Sea ice surface temperature retrieval from Landsat 8/TIRS: Evaluation of five methods against in situ temperature records and MODIS IST in Arctic region

• Published in: Remote sensing of environment 2020-10, Vol.248, p.111975, Article 111975
• Main Authors: Fan, Pei, Pang, Xiaoping, Zhao, Xi, Shokr, Mohammed, Lei, Ruibo, Qu, Meng, Ji, Qing, Ding, Minghu
• Format: Article
• Language: English
• Subjects: Air temperature; Arctic climates; Arctic sea ice; Arctic zone; Automatic weather stations; Bias; Buoy; Buoys; Climate studies; High resolution; Ice; Ice environments; Ice surface temperature; In situ measurement; Infrared imagery; Landsat; Landsat 8; Landsat satellites; Mass balance; MODIS; MODIS IST; Polar environments; Polar regions; Remote sensing; Root-mean-square errors; Satellite imagery; Sea ice; Sea ice temperatures; Skin; Skin temperature; Snow and ice; Spatial discrimination; Spatial resolution; Surface temperature; Surface-air temperature relationships; Temperature; Temperature data
• ISSN: 0034-4257; 1879-0704
• DOI: 10.1016/j.rse.2020.111975

Accurate and high-resolution sea ice surface temperature (IST) data is of great importance for Arctic climate studies. However, the validation of high-resolution IST data using in situ measurements in polar sea ice regions is lacking. This study assesses the accuracy of three split-window (SW) and two single-channel (SC) methods, based on Landsat 8 thermal infrared imagery at 100 m resolution over Arctic sea ice regions. The SW methods are proposed by Jin et al. (2015) (SW-Jin), Jiménez-Muñoz et al. (2014) (SW-JM), and Du et al. (2015) (SW-Du). The SC methods are proposed by Jiménez-Muñoz et al. (2014) (SC-JM) and Barsi et al. (2003, 2005) (SC-Barsi). IST data derived from 58 scenes of the Landsat 8 images were compared with coincident in situ ice skin temperatures and near-surface air temperatures, as measured by a combination of Ice Mass Balance (IMB) buoys, Snow and Ice Mass Balance Array (SIMBA) buoys, and automatic weather stations. SW-Du offers the best accuracy when compared with the skin temperature (bias: −1.06 K; root mean square error (RMSE): 2.08 K) and near-surface air temperature (bias: -0.98 K; RMSE: 2.17 K). SC-Barsi ranks second, with a bias of −1.55 K and RMSE of 2.40 K for the skin temperature. As for precision, IST from the Moderate Resolution Imaging Spectrometer (MODIS) has best performance (standard deviation (STD): 1.69 K), followed by SW-Du, SW-JM, and SC-Barsi (STD: 1.80 K, 1.82 K, and 1.85 K, respectively). The Landsat IST outperforms the MODIS IST in narrow lead areas, owing to its better spatial resolution, and SW-JM and SC-Barsi methods agree best with the MODIS IST in leads and marginal ice zone scenes, respectively. As all three SW methods are constrained by banding effects with different degrees in a lead scene, they are not recommended to be applied on an image scene with severe banding artifacts. The small bias (1.26 K) and high correlation (0.99) between skin temperature and near-surface air temperature prove the capability of using near-surface air temperature as a substitute for validating a satellite IST data if skin temperature data are not available. •IST from Landsat 8/TIRS images is accurate enough for sea ice related use.•Split window algorithm has better accuracy than single channel method.•Fine resolution IST images can capture small details on sea ice regions.•Banding artifacts in TIRS images still contaminate IST maps.