest. This enables stratification by key variables for wood supply forecasting, harvest planning, and infrastructure investment decisions. We propose satellite-based harvest detection and digital photogrammetry to continuously update the initial forest description. This methodology enables near real-time monitoring of planted forests at all scales and is adaptable to other regions with similar data availability.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100183"},"PeriodicalIF":5.7,"publicationDate":"2024-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099663","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}

引用次数: 0

Nonparametric quantification of uncertainty in multistep upscaling approaches: A case study on estimating forest biomass in the Brazilian Amazon

IF 5.7 Q1 ENVIRONMENTAL SCIENCES

Science of Remote Sensing

Pub Date : 2024-12-07 DOI: 10.1016/j.srs.2024.100180

Denis Valle , Leo Haneda , Rafael Izbicki , Renan Akio Kamimura , Bruna Pereira de Azevedo , Silvio H.M. Gomes , Arthur Sanchez , Carlos A. Silva , Danilo R.A. Almeida

The use of multistep upscaling approaches in which field data are combined with data from multiple remote sensors that operate at different spatial scales (e.g., UAV LiDAR, GEDI, and Landsat) is becoming increasingly popular. In these approaches, a series of models are fitted linking the information from these different sensors, often resulting in improved predictions over large areas. Quantifying the uncertainty associated with individual models can be challenging as these models may not generate uncertainty estimates (e.g., machine learning models such as random forest), a problem that is further exacerbated if the results from multiple models are combined within a multistep upscaling methodology. In this article, we describe a nonparametric conformal approach to quantify uncertainty. This approach is straight-forward to apply, is computationally inexpensive (differently from bootstrapping), and generates improved predictive intervals. Importantly, this methodology can be used regardless of the number of models adopted in the upscaling approach and the nature of the intermediate models, as long as the final model can generate predictive intervals. We illustrate the improved empirical coverage of the conformalized predictive intervals using simulated data for a two-step upscaling scenario involving field, UAV LiDAR, and Landsat data. This simulation exercise shows how increasing uncertainty in the first stage model (which relates biomass field data to UAV LiDAR data) leads to an increase in the severity of uncertainty underestimation by naïve predictive intervals. On the other hand, conformalized predictive intervals do not exhibit this shortcoming. Finally, we illustrate uncertainty quantification for a multistep upscaling methodology using data from a large-scale carbon project in the Brazilian Amazon. Our validation exercise using these empirical data confirms the improved performance of the conformalized predictive intervals. We expect that the conformal approach described here will be key for uncertainty quantification as multistep upscaling approaches become increasingly more common.

{"title":"Nonparametric quantification of uncertainty in multistep upscaling approaches: A case study on estimating forest biomass in the Brazilian Amazon","authors":"Denis Valle , Leo Haneda , Rafael Izbicki , Renan Akio Kamimura , Bruna Pereira de Azevedo , Silvio H.M. Gomes , Arthur Sanchez , Carlos A. Silva , Danilo R.A. Almeida","doi":"10.1016/j.srs.2024.100180","DOIUrl":"10.1016/j.srs.2024.100180","url":null,"abstract":"<div><div>The use of multistep upscaling approaches in which field data are combined with data from multiple remote sensors that operate at different spatial scales (e.g., UAV LiDAR, GEDI, and Landsat) is becoming increasingly popular. In these approaches, a series of models are fitted linking the information from these different sensors, often resulting in improved predictions over large areas. Quantifying the uncertainty associated with individual models can be challenging as these models may not generate uncertainty estimates (e.g., machine learning models such as random forest), a problem that is further exacerbated if the results from multiple models are combined within a multistep upscaling methodology. In this article, we describe a nonparametric conformal approach to quantify uncertainty. This approach is straight-forward to apply, is computationally inexpensive (differently from bootstrapping), and generates improved predictive intervals. Importantly, this methodology can be used regardless of the number of models adopted in the upscaling approach and the nature of the intermediate models, as long as the final model can generate predictive intervals. We illustrate the improved empirical coverage of the conformalized predictive intervals using simulated data for a two-step upscaling scenario involving field, UAV LiDAR, and Landsat data. This simulation exercise shows how increasing uncertainty in the first stage model (which relates biomass field data to UAV LiDAR data) leads to an increase in the severity of uncertainty underestimation by naïve predictive intervals. On the other hand, conformalized predictive intervals do not exhibit this shortcoming. Finally, we illustrate uncertainty quantification for a multistep upscaling methodology using data from a large-scale carbon project in the Brazilian Amazon. Our validation exercise using these empirical data confirms the improved performance of the conformalized predictive intervals. We expect that the conformal approach described here will be key for uncertainty quantification as multistep upscaling approaches become increasingly more common.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100180"},"PeriodicalIF":5.7,"publicationDate":"2024-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099662","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}

