Torres del Paine, Patagonia, Chile Alan Rhoades 1, Paul Ullrich 1,3, Colin Zarzycki 2, Hans Johansen 1, Zexuan Xu 1, and William Collins 1 Lawrence Berkeley National Laboratory 1, National Center for Atmospheric Research 2, University of California, Davis 3 1
Characteristics of Water in the West Estimates of Snowpack (SWE) based on: 1) Yearly estimates from DWR April 1 st SWE measurements 2) Long term average from VIC simulations Dettinger and Anderson, 2015 2
Characteristics of Water in the West Estimates of Snowpack (SWE) based on: 1) Yearly estimates from DWR April 1 st SWE measurements 2) Long term average from VIC simulations Water stores: 1) 23.5 MAF of storage from reservoirs 2) 17 MAF from April 1 st SWE storage (5 Oroville dams) = 72% of additional natural storage Dettinger from and SWEAnderson, 2015 3
Characteristics of Water in the West CA Snowpack Storage vs Reservoir Storage April 1 st Snow Water Equivalent (SWE) 50% of precipitation in 5 15 days Dettinger et al., 2011 4
Characteristics Characteristics of of Water Waterin inthe thewest West CA Snowpack Storage vs Reservoir Storage April 1st Snow Water Equivalent (SWE) ~40% of the Total Snow Accumulation in CA Largely due to specific extreme precipitation features in select months 50% of precipitation in 5 15 days Dettinger et al., 2011 Guan et al., 2013 Neiman, 2014 5
Global-to-Regional Downscaling Tool to Understand Mountain Hydroclimatology Variable Resolution in the Community Earth System Model (CESM) Sea Ice Prescribed 55km 14km Ocean SST Prescribed F AMIP CAM5 Component Set 28km 111km Grids constructed with SQuadGEN (Ullrich) Benefits: Global simulation (atmosphere ocean teleconnections) Increased resolution in specified areas (better topography) Decrease in model runtime and data storage ( smaller server usage) Eliminates multi model dataset needs (bias propagation) Merges regional and global modeling communities (scale awareness) Glimpse into the future of high resolution global climate modeling 6
Global-to-Regional Downscaling Tool to Understand Mountain Hydroclimatology 14 km 28km 111km Pilot VR CESM project in Chile column integrated water vapor figure adapted from Colin Zarzycki 7
Rhoades, A.M., Ullrich, P.A., Zarzycki, C.M., Johansen, H., Margulis, S., Xu, Z., and W.D., Collins (2017) A Variable-Resolution CESM Case Study of the Comparative Importance of Model Resolution and Microphysics in a Mountainous Region In Revision, J. of Hydrometeo. Identify and understand persistent VR CESM bias associated with changing model resolution and/or sub grid scale model parameterizations Grids SYPD: 5.05 CHCPY: ~13,000 Study Regions SYPD: 3.88 CHCPY: ~18,000 Server Cori NERSC Simulated Year Per Day (SYPD) Core-Hour Cost Per Year (CHCPY) Can VR CESM realistically represent historical precipitation, snowpack and surface temperature in mountainous regions? Differenced Topography (GTOPO30) SYPD: 1.96 CHCPY: ~33,500 SYPD: 0.49 CHCPY: ~193,000 8
Experimental Setup and Reference Datasets Experimental Setup Similarities Differences VR CESM MG1 and MG2 CAM5 CLM4 with 5 km year 2000 Surface Datasets Prescribed SST and Sea Ice Horizontal Resolution Prognostic vs Diagnostic Treatment of Cloud Hydrometeors Morrison and Gettelman microphysics developed for CESM1 (CESM2) in 2008 (2015) will be referred to as MG1 (MG2) Both MG1 and MG2 are 2 moment schemes (i.e., mass and number mixing ratio). Dataset Resolution Timeframe Hydroclimate Variables VR CESM (CAL_VR) MG1 28 km, 14 km, 7 km*, 3.5 km* 1999 2015 VR CESM (CAL_VR) MG2 28 km, 14 km 1999 2015 Precipitation, Snow Cover, SWE, Surface Temperature Precipitation, Snow Cover, SWE, Surface Temperature WRF_VR CESM MG1 28km, 14km, 7km 1999 2015 Precipitation, Snow Cover, SWE, Surface Temperature PRISM 4 km 1999 2015 Precipitation, Surface Temperature MODIS 5 km 2000 2015 Snow Cover SNSR 90 m 1999 2015 SWE *Refinements at nonhydrostatic scales are experimental, targeted over small regions (CA mountain region), mid latitudes (large scale dominated), and in DJF (limited convection). 9
