Quantifying Uncertainty in Future Extreme Precipitation Projections: A Dual Focus on Scaling Rates and Physical Changes = 미래 극한 강수 예측의 불확실성 정량화: 스케일링 속도와 물리적 변화에 대한 이중 관점

다국어 초록 (Multilingual Abstract)

This thesis tackles the uncertainty in future projections of extreme precipitation, focusing on both scaling rates with temperature and the projected changes. The analysis of scaling rates across global land regions reveals that the dominant source of uncertainty is the GCMs, while future emission scenarios contribute relatively little. In some areas, the scaling method also plays a significant role. Notably, using around nine GCMs is sufficient to obtain robust estimates in most regions, and CMIP6 models tend to show lower GCMs’ contribution to the uncertainty than CMIP5, reflecting possible improvements in model design. In terms of projected changes between the projection and historical periods, increases are evident globally, especially over densely populated regions. Decomposing extreme precipitation into thermodynamic and dynamic components shows that thermodynamic changes are consistently positive and tightly linked to warming, while dynamic changes are highly variable across regions. Dynamic uncertainty is dominated by internal variability throughout the projection period, whereas thermodynamic uncertainty grows over time as the contribution of model and scenario increase. Signal-to-noise ratios indicate that the projected thermodynamic changes are relatively robust, particularly in the tropics, while dynamic components remain highly uncertain, limiting their utility for regional adaptation planning. These findings highlight the need to improve the representation of atmospheric dynamics in climate models and to integrate uncertainty quantification into climate risk assessments.

국문 초록 (Abstract)

본 논문은 극한 강수의 미래 예측에서의 불확실성을 다루며, 온도에 따른 스케일링 비율과 예측 변화 모두에 초점을 맞춘다. 전 세계 육상 지역을 대상으로 한 스케일링 비율 분석 결과, 지배적인 불확실성의 원천은 일반순환모형(GCMs)이며, 미래 배출 시나리오는 상대적으로 작은 기여를 하는 것으로 나타났다. 일부 지역에서는 스케일링 방법 또한 중요한 역할을 한다. 특히 약 9개의 GCM을 사용하는 것으로 대부분의 지역에서 강건한 추정치를 얻기에 충분하며, CMIP6 모형은 CMIP5에 비해 GCM의 불확실성 기여가 낮게 나타나 모델 설계의 향상을 반영할 가능성이 있다. 예측 기간과 과거 기간 사이의 극한 강수 변화에 있어, 전 세계적으로 특히 인구 밀집 지역에서 증가가 뚜렷하게 나타난다. 극한 강수를 열역학적 성분과 역학적 성분으로 분해한 결과, 열역학적 변화는 일관되게 증가하는 경향을 보이며 온난화와 밀접하게 연관되어 있는 반면, 역학적 변화는 지역에 따라 매우 가변적이다. 예측 기간 전반에 걸쳐 역학적 불확실성은 내부 변동성에 의해 지배되며, 열역학적 불확실성은 모델 및 시나리오의 기여가 증가함에 따라 시간이 지남에 따라 커진다. 신호 대 잡음 비(signal-to-noise ratio)는 열역학적 변화가 특히 열대 지역에서 비교적 강건함을 시사하는 반면, 역학적 성분은 높은 불확실성을 유지하며 지역 적응 계획에서의 활용 가능성을 제한한다. 이러한 결과는 기후 모형에서 대기 역학의 표현을 개선할 필요성과 기후 위험 평가에 불확실성 정량화를 통합할 필요성을 강조한다.

목차 (Table of Contents)