引用次数: 0

Snow observation from space: An approach to improving snow cover detection using four decades of Landsat and Sentinel-2 imageries across Switzerland

IF 5.7 Q1 ENVIRONMENTAL SCIENCES

Science of Remote Sensing

Pub Date : 2024-12-06 DOI: 10.1016/j.srs.2024.100182

Charlotte Poussin , Pascal Peduzzi , Gregory Giuliani

Landsat and Sentinel-2 satellites offer significant advantages for monitoring snow cover over mountainous countries like Switzerland. Starting in the 1970s, Landsat data provides over 50 years of medium resolution imagery. However, the main limitation of optical imagery is cloud cover. Cloud obstruction is particularly challenging for Landsat and Sentinel-2 data, which have limited temporal resolutions. In this study we present the Snow Observation from Space (SOfS) algorithm composed of seven successive temporal and spatial techniques to reduce cloud coverage in the final snow cover products. We used long-term Landsat and Sentinel-2 datasets available from the Swiss Data Cube. The results indicate that the filtering techniques are efficient in reducing cloud cover by half while still leaving an average of less than 30% of cloud cover. The accuracy of the entire algorithm is evaluated over Switzerland, using in-situ measurements of 263 climate stations in the period 1984–2021. The validation results show an agreement between SOfS dataset and ground snow observations with an average accuracy of 93%.

{"title":"Snow observation from space: An approach to improving snow cover detection using four decades of Landsat and Sentinel-2 imageries across Switzerland","authors":"Charlotte Poussin , Pascal Peduzzi , Gregory Giuliani","doi":"10.1016/j.srs.2024.100182","DOIUrl":"10.1016/j.srs.2024.100182","url":null,"abstract":"<div><div>Landsat and Sentinel-2 satellites offer significant advantages for monitoring snow cover over mountainous countries like Switzerland. Starting in the 1970s, Landsat data provides over 50 years of medium resolution imagery. However, the main limitation of optical imagery is cloud cover. Cloud obstruction is particularly challenging for Landsat and Sentinel-2 data, which have limited temporal resolutions. In this study we present the Snow Observation from Space (SOfS) algorithm composed of seven successive temporal and spatial techniques to reduce cloud coverage in the final snow cover products. We used long-term Landsat and Sentinel-2 datasets available from the Swiss Data Cube. The results indicate that the filtering techniques are efficient in reducing cloud cover by half while still leaving an average of less than 30% of cloud cover. The accuracy of the entire algorithm is evaluated over Switzerland, using <em>in-situ</em> measurements of 263 climate stations in the period 1984–2021. The validation results show an agreement between SOfS dataset and ground snow observations with an average accuracy of 93%.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100182"},"PeriodicalIF":5.7,"publicationDate":"2024-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099666","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}

引用次数: 0

Quantifying forest stocking changes in Sundarbans mangrove using remote sensing data

