Object-based cloud and cloud shadow detection in Landsat imagery
Citations
2,469 citations
1,018 citations
Cites background or methods from "Object-based cloud and cloud shadow..."
...Most of the cloud detection algorithms are heavily dependent on the thermal band, as cloud pixels are much colder than clear-sky pixels (Huang, Thomas, et al., 2010; Irish et al., 2006; Zhu & Woodcock, 2012)....
[...]
...The pixels surrounding clouds and cloud shadows are masked out in the Fmask algorithm because many of these pixels may still be influenced by the thin edges of clouds and their shadows (Zhu & Woodcock, 2012)....
[...]
...%) for cloud detection over water (Zhu & Woodcock, 2012)....
[...]
...…that cloud shadows are usually dark in the Near Infrared (NIR) band, the original Fmask algorithm uses the difference between the NIR band and the flood-fill transformation (Soille, 1999; Soille, Vogt, & Colombo, 2003) of the same NIR band to extract potential cloud shadows (Zhu & Woodcock, 2012)....
[...]
...The Fmask algorithm matches clouds with their shadows based on similarity measurements (Zhu & Woodcock, 2012)....
[...]
981 citations
Cites methods from "Object-based cloud and cloud shadow..."
...The robust feature of the RIRLS method reduces the influence of ephemeral changes, or pixels affected by clouds, shadows, or snow that were not identified by Fmask. ρ̂ i; xð ÞRIRLS ¼ a0;i þ a1;i cos 2π Τ x þ b1;i sin 2π Τ x þ a2;i cos 2π ΝΤ x þ b2;i sin 2π ΝΤ x ð1Þ where, x Julian date I the ith Landsat Band T number of days per year (T = 365) N number of years of Landsat data a0,i coefficient for overall values for the ith Landsat Band a1,i,b1,i coefficients for intra-annual change for the ith Landsat Band a2,i,b2,i coefficients for inter-annual change for the ith Landsat Band ρ̂ i; xð ÞRIRLS predicted value for the ith Landsat Band at Julian date x based on RIRLS fitting....
[...]
...The cloud cover estimates are derived from a newly developed Fmask algorithm (Zhu & Woodcock, 2012) for Path 12 Row 31 between 1982 and 2011....
[...]
...All available Level 1 Terrain (corrected) (L1T) Landsat TM/ETM+ images for Worldwide Reference System (WRS) Path 12 and Row 31 with cloud cover less than 80% (based on the Fmask results) were used (Fig....
[...]
...For observations acquired after model initialization, only the Fmask algorithm is used. ρ 2; xð Þ−ρ̂ 2; xð ÞRIRLSN0:04 OR ρ 5; xð Þ−ρ̂ 5; xð ÞRIRLSb−0:04 ð2Þ where, x Julian date ρ(i,x) Observed value for the ith Landsat Band at Julian date x. ρ̂ i; xð ÞRIRLS Predicted value for the ith Landsat Band at Julian date x based on RIRLS fitting (Eq....
[...]
...Though the Fmask algorithm provides relatively accurate masks for clouds, cloud shadows, and snow, it is not perfect....
[...]
784 citations
Cites background from "Object-based cloud and cloud shadow..."
...Automated processes can also be implemented for cloud and shadow detection and subsequent masking (e.g. Fmask; Zhu and Woodcock, 2012; Zhu et al., 2015a)....
[...]
648 citations
Cites background or methods or result from "Object-based cloud and cloud shadow..."
...…active development by the creator, adoption from the academic community, physically meaningful attributes, ongoing development to utilize the Landsat 8 OLI cirrus band, development to support Sentinel-2 bands (Zhu et al., 2015) and a reported high cloud accuracy of 96.41% (Zhu and Woodcock, 2012)....
[...]
...The C programming language implementation of Function of Mask, known as CFMask, is a translation from Function of Mask, known as Fmask (Zhu and Woodcock, 2012; Zhu et al., 2015)....
[...]
...The C programming language implementation of Function of Mask, known as CFMask, is a translation from Function of Mask, known as Fmask (Zhu and Woodcock, 2012; Zhu et al., 2015)....
[...]
...We validated algorithms that already existed in both experimental and operational environments (Section 4), though we chose to implement Fmask (Zhu and Woodcock, 2012) in an operational environment, where we renamed it CFMask and created variants of it to produce cloud confidence bits, incorporate a cirrus test, remove the thermal tests from the algorithm, and add cirrus band tests to recover some accuracy from the lack of thermal data (Table 5)....
[...]
...• Function of Mask (Fmask) algorithm (Zhu and Woodcock, 2012)...
[...]
References
4,018 citations
"Object-based cloud and cloud shadow..." refers background in this paper
...Therefore, a morphological transformation called flood-fill is performed for Band 4 reflectance (NIR band) which brings the intensity values of dark areas that are surrounded by lighter areas up to the same intensity level as the surrounding pixels (Soille, 1999)....
[...]
1,389 citations
"Object-based cloud and cloud shadow..." refers methods in this paper
...For Landsat L1T images, Digital Number (DN) values are converted to TOA reflectances and BT (Celsius degree) with the LEDAPS atmosphere correction tool (Masek et al., 2006; Vermote & Saleous, 2007)....
[...]
1,198 citations
"Object-based cloud and cloud shadow..." refers background in this paper
...Though it works sometimes, most of the time it will inevitably include other dark surfaces that have similar spectral signatures (like topographic shadows or wetlands) and exclude cloud shadows that are not dark enough (Ackerman et al., 1998; Hutchison et al., 2009)....
[...]
...These sensors are usually equipped with more than one thermal band, or with water vapor/CO2 absorption bands, both of which are useful for thin semitransparent cloud detection (Ackerman et al., 1998; Derrien et al., 1993; Saunders & Kriebel, 1998)....
[...]
1,076 citations
"Object-based cloud and cloud shadow..." refers background in this paper
...The International Satellite Cloud Climatology Project-Flux Data (ISCCP-FD) data set estimates the global annual mean cloud cover is approximately 66% (Zhang et al., 2004)....
[...]
907 citations
"Object-based cloud and cloud shadow..." refers background in this paper
...However, if visible bands become saturated at a small value, for example 0.5 (Dozier, 1989), while NIR and SWIR bands do not (close to 1), it would make the absolute values of NDSI and NDVI much larger than 0, making probability of spectral variability lower for cloud pixels....
[...]
...5 (Dozier, 1989), while NIR and SWIR bands do not (close to 1), it...
[...]