Assessment of RegCM4 simulated inter-annual variability and daily-scale statistics of temperature and precipitation over Mexico
- First Online:
- 2.4k Downloads
The skill of a regional climate model (RegCM4) in capturing the mean patterns, interannual variability and extreme statistics of daily-scale temperature and precipitation events over Mexico is assessed through a comparison of observations and a 27-year long simulation driven by reanalyses of observations covering the Central America CORDEX domain. The analysis also includes the simulation of tropical cyclones. It is found that RegCM4 reproduces adequately the mean spatial patterns of seasonal precipitation and temperature, along with the associated interannual variability characteristics. The main model bias is an overestimation of precipitation in mountainous regions. The 5 and 95 percentiles of daily temperature, as well as the maximum dry spell length are realistically simulated. The simulated distribution of precipitation events as well as the 95 percentile of precipitation shows a wet bias in topographically complex regions. Based on a simple detection method, the model produces realistic tropical cyclone distributions even at its relatively coarse resolution (dx = 50 km), although the number of cyclone days is underestimated over the Pacific and somewhat overestimated over the Atlantic and Caribbean basins. Overall, it is assessed that the performance of RegCM4 over Mexico is of sufficient quality to study not only mean precipitation and temperature patterns, but also higher order climate statistics.
KeywordsInterannual variability Extremes Simulation Assessment
Daily scale temperature and precipitation extreme events are important due to their potential to cause life and economic losses (see, for example, Gosling et al. 2007; Poumadere et al. 2005; Brody et al. 2007), as well as damage to natural ecosystems (Easterling et al. 2000; Garrabou et al. 2009).
Central America has been identified as one of the most prominent climate change “hot spots” in the tropics (Giorgi 2006). In particular, Mexico is highly exposed and vulnerable to climate variability and tropical storms. De Alba and Andrade (2009) reported that between 1970 and 2006, about three hurricanes every 2 years reached the Mexican territory causing heavy damages and losses. For example in 2005, the Mexican Association of Insurance Institutions (AMIS) estimated that the economic losses due to hurricanes Emily, Stan and Wilma were about 2,282 million USD.
Previous studies examined different aspects of variability and extreme events over Mexico (Cavazos 1999; Cavazos and Rivas 2004). For instance, Cavazos (1999) studied the large-scale conditions associated with extreme precipitation events over Northeastern Mexico and Southeastern United States. A similar study for the city of Tijuana, Mexico, was carried out more recently (Cavazos and Rivas 2004). García-Cueto et al. (2010) investigated the duration and intensity of heat waves in the city of Mexicali (northern Mexico), and found that in 2006 there were 2.3 times more heat waves than in the decade of the 1970s. Arriaga-Ramírez and Cavazos (2010) calculated regional trends in annual and seasonal daily precipitation in Northwestern Mexico and Southwestern United States and found positive annual trends over the region.
Some studies assessed extreme events in climate change projections over some regions of Mexico and North America. For example, Diffenbaugh et al. (2008) found that regions of Northwestern Mexico and Southwestern United States might be highly vulnerable to climate change, mostly because of increases in variability of precipitation and temperature extremes. In fact, the impacts of climate change will be felt most strongly through changes in hydroclimatic variability and extremes rather than in mean values (IPCC 2007), particularly due to an increase in the intensity of storms together with a decrease in the frequency of rainy days (Trenberth 1999; Giorgi et al. 2011).
Both global and regional climate models (RCMs) can be used to simulate the changes in daily climate statistics in response to global warming, but their performance needs to be assessed prior to the production of climate projections. In particular, RCMs have been shown to provide reasonable representations of the tails of the daily-scale temperature and precipitation distributions in different contexts (e.g. Walker and Diffenbaugh 2009; Diffenbaugh et al. 2005). However, to date an analysis of variability and extremes in RCM simulations for Mexico has not been carried out. Therefore, in this paper we evaluate the skill of an RCM (RegCM4 described by Giorgi et al. 2012) in capturing mean patterns, variability and higher order statistics of daily-scale temperature and precipitation events over Mexico, and attempt to explain the causes of eventual mismatches between observations and simulation. We focus on interannual variability and on the annual 5 and 95 percentiles of daily temperature, 95 percentile of precipitation, frequency of occurrence of events of given intensities, maximum dry spell length and number of tropical cyclone days, quantities that are all important for impact studies. The aim of this assessment is to test the performance of RegCM4 as a tool for the study of climate variability and extremes over Mexico and their change under global warming conditions.
