Taking up yesterday's question, here's my answer. The graph above shows the data from Nemani et al, along with black data points that I took from their graph and their Excel trend line. The OLS estimate of the trend (assuming all the data points were independent and normally distributed) is 0.188 +/- 0.044 GT/yr of carbon (compared to the -0.05 +/- 0.1 trend for the 2000 to 2009 data).
It turns out it's reasonable to ignore the autocorrelation. For example, if you take the residuals from the trend line above, and then plot them against their neighbor at one year lag, here's the scattergram:
With a correlation coefficient of less than 10%, we are pretty safe in ignoring this. Then, the way you'd compare these two trends is to compute the difference in the slopes (0.188 - -0.05 = 0.238), and compare that to the combined error (sqrt(0.1^2+0.044^2) = 0.11. That would be 2.2 sigmas and so (just) statistically significant (p = 0.03).
So using the IPCC likelihood terminology, I think it's fair to say that "It is very likely that the previously demonstrated trend of increasing NPP has now slowed. It is possible that NPP is now actually decreasing, but the evidence for that is equivocal."