Meteorological stations - nearest neighbour - Thiessen Polygons

[Timo Brussee]

Dear SWAT users,

I have a watershed of 13420 km2, I have a number of stations that measure rainfall 43 to be exact.

These stations have large gaps and do not overlap in time. 

Correct me if I am wrong please, but:

As far as I know SWAT couples the nearest meteo. stations to a subbasin. If that stations does not hold observations for a certain period of time, these gaps are filled with monthly WGEN_user statistics. !So it does not search for another station located close by! I think this is a waste of accuracy if these gaps can also be filled with another station, that is within a range of X kilometers of that station.

Because I do have sometimes another station within a 10km range that could fill this gap...

What I could do as an alternative is use the Thiessen Polygon weighted mean area method to calculate a new value per subbasin. However due to the huge amount of stations (43) and subbasins (32) it is not possible for me to recalculate the new Thiessen Polygon areas if a station has gaps, and recalculate the thiessen polygon areas per region. (well it is, but I do not know how to programm this...)

What I did thusfar is calculate the Thiessen Polygon area percentages once! for one situation where I pretend all stations have observations. I did this in ArcGIS.

and use the output in the programm R (Rstudio) to calculate the weighted value per subbasin. If a station didn't have a value I would just weigh the other stations more heavy (not the correct way, but the easy way) so for an imaginary subbasin:

Stations 1: 10mm/day (25% area)

Station 2:  N.A. (25% area)

Station 3: 5mm day (50% area)

Subbasin X - outcome: 7.5mm/day (100%)

note that I did give a threshold value of at least 60% weight for these stations.

Could someone provide me their expert opinion on the best way to deal with these multiple Thiessen Polygons for stations with data gaps? or provide me with an alternative? - is there a programm that can do this if you provide subbasin area, station locations, and your data?

I would also like to know if my statement earlier about how SWAT deals with these meteo data gaps is true?

  

NOTE:

I added a map of the locations of my stations and a graph that shows how many stations (on average) cover a certain year with rainfall data.

So it could be that a year is covered by e.g. 16.834333 stations (so 16.8433*365 observations in that year by all stations together).

Best,

Timo

[Jim Almendinger]

Timo --

There used to be a add-on program to ArcSWAT called PCP_SWAT (by Xuesong Zhang, I believe), which would solve your problem.  It would take an array of precipitation stations, and for each day it would spatially interpolate the rainfall over the model basin, creating a daily rainfall record for the centroid of each subbasin.  Missing data from any station would be ignored and the spatial interpolation would be based on only the available data for that day. 

But this was several years ago -- it may only have worked for swat2009.  I don't know if it would work for swat2012.  And I don't see it offered on the SWAT web site.  I tried a google search myself but didn't find anything. 

Just in case anyone out there knows the current status of this program, please let Timo know -- I think it would solve his problem nicely. 

-- Jim

---

--
You received this message because you are subscribed to the Google Groups "ArcSWAT" group.
To unsubscribe from this group and stop receiving emails from it, send an email to arcswat+u...@googlegroups.com.
To post to this group, send email to arc...@googlegroups.com.
Visit this group at https://groups.google.com/group/arcswat.
For more options, visit https://groups.google.com/d/optout.

---

Spam
Not spam
Forget previous vote

--

Dr. James E. Almendinger
St. Croix Watershed Research Station
Science Museum of Minnesota
16910 152nd St N
Marine on St. Croix, MN  55047
tel: 651-433-5953 ext 19

[Timo Brussee]

Thanks for the quick reply Jim!

I do am using SWAT2012. For now I will just use this scenario as described below, if I come up with something better I will post it. In case other people encounter similar problems.

situation 1: regular Thiessen Polygon for all stations, allowing to fill the area if at least 60% of the subbasin area is covered by stations.

situation2: Similar but, I will delete stations that have a small coverage of data in time, and therefore "block" other stations from being part of a Thiessen Polygon. I will group stations (averaging the coordinates) and treating them as one if they are closeby each other and have medium coverage.  This way I have fewer stations but less gaps.

Situation 3: Calculate based on WGEN statistics.

If gaps in situation 1 <- fill with situation 2

If gaps in situation 2 <- fill with situation 3

Its the best and least time consuming method I can think of, not very scientific, but the nearest neighbouring method also isn't...

If anyone thinks of something better, or already has written a script for such a situation/developed a programm let me know.

Best and GL all with modelling!

Timo 

Op zaterdag 24 december 2016 12:14:16 UTC+1 schreef Timo Brussee:

[Cva Shrestha]

Timo

Sorry for late reply
i am not sure about the pcp-stat doing exactly wat jim is asking.
pcp_stat would rather prepare the wgen variables for generating -99(missing) values.

2.I am not sure if anyone could exactly explain how to use the precipitation stations.
There are a lot of issues with. I agree with you on ignoring stations with small data period but
again they could use to check other data when  necessary.

3.I would not suggest on averaging P station as things may not work as you expect.
This would be useful and again unnecessary if P station measure similary or same data. i.e low spatial variability
I would rather suggest you to increase the number of subbasin which would help SWAT to pick as many number of p -stations as possible.

4. finally in selecting the P -stations use yield data from your observed station
This should help you tremendously in selecting P station suiting the model

Thanks
Shiva