Unbiased ESTARFM (ubESTARFM) in R

DESCRIPTION

** Please refer to the GitHub repository for any updates. **

Fine spatial resolution land surface temperature (LST) data are crucial to study heterogeneous landscapes (e.g., agricultural and urban). Some well-known spatiotemporal fusion methods like the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM; Gao et al., 2006) and the Enhanced STARFM (ESTARFM; Zhu et al., 2010), which were originally developed to fuse surface reflectance data, may not be suitable for direct application in LST studies due to the high sub-diurnal dynamics of LST. To address this, we proposed a variant of ESTARFM, referred to as the unbiased ESTARFM (ubESTARFM), specifically designed to accommodate the high temporal dynamics of LST to generate fine-resolution LST estimates.

In ubESTARFM, we implement a local bias correction on the central pixel and similar fine-resolution pixels within the moving window using the mean value of corresponding coarse-resolution pixels as reference. By applying this linear scaling approach, we can scale the systematic biases of the fine-resolution data to a same level of the corresponding coarse-resolution data in each moving window, while maintaining the variation and spatial details of fine-resolution data.

The ubESTARFM algorithm is written in R. We recommend users use a multi-core processor that can allow ubESTARFM to run in parallel and to be more efficient. This collection contains some small spatial extent data for testing purposes. If you are interested in having a comprehensive assessment of ubESTARFM, please refer to the dataset published in the CSIRO Data Access Portal, which contains the full set of data (12 OzFlux sites across Australia for the period of 2013-2021) used in our RSE paper.

UPDATES IN THIS VERSION

• Fixed minor bugs in the algorithm.
• Added scripts for how we processed, fused and evaluated satellite LST (for reference purposes only).
• Updated OzFlux LST data using a new processing strategy, which does not consider the daylight saving time and does explicitly claim the 'seconds' timestep in the TOI (Time of Interests). Compared to the strategy used in our RSE paper, the new one is expected to better coincide with the satellite overpass time.

CITATION

Yu, Y., Renzullo, L. J., McVicar, T. R., Malone, B. P. and Tian, S., 2023. Generating daily 100 m resolution land surface temperature estimates continentally using an unbiased spatiotemporal fusion approach. Remote Sensing of Environment, 297, 113784. https://doi.org/10.1016/j.rse.2023.113784