Mapping twenty years of corn and soybean across the US Midwest using the Landsat archive

Sherrie Wang, Stefania Di Tommaso, Jillian M. Deines, David B. LobellScientific 2020 · DOI

Abstract

Field-level monitoring of crop types in the United States via the Cropland Data Layer (CDL) has played an important role in improving production forecasts and enabling large-scale study of agricultural inputs and outcomes. Although CDL offers crop type maps across the conterminous US from 2008 onward, such maps are missing in many Midwestern states or are uneven in quality before 2008. To fill these data gaps, we used the now-public Landsat archive and cloud computing services to map corn and soybean at 30 m resolution across the US Midwest from 1999-2018. Our training data were CDL from 2008-2018, and we validated the predictions on CDL 1999-2007 where available, county-level crop acreage statistics, and state-level crop rotation statistics. The corn-soybean maps, which we call the Corn-Soy Data Layer (CSDL), are publicly hosted on Google Earth Engine and also available for download online.

Introduction

The USDA’s Cropland Data Layer (CDL) is the industry standard for agricultural mapping in the United States. However, its historical coverage is inconsistent. Prior to 2008, the dataset is a patchwork of varying resolutions and missing states. This "historical hole" limits longitudinal studies on the impacts of policy changes, such as the 2007 Renewable Fuel Standard, or long-term climate shifts on crop distribution. To address this, we developed the Corn-Soy Data Layer (CSDL), a harmonized 20-year record (1999–2018) for 13 key Midwestern states responsible for approximately 90% of U.S. corn and soybean production.

Methodology

The CSDL was generated using Google Earth Engine (GEE) to process over 87,000 Landsat images. The classification pipeline relied on three core components:

  1. Harmonic Regression: To capture crop phenology, we applied Fourier transforms to spectral data, modeling the seasonal growth cycles of pixels rather than relying on static snapshots.
  2. Feature Selection: The model utilized the Green Chlorophyll Vegetation Index (GCVI), Near-Infrared (NIR), and Shortwave Infrared (SWIR) bands. While weather data was tested, it did not significantly enhance performance, leading to a more efficient spectral-only model.
  3. Random Forest Classification: A Random Forest model was trained on high-quality CDL data from 2008–2018 and then applied retrospectively ("hindcast") to the 1999–2007 period.

Results and Discussion

The CSDL demonstrates high fidelity to both pixel-level and aggregate statistics. It achieved over 85% pixel-level agreement with hold-out CDL data. At the county level, the correlation ($R^2$) with official NASS acreage statistics for the 1999–2007 test period was 0.913 for corn and 0.865 for soybeans.

A significant advantage of the CSDL is its consistency. Because it uses a uniform model and resolution across the entire time series, it avoids the noise and artifacts present in early CDL versions.

Comparison of CDL 2001 and CSDL 2001 showing the reduced noise in the CSDL product.

The dataset successfully captures major agricultural trends, such as the "western crawl" of corn and soybean expansion into the Dakotas and western Minnesota. However, challenges remain, particularly regarding cloud cover. In years with persistent cloudiness during the peak growing season (July–August), misclassification rates increase. Additionally, the model shows a slight bias toward over-predicting corn, the dominant crop in the training set.

Conclusion

The Corn-Soy Data Layer provides a critical missing link in American agricultural data. By leveraging the Landsat archive and cloud computing, we have created a consistent, high-resolution record that facilitates deeper analysis of two decades of land-use change. The CSDL is now publicly available for research and policy applications via Google Earth Engine.

Summary updated

Abstract

Field-level monitoring of crop types in the United States via the Cropland Data Layer (CDL) has played an important role in improving production forecasts and enabling large-scale study of agricultural inputs and outcomes. Although CDL offers crop type maps across the conterminous US from 2008 onward, such maps are missing in many Midwestern states or are uneven in quality before 2008. To fill these data gaps, we used the now-public Landsat archive and cloud computing services to map corn and soybean at 30 m resolution across the US Midwest from 1999-2018. Our training data were CDL from 2008-2018, and we validated the predictions on CDL 1999-2007 where available, county-level crop acreage statistics, and state-level crop rotation statistics. The corn-soybean maps, which we call the Corn-Soy Data Layer (CSDL), are publicly hosted on Google Earth Engine and also available for download online.
BibTeX
@inproceedings{wang2020mapping,
  author = {Sherrie Wang and Stefania Di Tommaso and Jillian M. Deines and David B. Lobell},
  title = {{Mapping twenty years of corn and soybean across the US Midwest using the Landsat archive}},
  booktitle = {Scientific},
  year = {2020},
  doi = {10.1038/s41597-020-00646-4},
}
Figure