This document describes how to create a bathymetry map using Bayesian kriging with limited data. It involves georeferencing available depth data points, importing them into the open-source R package GeoR, using Bayesian kriging simulations to predict depth values on a regular grid, and checking the accuracy of the resulting pseudo-bathymetry map. The example presented used depth data from Seychelles to create a 2km resolution map that had lower standard deviation than other remotely sensed data, demonstrating how Bayesian kriging can generate useful bathymetry estimates with coarse input data.

1. Creating Bathymetry Maps
With Coarse Data -
Bayesian Kriging Using
Open Source Tools
Hal Koike
University of Hawaii,
Hawaii Fisheries Cooperative Research Unit

2. Why do we need a
Bathymetry Map?
 Marine resource management is pushed
toward ecosystem based management (e.g.
linking with land development, marine
protected area)
 You need spatial data to fully understand
the ecosystem of your interest
 Species distribution for marine organisms is
known to be influenced by depth

3. Outside the United States…
Most countries do not have a spatial
data repository where bathymetry
data, land cover data, etc. is readily
available to be used for analysis.

4. If $$ is Limited,
What are the Options?
 Stick with what you have
 Create a pseudo-bathymetry map
Some budget friendly data covering the world
(bathymetry case)
 Navigational chart (low cost)
 MODIS (free)
 Hyperion (free)

5. Case for Seychelles…
What is available

6. What I need

7. Solution
Create a pseudo bathymetry map using
Bayesian Kriging option in GeoR (Rilbeiro jr.,
P.J. and Diggle, P.J. 2001)

8. What is GeoR?
 Created by Paulo J. Ribeiro Jr. and Peter J.
Diggle.
 One of the many packages available through
R-CRAN project
 Operated on R

9. Step 1. Georectification

10. Step 2. Enter the Depth Data

11. What it looks like after
entering all the points

12. Step 3. Import the Point
Data to Geo R

13. Step 4. Find your Range

14. Step 5. Run the Bayesian
Kriging Simulation
x <- seq(241472,403019,2000)
y <- seq(9449003,9559751,2000)
d1 <- expand.grid(x=x,y=y)
ex.bayes <-
krige.bayes(YourData,loc=d1,model=model.control(
cov.m="matern",kappa=0.5),prior=prior.control(phi.
discrete=seq(0,80000,l=10),phi.prior="reciprocal"))

16. Predicted Values

17. Predicted Values

18. Error of Predicted Values
(Estimation Variance)

19. Error of Predicted Values
(Estimation Variance)

20. Accuracy Check

21. Accuracy Comparison
Bathymetry Map Standard Deviation
SRTM 30 (1km grid) 76.89
Bayesian Kriging (2km grid) 9.00
Conventional Kriging (2km grid) 8.30
Statistically simulated bathymetry map had
less deviation then remotely sensed data