2.1 RegCM4 configuration
We use the hydrostatic, compressible, three-dimensional, regional climate model RegCM4 (Giorgi et al. 2012). This is the latest version of the modeling system originally developed by Giorgi et al. (1993a, b) and Pal et al. (2007). Several physics parameterizations are available in the model. For the present study, we employ the best performing configuration identified by Diro et al. (2012), which includes an enhanced radiative transfer scheme (Kiehl et al. 1996, Giorgi et al. 2012), a modified version of the planetary boundary layer scheme of Holtslag et al. (1990) (see Giorgi et al. 2012), a mixed convection scheme in which the parameterization of Emanuel (1991) is used over oceans and the scheme of Grell (1993) over land and the resolvable scale precipitation scheme of Pal et al. (2000). For the representation of land surface processes the model employs the Biosphere–Atmosphere Transfer Scheme (BATS) of Dickinson et al. (1993).
2.2 Precipitation, temperature and tropical storm observations
We use various datasets to evaluate the model performance. When assessing the model over the whole domain, we refer to precipitation data from the monthly datasets of the Global Precipitation Climatology Project (GPCP, Adler et al. 2003), and surface air temperature data from the University of Delaware (UDel, Legates and Willmott 1990a, b). The resolution of these datasets (GPCP and UDel) are 2.5° and 0.5°, respectively. For a more focused assessment, we analyze precipitation over seven sub-regions of Mexico: Baja California (BC), North-West Mexico (NW), Central-North Mexico (NM), North-East Mexico (NE), Central Mexico (CM), South Mexico (SM) and the Yucatan Peninsula (YP). Furthermore, we compare RegCM4 simulated precipitation and temperature statistics with Mexican daily observations from the CLICOM data set (Zhu and Lettenmaier 2007; Munoz-Arriola et al. 2009), which has a 0.12° resolution. Finally, we assess simulated tropical cyclone days against observations from the HURDAT best track database of the US National Weather Service National Hurricane Center (downloaded from the US National Weather Service National Hurricane Center online site: http://www.nhc.noaa.gov/pastall.shtml). Simulated wind fields are compared with the 1.5° resolution ERA-Interim (Dee et al. 2011) dataset.
2.3 Statistical metrics of precipitation and temperature performance
Interannual variability is measured in terms of interannual standard deviation of seasonal values for temperature and interannual coefficient of variation (standard deviation divided by the mean) for precipitation. In terms of extremes, the metrics considered here (all calculated for the period 1982–2008) are: Annual 5 and 95 percentiles of daily temperature (T05 and T95), annual 95 percentile of daily precipitation (P95) (Diffenbaugh et al. 2005), and maximum dry spell length (MDSL) defined as the maximum number of consecutive days in a year with precipitation <1 mm. Finally, in order to assess the RegCM4 performance in reproducing the distribution of precipitation intensity, we compare observed and simulated frequency of events exceeding specific precipitation thresholds.
We also compare the simulated and observed spatial distribution of tropical cyclones. Several algorithms have been developed to detect cyclones from reanalysis of observations or from atmospheric models (see for example Manabe et al. 1970; Bengtsson et al. 1982; Broccoli and Manabe 1990; Haarsma et al. 1993; Bengtsson et al. 1995; Tsutsui and Kasahara 1996; Vitart et al. 1997; Walsh 1997; Vitart and Stockdale 2001, Walsh et al. 2004). These methods in general monitor when some chosen dynamical and thermodynamical variables, for example sea level pressure, vorticity, wind speed or precipitation, exceed thresholds determined from observed tropical storm characteristics. These studies have shown that models can reproduce some aspects of storm climatologies, such as geographical and temporal distributions. However, it has been reported that the intensity of simulated tropical storms is weaker, and their spatial scale larger than observed because of low model resolution (Bengtsson et al. 1982; Vitart et al. 1997). Furthermore, threshold criteria taken from observed climatological values do not account for model biases and deficiencies. Camargo and Zebiak (2002) using their own detection algorithm demonstrated that the use of basin- and model-dependent threshold criteria improves the climatology and interannual statistics of model tropical cyclones. There is thus an element of customization in the selection of criteria for the identification of tropical storms in climate models.
