Author : Muhammad Nur Aidi

Summary : Weather data are very important for agricultural planning and other related activities. However, a weather station may not always be established in every location. Thus, weather data estimation is needed when either one or both of the following problems are involved: (1) estimating weather data for a location without a gaging station, and (2) forecasting weather data of a location with a weather station. Stochastic modeling of Philippine weather data using three geographical variables (longtitude, latitude, and altitude) and time was applied to deal with those problems. Inclusion of these four variables allowed for the spatial and temporal estimation of waether data. The derived stochastic model was used to generate sythetic weather data such as monthly averages, weekly values, or daily sequences. The approach was to determine a multivariate polynomial multiple regression model for monthly data to account for the spatial and temporal variabilities, and then to use the covariance properties to disaggregate monthly weather data to weekly values. Determination of the multivariate polynomial multiple regression model involved: (1) postulating a class of possible models; (2) screening of candidate models; and (3) diagnostic analysis and validation of the selected model. The fitted models passed the multivariate and univariate screenings based on the Wilks' Lambda, F, and t tests. The models were also found adequate for predicting of weather data given the geographical values and time. when used in conjuction with the disaggregation models for Philippine weather dat, they were also shown to be reasonably reliable for the estimation of weekly weather given the estimated monthly weather values.