IF 5.7 Q1 ENVIRONMENTAL SCIENCES

Science of Remote Sensing

Pub Date : 2024-12-04 DOI: 10.1016/j.srs.2024.100181

Yaqub Ali , M. Mahmudur Rahman

The Sundarbans, the world's largest mangrove ecosystem, faces significant challenges from forest stocking changes due to natural and anthropogenic factors. Scientific studies on these changes are not available. This study uses remote sensing techniques to quantify long-term changes in mangrove forest canopy height, aboveground biomass (AGB), and forest carbon stocks. Using Shuttle Radar Topography Mission (SRTM) and Global Ecosystem Dynamics Investigation (GEDI) LiDAR data sets, we assessed canopy height and forest stocking changes, and changes in AGB carbon fluxes over the last two decades in the Sundarbans mangrove. Calibrated SRTM data provided tree canopy height (TCH) estimates for 2000, while calibrated GEDI LiDAR data facilitated assessments of TCH for 2023. The findings show substantial changes in TCH, AGB, and carbon stock distribution in the Sundarbans mangrove between 2000 and 2023. TCH in the 5–10 m class notably increased from 58.3% in 2000 to 70.8% in 2023, while TCH above 15 m decreased, and those under 5 m regrew. Higher AGB carbon classes (>50 tons ha⁻¹) decreased, with only the lowest class (<50 tons ha⁻¹) increased, indicating notable forest carbon stock reduction due to deforestation and forest degradation. Approximately 1571 Kt of AGB carbon were lost over 23 years, which represents around 4% of the total stock. The driving forces of forest stocking changes could be the changes in the dynamic energy balance from the estuarine river system and the tidal waves, relative sea-level change, increases of salinity in various zones of Sundarbans mangrove, other anthropogenic factors, etc. This research provides valuable insights into Sundarbans mangrove dynamics, aiding global forest degradation and forest growth in understanding forest stocking change and their role in terrestrial carbon flux and global climate change. The results will be helpful for the forest manager in identifying the locations where there is forest degradation or enhancement of forest growing stock and planning any silvicultural operations that are needed in the forest. This is also useful for climate change scientists to understand probable man-made or natural driving forces of the changes in forest stocking in the Sundarbans mangrove forests. It underscores the urgency of integrating deforestation and forest degradation into climate strategies for effective carbon management and conservation efforts, that align with carbon sequestration goals, contributing to broader climate change mitigation strategies.

{"title":"Quantifying forest stocking changes in Sundarbans mangrove using remote sensing data","authors":"Yaqub Ali , M. Mahmudur Rahman","doi":"10.1016/j.srs.2024.100181","DOIUrl":"10.1016/j.srs.2024.100181","url":null,"abstract":"<div><div>The Sundarbans, the world's largest mangrove ecosystem, faces significant challenges from forest stocking changes due to natural and anthropogenic factors. Scientific studies on these changes are not available. This study uses remote sensing techniques to quantify long-term changes in mangrove forest canopy height, aboveground biomass (AGB), and forest carbon stocks. Using Shuttle Radar Topography Mission (SRTM) and Global Ecosystem Dynamics Investigation (GEDI) LiDAR data sets, we assessed canopy height and forest stocking changes, and changes in AGB carbon fluxes over the last two decades in the Sundarbans mangrove. Calibrated SRTM data provided tree canopy height (TCH) estimates for 2000, while calibrated GEDI LiDAR data facilitated assessments of TCH for 2023. The findings show substantial changes in TCH, AGB, and carbon stock distribution in the Sundarbans mangrove between 2000 and 2023. TCH in the 5–10 m class notably increased from 58.3% in 2000 to 70.8% in 2023, while TCH above 15 m decreased, and those under 5 m regrew. Higher AGB carbon classes (>50 tons ha⁻<sup>1</sup>) decreased, with only the lowest class (<50 tons ha⁻<sup>1</sup>) increased, indicating notable forest carbon stock reduction due to deforestation and forest degradation. Approximately 1571 Kt of AGB carbon were lost over 23 years, which represents around 4% of the total stock. The driving forces of forest stocking changes could be the changes in the dynamic energy balance from the estuarine river system and the tidal waves, relative sea-level change, increases of salinity in various zones of Sundarbans mangrove, other anthropogenic factors, etc. This research provides valuable insights into Sundarbans mangrove dynamics, aiding global forest degradation and forest growth in understanding forest stocking change and their role in terrestrial carbon flux and global climate change. The results will be helpful for the forest manager in identifying the locations where there is forest degradation or enhancement of forest growing stock and planning any silvicultural operations that are needed in the forest. This is also useful for climate change scientists to understand probable man-made or natural driving forces of the changes in forest stocking in the Sundarbans mangrove forests. It underscores the urgency of integrating deforestation and forest degradation into climate strategies for effective carbon management and conservation efforts, that align with carbon sequestration goals, contributing to broader climate change mitigation strategies.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100181"},"PeriodicalIF":5.7,"publicationDate":"2024-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099665","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}

