They seem to be desperate to defend their models, if they go to such length in declining the use of stations methodology. Although they accept that single stations might prove to be error-striken, they believe (no counter-checking here) that either the companies running them solve the problems by interpolating, or that it doesn't effect the median temperature data.
Then they go on in length about how climate models are physical in contrast to statistical models and thus are fairly unaffacted by temperature record changes and station errors.
I can't believe that, because they must have some boundary conditions (like starting temperatures) that have to be assembled from the real world. Sadly, they don't release the source codes to their models to verify this. The second problem, I have, is the difference between physical and statistical models, they draw upon.
They obviously think that all mechanics involved in world climate are physically well-known and cared about.
I want to draw upon an example that is smaller in scale, but even as chaotic as world climate, the process of chip removal during broaching or drilling processes. Although the physical laws governing the process are known, the chip creation itself is so chaotic that no prediction can hold. Yes, we can say that there are longer chips or smaller chips, if we know the material in use, but we can't predict what kind of chips we get.
The same we see in climate models, we can tell that there will be a change, either up or down given the natural cycle, but we can't predict the actual next step.
To claim that we have a full physical model of the whole world and its temperature and thermodynamic states is just an illusion. If we had, we could project the way the wind will change and cloud forming is going to happen, however, we cannot (as is stated by the IPCC report).
We indeed have some rudimentary understanding of what governs the earth climate system (like sun-cycles, cosmic rays, cloud forming, Heat Transfer, albedo), but we still don't have physical equations that are more than a close approximation of the 'now'-state.
The whole function to govern the temperature state, would be something like that:
DeltaT = f(cloud forming) * k1 + f(sun-energy) * k2 + f(GHG) * k3 + f(Aerosol) * k4 + f(albedo) * k5 etc.
And as real-climate has already demonstrated, those weighting factors are not known and can become utterly complex...