Based on these considerations and on the characteristics of the simulated cyclones by RegCM4 we selected a series of variables that allowed us to identify most tropical cyclones generated in the domain. Specifically, we define a cyclone-day for a particular location as a day in which a tropical cyclone passes through a grid-point of our domain. Tropical cyclone conditions are identified when, at a given grid point, the following conditions are satisfied at least once during the day: wind speed ≥21 m s−1, sea level pressure ≤1,005 hPa, daily precipitation rate ≥15 mm day−1. These days are then summed over the 27 year simulation to obtain the total number of cyclone days. The same variable thresholds are used to identify observed tropical storm data. In order to compare spatially observed and simulated tropical cyclones, we gridded the 6 hourly available observed HURDAT cyclone track information onto a 1° mesh covering the simulation domain. Following Weatherford and Gray (1988) the inner core of a cyclone extends from the center of the cyclone for a 111 km radius, so that, once we identify the center of the cyclone at a given grid point we consider the 8 surrounding 1-degree grid points to be part of the cyclone core. Although this procedure does not employ a formal cyclone detection and tracking algorithm, extensive visual tests showed that it is a simple way to identify most tropical depressions generated in the domain (see Sect. 3.5).
3 Results and discussion
3.1 Seasonal precipitation and temperature means over the Central-America CORDEX domain for the 1982–2008 period
The RegCM4 simulation captures adequately the patterns of seasonal precipitation over the Central America region (see Fig. 3). In DJF, a slight northeastward shift of the simulated Inter-Tropical Convergence Zone (ITCZ) over the eastern Pacific Ocean (see Fig. 3e, f) generates a wet bias over the western Mexican and Central-American coasts and a dry bias over the western South American coast (Fig. 3). In addition, an atypical northward wind is observed over northern South America (compared to the ERA-Interim reanalysis), causing a wet bias in this region and a dry bias over most of Brazil (Fig. 3e). For JJA, the northward excursion of the ITCZ is not well resolved by RegCM4, and thus we find a negative bias of precipitation in the Tropical North-eastern Pacific Ocean west of the Mexican coast. This may be associated with an anti-cyclonic circulation anomaly formed over this region (Fig. 3f), which moves the area of convergence further south than observed. In the case of the Atlantic Ocean north of Brazil, a northward wind anomaly may be partially causing a wet bias over this region.
For a more quantitative precipitation assessment, in Fig. 6 we show the annual cycle of observed (GPCP, CLICOM), ERA-Interim reanalysis and RegCM4 simulated precipitation for all sub-regions of Mexico (see Fig. 5). We find that RegCM4 is able to capture the annual cycle of precipitation over all regions, including regions of single and double rainy seasons separated by a summer dry period. Precipitation amounts are however overestimated over the SM, NE and CM regions and underestimated over the BC, NW and YP regions, while they are well reproduced over the NM region. We also note the good agreement between the CLICOM, GPCP and ERA-Interim datasets, although ERA-Interim gives the lowest precipitation amounts especially during the summer over the NW, CM and YP regions.
The underestimation of precipitation over Baja California (BC) during winter (see Fig. 5) may be related to the circulation biases, since the simulation shows a northward anomalous wind circulation (see Fig. 3) that may inhibit advection of cold and moist masses over the BC. Conversely, the precipitation underestimation over the NW region may be due to the weakened land-sea thermal contrast in the simulation during the rainy season, which in turn may induce a weakening of the North American Monsoon (Turrent and Cavazos 2009). Figure 4 shows a cold bias in JJA that may cause a weakening of the thermal low that forces the advection of moist air from the Pacific Ocean. The overestimation of precipitation over the mountainous areas of the NE, CM and SM regions is likely tied to the land surface and convection schemes used. For example, Diro et al. (2012) show that the use of BATS tends to produce relatively high precipitation amounts and that the simulation of precipitation over mountainous Central America is generally sensitive to the land and ocean flux surface physics schemes used in the model.
