Deep Learning Model Transfer in Forest Mapping Using Multi-Source Satellite SAR and Optical Images
Deep learning (DL) models are gaining popularity in forest variable prediction using Earth observation (EO) images. However, in practical forest inventories, reference datasets are often represented by plot- or stand-level measurements, while high-quality representative wall-to-wall reference data f...
Published in: | Remote Sensing |
---|---|
Main Authors: | , , , , |
Format: | Article in Journal/Newspaper |
Language: | English |
Published: |
MDPI AG
2023
|
Subjects: | |
Online Access: | https://doi.org/10.3390/rs15215152 https://doaj.org/article/6da10adb37b442adb1024198d3c16255 |
id |
ftdoajarticles:oai:doaj.org/article:6da10adb37b442adb1024198d3c16255 |
---|---|
record_format |
openpolar |
spelling |
ftdoajarticles:oai:doaj.org/article:6da10adb37b442adb1024198d3c16255 2023-12-10T09:54:13+01:00 Deep Learning Model Transfer in Forest Mapping Using Multi-Source Satellite SAR and Optical Images Shaojia Ge Oleg Antropov Tuomas Häme Ronald E. McRoberts Jukka Miettinen 2023-10-01T00:00:00Z https://doi.org/10.3390/rs15215152 https://doaj.org/article/6da10adb37b442adb1024198d3c16255 EN eng MDPI AG https://www.mdpi.com/2072-4292/15/21/5152 https://doaj.org/toc/2072-4292 doi:10.3390/rs15215152 2072-4292 https://doaj.org/article/6da10adb37b442adb1024198d3c16255 Remote Sensing, Vol 15, Iss 21, p 5152 (2023) vegetation mapping deep learning UNet synthetic aperture radar forest height interferometry Science Q article 2023 ftdoajarticles https://doi.org/10.3390/rs15215152 2023-11-12T01:35:59Z Deep learning (DL) models are gaining popularity in forest variable prediction using Earth observation (EO) images. However, in practical forest inventories, reference datasets are often represented by plot- or stand-level measurements, while high-quality representative wall-to-wall reference data for end-to-end training of DL models are rarely available. Transfer learning facilitates expansion of the use of deep learning models into areas with sub-optimal training data by allowing pretraining of the model in areas where high-quality teaching data are available. In this study, we perform a “model transfer” (or domain adaptation) of a pretrained DL model into a target area using plot-level measurements and compare performance versus other machine learning models. We use an earlier developed UNet based model (SeUNet) to demonstrate the approach on two distinct taiga sites with varying forest structure and composition. The examined SeUNet model uses multi-source EO data to predict forest height. Here, EO data are represented by a combination of Copernicus Sentinel-1 C-band SAR and Sentinel-2 multispectral images, ALOS-2 PALSAR-2 SAR mosaics and TanDEM-X bistatic interferometric radar data. The training study site is located in Finnish Lapland, while the target site is located in Southern Finland. By leveraging transfer learning, the SeUNet prediction achieved root mean squared error (RMSE) of <semantics> 2.70 </semantics> m and R 2 of 0.882, considerably more accurate than traditional benchmark methods. We expect such forest-specific DL model transfer can be suitable also for other forest variables and other EO data sources that are sensitive to forest structure. Article in Journal/Newspaper taiga Lapland Directory of Open Access Journals: DOAJ Articles Remote Sensing 15 21 5152 |
institution |
Open Polar |
collection |
Directory of Open Access Journals: DOAJ Articles |
op_collection_id |
ftdoajarticles |
language |
English |
topic |
vegetation mapping deep learning UNet synthetic aperture radar forest height interferometry Science Q |
spellingShingle |
vegetation mapping deep learning UNet synthetic aperture radar forest height interferometry Science Q Shaojia Ge Oleg Antropov Tuomas Häme Ronald E. McRoberts Jukka Miettinen Deep Learning Model Transfer in Forest Mapping Using Multi-Source Satellite SAR and Optical Images |
topic_facet |
vegetation mapping deep learning UNet synthetic aperture radar forest height interferometry Science Q |
description |
Deep learning (DL) models are gaining popularity in forest variable prediction using Earth observation (EO) images. However, in practical forest inventories, reference datasets are often represented by plot- or stand-level measurements, while high-quality representative wall-to-wall reference data for end-to-end training of DL models are rarely available. Transfer learning facilitates expansion of the use of deep learning models into areas with sub-optimal training data by allowing pretraining of the model in areas where high-quality teaching data are available. In this study, we perform a “model transfer” (or domain adaptation) of a pretrained DL model into a target area using plot-level measurements and compare performance versus other machine learning models. We use an earlier developed UNet based model (SeUNet) to demonstrate the approach on two distinct taiga sites with varying forest structure and composition. The examined SeUNet model uses multi-source EO data to predict forest height. Here, EO data are represented by a combination of Copernicus Sentinel-1 C-band SAR and Sentinel-2 multispectral images, ALOS-2 PALSAR-2 SAR mosaics and TanDEM-X bistatic interferometric radar data. The training study site is located in Finnish Lapland, while the target site is located in Southern Finland. By leveraging transfer learning, the SeUNet prediction achieved root mean squared error (RMSE) of <semantics> 2.70 </semantics> m and R 2 of 0.882, considerably more accurate than traditional benchmark methods. We expect such forest-specific DL model transfer can be suitable also for other forest variables and other EO data sources that are sensitive to forest structure. |
format |
Article in Journal/Newspaper |
author |
Shaojia Ge Oleg Antropov Tuomas Häme Ronald E. McRoberts Jukka Miettinen |
author_facet |
Shaojia Ge Oleg Antropov Tuomas Häme Ronald E. McRoberts Jukka Miettinen |
author_sort |
Shaojia Ge |
title |
Deep Learning Model Transfer in Forest Mapping Using Multi-Source Satellite SAR and Optical Images |
title_short |
Deep Learning Model Transfer in Forest Mapping Using Multi-Source Satellite SAR and Optical Images |
title_full |
Deep Learning Model Transfer in Forest Mapping Using Multi-Source Satellite SAR and Optical Images |
title_fullStr |
Deep Learning Model Transfer in Forest Mapping Using Multi-Source Satellite SAR and Optical Images |
title_full_unstemmed |
Deep Learning Model Transfer in Forest Mapping Using Multi-Source Satellite SAR and Optical Images |
title_sort |
deep learning model transfer in forest mapping using multi-source satellite sar and optical images |
publisher |
MDPI AG |
publishDate |
2023 |
url |
https://doi.org/10.3390/rs15215152 https://doaj.org/article/6da10adb37b442adb1024198d3c16255 |
genre |
taiga Lapland |
genre_facet |
taiga Lapland |
op_source |
Remote Sensing, Vol 15, Iss 21, p 5152 (2023) |
op_relation |
https://www.mdpi.com/2072-4292/15/21/5152 https://doaj.org/toc/2072-4292 doi:10.3390/rs15215152 2072-4292 https://doaj.org/article/6da10adb37b442adb1024198d3c16255 |
op_doi |
https://doi.org/10.3390/rs15215152 |
container_title |
Remote Sensing |
container_volume |
15 |
container_issue |
21 |
container_start_page |
5152 |
_version_ |
1784901529934757888 |