figure out uncertainty estimates for aggregate composite method predictions

[aappling-usgs]

see also #199

[aappling-usgs]

Thoughts from a few years ago:

We may find that a faster way to estimate the uncertainty in this sum is by Monte Carlo simulation, using the means and variances (or se.preds) of instantaneous fluxes to parametrically resample from those distributions, find the sum, and repeat until we have a population of sums from which we can estimate a distribution. See, for example, http://eprints.sics.se/2253/1/SICS-T--2002-01--SE.pdf ("Evaluating the CDF for m weighted sums of n correlated lognormal random variables", Lars Rasmusson, 2002, Swedish Institute of Compute Science, Report T2002:01, ISRN: SICS-T-2002/01-SE, ISSN:110-3154)