dc.description.abstract | Vietnam is one of the most important countries in producing rice in Asia as well as in the world. The majority of rice is produced in the Mekong Delta (MD) which was known as the rice bowl of Vietnam. Annually, it produces approximately a half of the country′s rice and account for more than 80% amount of rice export. The significance of rice crop yield estimation plays a critical role in agricultural management and policy development at regional and national scale. This study aims to develop an approach for rice crop yield prediction in the Mekong Delta, Vietnam using MODIS and ENVISAT ASAR data for rice crop seasons in 2007 and 2008. The data were processed through five main steps: (1) constructing time-series MODIS NDVI/EVI data, (2) noise filtering of the time-series NDVI/EVI data using the wavelet transform, (3) Fusion of MODIS and ENVISAT ASAR images, (4) developing a rice-crop yield prediction models, and (5) result verification using the root mean square error (RMSE), the mean absolute error (MAE), and the mean bias error (MBE).
In this study, an attempt has been made to study the potential of MODIS NDVI/EVI time-series data and ASAR images individually for the purpose of rice yield forecasting. Then, fusion data was also used as another case to estimate rice crop yield. At the same time, the combinations between NDVI-LST, EVI-LST and EVI-ASAR were also implemented to test if there is an improvement in the correlation and prediction results. The regression analysis between rice crop yield statistics and different parameters was implemented using linear and quadratic models.
From the regression analysis results, it was found that the statistical model-based can be successfully used for the purpose of rice yield estimation in the study area. The rice crop yield in MD could be better modeled using quadratic models compared to linear models. The quadratic model using combination of 2two variables (MODIS EVI and Backscattering coefficients) is the best one and gave more accurate prediction results than others, with correlation coefficients of 0.83 and 0.77 for the first and second crop in 2007 and R2 were 0.77 and 0.75 for crops in 2008, respectively.
The robustness of the established models was evaluated by comparisons between the predicted yields and crop yield statistic for 20 sampling districts in 2007 and 2008. The comparisons revealed satisfactory results obtained from the quadratic model using combination of MODIS EVI and Backscattering coefficients (derived from ASAR data) in both years. The percentage difference of the predicted from the actual yield is within acceptable limit (around 10%) and p-value < 0.05. The root mean square error (RMSE), mean absolute error (MAE) and mean bias error (MBE) were used to evaluate the prediction results. In 2007, The RMSE, MAE and MAE were 10.85%, 9.39% and -3.39% for winter-spring crop respectively. And for the summer-autumn crop, those parameters were 12.01%, 9.99% and 9.31 %. In 2008, for the first crop, the RMSE, MAE and MBE were 8.39%, 6.6%, and -0.54%. For the second crop, the RMSE, MAE and MBE were 8.96%, 7.29% and 0.45%. Those results were clear that there was a good correlation between the predicted yield and the rice yield statistics and the established model can be used to estimate rice crop yield in the study area.
In fact, there are many factors like pest, rice diseases, and the variations of climate conditions in the rainy season could lower the accuracy in rice crop yield prediction results. This study explored the potential of MODIS NDVI/EVI time-series and ASAR data for rice crop yield estimation in Mekong Delta before the harvesting period. The methods used in this study could be transferable to the other regions. | en_US |