3.2 Interannual variability
The second and third columns show the correlation (r) of CLICOM observations with ERA-Interim (ERA) and RegCM4 simulation for temperature (T) and precipitation (P) over each of the seven analyzed sub-regions shown in Fig. 5
r (ERA, CLICOM)
We find a different behavior in the northern versus the central and southern Mexican regions for precipitation variability (see Fig. 10). In the northern regions (BC, NM, and NW) the model and ERA-Interim reproduce well the CLICOM data in terms of both variability metric and correlation. By contrast the correlations are low and the interannual variability somewhat underestimated in the other regions, where however the anomalous year 2003 heavily affects the CLICOM variability value (see Table 1). Overall, if we do not consider the large variations occurred in the recent record over some regions, the model generally reproduces the observed interannual variability.
3.3 Daily precipitation intensity, P95 and MDSL over Mexico
3.4 T95 and T05 over Mexico
Similarly, the spatial pattern of T05 in RegCM4 is similar to that in CLICOM (see Fig. 14) with the coldest mean T05 values occurring in the northwestern mountains. The warmest T05 values (>15 °C) are consistent in CLICOM and RegCM4 and occur over the southern and southeastern states. The warm T05 bias in RegCM4 is partly due to its relatively coarse resolution over complex mountain areas, but it can also be affected by a northwestward wind anomaly during DJF in the Pacific Ocean, west of the Mexican coast (see Fig. 3), whereby cold air masses crossing Mexico from the northwest may not be adequately reproduced in the simulation.
3.5 Number of tropical cyclone days
Concerning the cyclone concentration, RegCM4 captures the area of large NCD density in the Equatorial Eastern Pacific, with a maximum of around 250 hurricane days within all the 1982–2008 period, but this area is smaller in RegCM4 than observed, indicating an underestimate of cyclone days. We analyzed two possible sources of this bias: surface wind circulation and vertical wind shear. Regarding the former, during summer the simulation shows an atypical atmospheric anticyclone at low levels compared with the ERA-Interim reanalysis, which may be causing humidity divergence and inhibiting cyclogenesis. The RegCM4 underestimation of the NCD in the Equatorial Pacific Ocean is also reflected in the dry bias over this region (see Fig. 3).
On the other hand, over the Atlantic Ocean RegCM4 appears to produce more cyclone days than observed, which may be related to a weaker vertical wind shear over the Atlantic Ocean in the simulation (Fig. 17). It is worth to notice that the overestimation of the NCD in the Intra-American Seas and the Gulf of Mexico seems to generate more land-falling tropical cyclones over the Yucatan peninsula and Eastern Mexico. We should mention that mismatches between observations and simulation may be due to differing numbers or duration of events, but our analysis cannot separate these two factors.
Some uncertainties in the comparison of simulated and observed cyclones are obviously due to our detection method. We experimented with different pressure, precipitation and wind threshold criteria and even though the calculated NCD varied, the general conclusions deriving from Figs. 16 and 17 were not substantially modified.
We evaluated the skill of the regional climate model RegCM4 in reproducing different statistics of the climate of Central America and more specifically Mexico, as a preliminary step before using this model for climate change projections over this region. The model was run using the Central America CORDEX domain specifications and lateral boundary conditions from ERA-Interim reanalysis for the period 1982–2008. We analyzed both climate means and higher order statistics relevant to impacts over the region, such as interannual variability and extremes. Furthermore, we analyzed the model performance in simulating statistics of tropical cyclones.
