17:15 〜 19:15
[ACG39-P11] Using low Earth orbit satellites and ground observations to create data-driven carbon flux models.
キーワード：carbon, remote sensing
NASA's MODIS observations have been an important part of remote sensing based approaches to understanding the carbon cycle given their extensive temporal and spatial coverage. This research creates gross primary production estimations at an Asia-wide level utilizing MODIS and eddy covariance tower observations, an SVR machine learning algorithm and data-driven upscaling. In an attempt to better understand the impact of NASA's adjustments to their MODIS data pre-processing methods in order to account for sensor degradation, these estimations are calculated for 3 MODIS collections: Collection 5, Collection 6, and Collection 6.1. First, using the trend of the anomaly of the normalized difference vegetation index derived from MODIS near-infrared and red reflectance band, it is shown that Collection 5 trends are much lower than those of Collection 6 and 6.1. Then, the outputs of the models are analyzed, and the trend of the anomaly of GPP from each collection is compared. Similarly to the MODIS data, the GPP estimations trend for Collection 5 is shown to be much lower than that of Collections 6 and 6.1. GPP estimations from the model outputs are then compared to the GPP estimations from the TRENDY suite of process-based climate models. Next steps are then discussed: this includes newer and increased ground observations, combining MODIS with other low Earth orbit satellites, and new machine learning methods, which will hopefully result in an improved model.