PREDICTION OF DROUGHT-FIRE RISK IN PEATLAND USING SPACE-TIME MODELING

• Main Author: Prastoro, Tarasinta
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84066
• Institution: Institut Teknologi Bandung

Peatlands in Indonesia are one of the largest peatlands in the world. Peatlands are useful as carbon storage and have high biodiversity. However, peatlands have the potential to endanger the environment in unstable conditions that occur when there is a clean release of carbon dioxide (CO2) gas. Dry peatlands will have contact with oxygen gas (O2) so that they become prone to burning. These fires are closely related to changes in land use that cause peatlands to degrade or dry out so that peatland conditions become unstable. To anticipate unstable conditions, indices can be used to measure peatland conditions, such as the Normalized Difference Vegetation Index (NDVI) and the Peat Fire Vulnerability Index (PFVI). PFVI uses water level, soil moisture, rainfall, and daily maximum temperature. Because water level and rainfall can change over time and are influenced by environmental conditions, the Generalized Space Time Autoregressive (GSTAR) space-time model is used to predict PFVI. In addition, the Autoregressive Integrated Moving Average (ARIMA) time series model is also used to predict maximum temperature. With the PFVI prediction results, a prediction of the drought-fire risk from peatlands can be obtained. GSTAR modeling consists of six stages. First, the GSTAR model order is identified by observing the movement of the Space Time Autocorrelation Function (STACF) and Space Time Partial Autocorrelation Function (STPACF) against time and spatial lag. Second, the GSTAR model parameter estimation is carried out using the Least Squares method. Third, a diagnostic test is carried out in the form of a stationarity test using the Inverse Autocovariance Matrix (IAcM) approach and a test for the fulfillment of the white noise assumption on the residuals using the Ljung-Box and Kolmogorov-Smirnov tests. Fourth, the best model is selected based on the Mean Square Error (MSE). Fifth, a model goodness-of-fit test is carried out which tests the significance of the parameters in the best model. Sixth, a prediction is made for some time into the future with the best model. NDVI measures the level of vegetation health using near infrared (NIR) radiation and red light radiation while PFVI measures the risk of drought-fire using water level, soil moisture, rainfall, and daily maximum temperature. To estimate the parameters in the PFVI equation, the PFVI value is fitted to the observed drought index (DIobs) which is a drought index that only depends on soil moisture. In this thesis, a requirement is added that the parameter has a positive value. In addition, modifications were made to the PFVI, which was initially daily, to every seven days. In this thesis, Jabiren Raya District, Pulang Pisau Regency, Central Kalimantan Province is used as an observation area. For modeling, four peatland locations were used, named Jabiren, Jabiren2, Jabiren5, and Jabiren7. The data used are data on water level, soil moisture, rainfall, and daily maximum temperature that are seven days apart from each other from February 20, 2021 to March 18, 2023. GSTAR modeling on water level and rainfall produces the best models for both variables, namely GSTAR(1;2) with the inverse distance weight matrix and GSTAR(1;0), respectively. This indicates that the water level of a location is influenced by all other locations while the rainfall of a location is only influenced by the location itself. ARIMA modeling on daily maximum temperature produces the best model, namely ARIMA(0,0,0). This indicates that the estimated daily maximum temperature will be constant. By fitting PFVI to DIobs through the Nelder-Mead optimization method, PFVI parameter estimates were obtained which were then used for PFVI prediction. Using the results of predictions of water level, rainfall, and daily maximum temperature, PFVI predictions were made for the next five weeks. Furthermore, the PFVI values at the four location points were used to create a PFVI contour map in part of the Jabiren Raya District. For the next five weeks, it was predicted that the areas around Jabiren and Jabiren5 had a low risk of drought-fires while the areas around Jabiren2 and Jabiren7 had a high risk of drought-fires. This contour map was then compared with the emergence of hotspots on May 15, 2021 and December 3, 2022. There was a contradiction in the PFVI mapping results which stated that observation areas close to hotspots had a low risk of drought-fires. Therefore, the PFVI value may not necessarily be able to capture the emergence of hotspots. In addition, the PFVI contour map was also compared with the NDVI contour. From this comparison, it was found that dense canopy density according to NDVI may still have a high risk of drought-fire.