|GUAN, KAIYU - University Of Illinois|
|LI, ZHAN - University Of Massachusetts|
|RAO, NAGRAJ - University Of The Philippines|
|XIE, DONGHUI - Beijing Normal University|
|ZENG, ZHENZHONG - University Of Princeton|
Submitted to: Remote Sensing of Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/25/2018
Publication Date: 6/7/2018
Citation: Guan, K., Li, Z., Rao, N., Gao, F.N., Xie, D., Zeng, Z. 2018. Mapping paddy rice area and yields over Thai Binh province in Viet Nam from MODIS, Landsat and ALOS-2/PALSAR-2. Remote Sensing of Environment. 1-15. https://doi.org/10.1109/JSTARS.2018.2834383.
Interpretive Summary: Rice is one of the most important food staples worldwide, particularly in Southeast Asia. Timely and reliable rice production estimates are important for crop management and administrative planning. The conventional method to estimate rice paddy area and yield is through administrative data collection, which is time-consuming and subjective. Satellite remote sensing data have been used in crop production estimation for decades. However, the successfulness depends on the availability of satellite images in optical bands, which are often obscured by frequent cloud coverage. This paper uses multiple remote sensing data, including the fused Landsat-MODIS data and L-band radar data to estimate planting area and yield for the Thai Binh Province in Vietnam for the second growing season of 2015. Results show that the fused Landsat-MODIS data provide frequent remote sensing images at field scale and significantly improves the crop yield estimation using the peak vegetation index. The remote sensing data fusion approach provides an effective and low-cost data set for rice production estimation, which is required by the National Agricultural Statistics Service and Foreign Agricultural Service.
Technical Abstract: Rice is the most important staple crop grown and consumed in Southeast Asia. Timely and reliable rice production estimates are therefore important in monitoring government development plans in the region and in mitigating the effects of extreme weather and climate change to address food insecurity. Estimating crop production requires mapping rice growing area and estimating crop yield, and satellite data has been long used for these purposes in developed countries. However, there remain significant challenges with utilizing satellite data for paddy rice particularly in Southeast Asia, due to significant cloud coverage leading to a scarcity of optical satellite data for such analysis. While radar data can penetrate through clouds, there is inconclusive and uncertain evidence on their applicability to estimate paddy rice yield. In this paper, we study the use of multiple sources of satellite data for mapping paddy rice area and estimating yield for the Thai Binh Province in Viet Nam for the second growing season of 2015. The two major satellite data used are: (1) a fusion data of surface reflectance by integrating Landsat and Moderate Resolution Imaging Spectroradiometer (MODIS) to generate high spatial resolution at more frequent time stamps, and (2) L-band radar backscatter data from the Advanced Land Observing Satellite 2 (ALOS-2) PALSAR-2 sensor at four time periods over the growing season. We first use these satellite data to map paddy rice fields and assess the performance of different data for the classification accuracy. We then utilize field-level yield data obtained through crop cutting activities to build an empirical model to estimate crop yield. Our findings indicate that although the NDVI time-series computed from Landsat-MODIS fusion reflectance data is not necessarily beneficial for paddy rice mapping (compared with only using the Landsat data), the fusion data significantly improves the crop yield estimation. Specifically, the use of the Landsat-MODIS fusion data allows us to find the peak value of vegetation index (VI) and derive the best empirical relationship between peak VI and crop cutting yield data. Our study also shows that L-band radar data has similar (or slightly lower) performance in paddy rice mapping as optical satellite data, but it has little contribution in crop yield estimation compared with the optical data. Finally, since we only have crop cutting yield data for one growing season, the yield model can at most capture the aboveground biomass change (variability?) through the peak VI. We suggest that crop cuts data from multiple years or at different regions are needed to account for inter-annual variation of Harvest Index during the reproductive stage.