Map Information:

Map Title:
Improving Interpolation - Bayesian Mapping
Name(s) of Author(s):
Gail Carter, Vikram Vyas, Saichiu Tong, Christpher Uchrin, Panos Georgopoulos
Organization:
NJDEP-DSRT & Rutgers EOSHI
Email Address:
Gail.carter@dep.state.nj.us
Telephone Number:
609-984-9654
Software Used to Make Map:
ArcGIS, Spatial & Geospatial Analyst, BMELib
Platform on Which
Software Was Used:
Windows NT
Plotter Used to Print Map:
Hewlett Packard DesignJet 1055
Data Sources:

RASA Model, KH data, statistics, professional judgment

Map Description:
Bayesian Maximum Entropy (BME) mapping is the next generation of interpolation methodology. Unlike current interpolation methods such as kriging, BME formally allows one to incorporate prior general information (general principles, physical laws, summary statistics) and soft information (interval information, expert judgements) into the mathematical analysis of the data set. BME interpolation without prior general or soft information is a krige, and accounts for spatial drift and autocorrelation in 3 directions. Recent books by Christakos (2000) and Christakos and Serre (2001) discuss modern spatial geostatistical methods in depth. Using the same initial hydraulic conductivity (Kh) dataset, an empirical isotropic semi-variogram was fitted to the data using ordinary kriging to produce the small map to the right. Next the same data was interpolated using BMElib software. Eight soft interval data points were taken from a U.S.G.S. Regional Aquifer Systems Analysis report (Martin, 1990) and used to fill data gaps. In addition summary statistics (probability density function) were incorporated as general prior information to inform and constrain the interpolation. Error analysis shows the BME interpolation to be greatly superior to the ordinary kriged interpolation. Over 36% of the BME error was between –1 to +1 feet. Less than 7% of the kriged interpolation met this same accuracy. In addition the overall error range for the BME (-30 to +30 feet) is less than half of the kriged error (-70 to +70).