Rice crop monitoring using new generation Synthetic Aperture Radar (SAR) imagery

Rice crop monitoring using new generation Synthetic Aperture Radar (SAR) imagery

[Abstract]: Rice cultivation systems in various countries of the world have been changing in recent years. These changes have been observed in the Mekong River
Delta, Vietnam, specifically in An Giang province. The changes in rice cultural practices have impacts on remote sensing methods developed for rice monitoring, in
particular, methods using new generation radar data. The objectives of the study were a) to understand the relationship between radar backscatter coefficients and
selected parameters (e.g. plant age and biomass) of rice crops over an entire growth cycle, b) to develop algorithms for mapping rice cropping systems, and c) to develop a rice yield prediction model using time-series Envisat (Environmental Satellite) Advanced Synthetic Aperture Radar (ASAR) imagery.

Ground data collection and in situ measurement of rice crop parameters were conducted at 35 sampling fields in An Giang province, Mekong River Delta, Vietnam. The average values of the radar backscattering coefficients that
corresponded to the sampling fields were extracted from the ASAR Alternative Polarisation Precision (APP) images (C band, spatial resolution of 30 m, and swath
width of 100 km). The temporal rice backscatter behaviour during the cropping seasons, including Winter Spring (WS), Summer Autumn (SA), and Autumn Winter (AW), were analysed for HH (Horizontal transmit and Horizontal receive), VV
(Vertical transmit and Vertical receive), and polarisation ratio data. In addition, the relationships between rice biomass and backscattering coefficient of HH, VV, and
polarisation ratio were established.

The methods were examined for rice identification and mapping in the study area by using ASAR APP and Wide Swath (WS) imagery. ASAR APP data were firstly used to determine the best method with high accuracy for rice delineation.
Then, the proposed method was applied for ASAR WS data (C band, 150 m spatial resolution, and 450 km swath width), covering the entire agricultural region of the An Giang province. Based on the discovered relationships between rice parameters and radar backscattering, a thresholding method applied for polarisation ratio and VV polarisation values of single-date ASAR APP data acquired in the middle of crop season was found to be the best method among various classification methods. Another threshold, i.e. the “normalised difference polarisation ratio (NDRa) index”, was formulated in this study for mapping the rice crops using ASAR APP image. The classification accuracy was assessed on the basis of the existing land use data and the published statistical data.

By using multiple regression analysis (rather than using an agrometeorological model found unsuitable for modern rice cultural practices), the correlation between backscattering coefficients of multi-date ASAR APP images
acquired during the crop season and the in situ measured yield was derived. The distribution maps of estimated rice yield were then produced based on that relationship. Consequently, rice production was finally estimated from these maps.

This study showed that the radar backscattering behaviour was much different from that of the traditional rice reported in previous studies, due to changes brought by modern cultural practices. HH, VV and HH/VV radar values were not significantly related to biomass (maximum r2 = 0.494) due to the effect of water management, plant density and structure. Using the polarisation ratio and VV data of rice fields during a long period of the rice season, the thresholding method based on empirical relationships demonstrated a relatively simple but effective tool to accurately derive the rice/non-rice classes. The results using Envisat ASAR APP
data acquired at a single date have provided the highest accuracy (99%) of provincial planted rice areas. To generate map of the rice area planted using three-date or twodate ASAR WS data, the integrated method (based on the temporal variation of the radar response and thresholding) yielded the highest accuracies of 99% and 95%,
respectively, at the provincial scale. This study developed a method to generate an accurate map of rice growing area before the end of crop season using single-date ASAR APP image taken in the middle of the rice cropping season. During this period, the difference between the HH and VV values is the highest. On the other
hand, the predictive model based on multiple regression analysis between in situ measured yield and polarisation ratios attained good results (97% accuracy) and thus
proved to be a potential tool for rice yield prediction.

This study concluded that time-series Envisat ASAR imagery can generate accurate maps of rice planted areas. Since radar backscattering coefficients were found uncorrelated with plant biomass in the study area, the use of SAR imagery for agro-meteorological (crop growth) modelling for rice yield prediction will be less reliable. Conversely, the use of statistical modelling (regression approach) was found highly accurate to generate rice production forecasts. Further work is needed to
examine and validate the rice mapping algorithm and statistical model-based method for rice yield estimation at other regions in the Mekong River Delta.