The model generally reproduced the spatial patterns of mean temperature and precipitation, as well as their interannual variability and extremes. The main deficiencies of the model were found over mountainous terrain, where precipitation was overestimated mainly due to the high frequency of low-precipitation events. The biases are related to anomalies in the reproduction of circulation patterns, such as shifts in the ITCZ position with its consequent effect on humidity advection, along with possible shortcomings in the parameterization of convection over high mountains. Uncertainties in observations associated with sparse station density as well as lack of under-catch gauge correction most likely add uncertainty in the evaluation of the model.
On the other hand, the RegCM4 reproduced well the phase of the mean annual cycle of precipitation for all the sub-regions of Mexico, capturing in particular the mid-summer drought over SM and YP regions. The model was also successful in reproducing the observed characteristics of interannual variability, particularly in the northern regions of Mexico. The spatial patterns of extreme statistics such as P95, T05 and T95 and MDSL were also captured. Although local biases were found, the comparison of observed and simulated precipitation intensity distribution showed a good agreement, with the exception of an overestimation of light precipitation events. The latter is a common problem in climate models related to the relatively coarse model resolution.
Concerning the occurrence of tropical cyclones, despite the relatively coarse model resolution and the simplicity of our detection method, we found that consistently with observations the model did generate realistic regions of tropical cyclone occurrence. The cyclone density was however underestimated in the Equatorial Pacific and overestimated in the Atlantic and Caribbean regions. Nevertheless, despite these biases, the model showed an encouraging performance in simulating tropical cyclones, which calls for detailed examination of this topic with improved detection and tracking algorithms.
Our analyses show that, in terms of climatological means as well as higher order statistics, RegCM4 is capable of providing realistic representations of Mexico and Central America climate. Therefore, in upcoming contributions, we shall apply this model to a series of climate-change projections over this region as part of the CORDEX program.
We thank Graziano Giuliani and Gulilat Tefera Diro for useful discussions and two anonymous reviewers for helpful suggestions. The Mexican federal agency CONACYT sponsored the first author through a PhD graduate-studies scholarship as well as CICESE research.
- Arriaga-Ramírez S, Cavazos MT (2010) Regional trends of daily precipitation indices in northwest Mexico and southwest United States. J Geophys Res 115:1–10Google Scholar
- De Alba E, Andrade R (2009) Evaluating the impact of the increase in hurricane frequency using an internal model, a simulation analysis. Preprints, 39th ASTIN Colloquium, Helsinki, FI, International Actuarial Association. Available online at: http://www.actuaries.org/ASTIN/Colloquia/Helsinki/Papers/S6_23_deAlba_Andrade.pdf
- Dickinson RE, Henderson-Sellers A, Kennedy PJ (1993) Biosphere–atmosphere transfer scheme (BATS) version 1E as coupled to the NCAR Community Model. In: NCAR Technical report. TN-387 + STR, NCAR, BoulderGoogle Scholar
- Giorgi F, Jones C, Asrar G (2009) Addressing climate information needs at the regional scale: the CORDEX framework. WMO Bull 58:175–183Google Scholar
- IPCC (2007) Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. In: Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL (ed) Cambridge University Press, CambridgeGoogle Scholar
- Kiehl J, Hack J, Bonan G, Boville B, Briegleb B, Williamson D, Rasch P (1996) Description of the NCAR community climate model (CCM3). In: NCAR Technical report. TN-420 + STR, NCAR, BoulderGoogle Scholar
- Munoz-Arriola F, Lettenmaier DP, Zhu C, Avissar R (2009) Water resources sensitivity of the Rio Yaqui Basin, México to agriculture extensification under multi-scale climate conditions. Water Resour Res 45: W00A20. doi:10.1029/2007WR006783
- Pal JS, Giorgi F, Bi X, Elguindi N, Solmon F, Gao X, Francisco R, Zakey A, Winter J, Ashfaq M, Syed F, Bell J, Diffenbaugh N, Karmacharya J, Konare A, Martinez-Castro D, Porfirio da Rocha R, Sloan L, Steiner A (2007) Regional climate modeling for the developing world: the ICTP RegCM3 and RegCNET. Bull Am Meteor Soc 88:1395–1409CrossRefGoogle Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.