Estimation of rice crop yield in Thailand using satellite data

Abstract

Occupying over 12% of the global cropland area, rice is the predominant crop in many
regions of the world. Southeast Asia alone accounts for 31% of the world’s rice harvesting area,
making this region vital for the food security of the growing global population. Current literature
in the field indicates that there are several factors impacting rice productivity, however there are
gaps pertaining to country-specific studies, namely the impact of climate change and challenges
regarding effective monitoring. Therefore, this study focuses on four research questions, they are:
(1) the climate parameters influencing rice productivity in Thailand; (2) the correlation between
rice biophysical variables and growth rate as a determinant to overall rice yield; (3) the potential
of satellite sensors for rice yield; and (4) the development of a regression model for rice yield
estimations.
For the first question, climate data (measured by two rainfall parameters and six crucial
temperature parameters) and rice yield data, which were collected at the provincial level
between the years 1981-2015, are used to determine the impacts of the climate on rice
productivity in Thailand. The result indicates a significance increasing/decreasing trend in the
mean minimum temperature, mean maximum temperature, and cumulative rainfall. The study
further investigates the importance of geographical variation by adopting spatial autocorrelation
(Moran’s I index). The result reveals that in 1992 there was a significant shift in cumulative rainfall
and the average temperature.
Furthermore, field experiments were conducted on rice crops in Thailand during the wet
season of 2017 to explore the correlation between rice biophysical variables and growth rate. The
temporality of rice biophysical variables is demonstrated by separating rice variety and irrigation
system. The leaf area index (LAI) peaks in the flowering stage and LAI development can be slightly
different depending on the rice variety and irrigation system. The correlation between yield and
other rice biophysical elements on a specific variety (RD41) is highly correlated to rice age, stem
density, height, chlorophyll contents, and wet and dry biomass. The correlation between yield,
and wet and dry biomass during the harvesting stage was the strongest.
To develop a rice yield prediction model, data collected from the time series of two
different satellite sensors: Sentinel-2 (optical) and Sentinel-1 (Synthetic Aperture Radar, or SAR)
were utilised. The vegetation indices (NDVI and EVI) and backscatter coefficient (sigma nought;
σ0
) usefully tracked rice phenology. The study furthers develop a linear regression model for rice
yield estimations based on different sensors and yields from in-situ measurements via Crop
Cutting Experiments (CCEs). The accuracy of the results is compared to official rice yields.
The correlation between vegetation indices, backscatter coefficient, and rice yield variables
is investigated in different growth stages and irrigation systems. Based on the simple regression
model for the optical sensors, the developed yield estimation model is correlated with NDVI in the
panicle stage (r = 0.37 and SEE = 0.70 tonnes/ha). While SAR (σ0
) is significant in the ascending
VV/VH ratio during the harvesting stage (r = 0.54 and SEE = 0.68 tonnes/ha). The findings suggest
remotely sensed data can be a good predictor for rice yield during the booting and mature stages

More information

Identifiers

ORCID for Jadunandan Dash: orcid.org/0000-0002-5444-2109

ORCID for Gareth Roberts: orcid.org/0009-0007-3431-041X

Catalogue record

Export record