PREDICTIVE MODEL FOR EARLY WARNING SYSTEM FOR HOTSPOT OCCURRENCE UTILIZING SPATIO-TEMPORAL AND PROBIT REGRESSION MODEL BASED ON PEATLAND CHARACTERISTICS AND PRECIPITATION

• Main Author: Caesar Suherlan, Bagas
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84181
• Institution: Institut Teknologi Bandung

Spatial mapping in hotspot modeling typically does not account for temporal effects. This limitation becomes significant when time information is a crucial factor, such as in early warning systems. The formulation of a probit regression model with covariates following a space-time process, such as a combination of the Generalized Space-Time Autoregressive (GSTAR) model and kriging methods, can address this issue by producing predictions of hotspot probabilities for future times. The GSTAR model functions as an extrapolator, providing time-based information, while the kriging model offers spatial estimates for the observation area, thereby completing the data set for subsequent input into the regression model to generate probability values. Furthermore, temporal information necessitates revising the assumption of independence among observations, requiring the Generalized Estimating Equation (GEE) method for estimating the parameters of the modified probit regression model. The results of this predictive model construction demonstrate a fairly good performance in mapping the risk of hotspot occurrence in Riau Province based on peatland characteristics and rainfall.