Assessing the relative effectiveness of statistical downscaling and distribution mapping in reproducing rainfall statistics based on climate model results

Abstract

To improve the level skill of climate models (CMs) in reproducing the statistics of daily rainfall at a basin level, two types of statistical approaches have been suggested. One is statistical correction of CM rainfall outputs based on historical series of precipitation. The other, usually referred to as statistical rainfall downscaling, is the use of stochastic models to conditionally simulate rainfall series, based on large-scale atmospheric forcing from CMs. While promising, the latter approach attracted reduced attention in recent years, since the developed downscaling schemes involved complex weather identification procedures, while demonstrating limited success in reproducing several statistical features of rainfall. In a recent effort, Langousis and Kaleris (2014) developed a statistical framework for simulation of daily rainfall intensities conditional on upper-air variables, which is simpler to implement and more accurately reproduces several statistical properties of actual rainfall records. Here we study the relative performance of: (a) direct statistical correction of CM rainfall outputs using nonparametric distribution mapping, and (b) the statistical downscaling scheme of Langousis and Kaleris (2014), in reproducing the historical rainfall statistics, including rainfall extremes, at a regional level. This is done for an intermediate-sized catchment in Italy, i.e., the Flumendosa catchment, using rainfall and atmospheric data from four CMs of the ENSEMBLES project. The obtained results are promising, since the proposed downscaling scheme is more accurate and robust in reproducing a number of historical rainfall statistics, independent of the CM used and the characteristics of the calibration period. This is particularly the case for yearly rainfall maxima. To date, the assessment of the overall performance of different GCM/RCM combinations in modeling Earth's past, present and future climate has been greatly facilitated by open-access databases of climate model simulations developed through projects like PRUDENCE (Prediction of Regional scenarios and Uncertainties for Defining EuropeaN Climate change risks and Effects, http://prudence.dmi.dk/), PCMDI/CMIP3 (Program for Climate Model Diagnosis and Intercomparison/Coupled Model Intercomparison Project -Phase 3, http:// www-pcmdi.llnl.gov), ENSEMBLES (ENSEMBLE-based Predictions of Climate Changes and their Impacts,

A novel method to estimate the maximization ratio of the Probable Maximum Precipitation (PMP) using regional climate model output

“…Therefore, it was decided to use RCM output as the main source of data to carry out this work due to the richness and completeness of this database. On the other hand, the quality of reproducing rainfall by climate models is model dependent [ Langousis et al ., ]. Consequently, to ensure data‐independency of the method it was decided to use multiple CRCM data sets driven by different GCMs.…”

A parametric approach for simultaneous bias correction and high‐resolution downscaling of climate model rainfall

“…Apart from the presence of biases, CM rainfall products are also characterized by intrinsic uncertainties originating from multiple sources, such as the emission scenario and initial conditions used (usually referred to as GCM intra‐model variability), the length of the calibration period, the insufficient resolution of local topographic features and, more importantly, the GCM/RCM combination used [see Giorgi and Francisco , ; Pan et al ., ; Räisänen , ; Déqué et al ., ; Lucarini et al ., ; Gleckler et al ., ; Foley , ; Chen et al ., ; Bastola et al ., ; Sulis et al ., ; Deidda et al ., ; Hasson et al ., ; Mascaro et al ., ; Fatichi et al ., , among others]. The latter factor (also referred to as inter‐model variability) has been identified as the most critical source of uncertainty in hydrologic simulations [see e.g., Giorgi and Francisco , ; Déqué et al ., ; Lucarini et al ., ; Kjellström et al ., ; Chen et al ., ; Sulis et al ., ; Deidda et al ., ; Langousis et al ., ].…”

“…The limited success of CM rainfall products in reproducing rainfall statistics (i.e., rainfall occurrence, amount, and frequency of extremes) at a basin level and at hydrologically relevant temporal scales (e.g., daily), and the increased importance of accurate precipitation estimates in determining hydrological budgets, the availability of water resources in space and time and flood risks, led to the development of statistical correction approaches (usually referred to as bias correction procedures) [see e.g., Themeßl et al ., ; Langousis et al ., , and references therein]. These are based on linear or more complex transformations (delta change method, power transformations, distribution mapping, etc.)…”

“…In its simplest form, distribution mapping is conducted by matching the quantiles of the empirical distributions obtained directly from the historical and CM data (also referred to as nonparametric distribution mapping) [see e.g., Michelangeli et al ., ; Stoll et al ., ; Themeßl et al ., ; Lafon et al ., ; Gutjahr and Heinemann , ], and remains sensitive to the length of the calibration period, the range of observed rainrates, and the presence of outliers [see e.g., Themeßl et al ., ; Lafon et al ., ; Langousis et al ., ]. Another drawback of the nonparametric Q‐Q correction regards regional frequency analysis of extreme rainfall, as one needs to develop separate maps of rainfall quantiles (also referred to as isohyetal maps) for each exceedance probability level (or equivalently return period) of practical interest.…”

Development of a time‐varying downscaling model considering non‐stationarity using a Bayesian approach

“…There are different methods available for the downscaling of precipitation and these have been used at various places across the world (Hewitson and Crane, 1996;Hellström et al, 2001;Fowler et al, 2007;Bordoy and Burlando, 2014). These methods can be divided into dynamical downscaling (Hewitson and Crane, 1996;Schmidli et al, 2007;Xue et al, 2014;Manor and Berkovic, 2015) and statistical downscaling (Wilby et al, 2002;Hessami et al, 2008;Langousis et al, 2015). Dynamical downscaling utilizes a regional climate model (RCM) and is based on mathematical conceptualization of physical processes (Rotach et al, 1997;Fowler et al, 2007;Laprise, 2008).…”