Investigation of climate dynamics over euro-mediterranean region using a variable- resolution global cesm model

Abstract

This thesis demonstrates the usability and added value of the Variable Resolution Community Earth System Model (VR-CESM) for climate investigations over the Euro-Mediterranean and it is the first work that employs VR-CESM for the region. The geographic location and complicated topography of Euro-Mediterranean leads to a regional climate governed by complex nonlinear interactions and cascade of feedbacks between multitude of scales, from global to local, and spatiotemporally highly varied climatic characteristics. Therefore, a modelling approach offering high resolution within a global context best fits to the regional climate investigations. Of particular concern, the region is a climate change hotspot bringing many projected climate related risks to the regional ecosystem and population which increases the need for better understanding the present-day climate and providing more realistic future projections. VR-CESM offers an alternative modeling approach across the global-to-regional model divide utilizing the Community Atmosphere Model's spectral element dynamical core (CAM-SE). It employs a global grid which is only refined over a limited area, thus substantially decreasing the computational demand while allowing global simulations with regional resolutions that are mostly unaffordable by uniform resolution GCMs. This technique can leverage today's parallel computing platforms almost to their fullest extent by offering near-perfect scalability. In this work, Euro-Mediterranean's climate is investigated at regional horizontal resolutions of 0.25° (28 km) and 0.125° (14 km) refined from a global resolution of 1° (111 km). An Atmospheric Model Intercomparison Project (AMIP) setup in which the atmosphere and land are the only active components is employed with out-of-the-box FHIST component set. All simulations are integrated over the period 1998-2014 and first two years are excluded from analyses considering the spin-up of the model. The performance of simulations is evaluated by comparison to a coarse resolution (quasi- uniform 1°) control simulation with an identical dynamics and similar physics against available observation-based datasets. Additionally, a parameter tuning experiment was conducted targeting top-of-the-model shortwave and longwave cloud forcings better matching the observations. The added value, as well as challenges, of VR-CESM and systemic biases associated with model resolution and/or sub-grid-scale parametrizations were identified and hypothesized. The improvements obtained are mainly related to a better representation of the complex topography of the region with higher resolution. Increasing the regional resolution to 0.25° generally yields considerable improvements over the quasi-uniform 1° control simulation, both for temperature and precipitation. Although doubling the regional resolution to 0.125° generally leads to modest improvements, pronounced ones are obtained in the representation of small-scale processes including that of extreme events. Model biases are addressed taking into account the model inherent ones and likely effects of the deficiencies in the reference data. Some persistent biases remain across all simulations and are discussed considering the mechanisms behind the related processes such as the synoptic scale moisture transport or land-atmosphere interactions. Variable resolution (VR) simulations represent the spatial variability of temperature with higher pattern correlations (between 0.97-0.99) than that for the control simulation, over the refined domain in all seasons. Increasing the resolution leads to apparent reductions in temperature biases. For instance, while 1° coarse resolution has large area-averaged surface temperature biases of 2.38, 1.82, 2.06, and 1.81°C for winter, spring, summer, and fall, respectively, the biases reduce to 1.38, 1.07, 1.67, and 1.13°C with 0.25° resolution. The total area with statistically significant biases in the control simulation, such as the warm biases exceeding 6°C over high mountains and coasts and cold biases over the southern regions exceeding 5°C locally, also decrease with VR simulations. Because precipitation is a process operating at multitude of scales and comprise strong spatiotemporal heterogeneity and that, dependency to convective parametrizations cannot be completely eliminated even at the highest resolution we used, simulation of precipitation is challenging, even more over complex topographies. Even though, spatial variability of precipitation is represented in better agreement with observations than the control simulation with pattern correlations between 0.91-0.95. Precipitation biases also reduce with increased resolution, especially in summer (i.e., from relative root-mean-square-error of 39.70, 29.49, 173.89, and 69.87%, respectively for winter, spring, summer, and fall for 1° to 34.63, 23.60, 133.07, and 50.28% for 0.25°). The improvements are more pronounced over regions with sharp topographical contrasts. For instance, the amplitudes of wet/dry biases over the windward/leeward sides of the high mountains and the dry biases over the coasts, which can exceed 2 mm/d, and precipitation bias contrast on the windward/leeward sides of high mountains, which can be as large as 7 mm/d for 1°, both reduce considerably with VR simulations. Also, the locations and expansions of upward air motions better coincide with the respective terrain upslopes and downslopes in VR simulations and orographically forced upward motions are strongest and most vertically expanded for the highest resolution. 0.125° performs the best in terms of daily maximum temperature extremes among all simulations except in fall (when it has a similar bias as in 0.25°). Both VR simulations outperform the quasi-uniform 1° one in summer and fall and 0.125° resolution simulation outperforms the 0.25° one in all seasons for extreme precipitation (e.g. biases for summer extreme precipitation for 1°, 0.25°, and 0.125° are 6.09, 5.36, 4.68 mm/d and 6.36, 5.63, 5.02 mm/d against CPC and E-OBS reference data, respectively). Overall, the results highlight that the VR-CESM is useful, even an attractive choice depending on the scientific purpose, for the present-day climate studies over Euro-Mediterranean and that the method can also be utilized for the future projections.