Search

[Display omitted] • Our algorithm improved LSP retrieval in ecosystems with low seasonal amplitude. • The detectability of LSP increases when the seasonal amplitude increases. • Evaluation results showed the reliability of the algorithm in various ecosystems. • These LSP retrievals could help to quantify ecological response to climate change. Land surface phenology (LSP), the study of the seasonal vegetation dynamics from remote sensing imagery, provides crucial information for plant monitoring and reflects the responses of ecosystems to climate change. The Moderate Resolution Imaging Spectroradiometer (MODIS) phenology product (MCD12Q2) provides global LSP information, but it has large spatial gaps in many regions, especially in ecosystems where rainfall influences phenology more than temperature. This study aimed to improve spatial coverage of LSP retrieval in these ecosystems. To do so, we used a regionally modified threshold algorithm for LSP retrievals, which were tested over continental Australia as it includes diverse landscapes of arid, mesic, and forest environments. We generated LSP metrics annually from 2003 to 2018 using satellite Enhanced Vegetation Index (EVI) time series at 500 m resolution, including the start, peak, end, and length of growing seasons, the minimum EVI value prior to and after the peak date, the seasonal maximum EVI value, the integral EVI value during the growing season (an approximation of productivity), and seasonal amplitude (maximum EVI value minus minimum EVI). Our regionally optimised algorithm improved the spatial coverage of LSP information in Australia from only 26 % of the continent to 70 % averaged across 16 years. Our results showed that the growing season amplitude was low (EVI < 0.1) over arid/semi-arid shrublands and savannas, tropical and subtropical savannas, and temperate evergreen forests, whose LSP metrics were captured by our regional algorithm and not by the global product. Some ecosystems, such as arid/semi-arid shrublands and savannas, showed more irregular phenology with low seasonal dynamics, and the growing seasons could skip a year or occur more than once in a year depending on climate conditions. Our algorithm was more sensitive to ecosystems with low seasonal amplitudes. We found that the detectability of LSP increases as the growing season amplitude increases, regardless of vegetation cover. Evaluation of the LSP metrics using eddy covariance flux tower measurements of gross primary productivity (GPP) demonstrated the reliability and accuracy of the algorithm. These improved LSP retrievals provide a greater understanding of the vegetation phenology across diverse ecosystems, especially savanna, shrubland, and evergreen forest ecosystems that cover more than 30 % of the land globally. The LSP provides essential information for ecological and agricultural studies such as quantifying bushfire fuel accumulation and forest carbon cycling, whilst enhancing our capacity for quantifying ecological responses to climate change. [ABSTRACT FROM AUTHOR]