Antarctic Landsat 8 single-channel sea surface temperature algorithm supporting data

Ocean heat transport drives approximately half of the ice mass loss in Antarctica and measurements near glacier ice fronts are crucial for understanding the processes driving that loss. However, field observations in these remote, harsh environments are exceedingly difficult to acquire and are concomitantly sparse. Remotely sensed sea surface temperature (SST) acquisitions consistently capture ocean heat transport in these regions offering a potential alternative to field observations. This alternative has been widely underutilized to date, in part because polar SST products have historically been considered less reliable than those at lower latitudes. SST products suffer from complex and poorly known sub-pixel-scale ice-ocean processes, mixed surface states that prevent clean ocean surface observations, and limited calibration and validation datasets, underscoring the need for improved methods in polar regions. Landsat\u2019s thermal infrared record, extending back to 1982, has not produced a dedicated SST algorithm, and no temperature product of any sort for polar regions. Yet with its relatively high spatial resolution and long time series, Landsat has substantial potential for observing partially ice-covered seas. Here, we provide data that supports the development and validation of an open-source, cloud-native Landsat 8 single-channel SST algorithm for West Antarctica that can be adapted and transferred to previous Landsat sensors that only had one thermal infrared band. Cross-calibration of this new SST product with MODIS SSTs found a root mean square error of 0.26 degrees C and yielded a 30% improvement upon the existing Landsat Surface Temperature algorithm used by the mission elsewhere around the world. Validation with seven iQuam Argo SST measurements over the last 12 years showed a strong, nearly one-to-one agreement with a calculated RMSE of 0.15 degrees C, although the scarcity of field observations prevented us from sampling the entire range of variability. The publicly available pipeline is fully open-source and replicable with the exception of the electromagnetic atmospheric propagation model (MODTRAN). The automated and cloud-optimized nature of our pipeline processes a new Landsat image from discovery to calibration in approximately 86 seconds, accelerating science by reducing data searches to mere seconds with only a few lines of Python code, eliminating manual downloads, and enabling near-instant replication when new collections are released.