MODIS phenology-derived, multi-year distribution of conterminous U.S. crop types

Abstract

Innovative, open, and rapid methods to map crop types over large areas are needed for long-term cropland monitoring. We developed two novel and automated decision tree classification approaches to map crop types across the conterminous United States (U.S.) using MODIS 250 m resolution data: 1) generalized, and 2) year-specific classification. The classification approaches use similarities and dissimilarities in crop type phenology derived from NDVI time-series data for the two approaches. The year-specific approach uses the training samples from one year and classifies crop types for that year only, whereas the generalized classification approach uses above-average, average, and below-average precipitation years for training to produce crop type maps for one or multiple years more robustly. We produced annual crop type maps using the generalized classification approach for 2001–2014 and the year-specific approach for 2008, 2010, 2011 and 2012. The year-specific classification had overall accuracies > 78%, while the generalized classifier had accuracies > 75% for the conterminous U.S. for 2008, 2010, 2011, and 2012. The generalized classifier enables automated and routine crop type mapping without repeated and expensive ground sample collection year after year. The resulting crop type maps for years prior to 2007 are new and especially important for long-term cropland monitoring and food security analysis because no other map products are currently available for 2001–2007.

Department

Natural Resources and the Environment

Publication Date

8-1-2017

Journal Title

Remote Sensing of Environment

Publisher

Elsevier B.V

Digital Object Identifier (DOI)

10.1016/j.rse.2017.06.033

Document Type

Article

Rights

© 2017 Elsevier Inc.

Link to Full Text

Share

COinS