Abstract
The extent of single- and multi-cropping systems in any region, as well as potential changes to them, has consequences on food security and land- and water-resource use, raising important management questions. However, addressing these questions is limited by a lack of reliable data on multi-cropping practices at a high spatial resolution, especially in areas with high crop diversity. In this paper, we develop and apply a relatively low-cost and scalable method to identify double-cropping at the field scale using satellite (Landsat) imagery. The process combines machine learning methods with expert labeling. The process evaluates multiple machine learning methods, including an image classification of a time-series, trained on data where cropping intensity labels were created by experts who are familiar with regional production practices. We demonstrate the process by measuring double-cropping extent in a part of Washington State in the Pacific Northwest United States-an arid region with cold winters and hot summers with significant production of more than 60 distinct types of crops including hay, fruits, vegetables, and grains in irrigated settings. Our results indicate that the current state-of-the-art methods for identifying cropping intensity-which apply simpler rule-based thresholds on vegetation indices-do not work well in regions with a high crop diversity and likely significantly overestimate double-cropped extent. Multiple machine learning models were applied on Landsat-derived vegetation index time-series data and were able to perform better by capturing nuances that the simple rule-based approaches are unable to. In particular, our (image-based) deep learning model was able to capture nuances in this crop-diverse environment and achieve a high accuracy (96-99% overall accuracy and 83-93% producer accuracy for the double-cropped class with a standard error of less than 2.5%) while also identifying double-cropping in the right crop types and locations based on expert knowledge. Our expert labeling process worked well and has potential as a relatively low-cost, scalable approach for remote sensing applications. The product developed here is valuable for the long-term monitoring of double-cropped extent and for informing several policy questions related to food production and resource use.