Remote Sensing of Autumn Phenology by Including Surface Soil Temperature: Algorithm Development, Calibration, and Validation

Abstract

Phenology exercises a critical control on annual terrestrial ecosystem carbon uptake and indicates interaction between climate and vegetation. Solely vegetation index is insufficient to accurately detect the end of growing season (EOS). Soil temperature (Ts) plays a modulating role in soil microbial functioning and plant growth, while its impact on EOS remains largely unknown. Hence, we compared the potential between Ts and air temperature (Ta) as the indicators of EOS by using flux data from 14 deciduous broadleaf forests, 24 evergreen needleleaf forests (ENF), 7 mixed forests, and 23 nonforests over Northern temperate and boreal regions (30°-60°N) for 2001-2014. The widely used NDVI-based double-logistic approach failed to capture EOS variability for these ecosystems, and we derived a new EOS algorithm with a soil temperature-based scaler, which improved the EOS modeling for all plant functional types. We found that Ts at different depths showed varied abilities for EOS modeling, and Ts at the 0-10 cm depth provided the best estimates of EOS in terms of both numbers of significant sites and the correlation coefficients (R). Estimated EOS occurred earlier by on average 2.9 days than the current MODIS phenology product for ∼56.5% pixels, especially for the ENF ecosystems (∼5.5 days). Our study suggests the usefulness of surface soil temperature for autumn leaf senescence phenology modeling, and that combination of environmental variables with the current modeling strategy can improve our understanding of autumn phenology with future climate change.