| Summary: | The original raw dataset used to generate this work contained a number of duplicate entries—roughly 7% of the total farm fields. The substantive majority of these were from one large farm that had conducted their operations in a way that caused duplication as a side effect in our data generation process. Unfortunately, as the error was in the raw dataset, its correction required a re-run of the entire data pipeline, resulting in numerous small downstream changes. With respect to the most important numbers, the accuracy of the classifier went down slightly from 91.5% to 91.2% measured in absolute terms but increased from 0.68 to 0.74 measured by kappa. The trend in cover cropped acres grew slightly stronger, and the yield effects in maize and soybean moved from 0.65% to 0.71% and 0.35% to 0.29% respectively. None of the overall conclusions of the work have materially changed. Below, we provide all changes to the applicable sections of the original manuscript in bold underscore (or strikethrough) where applicable, in addition to modified versions of the corresponding figures and supplementary materials.
|
|---|