Abstract
Abstract. Remote sensing in the optical domain is widely used in agricultural monitoring; however, such initiatives pose a challenge for developing countries due to a lack of high quality in situ information. Our proposed methodology could help developing countries bridge this gap by demonstrating the potential to quantify patterns of dry season rice production in Bangladesh. To analyze approximately 90,000 km2 of cultivated land in Bangladesh at 30 m spatial resolution, we used two decades of remote sensing data from the Landsat archive and Google Earth Engine (GEE), a cloud-based geospatial data analysis platform built on Google infrastructure and capable of processing petabyte-scale remote sensing data. We reconstructed the seasonal patterns of vegetation indices (VIs) for each pixel using a harmonic time series (HTS) model, which minimizes the effects of missing observations and noise. Next, we combined the seasonality information of VIs with our knowledge of rice cultivation systems in Bangladesh to delineate rice areas in the dry season, which are predominantly hybrid and High Yielding Varieties (HYV). Based on historical Landsat imagery, the harmonic time series of vegetation indices (HTS-VIs) model estimated 4.605 million ha, 3.519 million ha, and 4.021 million ha of rice production for Bangladesh in 2005, 2010, and 2015 respectively. Fine spatial scale information on HYV rice over the last 20 years will greatly improve our understanding of double-cropped rice systems, current status of production, and potential for HYV rice adoption in Bangladesh during the dry season.
Highlights
Rice is an important food staple for more than 2 billion people globally (Khush 2005; Muthayya et al 2014)
The harmonic time series (HTS)-vegetation indices (VIs) model estimated 4.021 million ha, 3.519 million ha, and 4.605 million ha of rice production at the national level for Bangladesh in 2015, 2010, and 2005, respectively, based on the closest values of LC8 and Landsat 5 (LT5) compared to the reference data
The LT5 analysis over-estimated rice areas in some years and under-estimated in others due to the implementation of a different cloud-masking algorithm from the one employed with LC8, which resulted in a lower number of cloud-free observations for fitting the harmonic time series model
Summary
Rice is an important food staple for more than 2 billion people globally (Khush 2005; Muthayya et al 2014). Mapping the extent of rice-growing areas, understanding diverse rice-farming systems, characterizing rice adoption or abandonment, and evaluating its potential for improvement is crucial to current and future food security goals, as well as environmental concerns such as greenhouse gas emissions (Kuenzer & Knauer 2013; Smith et al 2008; van Groenigen et al 2013; Whitcraft et al 2015). Researchers have identified remote sensing as one of the most effective methods to monitor rice production, especially at regional and national scales (Whitcraft et al 2015). Over the last three decades, different methods for rice mapping and monitoring have been developed using remote sensing data (McCloy et al 1987; Fang et al 1998; Dong et al 2016). The Landsat and Moderate Resolution Imaging Spectroradiometer (MODIS) constellations have been the most widely used because the spectral information they record is suitable for agricultural characteristics (Okamoto 1999; Whitcraft et al 2015)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