引用次数: 0

Assessing the reliability of woody vegetation structural characterisation from UAV-LS in a tropical savanna

IF 5.7 Q1 ENVIRONMENTAL SCIENCES

Science of Remote Sensing

Pub Date : 2024-12-03 DOI: 10.1016/j.srs.2024.100178

Rafael Bohn Reckziegel , Thomas Lowe , Timothy Devereux , Stephanie M. Johnson , Ellen Rochelmeyer , Lindsay B. Hutley , Tanya Doody , Shaun R. Levick

Terrestrial laser scanning (TLS) represents the gold standard in remote quantification of woody vegetation structure and volume, but is costly and time consuming to acquire. TLS data is typically collected at spatial scales of 1 ha or smaller, which limits its suitability for representing heterogeneous landscapes, and for training and validating satellite-based models which are needed for larger area monitoring. Advances in unoccupied aerial vehicle laser scanning (UAV-LS) sensors have recently narrowed the gap in quality between what TLS delivers and what can be acquired over larger areas from UAV platforms. We tested how well new nadir-forward–backward (NFB) UAV-LS technology can capture the structure of individual trees in a tropical savanna setting with a diversity of tree sizes and growth forms. UAV-LS data was acquired with a RIEGL VUX-120 LiDAR sensor mounted on a Acecore NOA hexacopter. Reference data was obtained with a RIEGL VZ-2000i TLS scanner using a multi-scan approach. Point clouds were segmented into individual trees and volumetrically reconstructed with RayCloudTools (RCT). We found no statistical difference between UAV-LS and TLS derived estimates of tree height, canopy cover, diameter, and wood volume. Mean tree height and DBH derived from UAV-LS were within 3% of the TLS estimate, and there was less than 1% deviation in stand wood volume. Our findings ease the advancements on the detailed monitoring of open forests, potentially achieving large-scale mapping and multi-temporal investigations. The open structure of savanna systems is well suited to UAV-LS sensing, but more research is needed across diverse ecosystems to understand the generality of these findings in landscapes with greater canopy closure or complex understorey conditions.

{"title":"Assessing the reliability of woody vegetation structural characterisation from UAV-LS in a tropical savanna","authors":"Rafael Bohn Reckziegel , Thomas Lowe , Timothy Devereux , Stephanie M. Johnson , Ellen Rochelmeyer , Lindsay B. Hutley , Tanya Doody , Shaun R. Levick","doi":"10.1016/j.srs.2024.100178","DOIUrl":"10.1016/j.srs.2024.100178","url":null,"abstract":"<div><div>Terrestrial laser scanning (TLS) represents the gold standard in remote quantification of woody vegetation structure and volume, but is costly and time consuming to acquire. TLS data is typically collected at spatial scales of 1 ha or smaller, which limits its suitability for representing heterogeneous landscapes, and for training and validating satellite-based models which are needed for larger area monitoring. Advances in unoccupied aerial vehicle laser scanning (UAV-LS) sensors have recently narrowed the gap in quality between what TLS delivers and what can be acquired over larger areas from UAV platforms. We tested how well new nadir-forward–backward (NFB) UAV-LS technology can capture the structure of individual trees in a tropical savanna setting with a diversity of tree sizes and growth forms. UAV-LS data was acquired with a RIEGL VUX-120 LiDAR sensor mounted on a Acecore NOA hexacopter. Reference data was obtained with a RIEGL VZ-2000i TLS scanner using a multi-scan approach. Point clouds were segmented into individual trees and volumetrically reconstructed with RayCloudTools (RCT). We found no statistical difference between UAV-LS and TLS derived estimates of tree height, canopy cover, diameter, and wood volume. Mean tree height and DBH derived from UAV-LS were within 3% of the TLS estimate, and there was less than 1% deviation in stand wood volume. Our findings ease the advancements on the detailed monitoring of open forests, potentially achieving large-scale mapping and multi-temporal investigations. The open structure of savanna systems is well suited to UAV-LS sensing, but more research is needed across diverse ecosystems to understand the generality of these findings in landscapes with greater canopy closure or complex understorey conditions.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"11 ","pages":"Article 100178"},"PeriodicalIF":5.7,"publicationDate":"2024-12-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143099660","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}

