File Download

There are no files associated with this item.

  Links for fulltext
     (May Require Subscription)
Supplementary

Article: Improved fine-scale tropical forest cover mapping for Southeast Asia using Planet-NICFI and Sentinel-1 imagery

TitleImproved fine-scale tropical forest cover mapping for Southeast Asia using Planet-NICFI and Sentinel-1 imagery
Authors
Issue Date24-Jul-2023
PublisherAmerican Association for the Advancement of Science
Citation
Journal of Remote Sensing, 2023 How to Cite?
Abstract

The  accuracy  of  existing  forest  cover  products  typically  suffers  from  “rounding”  errors  arising  from classifications that estimate the fractional cover of forest in each pixel, which often exclude the presence of large,  isolated  trees  and  small  or  narrow  forest  clearings,  and  is  primarily  attributable  to  the  moderate resolution of the imagery used to make maps. However, the degree to which such high-resolution imagery can mitigate this problem, and thereby improve large-area forest cover maps, is largely unexplored. Here, we developed  an  approach  to  map  tropical  forest  cover  at  a  fine  scale  using  Planet  and  Sentinel-1  synthetic aperture radar (SAR) imagery in the Google Earth Engine platform and used it to map all of Southeastern Asia’s forest cover. The machine learning approach, based on the Random Forests models and trained and validated using a total of 37,345 labels collected from Planet imagery across the entire region, had an accuracy of 0.937 and an F1 score of 0.942, while a version based only on Planet imagery had an accuracy of 0.908 and F1 of 0.923. We compared the accuracy of our resulting maps with five existing forest cover products derived  from  medium-resolution  optical-only  or  combined  optical-SAR  approaches  at  3000  randomly selected locations. We found that our approach overall achieved higher accuracy, and helped minimize the rounding  errors  commonly  found  along  small  or  narrow  forest  clearings  and  deforestation  frontiers  where isolated trees are common. However, the forest area estimates varied depending on topographic location and showed smaller differences in highlands (areas >300 m above sea level) but obvious differences in complex lowland landscapes. Overall, the proposed method shows promise for monitoring forest changes, particularly those caused by deforestation frontiers. Our study also represents one of the most extensive applications of Planet imagery to date, resulting in an open, high-resolution map of forest cover for the entire Southeastern Asia region.


Persistent Identifierhttp://hdl.handle.net/10722/331060
ISSN
2023 Impact Factor: 8.8

 

DC FieldValueLanguage
dc.contributor.authorYang, Feng-
dc.contributor.authorJiang, Xin-
dc.contributor.authorZiegler, Alan D-
dc.contributor.authorEstes, Lyndon D-
dc.contributor.authorWu, Jin-
dc.contributor.authorChen, Anping-
dc.contributor.authorCiais, Philippe-
dc.contributor.authorWu, Jie-
dc.contributor.authorZeng, Zhenzhong-
dc.date.accessioned2023-09-21T06:52:26Z-
dc.date.available2023-09-21T06:52:26Z-
dc.date.issued2023-07-24-
dc.identifier.citationJournal of Remote Sensing, 2023-
dc.identifier.issn2694-1589-
dc.identifier.urihttp://hdl.handle.net/10722/331060-
dc.description.abstract<p>The  accuracy  of  existing  forest  cover  products  typically  suffers  from  “rounding”  errors  arising  from classifications that estimate the fractional cover of forest in each pixel, which often exclude the presence of large,  isolated  trees  and  small  or  narrow  forest  clearings,  and  is  primarily  attributable  to  the  moderate resolution of the imagery used to make maps. However, the degree to which such high-resolution imagery can mitigate this problem, and thereby improve large-area forest cover maps, is largely unexplored. Here, we developed  an  approach  to  map  tropical  forest  cover  at  a  fine  scale  using  Planet  and  Sentinel-1  synthetic aperture radar (SAR) imagery in the Google Earth Engine platform and used it to map all of Southeastern Asia’s forest cover. The machine learning approach, based on the Random Forests models and trained and validated using a total of 37,345 labels collected from Planet imagery across the entire region, had an accuracy of 0.937 and an F1 score of 0.942, while a version based only on Planet imagery had an accuracy of 0.908 and F1 of 0.923. We compared the accuracy of our resulting maps with five existing forest cover products derived  from  medium-resolution  optical-only  or  combined  optical-SAR  approaches  at  3000  randomly selected locations. We found that our approach overall achieved higher accuracy, and helped minimize the rounding  errors  commonly  found  along  small  or  narrow  forest  clearings  and  deforestation  frontiers  where isolated trees are common. However, the forest area estimates varied depending on topographic location and showed smaller differences in highlands (areas >300 m above sea level) but obvious differences in complex lowland landscapes. Overall, the proposed method shows promise for monitoring forest changes, particularly those caused by deforestation frontiers. Our study also represents one of the most extensive applications of Planet imagery to date, resulting in an open, high-resolution map of forest cover for the entire Southeastern Asia region.<br></p>-
dc.languageeng-
dc.publisherAmerican Association for the Advancement of Science-
dc.relation.ispartofJournal of Remote Sensing-
dc.rightsThis work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.-
dc.titleImproved fine-scale tropical forest cover mapping for Southeast Asia using Planet-NICFI and Sentinel-1 imagery-
dc.typeArticle-
dc.identifier.doi10.34133/remotesensing.0064-
dc.identifier.issnl2694-1589-

Export via OAI-PMH Interface in XML Formats


OR


Export to Other Non-XML Formats