In modelling the error budget in terms of the relative uncertainty on an intensity
we are looking at something different than the error on the terrestrial albedo
delta_albedo / albedo
delta_(DS/BS) / (DS/BS)
We seem to be finding large delta_I/I’s and small delta_albedo/albedo. This can be because an error on BS could be compensating an error on the DS – i.e. the errors are not independent. In writing the error budget as
(d_I/I)^2= Sum_i [ (d_a(i)/a(i))^2],
where the a(i)’s are terms in the error budget we have made the assumption that the a(i)’s suffer errors that are independent so that various cross-terms in the derivatives are zero in the mean.
In model-fitting, such as ours, where some of the factors to be fitted are de-pendent (alfa and offset is an example) the error budget should be written differently, and we should look at this. Henriette’s estimate of d_DS/DS remains correct, I think, but it should be understood that some other terms appear in the full d_albedo/albedo error budget, which cancel. An example of this is seen in the post here.