引用次数: 0

Quantifying volumes of volcanic deposits using time-averaged ASTER digital elevation models 使用时间平均ASTER数字高程模型量化火山沉积物体积

IF 5.7 Q1 ENVIRONMENTAL SCIENCES

Science of Remote Sensing

Pub Date : 2024-12-01 DOI: 10.1016/j.srs.2024.100179

Ian T.W. Flynn, Daniel B. Williams, Michael S. Ramsey

Quantifying the volume of erupted volcanic material, particularly lava flows and domes, provides critical insights into the dynamics of an eruption. This in turn aids in future hazard modeling, mitigation, and response. However, acquiring the necessary topographic datasets to calculate volumetric change is difficult, especially for active volcanoes in remote regions. The Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) instrument has acquired global photogrammetric data since 2000, from which individual scene digital elevation models (DEMs) are created. We present a new straight forward method using ASTER DEMs to measure the volume of emplaced lava flows, domes, and tephra cones. We focus on five case studies that represent different eruption styles and products. For each of these we compare the results to those from previous studies that used alternative topographic datasets, such as synthetic aperture radar (SAR), airborne photogrammetry, or Light Detection and Ranging (LiDAR) measurements. These datasets, however, are expensive to acquire or lack the needed temporal resolution. We show that in nearly all cases, our volume results are either within the reported range for the eruption or ≤0.05 km³ of the previously reported value derived from SAR or LiDAR. The simplicity of the ASTER DEM approach combined with the global coverage of the data products enables more timely production of accurate volumetric data during and following an eruption, which can then be used to assess past and future eruption dynamics.

量化喷发的火山物质的体积，特别是熔岩流和圆顶，为火山喷发的动力学提供了关键的见解。这反过来又有助于未来的危害建模、缓解和响应。然而，获取必要的地形数据集来计算体积变化是困难的，特别是对于偏远地区的活火山。先进星载热发射和反射辐射计（ASTER）仪器自2000年以来获得了全球摄影测量数据，并由此创建了单个场景数字高程模型（dem）。我们提出了一种新的直接方法，使用ASTER dem来测量就位的熔岩流，圆顶和火山锥的体积。我们重点研究了五个代表不同喷发风格和产品的案例研究。对于每一项研究，我们都将结果与以前使用替代地形数据集的研究结果进行比较，例如合成孔径雷达（SAR）、航空摄影测量或光探测和测距（LiDAR）测量。然而，这些数据集的获取成本很高，或者缺乏所需的时间分辨率。我们表明，在几乎所有情况下，我们的体积结果要么在报道的喷发范围内，要么与先前报道的SAR或LiDAR得出的值≤0.05 km3。ASTER DEM方法的简单性与数据产品的全球覆盖相结合，可以在喷发期间和之后更及时地产生准确的体积数据，然后可以用来评估过去和未来的喷发动态。

{"title":"Quantifying volumes of volcanic deposits using time-averaged ASTER digital elevation models","authors":"Ian T.W. Flynn, Daniel B. Williams, Michael S. Ramsey","doi":"10.1016/j.srs.2024.100179","DOIUrl":"10.1016/j.srs.2024.100179","url":null,"abstract":"<div><div>Quantifying the volume of erupted volcanic material, particularly lava flows and domes, provides critical insights into the dynamics of an eruption. This in turn aids in future hazard modeling, mitigation, and response. However, acquiring the necessary topographic datasets to calculate volumetric change is difficult, especially for active volcanoes in remote regions. The Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) instrument has acquired global photogrammetric data since 2000, from which individual scene digital elevation models (DEMs) are created. We present a new straight forward method using ASTER DEMs to measure the volume of emplaced lava flows, domes, and tephra cones. We focus on five case studies that represent different eruption styles and products. For each of these we compare the results to those from previous studies that used alternative topographic datasets, such as synthetic aperture radar (SAR), airborne photogrammetry, or Light Detection and Ranging (LiDAR) measurements. These datasets, however, are expensive to acquire or lack the needed temporal resolution. We show that in nearly all cases, our volume results are either within the reported range for the eruption or ≤0.05 km<sup>3</sup> of the previously reported value derived from SAR or LiDAR. The simplicity of the ASTER DEM approach combined with the global coverage of the data products enables more timely production of accurate volumetric data during and following an eruption, which can then be used to assess past and future eruption dynamics.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"10 ","pages":"Article 100179"},"PeriodicalIF":5.7,"publicationDate":"2024-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142745839","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}

