Dynamical Downscaling Approach and Its Applications Yuqing Wang

Dynamical Downscaling Approach and Its Applications Yuqing Wang International Pacific Research Center (IPRC) and Department of Meteorology, University of Hawaii at Manoa, Honolulu, HI 96822, USA The 8 th International Summer Seminar on Climate System and Climate Change

Downscaling • Statistical Downscaling • Dynamical Downscaling

Outline 1. Regional Climates and Regional Climate Modeling 2. Regional Climate Modeling 1). RCM Development 2). Application to Dynamical Downscaling 3). Application to Climate Process Studies 3. Challenges and Some Issues 4). Regional Climate Predictability 4. Future Directions

What are Regional Climates? Climates vary regionally due to regional forcings, such as topography, land-water contrasts, urban and islands effects. These regional forcings can produce statistically predictable effects on the weather and climate anomalies. Examples of regional climates include: Orographic precipitation; Lake effects on precipitation anomalies; Urban heat islands; Sea breeze, etc. Some regional climates likely feed back onto and significantly affect the climate at larger scales, but lack of quantitative studies to date, however.

Long-term mean total rainfall in JJA (mm)

Complexity of Regional Climates • Geographic feature: Topography, continental scale land basin-scale ocean, coastline, large latitudinal coverage, landuse and vegetation cover. • Climatic features: Dominated by large-scale ocean-land atmosphere interaction, strong tropical-subtropicalextratropical interaction, rapid seasonal transition, large variability from diurnal, intraseasonal to interannual, decadal time scales. • Weather systems: Mesoscale nature of precipitation systems (Meiyu/Baiu fronts, low-level jests, etc. ), involving strong interactions between regional scale forcing (terrains, landsea contrasts, land-use), and large-scale circulation.

Resolution Issue 2 degree resolution 0. 5 degree resolution

Regional Modeling Approach • Modeling efforts using a limited-area high-resolution model driven at their lateral boundaries by observations/reanalysis or the output of a low-resolution global model – Dynamical downscaling. • Regional models can afford high-resolution in a limited domain and thus they can better resolve underlying topography, coastlines, land surface properties, and mesoscale circulations, improving simulations of regional climates, including their mean and variability. • At present, the results from a nested regional model do not feed back to affect its parent low-resolution global model. Therefore, regional modeling is expected to improve simulations at regional scales rather than the large-scale features that are expected to be resolved by GCMs.

RCM Development 1. Types of regional climate models 2. Nesting strategies 3. Verification and validation of RCMs

Components & Interactions In A Climate Model Lateral mixing DYNAMICA L CORE MODEL PHYSICS Ecosyste m Carbon cycle Ocean coupling PBL Vertical Mixing Surface Processe s Radiatio n Budget Moist Process es Land surface process Atmosphe ric aerosols & Chemistry Grid resolved Sub-grid scale

Types of Regional Climate Models A RCM is usually developed based on an existing meso-scale numerical model for short -term (24 -72 h) simulation or prediction, such as CWRF (NCAR), DARLAM (CSIRO), RSM (NCEP), RAMS (CSU). Spectral models: NCEP-RSM, MRI-RSM. Grid point models: Reg. CM 4, RAMS, i. RAM, CWRF. Stretched-grid global models.

Nesting Strategies: Large-Scale Control Lateral boundary nesting -- Sponge Bufferbuff Zone zone Inner domain Spectral nesting -- large-scale nudging The large-scale features are restored in the regional model domain by spectral nudging Two-way nesting An alternative: Global variable-resolution models for regional climate modeling Regional Model GCM

Verification of model simulations 1. Verification with independent dataset 2. Verification with data from driving fields Bias Care should be taken to use RMSE equivalent spatial scales Spatial Correlation Temporal Correlation Spatial standard deviation (SSD) Pattern recognition (EOF) Spectrum analysis Probability of occurrence of extreme climate eve

Validation of model physics (1) • Although most of the model physics parameterizations were tested offline, such as the radiation, cumulus convection, PBL, validation in RCMs is lacking partially due to the availability of high-resolution observational data. • Complex interactions and feedbacks between different physical processes add difficulties to the validation of model physics, such as feedbacks among cloud-radiation-convectioncirculation.

Validation of model physics (2) • Some model physics are region- or seasondependent, such as land/sea, cumulus convection, etc. Therefore conclusions might not be robust. • The advanced satellite data are useful for validating model physics, such as the cloud microphysics, cumulus convective parameterization, etc. • Validation of model physics should be performed under a wide range of climate regimes before the model is used to address climate change issues.

Applications of RCMs 1. Dynamical downscaling of climate-change projections, driving RCMs with GCM climate-change simulations; 2. Dynamical downscaling of seasonal to interannual climate prediction, driving RCMs with global model predictions; 3. Assessing regional climate change due to changing regional physiographic factors (e. g. , land use, sea-ice conditions etc. ). 4. Detailed climate process studies at regional scales, driving RCMs with historic atmospheric objective analysesreanalysis; 5. Reconstruction of regional-scale paleoclimatology, driving RCMs with GCM paleoclimatic simulations; 6. Reconstruction of recent-past states on the regional scale, driving RCMs with historic atmospheric objective analyses;

Regional Climate Modeling Simulatio n Projectio n prediction Large-Scale Forcing Dynamical downscaling Process Study REGIONAL MODELING Land use Land cover Paleoclimate Aeroso l forcin g Regional Climate Change SST Forcing Snow/ice cover change Land-atmosphere interaction Variability & Extreme events Predictabili ty Study External Interactive

Percentage change in mean annual snowpack in the western U. S. based on an ensemble of regional climate simulations for the present and future climate (20402060) as projected by a global climate model (Leung et al. 2004).

