Integrating remote sensing resources for developing an effective forecasting model in the royal rainmaking operation in upper northern provinces of Thailand

This research aimed at integrating data from remote sensing resources and machine learning for developing a forecasting model of successful royal rainmaking operation in the upper north provinces of Thailand. The Support Vector Machine (SVM), neuron network method, and decision tree (C4.5) were used for data integration and forecast modeling. The data were collected between 1 January 2012 to 31 December 2014 from four remote sensing resources: 1) Radiosonde data from the Meteorological Department; 2) Telemetering data from Hydro and Agro Informatics Institute; 3) Data from Meteorological Geostationary Orbit Satellite of Kochi University, Japan; and 4) Radiosonde data via SOND2 processing program from the Department of Royal Rainmaking and Agricultural Aviation. Dataset from all resources was totally 626 records. There were GPCM data in which 19 variables were selected from totally 57 variables; and data from MTSAT2 in which 109 variables were selected from totally 256 variables. The research findings are presented in three aspects: 1) Rainfall forecasting model, 2) Evaluation of successful opportunity model, and 3) Short term forecasting results. The evaluation of successful opportunity for rainfall forecasting in each province in the upper north of Thailand is also presented. It can be concluded that use of neuron network method with data from GPCM + MTSAT2 provides the most accurate rate for successful forecasting in daily royal rainmaking operation in the upper north of Thailand. However, for the rainfall forecasting (fall or not fall), it was found that use of data from MTSAT2 only is more accurate than use of data from GPCM only and GPCM + MTSAT2.