Open jesusff opened 1 month ago
This is from an old e-mail from @graziano-giuliani
On 2022-10-17, G. Giuliani wrote:
I have today produced the file for you using the RegCM pre-processing by using the RegCM model version used in the CORDEX-CORE experiment. Let me know if this works for you.
The RegCM uses the same CLM45 grid patch scheme as described in the CLM 4.5 description document:
https://www.cesm.ucar.edu/models/cesm1.2/clm/CLM45_Tech_Note.pdf
Note that the urban land unit is divided into density classes representing the tall building district, high density, and medium density urban areas (numurbl dimension, from high to low). The sum of the three is the overall Urban land unit. Look at page 28 of the NCAR Technical Note for a description of the Surface Heterogeneity and Data Structure.
It seems that for RegCM we are not considering only the impervious area.
I ping @gnikulin and @larsbuntemeyer, as this might affect the data request from CORDEX-CMIP6. Not sure how sfturf
entered the DR, but it might need further clarification.
Thanks for following up on this @jesusff
From the above, it seems indeed that what is in the RegCM sfturf* files does not have the same meaning as to what is in the REMO files (and what I used as benchmark datasets).
To re-iterate. For our CORDEX FPS URB analysis, we are interested in the impervious surface area (ISA). This is what our observed datasets represent (ESA, ESRI, LCZ), and also what REMO represents.
For RegCM, we might have to revise the sfturf* files that are currently available to us.
Let me quickly mock-up an example to show what we currently have, and what it should be?
I am focussing on the larger New York area (fugure below), using the mksrf_urban.nc
file, that presents all native CLMU parameters on a global grid (3600 x 7200 px).
Likely, our current sfturf* files present the sum of PCT_URBAN
over the three available urban density classes. Variable PCT_URBAN
represents the percent urban for each grid cell, per density class: the tall building district (TBD-0), and high density (HD-1), and medium density (MD-2) classes.
Each of these density classes has different urban parameters, that vary geographically according to 33 pre-defined regions, as explained in Jackson et al. (2010).
But, this PCT_URBAN
is not the same as the impervious surface area, because in CLMU, an urban grid cell consists - from a horizontal perspective) out of WTLUNIT_ROOF
(fraction of roof, unitless) and WTROAD_PERV
(fraction of pervious road, unitless)
Values for these variables, for the geographic region corresponding to our ROI:
ROI is in REGION_ID 18
**** WTLUNIT_ROOF ****
Value for density class 0: 0.75
Value for density class 1: 0.6000000238418579
Value for density class 2: 0.3499999940395355
**** WTROAD_PERV ****
Value for density class 0: 0.20000000298023224
Value for density class 1: 0.375
Value for density class 2: 0.692307710647583
The way I see it, the impervious surface area for RegCM should be calculated as follows:
# For each density class
ROAD_FRACTION = 1 - `WTLUNIT_ROOF`
IMPERVIOUS_ROAD_FRACTION = 1 - `WTROAD_PERV`
ISA = WTLUNIT_ROOF + IMPERVIOUS_ROAD_FRACTION
and then:
ISA_FINAL = PCT_URBAN_0 * ISA_0 + PCT_URBAN_1 * ISA_1 + PCT_URBAN_2 * ISA_2
with 0, 1, and 2 referring to the density classes.
See below plot of ISA per density class, and its sum (ISA_FINAL):
So, would ICTP be able to recompute the RegCM sfturf* files for all CORDEX domain grids, following the procedure above?
That would allow us to do a consistent assessment of the models in terms of their representation of impervious surface areas.
Thank you for the nice illustration, @matthiasdemuzere
Where did you get the mksrf_urban.nc
file? I assume it contains the parameters we need to apply for each region (WTLUNIT_ROOF, WTROAD_PERV) to weight the urban fractions and get the ISA. I think we (@JavierDiezSierra) produced the current RegCM stfurf out of the urb_2d
variable provided by @graziano-giuliani and we could do this weigthing if Graziano confirms that the urb_2d
is the PCT_URBAN
and that the model uses internally the same mksrf_urban.nc
file.
Regarding your formulas for ISA, you assign ROAD_FRACTION = 1 - WTLUNIT_ROOF
, but this is never used below. If WTROAD_PERV
is the pervious fraction of the road (non-roof areas) alone, then the second equation should read:
IMPERVIOUS_ROAD_FRACTION = ROAD_FRACTION*(1 - WTROAD_PERV)
right?
