Historical mapping of rice fields in Japan using phenology and temporally aggregated Landsat images in Google Earth Engine.

Mapping the expansion or reduction of rice fields is fundamental for food and water security, greenhouse gas emission accounting, and environmental management. The historical mapping of rice fields with satellite images is challenging because of the limited availability of remote sensing and training data from past decades. The use of phenology-based algorithms has been proposed for mapping rice fields because they can take advantage of rice fields' characteristic spectral signature during the transplanting phase and do not need training data. However, in order to employ phenology-based algorithms effectively for the historical rice mapping of large areas, we need to incorporate automatized methods able to deal with non-usable data (e.g., cloud cover) and with spatial inconsistencies in the number of available images for each pixel. Here we propose the combination of a pixel-based, phenological algorithm with the temporal aggregation of all available Landsat images to produce national level historical maps of rice fields in Japan from the 1980s onwards. We used temporally aggregated metrics (median, percentiles, etc.), derived from spectral indices of a large number of images within the Google Earth Engine, to minimize the issue of inconsistent image availability and reduce the effects of outliers in phenology-based algorithms. We produced seven rice field maps, for the periods 1985–89, 1990–94, 1995–99, 2000–04, 2005–09, 2010–14, and 2015–19. The overall map accuracies ranged from 83% to 95% when validated with visually interpreted aerial photography. We detected a 23% decrease in the area of rice fields at a country level, although the changes varied greatly among prefectures. Here we present the first freely available historical rice field maps of Japan from the 1980s onwards, together with the source code, and a web application that enables the exploration of the maps and data relating to the derived rice field area changes. The application of temporal aggregation is promising for dealing with the gap-filling of large amounts of satellite data, reducing the issue of data outliers and providing an effective use of the historical Landsat archive for phenology-based crop detection algorithms. Our maps could greatly help researchers, conservationists and policymakers studying the drivers and consequences of rice field changes, and our methods could be extrapolated to map rice fields at large scales in other regions of the world. [ABSTRACT FROM AUTHOR]