Estimating wet season-lowland rice production using remote sensing techniques: a case study in Lao PDR

Author: Vilon Viphongxay

Viphongxay, Vilon, 2020 Estimating wet season-lowland rice production using remote sensing techniques: a case study in Lao PDR, Flinders University, College of Science and Engineering

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact with the details.


Rice is an important crop that feeds more than half of the global population. Rice demand remains high, however, population growth, climate change, limited suitable land areas, and a lack of water challenge its availability. Access to fast and accurate information about rice production at lower spatial units, such as the district-level and lower, is crucial. This study explores the usefulness of freely available satellite images for mapping the extent of rice which, in turn, can be used to forecast rice production at lower spatial units. This study uses satellite imagery from Moderate Resolution Imaging Spectroradiometer (MODIS). Specifically, MOD09A1, the MODIS 8-day composite product that developed by selecting the least cloud contaminated data from the daily images, was used. All available imagery was downloaded from the Land Processes Distributed Active Archive Centre (LPDAAC) within the 2016 growing season, totalling 138 tiles (3 tiles by 46 time-slices). Additionally, Landsat 8 level-1 data, with a higher, 30-metre spatial resolution was used to spot-check the trend of the NDVI and LSWI in some key Agro-Ecological Zones. In pre-data processing step, the MODIS data was calibrated with the scaling factor provided with the manual and followed by the additional cloud removal. Atmospheric corrections and the exclusion of cloud cover and its shadow layers were carried out with the Landsat 8 level-1 data. The algorithm uses the NDVI, EVI, and LSWI to detect moisture levels in soil and vegetation at the time when rice crops were temporarily flooded. After that, all the irrelevant areas such as permanent water, forest, and steeply sloping layers were excluded. Then a phenology analysis of potential rice crops was carried out using the EVI to estimate the potential number of rice pixels. Finally, the study estimates district-level rice production using Simple Linear Regression while the Dasymetric mapping technique was used to interpolate rice yield at the district level. The study found that there were both over- and underestimations of the extent of rice crop areas in different locations within the study area when compared with the official figures. Underestimated rice crop area data was found in the Northern and Eastern upland provinces and this signals a limitation of the method. However, those provinces that have a relatively large scale of rice cultivation have an error of less than 10%, and these are located in low and flat land areas. Overall, the RMSE is 15,000 Ha, and the R2 value is 0.95.

Keywords: Lowland rice, MODIS, NDVI, EVI, LSWI, Dasymetric, Lao PDR

Subject: Earth Sciences thesis

Thesis type: Masters
Completed: 2020
School: College of Science and Engineering
Supervisor: David Bruce