2.1. Study Area Description

This study examines the Pampanga River Basin and the Pasig-Marikina-Laguna-Lake Basin, both located on Luzon Island, Philippines, as shown in Fig. 1. These basins are highly vulnerable to hydrometeorological hazards and are critical to the country’s agriculture, water resources, and urban infrastructure systems.

The Pampanga River Basin covers approximately 12,105 km\(^2\) in the Central Luzon region. It drains into Manila Bay and includes a broad alluvial floodplain with low elevations and flat terrain, especially in the lower areas. The basin also contains the Candaba Swamp, a seasonal wetland that helps retain floodwaters. In comparison, the Pasig-Marikina-Laguna-Lake Basin spans about 4,402 km\(^2\) in Metro Manila and nearby provinces. It includes Laguna de Bay (the largest lake in the Philippines), the Marikina River, and the Pasig River, which flows west into Manila Bay. This basin consists of both urban lowlands and hilly uplands, particularly within the Marikina watershed.

The Pampanga River Basin is primarily agricultural, with rice and sugarcane as the main crops, and is vulnerable to flooding and seasonal water shortages. The Pasig-Marikina-Laguna-Lake Basin is densely populated and highly urbanized, particularly in Metro Manila, which increases its risk of urban flooding during periods of extreme rainfall.

Both basins have a tropical monsoon climate, with a distinct wet season from May to October influenced by the southwest monsoon and frequent tropical cyclones, and a relatively dry season from November to April. Annual rainfall usually exceeds 2,000 mm, with significant year-to-year variation due to ENSO.

2.2. Climate Projections and Downscaling Approach

This study utilized climate projections from two versions of the MRI-AGCM ¹⁴,developed by the Japan Meteorological Agency. Recent global experiments with the MRI-AGCM multi-SST ensemble show that the model reproduces regional precipitation climatology with higher spatial accuracy than many CMIP5 models ⁵. In most tropical land areas, including Southeast Asia and the Maritime Continent, the MRI-AGCM projects increases in both mean and extreme rainfall toward the late 21st century, consistent with intensification of the Asian monsoon under global warming. The magnitude and spatial pattern of these projected increases closely match the multi-model CMIP5 ensemble, but with less inter-model variation, which indicates MRI-AGCM’s reliability for high-resolution rainfall projection studies in the region ⁵. In addition, a recent multi-model evaluation study ¹⁵ shows that MRI-AGCM3.2 reproduces seasonal-to-annual precipitation climatology and extreme-precipitation statistics as well as, or better than, a broad set of CMIP5/CMIP6 AMIP and HighResMIP atmospheric models, supporting its suitability for extreme-rainfall applications. MRI-AGCM3.2 is also among the highest-resolution global AGCM configurations in HighResMIP-class comparisons, which provides an advantage for resolving regional precipitation structures important for downscaling and basin-scale hazard assessment.

We used two versions of MRI-AGCM 3.2, which were further refined through downscaling to provide basin-scale climate projections. MRI-AGCM 3.2H has a horizontal resolution of approximately 60 km, and MRI-AGCM 3.2S operates at a finer resolution of approximately 20 km. Both models are widely used in CMIP5-based climate projection studies and can simulate large-scale atmospheric phenomena, such as monsoonal systems and tropical cyclone activity. Simulations covered a historical baseline period (1979–2003) and future climate scenarios under two Representative Concentration Pathways (RCPs): RCP 2.6 and RCP 8.5, which represent low and high greenhouse gas emission trajectories, respectively. For comparison with CMIP6 Shared Socioeconomic Pathways (SSP), RCP 2.6 is broadly comparable to SSP1-2.6, and RCP 8.5 is broadly comparable to SSP5-8.5 in terms of end-of-century radiative-forcing class, but they differ in emissions composition and transient evolution. Therefore, the correspondence is approximate, not exact ¹⁶.

To refine the coarse-resolution outputs of the GCMs for basin-scale impact analysis, we used dynamical downscaling with the WRF model version 3.7.1 ¹⁷. The WRF model settings followed those in Ushiyama et al. ⁹. However, the domain configuration used a wider area, with a \(110 \times 110 \times 40\) grid for the 15-km domain and a \(112 \times 112 \times 40\) grid for the 5-km domain. We also used three-dimensional variables (U: zonal wind, V: meridional wind, Q: specific humidity, T: temperature) at their original resolutions (S: 20 km, H: 60 km) instead of the 1.25-degree resolution. Our model configuration shows a significantly lower bias compared to the previous study ⁹ in reproducing monthly rainfall climatology over the Pampanga River Basin, as shown in Fig. 2.

