Remote sensing of vegetation phenology in the northern hemisphere from multi-channel passive microwave measurements of Chinese FengYun-3D satellite

Vegetation phenology regulates the inter-annual variations in carbon and water dynamics in terrestrial ecosystems, and serves as a key indicator of vegetation-climate interaction. Chinese FengYun-3D satellite multi-channel passive microwave measurements are responsive to seasonal changes of vegetation structure and internal water status, and have a daily global coverage under both clear and cloudy skies, offering valuable and complementary phenological information to optical-infrared measurements. However, no studies have yet explored their potential for global phenology extraction. Here we evaluated the capability of Normalized Emissivity Difference Vegetation index (NEDVI), which was derived from FengYun-3D X- and Ka- band microwaves, to extract forest and grass phenological dates in the Northern Hemisphere, including the start, end and length of the growing season (SOS, EOS and LOS). By testing three phenology models and two extraction methods at 31 flux sites from 2020 to 2022, NEDVI-derived SOS, EOS and LOS were found to be significantly correlated with those derived from in-situ gross primary production (GPP). No single model or extraction method can produce an absolutely superior accuracy in extracting phenological dates, while use of multi-model mean may greatly reduce the uncertainties. NEDVI-based relative threshold extraction method showed an overall bias of less than 2 days in the phenological dates when considering the multi-model average, outperforming the maximum rate of curvature method. Performances of the time series of NEDVI are overall better for the extraction of SOS than EOS, and are also comparable to those of MODIS and VIIRS global phenology products. Spatial patterns and latitudinal variations in NEDVI-derived phenology align with the two optical phenology products in the Northern Hemisphere. In evergreen forest types, NEDVI-derived SOS, EOS and LOS tend to present earlier, later and longer than those from MODIS and VIIRS, respectively. In addition, the product of NEDVI-derived growing season length and maximum carbon uptake capacity can better explain annual GPP variation across forest and grass sites (R² =0.43) compared to that of MODIS (R² =0.19) and VIIRS (R² =0.37). These findings demonstrate that FengYun-3D microwave NEDVI is promising for global retrievals of phenological dates. We also highlight that using the multi-model mean, instead of relying on a single model, can greatly enhance the robustness of microwave-based phenology extraction with lower uncertainty.