Publications

Varmaghani, A; Eichinger, WE (2016). Early-Season Classification of Corn and Soybean Using Bayesian Discriminant Analysis on Satellite Images. AGRONOMY JOURNAL, 108(5), 1880-1889.

Abstract
There are numerous applications that require crop classification as early as possible in the growing season. However, information about land cover from official land cover maps of the United States (cropland data layer [CDL] maps by the National Agricultural Statistics Service) are generally not available until aft er harvest. In the Upper Midwest, the primary rotating crops are corn (Zea mays L.) and soybean [Glycine max (L.) Merr.] (covering similar to 63% of Iowa) with an irregular annual rotation. This study investigated the feasibility of early-season classification of corn and soybean fields in Iowa by comparing the current and previous years' 30-m 16-d Landsat 8 images during the growing season to produce normalized difference vegetation index (NDVI) maps, along with the last-updated CDL land cover, to construct "agricultural units." We assigned a geometric weight to each unit by performing Bayesian discriminant analysis using the concept of a sliding threshold to categorize pixels. An examination of 8 yr of 250-m 16-d NDVI measurements from the Moderate Resolution Imaging Spectroradiometer (MODIS) over Iowa showed that late June is the most promising time for categorization. The geometrical model was tested on a 24- by 28-km(2) region in southwestern Iowa on 1 July 2014 (Day of the Year 182). There was an 86% agreement with the CDL data set (88 and 83% for corn and soybean, respectively, in the confusion matrix). This demonstrates that in spite of the complexity of crop behavior, a geometrical approach integrating probabilistic methods, previous statistical records, and map disaggregation into agricultural units can be a promising method for early-season crop classification.

DOI:
10.2134/agronj2015.0454

ISSN:
Feb-62