Predictive mapping based on geographic similarity: breaking away from statistical constraints

Authors: A-Xing Zhu*, Univ of Wisconsin, Jing Liu, Department of Earth Science, Santa Monica College, Santa Monica, California
Topics: Geographic Information Science and Systems, Spatial Analysis & Modeling, Quantitative Methods
Keywords: predictive mapping, spatial prediction, spatial interoplation, spatial statistics
Session Type: Paper
Day: 4/14/2018
Start / End Time: 2:00 PM / 3:40 PM
Room: Grand Chenier, Sheraton, 5th Floor
Presentation File: No File Uploaded

Existing predictive mapping methods often require soil samples to be sufficient enough to represent target-environment relationships throughout the study area. However, in geospatial modeling one of the key challenge issues for PSM application is the limited field samples available for the study area under concern. This paper presents a method for predictive mapping based on geographic similarity. With the assumption that similar environmental conditions have similar target variables, the new method uses the target-environment relationship at each individual sample location to predict properties of target variable at unvisited locations and estimate prediction uncertainty. Specifically, the environmental similarities of an unvisited location to a set of sample locations are used in a weighted average method to integrate the target-environment relationships at sample locations for prediction and uncertainty estimation. As a case study, the method was applied to map soil organic matter (SOM) content in the topsoil layer using two sets of soil samples. Compared with multiple linear regression (MLR), the new method produced a more accurate SOM map than MLR when the sample set was very limited in representing the study area, and achieved a comparable accuracy with MLR when the sample set can represent the study area better. In addition, the prediction uncertainty estimated by the new method was positively related to prediction residuals in both scenarios. This suggests that the new approach does not require samples to meet the conventional statistical requirements to perform accurate mapping, which is a welcome freedom for modeling spatial variation

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login