Short Description of processing steps from
EUMETSAT H-SAF Soil moisture time series
to
UHAM ICDC Soil moisture global maps
Product version: WARP5.5 R2.2
EUMETSAT-HSAF product number: H25-SM-OBS-4
UHAM-ICDC product version: 1.1 (Data files carry the file version numbers fv0.02)
Stefan Kern
CEN / CliSAP / ICDC
University of Hamburg
stefan.kern@uni-hamburg.de
##### November 10, 2016 #####
Documentation v1.1, issue 1.0, fv0.02
The soil moisture time series data of the EUMETSAT H-SAF product H25-SM-OBS-4 / H109 based on MetOp ASCAT data, product version WARP5.5 R2.2 are converted into geographic maps (cartesian grid) of the 5-day average/composite soil moisture (SM) distribution separately for ascending and descending overpasses.
Time series for all tiles are read, separated for ascending and descending overpasses and averaged / composited over 5 days on the original grid. Only data with known surface state flag (SSF) = 1 are used. Due to MetOp-ASCATs narrow swath overlap of adjacent swaths is quite small; hence it is rather a composite than an average and it is rarely averaged over data from 5 days.
Averaging is also conducted for soil moisture noise (SMN), frozen land probability and snow cover probability.
For remaining parameters values of the center day of the 5-day period are used. Processing flag (PF) and SSF are treated differently.
The 5-day-SSF is set to 4 (permanent ice) or 0 (unknown) if SSF=4 or SSF=0 in one of the 5 days at that location. The 5-day-SSF is set to 2 (temporarily frozen) or 3 (melt water on surface) if SSF=2 or SSF=3 in one of the 5 days at that location.
In addition there is a processing flag which indicates whether the sensitivity to soil moisture is less than 1% or which indicates that the soil moisture retrieval noise is above 50%. If these flags are set in combination with SSF=1 then the 5-day-SSF is - instead of kept as 1 - set to 5 (sensitivity less than 1%) and 6 (SMN > 50%), respectively.
The 5-day-PF is set to 4 (far too low SM), 8 (far too high SM), 10 (wet correction applied) or 12 (NRCS not usable) if PF=4 or 8 or 16 or 32 in one of the 5 days at that location. The 5-day-PF is set to 1 (too low SM) or 2 (too high SM) if PF=1 or PF=2 in one of the 5-days at that location; if both PF=1 and PF=2 occur within one 5 day period the value is considered dubious and the 5-day-PF is set to 12.
Two different 5-day SM distributions are given: one is based solely on nominally computed SM, the other one (called "extended") includes also those SM values which were negative (down to -25%, PF=1) or positive (up to 125%, PF=2) but set to 0% and 100%, respectively, and those for which the wet correction has been applied (PF=16, for Köppen BW climates). The same applies to SMN.
SSF and PF are combined into one flag PF* by setting PF* = ((5-day-PF) * 10) + 5-day-SSF.
This allows useful data for the following values of PF*:
1,5,6: nominal best retrieval, as 1 but with sensitivity < 1%, as 1 but with SMN > 50%
11,15,16: SM was negative but still > -25% and therefore set to 0%, as 11 but with sensitivity < 1%, as 11 but with SMN > 50%
21,25,26: SM was positive but still < 125% and therefore set to 100%, as 21 but with sensitivity < 1%, as 21 but with SMN > 50%
101,105,106: Wet correction applied, as 101 but with sensitivity < 1%, as 101 but with SMN > 50%
Note: The original PF excludes the combination "wet correction applied" and SM < 0% or SM > 100%.
Subsequently all data are interpolated into a cartesian grid of x- and y-dimensions of original grid. First grid cells with minimum distance to original grid positions are assigned the original data values. Second remaining gaps are filled by the average SM or SMN (or the corresponding other parameters except the flag) over a 3x3 grid box centered around the grid cell with missing data. This step is repeated 4 times and a 5th time using a 5x5 grid box. For the interpolation of the flag at least three valid flag values need to exist within a 3x3 grid box and of these the one with the highest count is used to fill the missing grid cell.
Note: The original data set is based on soil moisture time series. If soil moisture time series have substantial spatial variation in adjacent grid cells then soil moisture values of adjacent grid cells might differ from each other by an unexpectedly high amount.
Data are subsequently stored in netCDF.