Jump to content

User:Nlarsen3/sandbox/Environmental Mapping

From Wikipedia, the free encyclopedia

Given that Environmental mapping is very important for supporting policies or even to the decision making process. The use of the Bayesian spatial modeling system (REML), or the Integrated nested Laplace approximation (INLA), is to test gamma-k, elevation, slope, and topographic wetness index. Both have advantages and disadvantages to the both of them as stated in the article. 1) INLA is 1440 times slower compared to the REML system. With INLA the system is cheaper to calculate as frequent analysis vs to calculate the full marginal posteriors, "if the goal of the author's analysis is point predictions in the form of the marginal medians and quantification of uncertainty in the form of 5% and 95% quantiles, to achieve this goal, it is only necessary to calculate the quantities of interest instead of calculating the full posterior distributions. Results show that the approach using the calculation of full posteriors in INLA took more than 24 h, as reported by Huang et al. (2017), whereas the simulation approach using 10,000 independent samples from the posteriors took only around 10 min."[1] Both of the results were the exact same, just one took 24 hours while the other did the same calculations in only 10 minutes. (Fig. 2) 2) INLA spatial patterns would sometimes map discontinuous artifacts.  This is where SPDE models can overcome this challenge because they are “piece-wise continuous on the triangles of the mesh and predictions cannot be discontinuous unless there are discontinuous covariates in the model”. (Fig. 3)[2]

New article name goes here new article content ...


References

[edit]

“Environmental Mapping Using Bayesian Spatial Modelling (INLA/SPDE): A Reply to Huang Et Al. (2017).” NeuroImage, Academic Press, 27 Dec. 2017, www.sciencedirect.com/science/article/pii/S0048969717334861.

  1. ^ "ScienceDirect". www.sciencedirect.com. Retrieved 2019-02-25.
  2. ^ "ScienceDirect". www.sciencedirect.com. Retrieved 2019-02-25.
[edit]