Reconciling the inconsistency of annual temperature cycles modelled from Landsat and MODIS LSTs through a percentile approach

Abstract

ABSTRACT Land surface temperature (LST) is an important variable to understand surface energy fluxes, land–atmosphere interactions, and urban thermal environments. Time series analysis of LSTs through semi-physical models such as the annual temperature cycle (ATC) model has become critical for these understandings. However, studies are lacking in examining and reconciling the inconsistency of time series LST modelling results across spatial scales, weakening the reliability of these semi-physical models to characterize landscape thermal patterns. In this study, a percentile approach was used to reveal and reconcile discrepancies of ATC parameters estimated from Landsat (100 m) and Moderate Resolution Imaging Spectroradiometer (MODIS, 1000 m) LSTs. Results showed substantial differences across spatial scales for each of the ATC parameters, i.e. mean annual surface temperature (MAST), yearly amplitude of surface temperature (YAST), and revised phase shift (RPS), within the same land cover (e.g. 4.0 K difference between MAST estimated from Landsat LSTs and that from MODIS LSTs for grassland). The spatial distribution of ATC parameters estimated from MODIS LSTs across land cover types was quite different from that from Landsat LSTs. The percentile aggregation analysis suggested that the difference between MAST/YAST (and RPS) derived from MODIS LSTs and Landsat-aggregated values at the 25th (and 40th) percentile within a MODIS block was close to zero. Further regression analysis showed that differences in ATC parameters, particularly MAST and YAST, derived from different datasets could be reconciled. Our study offers new insights into understanding inconsistencies in and reconciliations of ATC parameters modelled at different spatial scales for quantifying landscape thermal patterns spatially and temporally.