[A-CG36] 衛星による地球環境観測

[ACG36-03] A new era of vegetation phenology monitoring: an application of third-generation geostationary satellite (Himawari/AHI) with snow-free vegetation index (NDGI)

*笹川 大河¹、李 偉³、山本 雄平³、市井 和仁³、奈佐原 顕郎² (1.筑波大学 理工情報生命学術院、2.筑波大学 生命環境系、3.千葉大学 環境リモートセンシング研究センター)

キーワード：植生フェノロジー、静止気象衛星、ひまわり、Phenological Eyes Network (PEN)

Vegetation phenology, such as start of the spring (SOS), end of the spring (EOS), start of the fall (SOF), and end of the fall (EOF), plays an essential role in carbon, nitrogen, and water cycle in terrestrial ecosystems. Some researchers have tried to detect these phenology with the time series of vegetation indices derived from remotely sensed data with artificial satellites. For the detection of vegetation phenology, NDVI (Normalized Difference Vegetation Index) or EVI (Enhanced Vegetation Index) has been widely used. However, some researchers reported that these vegetation indices must be strongly affected by the changes in the background of the canopy, especially snow covering and melting of the understories. It must cause significant noises (e.g., fake greening up) for phenology detection, so a new snow-free vegetation index named NDGI has been developed, and it showed better performance for the estimation of vegetation phenology with MODIS data. In addition to developing the new snow-free vegetation indices, the effectiveness of hyper-temporal remote sensing for vegetation phenology monitoring has been gradually revealed. Modern hyper-temporal satellite observation for vegetation phenology has been enabled by a series of third-generation geostationary satellites, which started from a Japanese geostationary satellite named "Himawari/AHI." Although Himawari/AHI provides the land surface information every 10 minutes, reducing the cloud's effects, only NDVI or EVI2 derived from Himawari/AHI has been applied for vegetation phenology monitoring. Therefore, this study aimed to derive the NDGI from the hyper-temporal dataset observed by the geostationary satellite (Himawari/AHI) and accurately detect the vegetation phenology of deciduous forests around East Asia. Firstly, we conducted atmospheric correction for the top of atmosphere reflectance value observed by the Himawari/AHI. Then, we calculated the time series of NDGI from our land surface reflectance dataset every 10 minutes from 9 AM to 3 PM (local time). Finally, we applied the double logistic curve fitting for the hyper-temporal dataset, which is partially cloud-masked, and estimated the timing of each vegetation phenology. For the accuracy assessment of our methods, we used in-situ observation datasets provided by the Phenological Eyes Network (PEN), one of the longest phenology monitoring networks in the world. Our results show that the combination of Himawari/AHI and NDGI more accurately estimated vegetation phenology compared to previous methods. Also, we found that the timing of spring phenology was detected more precisely than that of fall phenology. It must be because of the heterogeneity of individual trees during the fall phenology period. Our results anticipate that the combination of hyper-temporal datasets obtained from geostationary satellites and snow effect-free indices, such as NDGI, can be a powerful way to detect vegetation phenology, and we can apply this method to other geostationary satellites, such as GeoKompsat-2A/AMI, MTG-I/FCI.