Effects of Increased Resolution Over Complex Terrain on Mountain Hydroclimatology DJF climate averages for MG1 simulations show a clear windward/leeward mountain precipitation bias which propagates into the other hydroclimate variables. Here the diagnostic treatment of hydrometeors requires precipitation to fall out immediately. 28 km 14 km 7 km 3.5 km Difference from Observed/Reanalysis Precipitation (mm/day) Snow Cover (%) SWE (mm) Surface Temperature (K) DJF seasonal Pearson Pattern Correlations in MG1 were 0.5 to 0.7 across precipitation, snow cover, and SWE and 0.9 for surf. temp., although a systemic cold bias is present 10
Effects of Prognostic Hydrometeors (MG2) on Mountain Hydroclimatology DJF climate averages for MG2 simulations most closely represented the observed precipitation, snow cover, and SWE results by alleviating bias to within +2% of PRISM, +6% of MODIS, and 5% of SNSR. 28 km 28 km 14 km 14 km Difference from Observed/Reanalysis Precipitation (mm/day) Snow Cover (%) SWE (mm) Surface Temperature (K) DJF seasonal Pearson Pattern Correlations in MG2 across all of the hydroclimate variables more closely matched observed by +0.12 (0.73) compared to MG1 (0.61) 11
Daily Climate Lifecycles and Mountain Windward/Leeward Distributions The least average (range) bias in accumulated PRECT was found for MG2 in the California Mountain Region SWE accumulation cycle closely matched Landsat era until the historical peak accumulation months of March to April, where SWE steadily became low biased which may indicate that spring dependent melting factors in the model may be too strong and inhibits SWE persistence. Systemic cold bias found with ensemble average (range) of bias of 2.2 K (7.7 K) which persists in DJF and generally becomes worse with refinement of model horizontal resolution (likely due to resolving more elevation classes). Too Wet Too Dry The distributions of PRECT, FSNO and SWE in the windward/leeward of the Sierra Nevada in MG2 was much improved compared with MG1 Observed Windward/Leeward Ratios Ensemble Ensemble Average MG1 Average MG2 PRECT 2.9 8.79 2.8 FSNO 1.2 2.67 1.0 SWE 2.2 10.68 1.6 12
Daily Climate Lifecycles and Mountain Windward/Leeward Distributions The least average (range) bias in accumulated PRECT was found for MG2 in the California Mountain Region Too Wet Too Dry The distributions of PRECT, FSNO and SWE in the windward/leeward of the Sierra Nevada in MG2 was much improved compared with MG1 Windward/Leeward Ratios SWE accumulation cycle closely matched SNSR until the historical peak accumulation months of March to April Observed Ensemble Ensemble Average MG1 Average MG2 PRECT 2.9 8.79 2.8 Systemic cold bias found with ensemble average (range) of bias of 2.2 K (7.7 K) which persists in DJF and generally becomes worse with refinement of model horizontal resolution (likely due to resolving more elevation classes). FSNO 1.2 2.67 1.0 SWE 2.2 10.68 1.6 13
Daily Climate Lifecycles and Mountain Windward/Leeward Distributions The least average (range) bias in accumulated PRECT was found for MG2 in the California Mountain Region Too Wet Too Dry The distributions of PRECT, FSNO and SWE in the windward/leeward of the Sierra Nevada in MG2 was much improved compared with MG1 Windward/Leeward Ratios SWE accumulation cycle closely matched SNSR until the historical peak accumulation months of March to April Observed Ensemble Ensemble Average MG1 Average MG2 PRECT 2.9 8.79 2.8 FSNO 1.2 2.67 1.0 Systemic cold bias found with ensemble average (range) of bias of 2.2 K (7.7 K) SWE 2.2 10.68 1.6 14
