Climate Change Projections of Temperature and Precipitation in Chile Based on Statistical Downscaling
Total Page:16
File Type:pdf, Size:1020Kb
Climate Dynamics (2020) 54:4309–4330 https://doi.org/10.1007/s00382-020-05231-4 Climate change projections of temperature and precipitation in Chile based on statistical downscaling Daniela Araya‑Osses1,2 · Ana Casanueva3,4 · Celián Román‑Figueroa2,5 · Juan Manuel Uribe1 · Manuel Paneque1 Received: 7 August 2019 / Accepted: 5 April 2020 / Published online: 9 April 2020 © The Author(s) 2020 Abstract General circulation models (GCMs) allow the analysis of potential changes in the climate system under diferent emis- sions scenarios. However, their spatial resolution is too coarse to produce useful climate information for impact/adaptation assessments. This is especially relevant for regions with complex orography and coastlines, such as in Chile. Downscaling techniques attempt to reduce the gap between global and regional/local scales; for instance, statistical downscaling methods establish empirical relationships between large-scale predictors and local predictands. Here, statistical downscaling was employed to generate climate change projections of daily maximum/minimum temperatures and precipitation in more than 400 locations in Chile using the analog method, which identifes the most similar or analog day based on similarities of large- scale patterns from a pool of historical records. A cross-validation framework was applied using diferent sets of potential predictors from the NCEP/NCAR reanalysis following the perfect prognosis approach. The best-performing set was used to downscale six diferent CMIP5 GCMs (forced by three representative concentration pathways, RCPs). As a result, minimum and maximum temperatures are projected to increase in the entire Chilean territory throughout all seasons. Specifcally, the minimum (maximum) temperature is projected to increase by more than 2 °C (6 °C) under the RCP8.5 scenario in the austral winter by the end of the twenty-frst century. Precipitation changes exhibit a larger spatial variability. By the end of the twenty-frst century, a winter precipitation decrease exceeding 40% is projected under RCP8.5 in the central-southern zone, while an increase of over 60% is projected in the northern Andes. Keywords Statistical downscaling · Predictors · Climate change · GCMs · Temperature · Precipitation 1 Introduction Many studies have demonstrated that climate change afects Electronic supplementary material The online version of this article (https ://doi.org/10.1007/s0038 2-020-05231 -4) contains the development of countries and impacts the environment supplementary material, which is available to authorized users. irreversibly. One consequence of climate change is the modifcation of the statistical distribution of atmospheric * Manuel Paneque patterns (Sanabria et al. 2009). This could largely afect the [email protected] world’s population, as increasing temperatures and precipita- 1 Facultad de Ciencias Agronómicas, Universidad de Chile, tion fuctuations might have an efect on water availability Santa Rosa 11315, 8820808 La Pintana, Santiago, Chile and crop production in the future (Kang et al. 2009). Thus, 2 Bionostra Chile Research Foundation, Almirante Lynch knowledge of the current and projected future global climate 1179, 8920033 San Miguel, Santiago, Chile conditions is fundamental for the determination of vulner- 3 Federal Ofce of Meteorology and Climatology, MeteoSwiss, abilities and the development of climate change adaptation Zurich Airport, 8058 Zürich, Switzerland strategies (Magaña et al. 2000). 4 Meteorology Group, Department of Applied Mathematics General circulation models (GCMs) are the most reliable and Computer Sciences, University of Cantabria, tools to assess climate evolution under diferent anthropo- 39005 Santander, Spain genic forcings such as greenhouse gas emissions, and thus 5 Doctoral Program in Sciences of Natural Resources, provide an estimation of possible future climates based on Universidad de La Frontera, Av. Francisco Salazar 01145, socio-economic and demographic factors corresponding to 4811230 Temuco, Chile Vol.:(0123456789)1 3 4310 D. Araya-Osses et al. diferent emissions scenarios (Cabré 2011). GCMs are based and vegetation coverage (Breshears et al. 2005). Chile is on physical principles that numerically describe the climate a very diverse country from an orographic and physical system and can reproduce the observed characteristics of standpoint. Latitude, altitude, and the infuence of the Pacifc the current climate and past climate changes (Solomon et al. Ocean and continentality are among the major climate driv- 2007). However, the low spatial resolution of these global ers in the region, as along with the infuence of the Ama- models (i.e., 150–300 km) limits their