Application of Remote Sensing and Google Earth Engine for Agricultural Mapping in South Asia



Journal Title

Journal ISSN

Volume Title



Agricultural land use is one of the dominant land use types in South Asia (SA). A majority of SA population depends on agriculture for their livelihood. The agricultural activities and food production in SA are tightly related to the poverty level in many SA countries and has broader impact on global food security, climate change, economics, etc. To feed the growing population in SA with most of the land suitable for agriculture already cultivated, crop intensification and transitions from traditional agriculture to non-crop, cash crops, and fishery are expected. These land use and land cover changes have profound impact on the regional and global food security and economics. Thus, timely producing agricultural land use data products with remote sensing technique are very important to achieve sustainable agriculture and monitoring such agriculture land use changes. However, the application of remote sensing in SA faces many challenges, e.g., persistent high cloud covers during monsoon seasons, very limited availability of ground truth samples, and excessively small and irregularly shaped agriculture fields, etc. Thus, this dissertation study aims to design algorithms and tools to help monitoring agriculture land use changes in SA using remote sensing images and provide insight on addressing such challenges. As Google Earth Engine (GEE) is becoming increasingly popular in the remote sensing community, this dissertation also explores utilizing GEE’s potential for operational agriculture mapping and designing complex data processing workflows. To achieve the overall goal, this dissertation presents research for three objectives. Specifically, this dissertation first presents a novel workflow for inland fishpond mapping using spectral and spatial information derived from remote sensing images. This workflow was implemented on GEE and was tested in a case study in Singra Upazila in Bangladesh. The results showed that the method successfully detects fishponds with an F1 score of 0.64. Next, a GEE-based workflow that combines MODIS Terra and Aqua data and uses Harmonic Regression to reconstruct time-series Normalized Difference Vegetation Index (NDVI) for crop intensity mapping was presented. The method was used for crop intensity mapping in Bangladesh 2010 and showed a national average crop intensity of 1.66. Lastly, a GEE-based web application named RiceMapEngine was developed to provide a one-stop experience of rice mapping to higher-level officials and decision makers. This application was demonstrated in rice mapping for Chitwan district in Nepal. The result showed that this application can successfully help produce early-season and post-season rice maps using GEE with very easy-to-use interfaces.



Crop intensity mapping, Fishpond mapping, Google Earth Engine, Paddy rice mapping, Remote sensing, South Asia