Linking climate change to urban storm drainage system design: An innovative approach to modeling of extreme rainfall processes over different spatial and temporal scales

Highlights

• An original spatio-temporal statistical downscaling procedure was proposed.

• An original procedure for modeling extreme rainfall processes over a wide range of spatial and temporal scales.

• It is feasible to establish the linkages between GCM climate projections and rainfall extremes at a local site.

• A robust procedure for accurate assessment of climate change impacts on local rainfall extremes.

• An illustrative application using 21 GCMs and extreme rainfall data for Ontario region was presented.

Abstract

The main challenge in the assessment of climate change impacts on the estimation of extreme rainfalls (ERs) for urban drainage systems design is how to establish the linkages between the climate projections given by Global Climate Models (GCMs) at global scales and the observed extreme rainfalls at a given local site. Downscaling approaches have been proposed in many previous studies to downscale global-scale GCM daily information to regional-scale daily climate projections. However, these daily downscaled data are still considered too coarse in both spatial and temporal resolutions and hence not suitable for climate change impact studies at a local site or for small urban watersheds. The present study proposes therefore an innovative statistical downscaling (SD) approach for establishing the linkage between daily extreme rainfalls at regional scales and daily and sub-daily extreme rainfalls at a local (point) scale. The feasibility and accuracy of the proposed method were assessed for a case study in Ontario (Canada) using observed ER data from seven raingauges and climate simulation outputs from 21 GCMs that have been downscaled by NASA to a regional 25-km scale for the RCP 4.5 scenario. Results based on various graphical and numerical comparison criteria have indicated the feasibility and accuracy of the proposed SD approach. In addition, a robust assessment of the climate change impacts on the ERs for urban drainage system design was performed using a series of statistical tests in sequence to evaluate the significant changes of rainfalls among different time periods. It was found that significant increases by 8% to 18% in extreme design rainfalls of return periods up to T = 25 years, and insignificant increases by 3% to 8% in the 50-year and 100-year design rainfalls for many locations, except for one location with a significant increase of 18%. The confidence intervals were also computed for these estimated design rainfalls with widths varying from 5% to 22%.

Introduction

The design of urban storm drainage systems requires a “design storm” that represents the time distribution of rainfalls within an extreme storm event. More specifically, in engineering practice, the design storm of a specified exceedance probability and duration is commonly estimated from the extreme rainfall intensity-durationfrequency (IDF) relations (Hershfield, 1961, Chow, 1964, WMO, 2009). However, in recent years, climate change has been recognized as having a profound impact on the hydrologic cycle at different temporal and spatial scales. The temporal scales could vary from a very short time interval of a few minutes (for urban water cycle) to a yearly time scale (for annual water balance computation). The spatial resolutions could be from a few square kilometers (for urban and rural watersheds) to several thousand square kilometers (for large river basins). In particular, the intensity and frequency of extreme precipitation events in most regions will be likely increased in the future (Alexander et al., 2006, Lenderink and van Meijgaard, 2008, Kharin et al., 2013, Shephard et al., 2014, Zhang et al., 2017). Hence, there exists an urgent need to assess the possible impacts of climate variability and climate change on the IDF relations in general and on the design storm in particular for improving the design of urban drainage systems in the context of a changing climate (Willems et al., 2012, CSA, 2012, Madsen et al., 2014, Simonovic et al., 2016).

To achieve this, the projected annual maximum rainfall series for different rainfall durations from short time scales (e.g., a few minutes) to long time intervals (e.g., one day or longer) under different possible climate change scenarios are required. Consequently, global climate models (GCMs) have been extensively used currently to provide projected precipitations for future periods (e.g., next hundred years) based on different Representative Concentration Pathways (RCP) scenarios. These models have been recognized to be able to represent reasonably well the main features of the global distribution of basic climate parameters (Lambert and Boer, 2001), but could not reproduce well details of regional climate conditions at temporal and spatial scales of relevance to hydrological impact studies (Nguyen and Nguyen, 2008). This is because outputs from GCMs are usually at resolutions that are too coarse (generally greater than 200 km and mostly available at a daily time scale) and are not suitable for many climate change impact studies in urban areas. A new rainfall modeling approach is thus needed to establish an accurate linkage between climate projections from GCMs and extreme rainfall (ER) processes at a local site of interest.