Regarding the naming for the "urban fraction" files that we plan to publish, we would need a new variable, as sfturf
does not fit our case, then.
The area type for impervious surfaces is not included in CF. We would need to ask for inclusion of this area type into CF, so we could define e.g.:
"sftimf": {
"frequency": "fx",
"standard_name": "area_fraction",
"units": "%",
"cell_methods": "area: mean where impervious",
"cell_measures": "area: areacella",
"long_name": "Percentage of the Grid Cell Occupied by impervious surface areas",
"comment": "not in CMIP or in CF",
"dimensions": "longitude latitude",
"out_name": "sftimf",
"type": "real"
}
However, I can imagine impervious surfaces (rock) which are not urban, so we might need both variables to obtain what we mean. Or, we could define the impervious surfaces already intersected with the urban areas, which I see it is typically termed "constructed" (e.g. here) or "artificial" (here) impervious surface area (to distinguish it from other natural impervious surfaces)
Also, sfturf
is a TIER2 variable for CORDEX, meaning that it could be inconsistent across RCMs, but we could try to define (say) sftimf
as consistently as possible.
Regarding your formulas for ISA, you assign
ROAD_FRACTION = 1 - WTLUNIT_ROOF
, but this is never used below. IfWTROAD_PERV
is the pervious fraction of the road (non-roof areas) alone, then the second equation should read:IMPERVIOUS_ROAD_FRACTION = ROAD_FRACTION*(1 - WTROAD_PERV)
right?
@jesusff Indeed, I made a mistake in writing down the equations. What you have here is correct.
Honestly I cant remember where i got the *.nc file from, I guess I took it from the cesm source code?
Some more info is in the user guide: https://escomp.github.io/ctsm-docs/versions/release-clm5.0/html/tech_note/Urban/CLM50_Tech_Note_Urban.html?highlight=urban
I assume the netcdf file I refer to is the same as mentioned in Fig. 2.15.3, urban_properties_data.05deg.nc? Perhaps that can help your search (I am on holidays without laptop at the moment)?
The term “urban fraction” is misleading. I think what is meant here is the "fraction of impervious surfaces". It is also important to use this term clean and consistent in this paper. For background - in REMO we consider an “impervious surface” with high roughness length and albedo, zero vegetation and no water holding capacity, which is averaged with the parameter values from vegetated surfaces of an urban area. In addition, we use a tile approach for land and water surfaces of an urban area (see also above). An urban area as such is not defined.
I think I see your point, but we should then move fast on this. For CORDEX-CMIP6, this variable has already been requested as
sfturf
(Percentage of the Grid Cell Occupied by Urban Area), defined as an area_fraction with cell methodarea: mean where urban
. The urban area type is defined quite loosely in CF (https://cfconventions.org/Data/area-type-table/current/build/area-type-table.html) as "Urban land is comprised of areas where much of the land is covered by structures. Included in this category are cities, towns and villages."Is there a publicly available resource discussing this impervious vs urban area? If I see your point correctly, you mean that the fraction should include only the impervious areas, and not the urban green or blue areas. What about the open or sparsely built LCZ types which assume abundance of pervious areas between buildings? Is there an assumed impervious area fraction defined for each of them @matthiasdemuzere ? Sorry for my ignorance
Could this different interpretation be behind the differences between REMO and RegCM "urban" fractions? RegCM is not using LCZs, but they have some classes (e.g. medium density) which likely include green areas.
In my assessment, I explicitly refer to impervious surface area, as this is what I look into. The part of the surface that is impervious, such as building, streets, and other impervious man-made structures.
This is what all three observational datasets represent, including the LCZs. As for the latter, each LCZ label is converted to their corresponding impervious surface area.
REMO clearly represents the same. And I assumed the sfturf files for RegCM also represent the same? RegCM uses the Jackson et al data, that consist of three urban classes (HD, MD, LD). Each of these classes have a range of parameters that vary spatially according to 33 regions. These classes all contain horizontal impervious surfaces, which I assumed are represented in the sfturf files.
@GLangendijk I am not sure who made these RegCM files. Perhaps it would be good to confirm the above is actually the case? To make sure all datasets represent the same thing?