引用次数: 0

Global soil moisture mapping at 5 km by combining GNSS reflectometry and machine learning in view of HydroGNSS 考虑到 HydroGNSS，结合全球导航卫星系统反射测量法和机器学习绘制 5 千米全球土壤湿度图

IF 5.7 Q1 ENVIRONMENTAL SCIENCES

Science of Remote Sensing

Pub Date : 2024-11-16 DOI: 10.1016/j.srs.2024.100177

Emanuele Santi , Davide Comite , Laura Dente , Leila Guerriero , Nazzareno Pierdicca , Maria Paola Clarizia , Nicolas Floury

The potential of GNSS reflectometry (GNSS-R) for the monitoring of soil and vegetation parameters as soil moisture (SM) and forest aboveground biomass (AGB) has been largely investigated in recent years.

In view of the ESA's HydroGNSS mission, planned to be launched in 2024, this study has explored the possibility to map SM at global scale and relatively high resolution of about 0.05° (corresponding approximately to 5 Km) using GNSS-R observations, by implementing and comparing two retrieval algorithms based on machine learning techniques, namely Artificial Neural Networks (ANN) and Random Forest Regressors (RF). Waiting for HydroGNSS commissioning and operation, the NASA's Cyclone GNSS (CyGNSS) land observations have been considered for this scope. Taking advantage of the versatility of both machine learning techniques, several combinations of input data, including CyGNSS observables and auxiliary information, have been exploited and the role of GNSS-R and auxiliary data has been assessed. Given the lack of global SM data at 0.05° resolution, the following novel strategy has been implemented to establish the training set: as first, training has been carried out at lower resolution by considering as target the SMAP SM on EASE-Grid 36 km. Then the trained algorithms have been applied to CyGNSS data at 0.05° to obtain global SM maps at this resolution. Finally, the SM at 0.05° has been validated against ISMN, to keep training and validation as much independent as possible. The two retrieval techniques exhibited similar accuracies and computational cost, with correlation coefficient R ≃ 0.9 between estimated and target SM computed globally, and RMSE ≃ 0.05 (m³/m³). Moreover, the SM maps at 0.05° revealed some finer details and small-scale patterns that are not shown by the original SMAP SM data at 36 km. Regardless of the ML technique applied, this study confirmed the promising potential of GNSS-R for the global monitoring of SM at improved resolution with respect to SM products available from microwave satellite radiometers.

