HISTORICAL RAINFALL RECONSTRUCTION IN PERIOD OF 1900-2010 FOR EXTREME CLIMATE EVENTS ANALYSIS OVER INDONESIA RAGION

• Main Author: Wati, Trinah
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/71266
• Institution: Institut Teknologi Bandung

Extreme weather events and climate anomalies are rising in both quantity and intensity due to climate change, leading to higher risks for the human and natural systems. Thus, analyzing hazards is essential in managing those risks. There are two approaches in climate change analysis i.e., top-down, based on projection scenarios of GCM simulations, and bottom-up, based on historical climate analyses. In the latter, long-term precipitation data up to 100 years are crucial for understanding and analyzing hydrometeorological hazards such as floods and drought. However, availiable observational data does not always meet the requirement. In Indonesia, observational data are usually limited due to the length of data, the density of the network, quality, and consistency when associated with extreme events. On the other hand, the availability of GCM simulations allows the development of extended long-term reanalysis datasets but with coarse spatial resolution. This study aims to develop a method to reconstruct historical rainfall data for the 1900-2010 period over Indonesia using ERA-20C global reanalysis dataset. The main methods involve retrospective application of Constructed Analogues statistical downscaling (CA-SD). In order to apply the CA-SD method, a precipitation database was selected from several candidates of datasets. Performance analysis was carried out on eight research product precipitation datasets in the Indonesian region. The datasets have a horizontal resolution between 0.1° and 0.25° from 2003 to 2015, compared to the observed daily rainfall at 133 stations. The comparison uses 13 statistical metrics grouped into three types i.e., distribution, time sequence, and representation of extreme values. By applying the Summation of Rank (SR), the results show that MSWEPv2 has the best performance based on the research product dataset for applications of climatology and hydrology in Indonesia, and hence it used as a database for the CA-SD rainfall reconstruction. The CA-SD method employs ERA-20C's zonal (U) and meridional (V) wind parameters at 850 hPa as predictors for rainfall reconstruction. The wind vectors are converted into scalar field variables i.e., the stream function ? and potential velocity ? for convenience in the diagnosis steps using cosine similarity. A multivi window approach is also applied so that 14 candidate ensemble members are obtained from seven spatial windows representing key areas of the Asia-Australian monsoons. This allows probabilistic estimation for the rainfall rainfall reconstruction. prediction or estimation. Sensitivity tests are also carried out by applying Singular Value Decomposition (SVD) method, by which three predictors i.e., the ? of windows 2 and 6 and ? of windows 7 are excluded. Furthermore, four schemes namely MLR (multiple linear regression), WMEAN (weighted average), PC-MLR (multiple linear regression with principal component analysis/PCA) and PC-WMEAN (weighted average with PCA) calculation are compared for (backward) prognostic steps. The results indicate that the MLR and PC-MLR schemes are more suitable for capturing extreme daily events than the WMEAN and PC-WMEAN schemes. The developed CA-SD method is applied for daily rainfall data reconstruction of the 1900-2010 period but first tested using a leave one year out approach in the 1979-2010 period. The results show that the reconstructed climatological mean is overestimated compared to MSWEP but underestimated compared to station observation data. More comprehensive evaluation is also performed using several determinisctic and probabilistic metrics including CRPS, SPREAD, correlation, RMSE, BSS and AUC ROC at certain thresholds of rainfall events. The results of CRPS, SPREAD and ensemble mean correlations showed an increase with temporal aggregation. Based on BSS and AUC ROC, the reliability of estimated 5 and 20 mm rain events is better than rain days (>0.5 mm), and the reliability of 50 and 100 mm rain events is the worst compared to other rains. The reliability of estimation based on probabilistic verification also improves with temporal aggregation. The Bayesian Model Averaging (BMA) calibration on 10-day accumulated rainfall increased the SPREAD and RMSE values but did not affect the CRPS value. The results of BMA calibration on 10-day rainfall with the threshold of 200 and 300 mm, show improvement of estimation reliability at 65% to 91% of observation stations and most of gridded in Java Island. The reconstructed rainfall data of 1900-2010 period has been produced in this study, which is potentially usable for climatological and hydrological studies that requires data in monthly and decad (10-day) time scales. In this study, it is demonstrated that the severe drought event over Java Island during the period of 1960 to 1970 can be studied in more details. The spatial and temporal distribution of Standard Precipitation Indices (SPIs) reveal that the drought had occurred due to lack of rainfall during the rainy seasons indicating suppression of monsoons. Global connections of the events are not further studied but such drought of such severity is important to study about its possible reoccurrence in the future due to climate change.