TY - JOUR
T1 - Evaluation of four image fusion NDVI products against in-situ spectral-measurements over a heterogeneous rice paddy landscape
AU - Kong, Juwon
AU - Ryu, Youngryel
AU - Huang, Yan
AU - Dechant, Benjamin
AU - Houborg, Rasmus
AU - Guan, Kaiyu
AU - Zhu, Xiaolin
N1 - Funding Information:
This research was conducted with support of the Korea Environment Industry & Technology Institute (KEITI) through its Urban Ecological Health Promotion Technology Development Project funded by the Korea Ministry of Environment (MOE) (2019002760002). English proofreading was supported by the Research Institute of Agriculture and Life Sciences, Seoul National University. We have made our in situ data openly accessible via http://environment.snu.ac.kr/landscape_spectral/ and expect that they will be useful for evaluating future image fusion products.
Publisher Copyright:
© 2020 Elsevier B.V.
PY - 2021/2/15
Y1 - 2021/2/15
N2 - Satellite image fusion methods that improve spatial and temporal resolution have significant potential to advance understanding of ecosystem dynamics in space and time. However, systematic evaluations of image fusion methods against in situ spectral data are lacking. Here, we used a suite of in situ spectral data collected at 60 elementary sampling units (10 × 10 m) covering 15 Landsat pixel (30 × 30 m) plots and one Moderate Resolution Imaging Spectroradiometer (MODIS) pixel (250 × 250 m) throughout the entire growing season in a heterogeneous rice paddy landscape to evaluate four state-of-the-art image fusion NDVI products. They include the Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM), Flexible Spatiotemporal DAta Fusion (FSDAF), SaTellite dAta IntegRation (STAIR), and the CubeSat Enabled Spatio-Temporal Enhancement Method (CESTEM); the former three blended Landsat and MODIS data, whereas the latter combined CubeSats, Landsat, and MODIS observations. All fusion products showed strong linear relationships against in situ data when combining all spatial and temporal observations (R2: 0.73 to 0.93) although there were partly negative biases (–1% to –9%). These biases resulted from forcing data to image fusion algorithms, such as Landsat (–4%) and MODIS (–7%). Performance difference between fusion methods were considerably larger for spatial than for temporal variation. Furthermore, Landsat NDVI explained only 17–22% of spatial variation against in situ spectral data, which can be translated into weak performance of image fusion products to predict spatial variability in NDVI. Image fusion products that relied on spatial interpolation showed large biases (–15% to –30%) for a vegetation plot surrounded by mixed land cover plots. Our results highlight key sources of uncertainty and will be instrumental in improving satellite image fusion methods to monitor land surface phenology in space and time.
AB - Satellite image fusion methods that improve spatial and temporal resolution have significant potential to advance understanding of ecosystem dynamics in space and time. However, systematic evaluations of image fusion methods against in situ spectral data are lacking. Here, we used a suite of in situ spectral data collected at 60 elementary sampling units (10 × 10 m) covering 15 Landsat pixel (30 × 30 m) plots and one Moderate Resolution Imaging Spectroradiometer (MODIS) pixel (250 × 250 m) throughout the entire growing season in a heterogeneous rice paddy landscape to evaluate four state-of-the-art image fusion NDVI products. They include the Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM), Flexible Spatiotemporal DAta Fusion (FSDAF), SaTellite dAta IntegRation (STAIR), and the CubeSat Enabled Spatio-Temporal Enhancement Method (CESTEM); the former three blended Landsat and MODIS data, whereas the latter combined CubeSats, Landsat, and MODIS observations. All fusion products showed strong linear relationships against in situ data when combining all spatial and temporal observations (R2: 0.73 to 0.93) although there were partly negative biases (–1% to –9%). These biases resulted from forcing data to image fusion algorithms, such as Landsat (–4%) and MODIS (–7%). Performance difference between fusion methods were considerably larger for spatial than for temporal variation. Furthermore, Landsat NDVI explained only 17–22% of spatial variation against in situ spectral data, which can be translated into weak performance of image fusion products to predict spatial variability in NDVI. Image fusion products that relied on spatial interpolation showed large biases (–15% to –30%) for a vegetation plot surrounded by mixed land cover plots. Our results highlight key sources of uncertainty and will be instrumental in improving satellite image fusion methods to monitor land surface phenology in space and time.
UR - http://www.scopus.com/inward/record.url?scp=85098084686&partnerID=8YFLogxK
U2 - 10.1016/j.agrformet.2020.108255
DO - 10.1016/j.agrformet.2020.108255
M3 - Journal article
AN - SCOPUS:85098084686
SN - 0168-1923
VL - 297
JO - Agricultural and Forest Meteorology
JF - Agricultural and Forest Meteorology
M1 - 108255
ER -