direct use in regional zon in the northern zone, westerly winds in the southern or local impact models, which require higher resolution zone, and the Humboldt current system, which is driven (Amador and Alfaro 2009). by the South Pacifc Subtropical high-pressure cell (Sarri- Statistical downscaling (SD) and dynamical downscal- colea 2017a; Gutiérrez et al. 2016). Thus, the application of ing (DD) are two commonly used approaches to bridge the downscaling methods is particularly challenging in Chile gap between coarse GCMs and local impacts (Amador and due to its large regional spatio-temporal variabilities, which Alfaro 2009). DD is often performed through regional cli- are evident not only regarding temperature and precipitation mate models (RCMs), which solve the governing equations but also for region-specifc climate impacts. of the atmosphere in a limited spatial domain, subject to Downscaling techniques have been applied in Chile for initial conditions from GCMs (or reanalysis), on a higher the determination of future climate based on the scenarios resolution than the driving GCMs (e.g., 12–50 km). They of the Fourth Assessment Report of the Intergovernmental are computationally expensive and include certain param- Panel on Climate Change (Solomon et al. 2007) through eterizations to represent the processes occurring at a higher the application of (1) DD for the entire territory (Fuenza- resolution than the grid space (Amador and Alfaro 2009). lida 2007; Garreaud 2011; Santibañez 2014) and (2) SD for SD techniques establish empirical relationships between specifc regions (Souvignet et al. 2010; Fiebig-Wittmaack predictors (large-scale variables) and predictands (local et al. 2012). Souvignet et al. (2010) used multiple linear variables; Zorita and von Storch 1999; Amador and Alfaro regression with the SDSM (Statistical DownScaling Model) 2009; Gutiérrez et al. 2013), featuring the advantages of low software for the HadCM3 model and A2a and B2a emis- computation requirements and providing information with sions scenarios, considering diferent predictors for each the same spatial resolution as the observational input data, meteorological station in the Upper-Elqui watershed in that is, as high-resolution grids or local points (Ruiz 2007). Coquimbo. Fiebig-Wittmaack et al. (2012) employed a sto- However, statistical downscaling assumes that the empiri- chastic weather generator in the Elqui river basin based on cal relationships deduced for the present climate are valid the CGCM3 model for SRES A2 and B2 scenarios (Long for the future climate (stationarity assumption), requires Ashton Research Station Weather Generator). To our knowl- sufciently long and high-quality observational series, and edge, SD has been used to produce climate change projec- provides results only for variables of which one has observed tions at local or basin levels only, whereas country-level data (Gaertner et al. 2012). Moreover, the set of predictors applications have not been attempted. Given the need for that best explains the objective variable need to be optimized high-resolution projections to establish a baseline for sub- for each predictand, location, season, etc. (Casanueva et al. sequent impact-oriented studies, this study aimed to develop 2016). SD could be classifed by technique (Wilby and Wig- climate change projections of minimum/maximum tempera- ley 1997) into regression method, weather type approaches ture and precipitation in Chile based on statistical downscal- and stochastic weather generators, and also by approach ing under a perfect prognosis approach of diferent GCMs (Rummukainen 1997) into Perfect prognosis (PP) and Model under RCP2.6, RCP4.5, and RCP8.5 emissions scenarios. Output Statistics (MOS). Regression models and weather The paper is organized as follows. Material and methods type approaches establish a relationship between observed are described in Sect. 2 and results are presented in Sect. 3. large-scale predictors and observed local-scale predictands Section 4 includes a discussion of the results and Sect. 5 (Maraun et al. 2010). The perfect prognosis approach is the summarizes the main conclusions. application of this relationship to predictors from GCM in a weather forecasting context (Maraun et al. 2010). MOS contrary to PP calibrates the statistical relationship between 2 Materials and methods predictor and predictand by using simulated predictors and observed predictand (Maraun et al. 2010). 2.1 Study area and datasets Climate change projections over Chile can provide essen- tial information for the establishment of mitigation and 2.1.1 Study area and Chilean climate adaptation policies. In particular, high-resolution projections enable the assessment of possible risks and vulnerabilities This study was carried out in continental Chile, which is related to specifc