Exploring the Role of Land Cover Composition in Land Surface Phenology: an case study in the 2002 Ponil Complex Fire, New Mexico

Abstract:

Land surface phenology (LSP) characterizes the timing and magnitude of seasonal dynamics in the area-integrated vegetation communities observed for an entire satellite pixel. Although many studies have modelled LSP with environmental factors (climate and topography), little attention has been paid on the role of land cover composition in LSP variation. To fill this gap, this study uses the Generalized Boosted Regression Modeling, a multivariate machine learning approach, to assess the contributions of land cover composition as well as climate and topography on LSP shifts after the 2002 Ponil Complex Fire. Specifically, LSP metrics were start (SOS) and end (EOS) of growing season and annual greenness maximum (GMax) and minimum (GMin) that were derived from daily 500m MODIS data from 2001-2018. Environmental factors were: (1) seasonal precipitation and maximum and minimum temperature calculated from 1 km Daymet data from 2001-2018, (2) topographic parameters (elevation, slope, and aspects) derived from 30m Shuttle Radar Topography Mission data, (3) land cover composition variables [vegetation fractional coverage (VFC) and tree proportion (TP)] calculated from 3 m PlanetScope images in 2018, and (4) VFC and TP from 2001-2017 estimated from MODIS GMax and GMin/GMax based on a robust linear relationship model. These environmental variables were applied to model spatial LSP variability in 2018 and interannual LSP variability from 2001-2018. Results shows that TP was the main driver of spatial variability of both SOS and EOS with relative importance of 22% and 42%, respectively. For the interannual variability, the main driver was precipitation in March-May for SOS (31%) and daily maximum temperature in September-November for EOS (30%). Land cover composition was also provided significant contribution, which was 7% (ranking 8/8) of TP and 9% (ranking 4/8) of VFC to SOS, and 13% (ranking 2/8) of TP and 10% (ranking 4/8) of VFC to EOS. Moreover, with the increase of TP, SOS increased and EOS decreased in both spatial and interannual models. Overall, this study demonstrates the importance of land cover composition in modelling LSP variation.