Remote sensing using satellite imagery can be used to map field-level characteristics, including crop type, yield, and phenology, for agricultural areas around the world. Such data can be critical for modeling pest outbreaks and dynamics as it provides much needed information about pest habitat type and quality at landscape scales. Here we focus on the state of the art methods to map crop phenology, particularly in smallholder agricultural systems, where field sizes are small ( < 2 ha) and are typically smaller than the resolution of historical and readily-available satellite imagery. Considering crop phenology, we map several key characteristics, including the start of the season (SOS), the end of the season (EOS), and the overall length of the growing season (LOS). We compare methods that rely on three different satellite sensors that have tradeoffs considering the spatial resolution, temporal resolution, and years of coverage available. Specifically, we compare methods using Moderate Resolution Imaging Spectroradiometer (MODIS) data, with a spatial resolution of 250 meters, a temporal resolution of 1 day, and availability from 2000 onwards, Landsat data, with a spatial resolution of 30 meters, a temporal resolution of 16 days, and availability from the 1970s onwards, and Sentinel-2 data, with a spatial resolution of 10 meters, a temporal resolution of 5 days, and availability from 2015 onwards. We also examine the ability to use fusion algorithms, such as ESTARFM, to fuse the benefits of higher spatial resolution data (e.g., Landsat) with higher temporal resolution imagery (e.g., MODIS). We apply our models to several different smallholder systems, including maize fields in Mexico, maize fields in Ethiopia, and wheat fields in India, to identify the best generalizable approach for mapping crop phenology in smallholder systems across the globe.
