TY - JOUR
T1 - Assessing the accuracy of Sentinel-2 instantaneous subpixel shorelines using synchronous UAV ground truth surveys
AU - Pucino, Nicolas
AU - Kennedy, David M.
AU - Young, Mary
AU - Ierodiaconou, Daniel
N1 - Publisher Copyright:
© 2022 Elsevier Inc.
PY - 2022/12/1
Y1 - 2022/12/1
N2 - Due to recent technological advancements in the field of cloud-based satellite remote sensing, the barriers to global analysis ready dataset access and processing have lowered drastically increasing the opportunity for large-scale long-term satellite-derived shoreline monitoring solutions. However, validation studies employing synchronous and independent in-situ instantaneous shorelines at scale are limited. Thanks to a unique dataset of 47 synchronous UAV photogrammetric surveys in 15 coastal locations in Victoria, Australia, over 3 years, we validated Sentinel-2 instantaneous subpixel shorelines extracted using the most common water indices, CoastSat and a tidal-balanced convolutional neural network (Unet+++). We assessed the accuracy of the water level estimation using 9 wave run-up models and wave data sourced from a nearshore wave buoy network, a regional hindcast model and a global altimeter wave dataset. This paper also presents the first validation of Deep Learning-derived shorelines using synchronous independent in-situ data. We found the best overall performance and least threshold-sensitive shoreline extraction approach to be the WI index, whilst NDWI had the poorest performances despite being the most utilized across the literature. We found that the thresholdless Unet+++ solution coupled with an ensemble wave run-up model and the altimeter wave data is an adequate trade-off in terms of accuracy, adaptability and wave data access for global scale shoreline studies. Moreover, we highlighted the very high spatial variability of beach-scale shoreline extraction performances and found that generally, the best SDS extractor only performs best along 30–40% of a coastline. Our results suggest that future satellite-derived shoreline approaches should investigate the relationships between scene-dependent variables and shoreline accuracies in order to better understand the causes of beach-scale performance variations and increase the accuracy of satellite-derived shorelines.
AB - Due to recent technological advancements in the field of cloud-based satellite remote sensing, the barriers to global analysis ready dataset access and processing have lowered drastically increasing the opportunity for large-scale long-term satellite-derived shoreline monitoring solutions. However, validation studies employing synchronous and independent in-situ instantaneous shorelines at scale are limited. Thanks to a unique dataset of 47 synchronous UAV photogrammetric surveys in 15 coastal locations in Victoria, Australia, over 3 years, we validated Sentinel-2 instantaneous subpixel shorelines extracted using the most common water indices, CoastSat and a tidal-balanced convolutional neural network (Unet+++). We assessed the accuracy of the water level estimation using 9 wave run-up models and wave data sourced from a nearshore wave buoy network, a regional hindcast model and a global altimeter wave dataset. This paper also presents the first validation of Deep Learning-derived shorelines using synchronous independent in-situ data. We found the best overall performance and least threshold-sensitive shoreline extraction approach to be the WI index, whilst NDWI had the poorest performances despite being the most utilized across the literature. We found that the thresholdless Unet+++ solution coupled with an ensemble wave run-up model and the altimeter wave data is an adequate trade-off in terms of accuracy, adaptability and wave data access for global scale shoreline studies. Moreover, we highlighted the very high spatial variability of beach-scale shoreline extraction performances and found that generally, the best SDS extractor only performs best along 30–40% of a coastline. Our results suggest that future satellite-derived shoreline approaches should investigate the relationships between scene-dependent variables and shoreline accuracies in order to better understand the causes of beach-scale performance variations and increase the accuracy of satellite-derived shorelines.
KW - Accuracy
KW - Deep learning
KW - Satellite
KW - Shoreline
KW - UAV
KW - Validation
UR - http://www.scopus.com/inward/record.url?scp=85138809308&partnerID=8YFLogxK
U2 - 10.1016/j.rse.2022.113293
DO - 10.1016/j.rse.2022.113293
M3 - Article
AN - SCOPUS:85138809308
SN - 0034-4257
VL - 282
JO - Remote Sensing of Environment
JF - Remote Sensing of Environment
M1 - 113293
ER -