The objectives of this research were to propose a model with some functions including CAR spatial effects and to select a proper model for predicting rince and rubber yields in southern provices of Thailand. A linear mixed model (LMM) with spatial effects and dummy varables and Fourier seasonal effects was proposed. Bayesian estimation was used and the estimated rice and rubber yields were used for prediction. The dependent variables were rice and rubber yields in southern provinces of Thailand. Factors considered were rain temperature soil fertilizer soil organic matter trend and also sesonal effects. The results show that rain temperature soil fertilizer soil organic matter trend and sesonal effects influence on rice and rubber yields. The proposed model with Fourier seasonal term is the most appropriate comparing to the model with dummy variable for season effects and the Holt-Winters model. Therefore the proposed model with Fourier sesaonal term should be the first consideration for prediction.

The objectives of this research were to propose a spatial analysis using Bayesian generalized linear mixed model for predicting drought in Thailand. The spatial effect was assumed to be a conditional autoregressive model (CAR model). The dependent variables was a drought or non-drought occurrence which was assumed to have a Bernolli distribution with logistic link function. Factors considered were rain temperature number of rainy days number of large reservoirs and number of medium reservoirs. The results show that the rain temperature number of rainy days number of large reservoirs and number of medium reservoirs influence on the drought occurrence. This study is useful for planning making decision preventing and remedying the drought problem.

The objective of this study was to construct the black pepper prices forecasting model by 4 statistical methods: Box-Jenkins method Holt’s exponential smoothing method Brown’s exponential smoothing method and damped trend exponential smoothing method. Time series of monthly black pepper prices which were gathered from the website of the Office of Agricultural Economics during January 2005 to December 2018 of 168 observations were divided into 2 datasets. The first dataset had 156 observations from January 2005 to December 2017 for constructing the forecasting models. The second dataset had 12 observations from January to December 2018 for comparing the accuracy of the forecasts via the criteria of the lowest mean absolute percentage error (MAPE) and root mean squared error (RMSE). Research findings indicated that for all forecasting methods that had been studied the most accurate method was damped trend exponential smoothing method (MAPE = 6.5773 RMSE = 10.3775)

Weather forecasts are very necessary to predict the future weather conditions. The important thing in forecasting is the model and variables. The model used in the forecasting must be accurate and effective. In addition the variables used in forecasting must be comprehensive and appropriate to be able to represent the forecast. The initial conditions for forecasting must be appropriate and reliable. There are many methods to generate the initial conditions in forecasting. To generate the initial conditions for winter monsoon prediction with Kalman Filtering method which is a method used to predict the status of various systems and able to create values of variables used in forecasting to have a variety of similarities can be used as a starting variable in forecasting very well. In this research the winter monsoon prediction by the Shallow Water Model and validating the initial conditions by Root Mean Square Error (RMSE). The result of weather prediction in the winter monsoon can be used the Kalman Filtering method for generate the initial conditions.

The purpose of this research was to construct the appropriate forecasting model for the prices of field corn in Thailand using four time series analysis methods including Box-Jenkins method Holt’s exponential smoothing method damped trend exponential smoothing method and combined forecasting method. Time series data from the website of the Office of Agricultural Economics with total 227 months during January 1997 to November 2015 are used and divided into two series. The first 216 months from January 1997 until December 2014 used to build the forecasting models and the last 11 months from January until November 2015 used to compare the forecasting methods accuracies via the criteria of the lowest mean absolute percentage error and root mean squared error. Research finding indicated that combined forecasting model was the most accurate.