USDA began using satellite imagery to map crop acreage in the 1970s, but did not make the data public until the late 1990s when it developed the Cropland Data Layer (CDL). The CDL estimates which crop is planted on each 30mx30m pixel in a state. It is now available for every state in the lower 48 for each year since 2009.
Satellites take multispectral images of the earth's surface, and scientists use machine learning to match the spectral profiles of the images to known spectral profiles of crops. Currently, the images underlying the CDL are collected from Landsat 8.
The picture to the right shows pistachio acreage as estimated by the CDL in 2008 and 2019. Note the massive growth, especially in Fresno, Madera, and Tulare Counties.
You can view these data in my new CDL data block. Slide the bar back and forth to toggle between years, or choose different years or crops. (Huge kudos to Seunghyun Lee for creating this app.)
The CDL has been used in economic research on land use in the Midwest, including by UC Davis ARE graduates Nathan Hendricks and Matthieu Stigler. The high spatial resolution of the data is crucial for understanding crop rotation and the environmental effects of biofuel policies, among other topics.
California has a much wider array of crops than the Midwest, which reduces the accuracy of the CDL. In Iowa in 2019, the CDL correctly identified 95% of the corn and soybean pixels in the evaluation dataset, whereas in California it correctly identified only 80% of almond and pistachio pixels and 90% of grape and alfalfa pixels.
This low accuracy raises challenges for using the CDL in research in California, especially for field-level analysis.
However, in aggregate for the state, the acreage implied by the CDL is quite similar to the official data published by CDFA for many commodities, including grapes, rice, walnuts, pistachios and tomatoes. It is less accurate for the other major CA crops, as shown in the plot below. (To make these comparisons for yourself, go to my California Crops and CDL data blocks.)
This coming year, Dalia Ghanem, Alex Aue, Debashis Paul, and I will be working with funding from CeDAR to develop an accurate field-level California crop map from the CDL and quantify its measurement error.