The paper presents a very interesting analysis on solar radiation measurements at the Netherlands.
The use of these long term time series together with the mathematical approach makes the paper very interesting and suitable for publication in ACP.

However, the "innovative" point which is the proxy approach on presenting and interpreting the data could be also a disadvantage for the paper quality if some more analysis is not presented. Mainly comparing the actual measurement results with the proxy results.

For example right now it is mentioned that the all sky measurements and proxy all sky computations agree within 5% but also all the results shown (e.g. in figure 4) point out to (50 year) changes of that (~5%) order.

More specific comments.

abstract and table 2:

The Trends (per decade) in fractional cloudiness (abstract and table) and Fractional cloudiness term of Equation 3 (table) have to be clarified. This is because in the description of data used is not clear if that is octas (it is not but it is not clear), or a number from 0 to 1 ( thus this 0.0097 can be translated to a 0.97% per decade change at least in the abstract). In another section this is expressed as cloud cover percentages as a function of cloudiness based on the 2010 Boers paper and then grouping the 0-1 and 7-8 octas conditions. So in the metadata sections the way that this fractional cloudiness is calculated could be more clear.

Proxy functions and data.
The fit of hourly values of a certain octa value with solar zenith angles.
So for each of these points you are using a cloud observation (or measurement after 2002) and a solar zenith angle. Since solar zenith angles are not linearly changing with time, you have to clarify which one is it used. In addition, one hour solar radiation averages attributed in one cloud observation could end up in quite complex "shapes" of these fits. Especially for 3-6 octas where more or less the data points within a certain octa and a certain solar zenith angle could be easily a bimodal distribution based on the sun obscureness by clouds or not. Could you add some comments on this issue and on the proxy functions retrieval ?
I think it would be necessary to show a figure of actual (all sky) measurements and proxy (all sky data) as a function of time. This in order to provide a link on your study and other (most) studies that interpret their results using actual measurements. Also in order to quantify the results based on the fact that percentage changes of the 50 year dataset are in the same magnitude with proxy and actual measurement yearly averages.

I miss something about the instrument calibration. Since we are talking about instruments starting operating in the 60's a comment on the instrument uncertainty should be included especially in order to be compared with the actual solar radiation changes that are presented for the 50 year period.

Figure 4 is very interesting and the results coming from it are also unique. Especially the fact that even if fractural cloudines/cloud fraction term, increases/decreases the all sky proxy term is increasing due to the cloud base term or thinner clouds.
My question for this figure is:
Since all sky proxies are calculated (and more or less represent the actual measurements), plus the cloud fraction term is calculated based on (independent than the pyranometers) cloud observation, plus the clear sky term is also based on the G (umbrella function); could it be that the cloud base term is simply the term (all sky proxy) – (clear sky proxy) – (cloud fraction term) ? To try to be more clear: if someone would have to re-evaluate the cloud observation data in the future, would that affect also the cloud base term ? (since all sky term is radiation and in principle it will not change and the clear sky term will also not change much).