Water mask
Although European water bodies are relatively stable, intra- and inter-annual water body dynamics can occur as a consequence of tides, weather events, floods, and climate change. As such, the long time series of AVHRR water body data can also serve as basis for detailed analyses on weather-, climate- and human induced dynamics of water bodies and surface water.
The derivation of a Water Mask is one of the first steps within the TIMELINE processing, as many subsequent processors, which process only land pixels, rely on the water mask as an input.
Water bodies are derived using a dynamical threshold approach from Top of Atmosphere (TOA) reflectance data of bands 1 and band 2, brightness temperatures of band 4, the solar zenith angle, a pre-calculated cloud mask, and a global reference water mask.
Considering larger areas - as covered by the AVHRR swaths -, water surfaces show a relatively strong diversity in reflection properties. Such, the mean reflection in the near infrared (NIR) part of the spectrum varies with latitude, longitude and view conditions within one observation.
To account for the varying reflection properties, a dynamical thresholding method (Klein et al. 2015) was adapted to AVHRR and extended by a module that allows flexible threshold classification. Further reflectance tests as well as a cloud shadow test and a dynamic local temperature test are part of the water body processor. A full description of the processing flow is found in Dietz et al. (2017).
The methodology was tested on the base of 57 Landsat reference datasets. Water bodies were derived from the Landsat scenes using FMASK (Zhu et al. 2015) and a manually correction in case of e.g. cloud shadows. The assessment revealed an overall accuracy 95% of the AVHRR water body detection. Detected errors were partly due to bad geolocation of the respective AVHRR datasets. It can be assumed that overall accuracy of the water detection algorithm is even higher. Note however, that the TIMELINE water mask does not include sun glint areas.
References:
Dietz, A; Klein, I; Gessner, U; Frey, C; Künzer, C & Dech, S (2017) Detection of Water Bodies from AVHRR Data — A TIMELINE Thematic Processor. Remote Sensing, 9 (1): 57.
Klein, I.; Dietz, A.; Gessner, U.; Dech, S.; Künzer, C. Results of the GlobalWaterPack: A novel product to assess inland water body dynamics on a daily basis. Remote Sens. Lett. 2015, 6, 78–87.
Zhu, Z; Wang, Shixiong; Woodcock, CE (2015): Improvement and expansion of the Fmask algorithm: cloud, cloud shadow, and snow detection for Landsats 4-7, 8, and Sentinel 2 images. Remote Sensing of Environment 159: 269-277.