Although the improved resolution and physical representation enhance the results, the downscaled WRF output still shows systematic biases, especially in precipitation magnitudes. To correct this, we applied statistical bias correction using the Quantile Mapping (QM) method, following Ushiyama et al. ⁹. This method adjusts the cumulative distribution function (CDF) of the model outputs to match that of observed rainfall data from 38 ground-based gauges collected by the Philippine Atmospheric, Geophysical and Astronomical Services Administration (PAGASA) and the Metro Manila Development Authority (MMDA).

The complete modeling workflow includes selecting GCM outputs (MRI-AGCM 3.2H and 3.2S), downscaling with WRF to 5 km resolution, applying QM bias correction, and computing rainfall climatology, extreme precipitation indices (such as heavy rain or drought indicators), and frequency analysis to calculate design rainfall. This methodological framework provides high-resolution, physically consistent, and statistically reliable climate projections that support disaster risk reduction and water resource planning in the Philippines.

2.3. Heavy Rainfall and Drought Indicators, and Design Rainfall Calculation

Three categories of hydroclimatic indicators were derived from the bias-corrected daily rainfall data set to evaluate future changes across the study basins. First, basin-averaged rainfall climatology was assessed by calculating monthly and annual mean precipitation, as well as the spatial distribution of annual mean rainfall, to understand spatial variability in changes. Second, rainfall-extreme indicators were analyzed. For heavy rainfall, the 95th and 99th percentile rainfall values were calculated, and the number of days exceeding 50 mm/day was determined. For drought, the consecutive dry days (CDD) and consecutive wet days (CWD) indices were calculated using a 1 mm/day rainfall threshold, following definitions from the Expert Team on Climate Change Detection and Indices (ETCCDI) ¹⁸. These were calculated for both the historical baseline (1981–2000) and future period (2075–2100) to evaluate changes in drought persistence and wet spell duration. Lastly, to understand the impact of changes in heavy rainfall on the design rainfall that is typically used as design criteria for infrastructures, the annual maximum rainfall (in multiple durations 1-day to 5-day) were calculated and frequency analysis was done using Gumbel distribution.

Given the limited duration of available annual maximum rainfall records (25 years), the Gumbel distribution is a statistically sound and practical choice for extreme value analysis in hydrology. As a special case of the Generalized Extreme Value (GEV) distribution with a shape parameter (ξ) of zero, the Gumbel model is often preferred when the data length is insufficient to estimate tail behavior reliably ^19,20. Unlike the GEV distribution, which requires stable estimation of a third (shape) parameter, the Gumbel model performs well with 25–30 years of annual maxima in data-limited regions ²¹. The GEV distribution, while more flexible, introduces uncertainty through its third parameter, which is particularly problematic for short series where the shape parameter is poorly constrained. In these cases, the added complexity of GEV may result in overfitting or unstable extrapolations ²². The Gumbel distribution has proven effective in estimating moderate return periods (e.g., 10–50 years), making it a widely adopted model for infrastructure design rainfall and disaster planning.

3.1. Historical Model Evaluation

The evaluation shows that the MRI-AGCM3.2-driven WRF simulations capture the seasonal rainfall cycle in both the Pampanga River and Pasig-Marikina-Laguna-Lake Basins, with a clear wet-season peak during June-July-August-September that matches observations. In both basins, the models reproduce the timing of monsoon onset and peak rainfall well, although the original (OR) simulations display some amplitude biases, especially during peak monsoon months. Bias correction (BC) improves agreement with observed monthly totals by reducing wet-season overestimation and dry-season bias. This improvement is more significant in the Pampanga River Basin, where BC simulations better match the observed amplitude and phase of the seasonal cycle. In the Pasig-Marikina-Laguna-Lake Basin, the corrected runs slightly underestimate peak monsoon rainfall but still maintain a realistic seasonal pattern. For rainfall extremes, the frequency-of-occurrence plots indicate that both downscaled GCMs represent the general shape of the observed distribution, including the rapid decrease in relative frequency at higher rain rates. However, the original simulations under-represent very high basin-mean rain rates (greater than 150–200 mm/day), revealing a limitation in capturing the upper tail. Bias correction partially adjusts the distribution and improves agreement with observed mid-to-high intensity bins, but some differences persist at the most extreme values. Overall, the simulations show credible skill in reproducing basin-scale rainfall climatology and daily rainfall distributions, supporting their use for climate change assessment. Remaining differences in the extreme tail highlight the need for explicit evaluation of heavy-rainfall statistics when interpreting projected changes in mean and extreme precipitation for both study basins.

