[GRASS-user] Estimating Albedo from Landsat

Hamish hamish_b at yahoo.com
Wed Sep 15 23:04:13 EDT 2010

Nikos wrote:
> # with "-f"
> not sure about the real differences with the previous command. Look fine
> though :-)

the paper states that the filtering will convert pixels which have 5 of 8
surrounding cells as clouds into a cloud pixel as well, thus removing small
holes. I have not checked if the source code exactly follows this method
or not though.

someone reading this can probably correct me if I am wrong, but AFAIU the
surface temperature from band 6's IR is the skin temperature, so only the
outer micron of a rock sitting in the sun can be quite a bit hotter than
the nearby air temperature.

`r.univar -e percent=0.5,99.5` can be better to filter out some outliers
from the band 61 temperature data versus just looking at the min,max.

the algorithm does not pick up well thin cirrus clouds, and the temperature
up there can be quite cool. also I noticed some cells at the thin cloud
edges were missed. The visual 3,2,1 d.rgb picture is easy to see with the
human eye if there are clouds or not. Apparently the MODIS Aqua/Terra
satellites can detect the right wavelength to pick up the thin cloud
signature, while Landsat doesn't record that data.

leaving off -s,-f doesn't really help with the North Carolina lsat7_2000
data. leaving off -2 as well removes a lot of cloud cells, but seems to
be artifacts following roads, etc.  ?

probably more refinement needed in the TOAR step? currently:
GRASS65> i.landsat.toar band=lsat7_2000 sensor=7 date=2000-03-31 \
  product_date=2000-07-02 -v solar_elevation=51.524652 gain=HHHLHLHHL

maybe i.attcor could help, or that NC data is not really raw DN but
already processed into QCAL values, or..??



More information about the grass-user mailing list