John,
A few things going on here:
The background does not pick up well on the temperature gradients here, which is to be expected with a 3 km system. The overall pattern still originates with the background field, and although we have made adjustments to give obs more weight when they differ from the background and are in narrow valleys like this, it doesn't help that there are many other obs nearby that have an influence here. You can see in the attached deck that the background field is based pretty much on the terrain here, there are no other local influences.
A note about the land/water mask: It's not a hard wall, and there is some cross-influence. This is by design; at one point we did have a hard wall and that led to zero-increment lines along the coast, making for a very odd looking analysis. If you look towards the end of the deck, you will see where the boundary is.
It looks like what happened is that BFXC1 got caught between two increment areas, and because of that the temperature of that grid box did not change very much. A few other things going on:
It appears that one of the NTBC1 obs has an overlapping location with a Weatherflow ob; this is the location that will stay unflagged. That location is 34.4 N, 116.69 W. This is due to a rounding issue in the making of these files, a more precise location actually goes into the system. The NTBC1 location to the east will be flagged.
One of the obs west of the city, UP619, is from the Union Pacific Railroad. Obs from this network have a history of being much warmer than their surroundings; this appears to be because they are cited close to the ground near the tracks to detect heat kinks. Many offices have asked us to remove UPRR obs, we can remove this one too, if you want. C9548 also appears to have some influence and, for this cycle, came in well above the background value.
In the last slide I have ob/analysis/background values for several obs and mask values at their respective ob points. It's worth pointing out that for many of these grid boxes, the grid is part land, part water. That can lead to obs being washed out a bit, especially when many obs are bundled so close together as they are here.
I point out that the area we are talking about is very small spatially, and there are likely at least a few cases of several obs of differing values being in the same grid box. Obviously a grid box value isn't going to match all of them. For example, KP44 is 1 km west of XLED/NTBC1.
When you have a couple grid boxes of distance to work with, you can get some improvement. For example, the distance between the points where increments are strongest is only about 6 km (< 3 grid boxes), and the temperature distance between them moves from 12 F to 17 F between background and analysis; their observed difference was about 20 F.
In short, when we are looking at this sort of scale, we are running up against what the RTMA is capable of resolving.
Hopefully this helps,
Steve
