From: nico r. <nic...@gm...> - 2009-03-14 11:39:11
|
I am having a confusion with my indicator simulation results from GEMS. I am simulating (SISIM) biomass distribution, generated 100 realization over around 19000 grid locations. I have extrapolated the lower tail and upper tail to the lowest value (which is 2) and the highest value (139) in the biomass distribution. All 100 simulated maps have values ranging from 2 to 139. I generated E type map, which is the average of the 100 simulations, and the E type map has a lot narrower range of predictions (predictions range from 10 to 45). The simulation grids in this case do not coincide with the data locations, therefore, I can imagine my simulated E type map having lower range in predictions. However, I assume that if the simulation grids would also have the true data locations, then kriging being an exact interpolator, I would presume that my E type map would show predictions from 2 to 139. To check this further, I made another set of simulations. But in this case, the simulation grid also has the true data locations. In the results, while the simulated maps have ranges from 2 – 139, the e type map still has lower range in predictions. I checked the simulated values for the true data locations (which were in the simulation grid), and the simulated values are different from the true value here. And in this case, as we could expect, the E type map does not re-produce the true value for these locations. Kriging being exact interpolator, I was expecting that in these true data locations, the all simulated values should be the true biomass data; in this case the E type map would show the true observed biomass values, and the variance would be zero. I must be making some mistake in doing the analyses. Do you have any thought what could be going wrong. Thanks for your time Sanjay |