Page Banner

United States Department of Agriculture

Agricultural Research Service

Title: Generating Spatially Correlated Fields with a Genetic Algorithm

Authors
item Pachepsky, Yakov - DUKE UNIVERSITY
item Timlin, Dennis

Submitted to: Mathematical Geology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: December 10, 1997
Publication Date: N/A

Interpretive Summary: Precision agriculture requires a detailed knowledge of soil variability. Crop simulation models can be used to carry out simulations to obtain a range of crop yields and the expected spatial distribution with some knowledge of the spatial distribution of soil properties such as soil water holding capacity. Since soil properties are not measured at all the locations for which simulated yields are needed, some method of estimating soil properties at a large number of locations is needed. This method will also preserve the overall measured spatial pattern of the distribution of soil properties. The results of this study will allow the use of crop simulation models to obtain an expected spatial distribution of crop yields. This will help farmers better manage application of chemicals and other agricultural inputs to obtain higher yields with lower cost and less impact on the environment.

Technical Abstract: Numerically generated realizations of random fields are used to estimate the natural variability of soil and other geological properties. The spatial variability can be summarized in the form of a semi-variogram. When the generated realizations are used as inputs for simulations with a deterministic model, it is desirable to make the semi-variogram of the generated field close to that of the measured field. The objective of this study was to evaluate an optimization method to generate a spatially correlated distribution of a soil property that has a semi-variogram with a specified range, sill, and nugget. We describe the use of a genetic algorithm (GA) for this objective. In unconditioned simulations, statistics of the GA-generated realizations were significantly closer to the input ones than those from sequential Gaussian simulations. Distributions of generated values at a particular node over sequential realizations were close to the normal distribution. The GA is very computationally intensive and may not be suitable for fine grids. The sequential Gaussian algorithm conditioned with GA-generated values on a coarse grid can produce a set of realizations with similar statistics for the fine grids embedding the coarse one.

Last Modified: 4/23/2014
Footer Content Back to Top of Page