鉴于计划于 2024 年发射的欧空局 HydroGNSS 飞行任务，本研究探讨了利用 GNSS-R 观测数据绘制全球尺度和相对较高分辨率（约 0.05°，相当于约 5 千米）的土壤和植被参数图的可能性。05° （大约相当于 5 公里）的 SM 地图的可能性，具体方法是实施和比较两种基于机器学习技术的检索算法，即人工神经网络（ANN）和随机森林回归器（RF）。在等待水文全球导航卫星系统调试和运行期间，美国国家航空航天局的旋风全球导航卫星系统（CyGNSS）陆地观测数据被考虑用于这一范围。利用这两种机器学习技术的多功能性，对包括 CyGNSS 观测数据和辅助信息在内的多种输入数据组合进行了开发，并对 GNSS-R 和辅助数据的作用进行了评估。鉴于缺乏 0.05°分辨率的全球 SM 数据，采用了以下新策略来建立训练集：首先，将 EASE-Grid 36 公里上的 SMAP SM 作为目标，在较低分辨率下进行训练。然后，将训练好的算法应用于 0.05°的 CyGNSS 数据，以获得该分辨率的全球 SM 地图。最后，0.05°的SM与ISMN进行了验证，以尽可能保持训练和验证的独立性。两种检索技术显示出相似的精确度和计算成本，全球计算的估计 SM 与目标 SM 之间的相关系数 R ≃0.9，RMSE ≃0.05（m3/m3）。此外，0.05° 的 SM 地图显示了一些更精细的细节和小尺度模式，而 36 公里处的 SMAP 原始 SM 数据没有显示这些细节和模式。无论采用哪种多层面技术，这项研究都证实，与微波卫星辐射计提供的 SM 产品相比，GNSS-R 在提高分辨率对 SM 进行全球监测方面具有巨大潜力。

{"title":"Global soil moisture mapping at 5 km by combining GNSS reflectometry and machine learning in view of HydroGNSS","authors":"Emanuele Santi , Davide Comite , Laura Dente , Leila Guerriero , Nazzareno Pierdicca , Maria Paola Clarizia , Nicolas Floury","doi":"10.1016/j.srs.2024.100177","DOIUrl":"10.1016/j.srs.2024.100177","url":null,"abstract":"<div><div>The potential of GNSS reflectometry (GNSS-R) for the monitoring of soil and vegetation parameters as soil moisture (SM) and forest aboveground biomass (AGB) has been largely investigated in recent years.</div><div>In view of the ESA's HydroGNSS mission, planned to be launched in 2024, this study has explored the possibility to map SM at global scale and relatively high resolution of about 0.05° (corresponding approximately to 5 Km) using GNSS-R observations, by implementing and comparing two retrieval algorithms based on machine learning techniques, namely Artificial Neural Networks (ANN) and Random Forest Regressors (RF). Waiting for HydroGNSS commissioning and operation, the NASA's Cyclone GNSS (CyGNSS) land observations have been considered for this scope. Taking advantage of the versatility of both machine learning techniques, several combinations of input data, including CyGNSS observables and auxiliary information, have been exploited and the role of GNSS-R and auxiliary data has been assessed. Given the lack of global SM data at 0.05° resolution, the following novel strategy has been implemented to establish the training set: as first, training has been carried out at lower resolution by considering as target the SMAP SM on EASE-Grid 36 km. Then the trained algorithms have been applied to CyGNSS data at 0.05° to obtain global SM maps at this resolution. Finally, the SM at 0.05° has been validated against ISMN, to keep training and validation as much independent as possible. The two retrieval techniques exhibited similar accuracies and computational cost, with correlation coefficient R ≃ 0.9 between estimated and target SM computed globally, and RMSE ≃ 0.05 (m<sup>3</sup>/m<sup>3</sup>). Moreover, the SM maps at 0.05° revealed some finer details and small-scale patterns that are not shown by the original SMAP SM data at 36 km. Regardless of the ML technique applied, this study confirmed the promising potential of GNSS-R for the global monitoring of SM at improved resolution with respect to SM products available from microwave satellite radiometers.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"10 ","pages":"Article 100177"},"PeriodicalIF":5.7,"publicationDate":"2024-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142706863","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}

引用次数: 0

Coastal vertical land motion across Southeast Asia derived from combining tide gauge and satellite altimetry observations 结合验潮仪和卫星测高仪观测得出的东南亚沿海陆地垂直运动数据