To refine the GCM coarse grid resolution climate projection data to much finer spatial resolutions (regional or local scales) for reliable assessment of climate change impacts, different downscaling methods have been proposed (Wilby et al., 2002, Fowler et al., 2007, Maraun et al., 2010, Gooré Bi et al., 2017). These methods can be classified into two broad categories: dynamical downscaling (DD) and statistical downscaling (SD). The DD techniques involve the extraction of regional scale information from large-scale GCM data based on the modeling of regional climate dynamical processes (Laprise, 2008, Xue et al., 2014, Xu et al., 2019). These models use physical principles to reproduce local climates, thus, are comprehensive physical models, but are computationally intensive. Whereas the SD techniques rely on the empirical relationships between observed (or analyzed) large-scale atmospheric variables and observed (or analyzed) surface environment parameters (Wilby et al., 2002, Nguyen and Nguyen, 2008, Bürger et al., 2012, Bürger et al., 2013, Werner and Cannon, 2016). The SD methods are thus flexible to adapt to specific study purposes and inexpensive computing resource requirement (Nguyen and Nguyen, 2007). Both downscaling methods therefore have their strengths and limitations (Wilby et al., 2002, Teutschbein et al., 2011), and both are currently used for downscaling the GCM outputs to the regional scales (approximately 10 to 50 km resolutions) for climate-related studies. There are many different downscaled climate projection data sources available at the national or global level provided by different organizations. For example, in North America, just to list a few, the NA-CORDEX dataset available at roughly 25 or 50 km resolution covering most of North America (NA-CORDEX, 2018), NEX-GDDP dataset available at about 25 km resolution covering the entire globe (NEX-GDDP, 2018), the NEX-DCP30 dataset available at approximately 1 km resolution for the conterminous United States (NEX-DCP30, 2018), PCIC statistically downscaled climate scenarios available at roughly 10 km for Canada (PCIC, 2018). However, downscaled data from these methods are still considered bias when comparing to observed data at a local site in a same grid cell and need to be bias-corrected. A bias-correction method is therefore required to correct the data before it can be used for impact assessments and adaptation studies (Willems et al., 2012).

In addition to the spatial downscaling, the temporal downscaling is also required to derive the distributions of sub-daily extreme rainfalls from that of the daily values since the climate projections are normally available at the daily scale. Several approaches have been developed in the literature, such as the chaotic method, the scale-invariance approach, the point-process model, the neural networks techniques (Sivakumar et al., 2001, Nguyen et al., 2002, Coulibaly et al., 2005, Marani and Zanetti, 2007, Nguyen et al., 2007, Lee and Jeong, 2014, Herath et al., 2016). Among these methods, the scale-invariance (or scaling) concept has increasingly become a new methodology in the analysis and modeling of various extreme hydrological processes across a wide range of temporal scales (Gupta and Waymire, 1990, Sposito, 1998, Hubert, 2001, Veneziano and Furcolo, 2002, Veneziano and Lepore, 2012, Lovejoy and Schertzer, 2012). Scale invariance implies that the distributions and statistical properties of ERs over different time scales are related to each other by an operator involving only the scale ratio and the scaling exponent (Gupta and Waymire, 1990). In particular, the scaling method has been used in the construction of the IDF relations for the current and future climates (Burlando and Rosso, 1996, Nguyen et al., 2002, Yu et al., 2004, Bougadis and Adamowski, 2006, Nguyen et al., 2007, Blanchet et al., 2016, Ghanmi et al., 2016). More specifically, several scaling models have been proposed in the literature, such as the scaling Gumbel (GUM) model based on the non-central moments and the probability weighted moments (Menabde et al., 1999, Yu et al., 2004). The scaling Generalized Extreme Values (GEV) model based on the non-central moments (Nguyen et al., 1998). More recently, a novel scaling probability weighted moments-based GEV model has been shown to outperform other existing scaling models (Nguyen and Nguyen, 2018).

In view of the above issues, the present paper proposes an innovative SD approach that is able to establish the linkage between climate change information to extreme rainfall estimation for urban storm drainage systems design; a difficult and challenging task in current engineering practices. This SD approach was based on a new procedure for modeling the ER processes over different spatial and temporal scales. The feasibility and accuracy of the proposed approach were assessed for a case study in Ontario (Canada) using the IDF data from a network of seven raingauges. These data are provided in Sections 2. Details of the proposed methodology are described in Section 3. Results are presented and discussed further in Section 4. Finally, a summary of the research findings is provided in Section 5.