The land and ocean instrumental temperature data is a measure of something with a mean error of about +/- 0.125 C degrees. If all the data where measured in exactly the same way, that data would be a uniform measure of whatever it is measuring. When you mix methods you begin to add a different kind of uncertainty that is not included in the standard published error. Part of the problem with "Global" land and ocean data is that land is measured with a (Tmax +Tmin)/2 average where Tmax and Tmin have different influences. Oceans are measured as a simple average of spot readings at a location that has less variability or noise. You have apples and oranges if you can't tease out the "other" factors that are being measured.
So how well can models be performing if they cannot find simple anomalies like land use impact.
Note: The 15 year lag is an indication that land drying and general deterioration over time has the more significant impact.