IF 5.7 Q1 ENVIRONMENTAL SCIENCES

Science of Remote Sensing

Pub Date : 2024-11-12 DOI: 10.1016/j.srs.2024.100176

Dongju Peng , Grace Ng , Lujia Feng , Anny Cazenave , Emma M. Hill

Vertical land motion (VLM) is complex in Southeast Asia because this region is subject to a range of natural processes (e.g., earthquakes) and anthropogenic activities (e.g., groundwater withdrawal) that can change land heights. To aid in coastal management, long-term observations of VLM are as crucial as observations for climate-induced sea surface height changes; however, such long-term observations are sparse for Southeast Asian coasts. To fill this observational gap, here we derive monthly VLM time series from 1993 to 2020 at 50 coastal sites across Southeast Asia by combining tide-gauge records and newly generated satellite altimetry observations. These altimetry observations are reproduced sea-level products using new altimetry standards and more accurate geophysical corrections. Our 27-year-long VLM dataset shows high spatial variability and non-linear temporal changes in VLM across Southeast Asia. We identify several major sources that dominate the regional land-height changes, which include large subsidence due to groundwater extraction in Manila and Bangkok, land uplift in Indonesia and subsidence in Thailand from postseismic deformation resulting from the sequence of large Sumatran earthquakes since 2004, and land subsidence as a result of sediment compaction in Malaysia. Those signals are quantitatively or qualitatively consistent with observations from other sources. This VLM dataset can be used to advance our understanding of the physical mechanisms behind land-height changes and to improve sea level projections in the region.

在东南亚，陆地垂直运动(VLM)是一个复杂的问题，因为该地区受到一系列自然过程(如地震)和人为活动(如抽取地下水)的影响，这些都会改变陆地高度。为了帮助沿岸管理，对 VLM 的长期观测与对气候引起的海面高度变化的观测一样重要，但对东南亚沿岸的长期观测却很少。为填补这一观测空白，我们结合验潮仪记录和新生成的卫星测高观测数据，在东南亚 50 个沿岸站点建立了 1993-2020 年的月 VLM 时间序列。这些测高观测数据是利用新的测高标准和更精确的地球物理修正再现的海平面产品。我们长达 27 年的 VLM 数据集显示，整个东南亚地区的 VLM 具有很高的空间变异性和非线性时间变化。我们确定了主导区域陆地高度变化的几个主要来源，其中包括马尼拉和曼谷因抽取地下水而导致的大规模沉降、印度尼西亚的陆地隆起和泰国自 2004 年以来苏门答腊岛系列大地震导致的震后变形引起的沉降，以及马来西亚沉积物压实导致的陆地沉降。这些信号在数量或质量上与其他来源的观测结果一致。这个甚低层地貌数据集可用于促进我们对陆地高度变化背后的物理机制的了解，并改进对该地区海平面的预测。

{"title":"Coastal vertical land motion across Southeast Asia derived from combining tide gauge and satellite altimetry observations","authors":"Dongju Peng , Grace Ng , Lujia Feng , Anny Cazenave , Emma M. Hill","doi":"10.1016/j.srs.2024.100176","DOIUrl":"10.1016/j.srs.2024.100176","url":null,"abstract":"<div><div>Vertical land motion (VLM) is complex in Southeast Asia because this region is subject to a range of natural processes (e.g., earthquakes) and anthropogenic activities (e.g., groundwater withdrawal) that can change land heights. To aid in coastal management, long-term observations of VLM are as crucial as observations for climate-induced sea surface height changes; however, such long-term observations are sparse for Southeast Asian coasts. To fill this observational gap, here we derive monthly VLM time series from 1993 to 2020 at 50 coastal sites across Southeast Asia by combining tide-gauge records and newly generated satellite altimetry observations. These altimetry observations are reproduced sea-level products using new altimetry standards and more accurate geophysical corrections. Our 27-year-long VLM dataset shows high spatial variability and non-linear temporal changes in VLM across Southeast Asia. We identify several major sources that dominate the regional land-height changes, which include large subsidence due to groundwater extraction in Manila and Bangkok, land uplift in Indonesia and subsidence in Thailand from postseismic deformation resulting from the sequence of large Sumatran earthquakes since 2004, and land subsidence as a result of sediment compaction in Malaysia. Those signals are quantitatively or qualitatively consistent with observations from other sources. This VLM dataset can be used to advance our understanding of the physical mechanisms behind land-height changes and to improve sea level projections in the region.</div></div>","PeriodicalId":101147,"journal":{"name":"Science of Remote Sensing","volume":"10 ","pages":"Article 100176"},"PeriodicalIF":5.7,"publicationDate":"2024-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142658607","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}

引用次数: 0