Satellite monitoring of cropland-related carbon sequestration practices in North Central Montana

Loading...
Thumbnail Image

Date

2008

Journal Title

Journal ISSN

Volume Title

Publisher

Montana State University - Bozeman, College of Agriculture

Abstract

This study used an object-oriented approach in conjunction with the Random Forest algorithm to classify agricultural practices set forth in carbon contract agreements associated with the Chicago Climate Exchange (CCX), including tillage (till or no-till (NT)), conservation reserve (CR), and crop intensity. The object-oriented approach allowed for per-field classifications and the incorporation of contextual elements in addition to spectral features. Random Forest is an advanced classification method that avoids data over-fitting and incorporates an internal classification accuracy assessment. Landsat satellite imagery was chosen for its continuous coverage, cost effectiveness, and image accessibility. Classification (2007) results included producer's accuracies of 91% for NT and 31% for tillage when applying Random Forest to image-objects generated from a May Landsat image. Low classification accuracies likely were attributed to the misclassification of conservation-based tillage practices as NT. Crop and CR lands resulted in producer's accuracies of 100% and 90%, respectively. Crop and fallow producer's accuracies were 95% and 82% in the 2007 classification; misclassification within the fallow class was attributed to pixel-mixing problems in areas of narrow (>100 m) strip management. A between-date normalized difference vegetation index approach was successfully used to detect areas "changed" in vegetation status between the 2007 and prior image dates; classified "changed" objects were then merged with "unchanged" objects to produce final classification maps of crop versus fallow. Resulting statistics showed that 22% of lands classified as CR had occurred outside of the Conservation Reserve Program (CRP). Field survey results were applied for tillage analysis because of low image classification rates and indicated that 56% of the evaluated region was under NT in 2007, with 44% practicing some form of tillage. Crop intensity estimates indicated that only 5% was under continuous cropping. These observations show the potential for the increased NT and continuous cropping. The application of carbon sequestration estimates to the land use data predict that approximately 59,497 t C yr⁻¹ might be sequestered through the universal adoption of NT and a 1.0 rotation (continuous cropping). Financial incentives through carbon credit programs might motivate land managers to make these management changes and to maintain CR lands.

Description

Keywords

Citation

Copyright (c) 2002-2022, LYRASIS. All rights reserved.