3.2. Rainfall Climatology Trends

The dynamically downscaled and bias-corrected climate projections show clear spatial and temporal trends in rainfall climatology across the Pampanga River and Pasig-Marikina-Laguna-Lake Basins under future climate scenarios. Analyses used 5 km resolution outputs for 1979–2003 (historical baseline) and 2075–2100 (future), based on MRI-AGCM 3.2H and 3.2S under RCP 2.6 and RCP 8.5.

3.2.1. Basin-Averaged Monthly Climatology and Frequency of Heavy Rainfall Events (\(\geq 50\) mm/day)

Figure 3 shows that the wet season intensifies under future scenarios, especially under RCP 8.5. In both basins, rainfall peaks during the southwest monsoon period (May to October), with projected monthly totals higher and more concentrated in the middle of the season (June to August). Under RCP 2.6, seasonal patterns remain similar to the historical period, with only moderate increases in rainfall during the early monsoon months. These changes indicate a shift toward stronger monsoonal rainfall and potentially shorter, more intense wet seasons. The projected changes in monthly rainfall climatology are closely linked to the frequency of heavy rainfall events. As monthly rainfall increases, particularly during the southwest monsoon season (June to September), the number of days with precipitation exceeding 50 mm also rises. This pattern shows that the increase in monsoonal rainfall under RCP 8.5 results from more intense rainfall episodes rather than simply more frequent wet days. As a result, seasonal peaks in total rainfall align with months that have the highest counts of heavy-rain events, reinforcing the connection between changes in mean climate and extremes. Fig. 4 demonstrates this relationship, showing that months with the greatest rainfall increases under future scenarios also experience the largest rises in the frequency of heavy-rain events exceeding 50 mm/day. Changes in the monthly occurrence of heavy rainfall events (\(>50\) mm/day) consistently intensify during the southwest monsoon months, matching the increases in monthly mean rainfall shown in Fig. 3. For both basins, MRI-AGCM 3.2 simulations show that the frequency of heavy rainfall rises significantly from May to October under RCP 8.5, with the largest increases between June and August. Under RCP 2.6, projected changes are smaller and remain closer to historical levels, indicating that strong mitigation could limit the frequency of extreme rainfall.

In the Pampanga River Basin, the number of heavy-rain days increases throughout the wet season, peaking around July to September across all aggregation periods (1–4 days). The increase is most pronounced for multi-day events (3-day and 4-day totals), indicating a higher likelihood of prolonged heavy rainfall episodes and potential flooding in downstream areas.

In the Pasig-Marikina-Laguna-Lake Basin, basin-wide increases in heavy rainfall frequency are observed. From March to November, heavy rainfall events become more frequent, with the most significant increases occurring in June–September under RCP 8.5. The consistent rise across all aggregation durations indicates stronger clustering of extreme rainfall events during the monsoon period, aligning with projected increases in mean rainfall for the basin (Fig. 3). These findings indicate that climate change in both basins leads to greater seasonal rainfall totals and more frequent high-intensity events, which may increase flood hazards in the coming decades.

3.3. Changes in Extreme Rainfall Events

Analysis of extreme rainfall events shows a consistent increase in both intensity and frequency under future climate scenarios, especially under RCP 8.5. Key indices are daily rainfall thresholds (\(\geq 50\) mm/day), percentile-based extremes (95th and 99th percentiles), and design rainfall estimates for 1-in-5, 1-in-30, and 1-in-100-year return periods.

Spatial patterns of extreme rainfall intensity at the 95th and 99th percentiles (Fig. 5) and the frequency of heavy rainfall events exceeding 50 mm/day (Fig. 6) show consistent intensification under future climate scenarios. In both basins, MRI-AGCM 3.2 simulations indicate an overall increase in high-percentile rainfall, especially under RCP 8.5. The 95th-percentile maps show widespread increases across both basins, while the 99th-percentile fields reveal localized amplification in the lower Pampanga River Basin and the central to eastern areas of the Pasig-Marikina-Laguna-Lake Basin. These hotspots align with areas of stronger monsoon moisture convergence and orographic enhancement, indicating a physical connection between intensified monsoon dynamics and localized extremes.

The spatial analysis of heavy rain frequency in Fig. 6 supports this intensification. The annual mean and maximum number of days with rainfall greater than 50 mm per day both increase across the study area, with the largest rises under RCP 8.5. In the Pampanga River Basin, extreme-rain days become more frequent in the lower western catchment near the Manila Bay outlet, while the Pasig-Marikina-Laguna-Lake Basin shows a more uniform increase across its urban and eastern highland zones. Under RCP 2.6, increases are modest, indicating that mitigation can substantially limit the rise in heavy-rainfall frequency. These results show that climate change is likely to increase both the intensity and frequency of extreme rainfall events, supporting earlier findings of greater monsoon rainfall concentration and suggesting a compounded rise in flood hazard potential in both basins.