• I. Introduction 1
• 1. Background 1
• 2. Research Goals and Objectives 6
• II. Literature Review 7
• 1. Global Circulation Models 7
• 2. The Uncertainty of Future Projections 8
• III. Study I: Uncertainty Decomposition of Scaling Rates 11
• 1. Introduction 11
• 2. Objectives 13
• 3. Data and Methods 14
• 3.1. Future Emission Scenarios 15
• 3.2. Global Circulation Models (GCMs) 16
• 3.3. Scaling Methods 19
• 3.4. Sobol' Sensitivity Analysis 21
• 3.5. Framework for Quantifying Numbers of GCMs 22
• 4. Results 23
• 4.1. Uncertainty Decomposition of the Projected Scaling Rates 23
• 4.2. GCM Quantity in the Uncertainty Decomposition Framework 30
• 4.3. A Comparison Between CMIP5 and CMIP6 37
• 5. Discussions 47
• 6. Conclusions of Study I 49
• IV. Study II: Uncertainty Decomposition of Changes in Extreme Precipitation 51
• 1. Introduction 51
• 2. Objectives 53
• 3. Data and Methods 54
• 3.1. Future Emission Scenarios and GCMs 55
• 3.2. Physical Diagnostic Model 57
• 3.3. Uncertainty Decomposition Framework 58
• 3.4. Signal-to-Noise Ratio (SNR) 59
• 4. Results 60
• 4.1. Changes of Rx1Day and the Contributions from TH and DY 60
• 4.2. Uncertainty of Changes and the Decomposition 64
• 4.3. SNR Analysis 69
• 5. Discussions 73
• 6. Conclusions of Study II 76
• V. Overall Conclusions 79
• References 82
• Abstract* 90
• 국 문 초 록 92
• Acknowledgment 94
• List of Figures
• Figure 1. The overview of the study design. 15
• Figure 2. Geographic distribution of 26 regional areas utilized in this study. The regional definitions are adopted from the IPCC Special Report on Managing the Risks of Extreme Events and Disasters to Advance Climate Adaptation (SREX). 16
• Figure 3. The average of estimated scaling rates and their spread (difference between the 90th and 10th quantile) across 153 ensembles for the entire simulation period (2041-2090): (a) average scaling rates for grids, (b) average scaling rates for regions, (c) spread for grids, and (d) spread for regions. 25
• Figure 4. Global uncertainty contributions from each source: (a) GCMs, (b) future emission scenarios, (c) scaling methods, (d) interaction between GCMs and future emission scenarios, (e) interaction between GCMs and scaling methods, and (f) interaction between future emission scenarios and scaling methods. The values in the box at the bottom left of each panel represent the global average across all grid points. 27
• Figure 5. Uncertainty contribution from all sources for six regions: (a) AMZ, (b) CGI, (c) NAS, (d) SSA, (e) SAF, and (f) NAU; along with (g) the global average. In each panel, colors represent the uncertainty contribution from different sources: blue for GCMs, purple for future emission scenarios, red for scaling methods, yellow for interactions between GCMs and future emission scenarios, green for interactions between GCMs and scaling methods, and pink for interactions between future emission scenarios and scaling methods. 29
• Figure 6. Same as Figure 5, but for the remaining 20 regions. 30
• Figure 7. Uncertainty contribution from varying numbers of GCMs, ranging from 2 to 17. (a) Average uncertainty contribution over 31 simulation periods from GCMs of 6 regions that were presented in Figure 4. Each scatter point represents the average values of all possible combination pairs taken from the 17 GCMs. The shaded red line denotes the 5% of uncertainty contribution from the 17 GCMs, represented by red points. (b) Globally averaged uncertainty for each simulation period. The hatch in each cell of the heat map indicates values within 5% of uncertainty contribution from the 17 GCMs. 32
• Figure 8. Same as Figure 7a, but for the remaining 20 regions. 33
• Figure 9. Distribution of uncertainty contributions across six regions: (a) AMZ, (b) CGI, (c) NAS, (d) SSA, (e) SAF, and (f) NAU for different numbers of GCMs (2, 5, 9, 10, 12, 14, 16): Each bar represents a sample from combinations within each group of each number of GCM. 35
• Figure 10. Same as Figure 9, but for the remaining 20 regions. 36
• Figure 11. Same as Figure 3, but for CMIP5. 40
• Figure 12. Same as Figure 4, but for CMIP5. 41
• Figure 13. The differences of the uncertainty contribution of each source between CMIP6 and CMIP5 (CMIP6-CMIP5). (a) GCMs, (b) future emission scenarios, (c) scaling methods, (d) interaction between GCMs and future emission scenarios, (e) interaction between GCMs and scaling methods, and (f) interaction between future emission scenarios and scaling methods. Purple color indicates negative values, where an uncertainty source contributes in CMIP5 more than in CMIP6, while orange indicates the opposite. The values in the box at the bottom left of each panel represent the global average across all grid points. 42
• Figure 14. Same as Figure 6, but for 26 regions of CMIP5. 45
• Figure 15. Differences in uncertainty contributions between CMIP6 and CMIP5 (CMIP6 - CMIP5). The first column illustrates regional differences, with bars indicating the average across all simulation periods and error bars representing the magnitude of the standard deviation. The second column depicts global averages for each simulation period, with a line plot showing the average value and a shaded red area indicating the magnitude of the standard deviation. 46
• Figure 16. Spatial and temporal changes in Rx1Day under SSP1-2.6 (a, b), SSP2-4.5 (c, d), and SSP5-8.5 (e, f). Panels (a), (c), and (e) show spatial averages across all years and GCMs for each scenario at each grid point. Panels (b), (d), and (f) display temporal trends, computed as averages over all grids and GCMs for each scenario by year, followed by linear regression. The orange text indicates the p-value of the regression. 63
• Figure 17. Spatial and temporal changes in TH and DY components under SSP1-2.6 (a, b, c), SSP2-4.5 (d, e, f), and SSP5-8.5 (g, h, i). Panels (a, d, g) and (b, e, h) show spatial averages across all years and GCMs for each scenario at each grid point for TH and DY respectively. Panels (c), (f), and (i) display temporal trends, computed as averages over all grids and GCMs for each scenario by year, followed by linear regression. The left-y axis for TH represented purple while the right-y for DY represented orange color. The texts indicate the p-value of the regression. 64
• Figure 18. Uncertainty contributions from the three sources for the global average over time (a, c, e), and the corresponding total uncertainties (b, d, f). Panels (a, b) represent Rx1Day changes, (c, d) show its TH component, and (e, f) illustrate the DY component. 67
• Figure 19. Global uncertainty contributions from all sources. Each large panel represents Rx1Day changes (first), TH (second), and DY (last). Each column corresponds to a different source of uncertainty, and each row shows a different time slice: Near Future (20212040), Mid Future (20512070), and Far Future (20812100). 68
• Figure 20. Spatially distribution of SNR for TH (first column) and DY (second column). Each row shows a different time slice: Near Future (20212040), Mid Future (20512070), and Far Future (20812100). 71
• Figure 21. Spatially distribution of how noise in TH (first column) and DY (second column) would affect the signal of Rx1Day changes. Each row shows a different time slice: Near Future (20212040), Mid Future (20512070), and Far Future (20812100). 72
• List of Tables
• Table 2. Summary of the seventeen CMIP6 GCMs utilized study I. 18
• Table 3. Summary of the ten CMIP5 GCMs utilized in study I. 37
• Table 4. Summary of the twenty CMIP6 GCMs utilized in study II. 56