Estimation of subpixel MODIS water temperature near coastlines using the SWTI algorithm

- This is the first SWTI (Sharpening Water Thermal Imagery) application to MODIS.
- SWTI disaggregates at 250 m the majority of MODIS water pixels near coastlines.
- Results are validated using contemporary ASTER sea surface temperature images.
- Temperature errors introduced by SWTI are < 1 K.
- SWTI could be easily applied automatically to the standard MODIS products.

Satellite derived water surface temperature maps are widely used in many environmental studies and applications. The Moderate Resolution Imaging Spectroradiometer (MODIS) is among the widely used sensors in this field and sea surface temperature (SST) is one of the standard quantities derived from MODIS imagery. However, MODIS SST maps have limited applications in near-shore and coastal environments due to inadequate spatial resolution of 1 km. This problem means that the MODIS pixels closer than 1 km from the shore are mixed pixels, i.e. they include by both water and land, and must be discarded from the SST map. In this work SWTI (Sharpening Water Thermal Imagery) methods were applied to MODIS thermal imagery for the first time. The information required by SWTI regarding cover fractions and perpendicular vegetation index was obtained from the MODIS images in the Visible-Near Infrared bands at a spatial resolution of 250 m. In this way, the SST MODIS maps were extended to a minimum distance of 250 m from the shore. The SWTI results were evaluated using as a reference the SST computed from two ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) images acquired simultaneously to the MODIS images and covering the same areas. The applied validation methodology provides an evaluation of the deviations introduced by SWTI separated from the pre-existing differences between MODIS SST and ASTER SST upscaled to 250 m. For sea coast environments, SWTI was able to compute the SST of more than 80% of the pixels close to the shore at a spatial resolution of 250 m. This represents an increase of 67% compared to the number of pixels obtainable using a simple downscaling method based on polynomial interpolation; in areas with lagoons and estuaries the increases were + 70% and + 60% respectively. The ASTER SST comparison showed that the SST bias and the unsystematic deviation introduced by SWTI were ∆S ≤ 0.45 K and σ(ϵS) ≤ 0.88 K respectively, corresponding to a total deviation TD ≤ 0.97 K. SWTI is written in the IDL language and could be adapted for automatic application to MODIS images.