Mean winter (DJF) surface temperature change in the western U. S. simulated by the NCAR/DOE Parallel Climate Model, PCM (left panel), and the Penn State/NCAR MM 5 (right panel), driven by the PCM. Temperature changes were calculated as the difference between the ensemble simulation of the future climate (2040 -2060) following a business as usual emission scenario and the control climate (Leung et al. 2004).

Spatial distribution of rainfall change in JJA 1998 between ensembles with current and reforested vegetation covers in the Indochina peninsula (94 o-109 o. E, 9 o-19 o. N) using the IPRC-Reg. CM. The hatching is for statistically significant areas at 90 % confidence level. The contours show the orography at the 500, 1000, 1500, 2000, 3000, 4000, and 5000 m heights. (Sen et al. 2004).

Column-integrated CLW Downward SW radiation flux Differences of vertically integrated liquid water content (10 -2 mm, left panel) and downward shortwave radiation flux at the sea surface (W m-2, ringt panel) between the control and No-Andes runs using IPRC-Reg. CM, averaged for ASO 1999. Contour interval is 10 -2 mm and values greater than 10 -2 mm are shaded in the left panel. Contour interval is 15 W m-2 and values less than -15 W m-2 are shaded in the right panel. (Xu et al. 2004).

Regional Climate Predictability • Predictability of the first kind: Initial value problem To address how uncertainties in the initial state of the climate system affect the prediction of a later state. • Predictability of the second kind: Boundary value problem To address how the predictability limit of the climate system responds to changes in the boundary conditions (SST, soil moisture; • Climate Predictability: A mixed initial and anthropogenic forcing, etc. ) boundary value problem since the initial conditions of slow evolving values (such as SST,

Regional Climate Model Predictability • Uncertainties in driving fields Garbage-in garbage-out issue (Giorgi and Mearns 1999); Some improvements are still possible in the region with strong lower boundary forcing that is lack from the coarse resolution GCM. • Uncertainties in the nested RCMs Unphysical treatment of lateral boundary conditions; Inconsistency in dynamics and physics between RCM and GCM; Unrealistic representation of physical processes; Internal flow-dependent instabilities of the chaotic • RCM climate. Predictability system. Region-dependent, season-dependent; Different variables have different predictabilities,

3. Challenges and Some Issues • • • Resolution issue Physical parameterization issue Model evaluation and diagnostics issue Model domain size and location issue Issue on modeling extreme climate events • Dynamical downscaling issue • RCM intercomparison issue • Issues related to climate process studies

(1) Model Resolution Issue • What are the relative merits of increasing spatial resolution versus more accurate or sophisticated physics if the resolution reaches some level? • Global models can be run at resolutions equivalent to our current RCM resolution 5 years ago at least for time slice experiments. • It is not clear Whether we still need RCMs or RCMs will simply move to cloud-resolving models (CRMs)

(2) Physical Parameterization Issue • Whether and how can we consider the dependency on both spatial resolution and timestep of the model’s physical parameterizations? Parameterization for RCM is much more difficult than that for large-scale global models since we try to resolve more detailed physics.

(3) Model Evaluation and Diagnostics Issue • What datasets do we need to evaluate climate simulations at regional scales? • What sophisticated diagnostic methods can be employed to evaluate model performance? Special attention should be paid to diagnosing internal variability and uncertainties of RCMs and the added values from RCMs to understand model behavior and improve RCM predictability! Extensive diagnostics of the sources of model biases is critical to model improvements.

(4) Model Domain Size and Location Issue • Is there an optimal domain size with which an RCM can achieve the best simulation of regional climate? If the domain is too large the solution for largescale features might drift from the driving GCM or observations. If the domain is too small the RCM might not allow development of internal dynamics. How to quantify these two criteria? This seems resolution-dependent and needs to be studied

(5) Issue on Modeling Extreme Climate Events RCMs are better than the driving GCM in simulating the probability distribution of occurrence of extreme climate events (droughts, floods, summer hot waves, winter snowstorms, etc. ). This advantage needs to be further tested and demonstrated in different regions and seasons. Current RCMs have not shown promising capability in simulating all aspects of tropical storms, including their frequency, track, and intensity, but it is quite challenging due to our very limited understanding of physical processes controlling the TC genesis and intensity.

(6) Dynamical Downscaling Issue • Can RCMs be used to assess climate change issues before the models are comprehensively tested and verified for quite different climate regimes? Probably not, but people couldn’t wait. Testing and verifying the RCM simulation for a range of climatic regimes (e. g. , mid-latitude, tropics, summer, winter, seasonal transition, interannual variability) are critical before applying the RCM to climate change projections.

(7) RCM Intercomparison Issue • What can we learn from model inter-comparison projects? *To identify model weakness and uncertainties; *To infer predictability of different climate regimes. *Limited regions with only very limited models involved (some of them were not independently developed); *Comparisons are made mostly for final products generated by complex physical parameterizations, less attention was given to diagnosing the processes leading to the differences among different models; *No much attention has been given to the strength and weakness of different individual physics parameterization schemes;

(8) Climate Process Study Issue • Do we need ensemble in regional climate modeling? Ensemble simulations are strongly recommended in future studies to both understand regional climate processes and detect signals of climate change and sensitivity due to the chaotic nature of the atmospheric motion. The use of small model domain however show small variance between simulations with different initial conditions due to strong control by large-scale driving fields.

4. Some future directions • To continue the application of RCM to dynamical downscaling; • To contribute to the development of model physical parameterizations; • To help reduce common biases in GCMs and develop the super physical parameterization ensemble for GCMs; • To conduct climate process studies to improve our understanding of different climate processes at regional scales; • To study the contribution of high frequency weather and meso-scale signals to the larger-scale circulation (upscaling) with the stretched-grid global models.

Thank you for your attention!