I've improved the QMED analysis thing. By default, I have set this up such that it uses multiple donor catchment and weights the adjustment factors using an inverse distance weighting (IDW) scheme with a power of 3. Alternative schemes are also available including equal weighting and simply taking the nearest only. I have verified the IDW weighting against the science report dataset, it peforms slightly better than the "nearest only" approach.
I have included a default cut-off of the first 20 catchments within 500 km. For IDW with power 3 this limit does not do much, it's just for efficiency to drop off the catchments further away. For lower weights and equal weighting the story is different, obviously.
Note that QMED donor selection should just be on geographic distance. Otherwise you're trying to improve the underlying regression model (which uses AREA, SAAR, FARL, BFI, URBEXT) with the most likely effect you're making things worse.
I've improved the QMED analysis thing. By default, I have set this up such that it uses multiple donor catchment and weights the adjustment factors using an inverse distance weighting (IDW) scheme with a power of 3. Alternative schemes are also available including equal weighting and simply taking the nearest only. I have verified the IDW weighting against the science report dataset, it peforms slightly better than the "nearest only" approach.
I have included a default cut-off of the first 20 catchments within 500 km. For IDW with power 3 this limit does not do much, it's just for efficiency to drop off the catchments further away. For lower weights and equal weighting the story is different, obviously.
Note that QMED donor selection should just be on geographic distance. Otherwise you're trying to improve the underlying regression model (which uses AREA, SAAR, FARL, BFI, URBEXT) with the most likely effect you're making things worse.