3.3.1. Changes in Annual Maximum Rainfall

The observed increase in both the frequency and spatial extent of heavy rainfall events leads to higher annual maximum precipitation. To quantify these changes and assess their implications for design applications, basin-averaged annual maximum rainfall for multiple accumulation durations (1–5 days) was analyzed, as shown in Fig. 7. This analysis directly measures how projected climate change affects the magnitude and recurrence of extreme rainfall events, linking the statistical behavior of observed extremes to design rainfall estimates under future scenarios.

Figure 7 shows the distributions and empirical cumulative distribution functions (CDFs) of annual maximum precipitation for both basins under historical and future scenarios. The boxplots indicate a consistent rise in median and upper-quartile rainfall across all aggregation periods under RCP 8.5, suggesting an overall increase in extreme rainfall events. The empirical CDFs display a rightward shift in the distribution, meaning that for any given non-exceedance probability, future rainfall totals are higher than those in the past. This shift becomes more pronounced with longer event durations, especially for 3-day and 4-day aggregations, indicating a greater likelihood of prolonged heavy rainfall episodes. Under RCP 2.6, the distributions show only modest or minimal changes, suggesting that lower-emission pathways can limit increases in extreme rainfall intensity.

3.4. Changes in Drought Indicators

To assess potential changes in drought characteristics, we analyzed two key indices: CDD and CWD, as summarized in Fig. 10. The projections show a clear intensification of dry conditions, especially under the high-emission RCP 8.5 scenario.

In the Pampanga River Basin, future climate scenarios indicate a consistent increase in maximum CDD throughout the basin, with the largest changes in the northern and central regions. Under RCP 8.5, several areas experience dry spells exceeding 5 consecutive days compared to the historical baseline. At the same time, CWD shows a declining trend, reflecting shorter wet periods and reduced persistence of rainfall events. These changes indicate a higher risk of agricultural droughts, especially during the transition months between wet and dry seasons.

In the Pasig-Marikina-Laguna-Lake Basin, similar patterns are observed. The central and eastern sectors show the greatest increases in CDD, particularly under MRI-AGCM 3.2S simulations. CWD decreases are less spatially uniform but remain common, especially in the lower catchment near Laguna Lake. These changes are critical for urban water supply and ecosystem services because they affect both runoff and groundwater recharge.

Overall, the projections indicate a dual hazard: more intense short-duration floods due to extreme rainfall and longer dry periods with potentially greater socio-economic consequences. The combined changes in wet and dry spells highlight the need for integrated flood-drought risk management frameworks in both basins, considering spatiotemporal variability and sector-specific vulnerabilities.

3.5. Comparison Across Climate Models

The use of two global climate models, MRI-AGCM 3.2H (60 km resolution) and MRI-AGCM 3.2S (20 km resolution), allowed a comparative assessment of projection uncertainty in flood and drought indicators. Both models generally agree on the direction of change under future climate scenarios, with projections indicating intensified rainfall extremes and longer dry spells. However, they differ in the magnitude, spatial distribution, and seasonal timing of projected effects.

MRI-AGCM 3.2S consistently projects higher frequencies of 50 mm/day rainfall events, especially in localized hotspots such as the southern Pampanga River Basin and eastern Metro Manila. This likely results from its finer spatial resolution, which better captures mesoscale convective processes, orographic enhancement, and localized storm systems. In contrast, MRI-AGCM 3.2H produces smoother spatial patterns and slightly lower estimates of design rainfall for high return periods (e.g., 1-in-100-year events).

For drought indicators, both models project increased maximum CDD, but MRI-AGCM 3.2S indicates more severe and spatially variable drought conditions, particularly in transitional zones between wet and dry climates. Reductions in CWD are generally more pronounced in MRI-AGCM 3.2H, although these differences are less significant than those observed for CDD.

Inter-model discrepancies highlight the need to consider model structural uncertainty when using downscaled climate projections for disaster risk assessments. High-resolution models such as 3.2S provide greater detail and local realism, which are important for urban flood planning, but they may also be more sensitive to initial conditions and internal variability. In contrast, coarser models such as 3.2H may under-represent extremes but offer smoother trends that are useful for long-term regional planning.

Using both models increases the study’s robustness by capturing a plausible range of climate futures and supplying essential information for risk-informed adaptation. This approach also demonstrates the importance of ensemble methods, where multiple GCMs and downscaling pathways are combined to better quantify uncertainty and support resilient infrastructure design.