Publications

Gao, F; Anderson, MC; Zhang, XY; Yang, ZW; Alfieri, JG; Kustas, WP; Mueller, R; Johnson, DM; Prueger, JH (2017). Toward mapping crop progress at field scales through fusion of Landsat and MODIS imagery. REMOTE SENSING OF ENVIRONMENT, 188, 9-25.

Abstract
The ability to regionally monitor crop progress and condition through the growing season benefits both crop management and yield estimation. In the United States, these metrics are reported weekly at state or district (multiple counties) levels by the U.S. Department of Agriculture (USDA) National Agricultural Statistics Service (NASS) using field observations provided by trained local reporters. However, the ground data collection process supporting this effort is time consuming and subjective. Furthermore, operational crop management and yield estimation efforts require information with more granularity than at the state or district level. This paper evaluates remote sensing approaches for mapping crop phenology using vegetation index time-series generated by fusing Landsat and MODIS (Moderate Resolution Imaging Spectroradiometer) surface reflectance imagery to improve temporal sampling over that provided by Landsat alone. The case study focuses on an agricultural region in central Iowa from 2001 to 2014. Our objectives are 1) to assess Landsat-MODIS data fusion results over cropland; 2) to map crop phenology at 30 m resolution using fused surface reflectance data; and 3) to identify the relationships between remotely sensed crop phenology metrics and the crop progress stages reported by NASS. The results show that detailed spatial and temporal variability in vegetation development across this landscape can be identified using the fused Landsat-MODIS data. The mean difference (bias) in Normalized Difference Vegetation Index (NDVI) between actual Landsat observations and the fused Landsat-MODIS data, generated for Landsat overpass dates, is in the range of 0.011 to 0.028 for every year. The derived phenological metrics show distinct features for different crops and natural vegetation at field scales. Strong correlations are observed between remotely sensed phenological stages, based on NDVI curve inflection points, and the observed crop physiological growth stages from the NASS Crop Progress (CP) reports. The green-up dates detected from remote sensing data typically occurred during crop vegetative stages when 2-4 leaves were developed for both corn and soybeans, or about 1-3 weeks after the reported emergence dates when the plant were first visible to ground based observers. Despite being a lagging indicator, remotely sensed green-up can be used effectively to backcast emergence, e.g. as input to spatially distributed crop models. The differences in green-up date between corn and soybean were 8-10 days, consistent with the offset in emergence dates reported by NASS at district level. The reported harvest dates were typically about 2-3 weeks after the dormancy stage was detected via remote sensing for corn and about 1-2 weeks for soybeans. This suggests that probable harvest times for individual fields may be predicted 1-3 weeks ahead using remote sensing data. The results suggest that crop phenology and certain growth stages at field scales (30 m spatial resolution) can be linked and mapped by integrating imagery from multiple remote sensing platforms. Published by Elsevier Inc.

DOI:
10.1016/j.rse.2016.11.004

ISSN:
0034-4257