Page 187 - Special Topic Session (STS) - Volume 1

STS 423 Avery S. et al.
Crop Yield Modelling
The CDL is also a foundational element for remotely-sensed yield
estimation within NASS. Operational efforts to date have focused on corn and
soybean yields within the Corn Belt at county and state-levels (Johnson, 2014).
Research efforts are ongoing for other crops and areas of the U.S. (Johnson,
2016). Because crop progress, condition, and yield are dynamic, high revisit
rate optical and thermal satellite imagery is required. Weekly observations are
minimally sought and the best source for that type of information has been
from the Moderate Resolution Imaging Spectroradiometer (MODIS), which
has been in existence for nearly two decades. MODIS, while temporally ideal,
is compromised spatially as it is only 250 meters in resolution, and is a
challenge to identify with sufficient map precision field crop types and their
boundaries. Thus, the 30 meter spatial resolution CDL fills this need.
Having highly accurate field-level crop type information ultimately allows
one to isolate or mask the MODIS observations to only crop specific areas.
This provides a clean signal of the vegetation profile throughout the growing
season and improves yield model performance. Because the CDL is generated
within season it can be leveraged as early as the August Crop Production
report. More generally, the CDL can also be used to create year agnostic crop
type or area masks (Johnson, 2012) by integrating the data over several
seasons. This is useful for scenarios, such as needing to mask MODIS data that
exists prior to the 2008 availability of national-level CDLs or when needing a
crop predictive layer within season even before the CDL is available. This CDL-
derived crop information is also useful for simplified yield modelling
circumstances where managing year-specific crop maps is unwieldy.

Area Frame Stratification
A new automatic area frame stratification method (Boryan et al., 2014b;
Boryan and Yang, 2017) was recently developed and implemented for NASS
operations based on the Cultivated Layer. The NASS state-level area frames
are stratified based on percent cultivated cropland within NASS Primary
Sampling Units (Figure 3) and are used to select samples for NASS’s annual
JAS. For more than fifty years, the traditional area frame stratification method
was conducted using visual interpretation of aerial photography or satellite
data (Cotter et al., 2010). Research findings show that using the automated
stratification method, based on the CDL, significantly improves area sampling
frame stratification accuracies in intensively cropped areas (>75% cultivation)
and overall stratification accuracies when compared to traditional stratification
based on visual interpretation of aerial photography or satellite data, while
reducing the cost of area frame construction (Boryan et al., 2014b).

176 | I S I W S C 2 0 1 9