HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Figures Only Full Text Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Chen, F. Articles by Luvall, J. C. Search for Related Content PubMed Articles by Chen, F. Articles by Luvall, J. C. Agricola Articles by Chen, F. Articles by Luvall, J. C. Related Collections Carbon Sequestration Remote Sensing

SOIL & WATER MANAGEMENT & CONSERVATION

Mapping Soil Organic Carbon Concentration for Multiple Fields with Image Similarity Analysis

^a Dep. of Crop and Soil Sciences, Univ. of Georgia, Athens, GA 30602
^b Global Hydrology and Climate Center, NASA, Huntsville, AL 35806

^* Corresponding author (fchen{at}uga.edu).

Remotely sensed imagery with high spatial resolution has been used to map soil organic carbon (SOC) concentrations at a field scale with greatly increased accuracy and reduced cost compared with grid sampling. The procedure, however, requires each crop field to be sampled and mapped separately. The purpose of this study was to determine if cost could be reduced further by grouping a number of crop fields based on their image similarity, and then mapping them together as one group. Ten crop fields with a bare soil surface were selected from a 2000 NASA ATLAS image. The similarity among these fields was examined with the Ward neural network system (WNNS) using the image histogram features extracted from the image for each field. Seven fields were placed into two groups based on the coefficient of determination (R²) values computed from WNNS, with one group consisting of three fields and the second consisting of four fields. Soil samples were taken from the seven fields along with their global positioning system locations and were divided into two data sets, with one for model development and the other for result checking. Models for mapping SOC concentrations were developed for each group of fields using a single procedure. The resulting maps were checked based on soil sample sets that were not used in model development and showed good agreement between mapped values and lab-determined values, with r² values of 0.80 for one group of fields and 0.77 for the second group of fields. The models were greatly improved compared with the model developed for all seven fields (R² was 0.87 and 0.91 for two groups vs. 0.63 for all fields and RMSE was 0.108 and 0.143 vs. 0.219 of SOC percentage). The model developed with similarity grouping was also compared with the model for field-by-field mapping and showed close agreement (R² was 0.87 for Group 1 vs. 0.89 for Field 2 only in Group 1 and RMSE was 0.108 vs. 0.119 for the same field).

Abbreviations: ANN, artificial neural network • MLP, multilayer perception • SOC, soil organic carbon • WNNS, Ward neural network system