Daily Climate Lifecycles and Mountain Windward/Leeward Distributions The least average (range) bias in accumulated PRECT was found for MG2 in the California Mountain Region Too Wet Too Dry The distributions of PRECT, FSNO and SWE in the windward/leeward of the Sierra Nevada in MG2 was much improved compared with MG1 Windward/Leeward Ratios SWE accumulation cycle closely matched SNSR until the historical peak accumulation months of March to April Observed Ensemble Ensemble Average MG1 Average MG2 PRECT 2.9 8.79 2.8 FSNO 1.2 2.67 1.0 Systemic cold bias found with ensemble average (range) of bias of 2.2 K (7.7 K) SWE 2.2 10.68 1.6 15
Rhoades, A.M., Ullrich, P.A., Zarzycki, C.M., Johansen, H., Margulis, S., Xu, Z., and W.D., Collins (2017) A Variable-Resolution CESM Case Study of the Comparative Importance of Model Resolution and Microphysics in a Mountainous Region In Revision, J. of Hydrometeo. Conclusions Motivation: VR CESM is a useful tool to understand resolution and microphysics impacts in CESM, especially in mountains Goals: Identify and understand persistent VR CESM bias associated with changing model resolution (28km, 14km, 7km, and 3.5km), sub grid scale model parameterizations (28km and 14km), and/or nonhydrostatic dynamical core (28km, 14km, 7km). 16
Rhoades, A.M., Ullrich, P.A., Zarzycki, C.M., Johansen, H., Margulis, S., Xu, Z., and W.D., Collins (2017) A Variable-Resolution CESM Case Study of the Comparative Importance of Model Resolution and Microphysics in a Mountainous Region In Revision, J. of Hydrometeo. Conclusions Motivation: VR CESM is a useful tool to understand resolution and microphysics impacts in CESM, especially in mountains Goals: Identify and understand persistent VR CESM bias associated with changing model resolution (28km, 14km, 7km, and 3.5km), sub grid scale model parameterizations (28km and 14km), and/or nonhydrostatic dynamical core (28km, 14km, 7km). Results in California Mountainous Region: Increased refinement in model horizontal resolution to more realistically represent topographic and surface characteristics did not alleviate mountain hydroclimate bias tendencies in VR CESM The use of a nonhydrostatic dynamical core did not alleviate bias in most hydroclimate variables, save for surface temperature. Forcing dataset plays a larger role in mountain hydroclimate bias tendencies in WRF (NCEP vs VR CESM) 17
Conclusions Rhoades, A.M., Ullrich, P.A., Zarzycki, C.M., Johansen, H., Margulis, S., Xu, Z., and W.D., Collins (2017) A Variable-Resolution CESM Case Study of the Comparative Importance of Model Resolution and Microphysics in a Mountainous Region In Revision, J. of Hydrometeo. Motivation: VR CESM is a useful tool to understand resolution and microphysics impacts in CESM, especially in mountains Goals: Identify and understand persistent VR CESM bias associated with changing model resolution (28km, 14km, 7km, and 3.5km), sub grid scale model parameterizations (28km and 14km), and/or nonhydrostatic dynamical core (28km, 14km, 7km). Results in California Mountainous Region: Increased refinement in model horizontal resolution to more realistically represent topographic and surface characteristics did not alleviate mountain hydroclimate bias tendencies in VR CESM The use of a nonhydrostatic dynamical core did not alleviate bias in most hydroclimate variables, save for surface temperature. Forcing dataset plays a larger role in mountain hydroclimate bias tendencies in WRF (NCEP vs VR CESM) The use of a new physics parameterization (MG2) coupled with resolving topography and surface characteristics at 14km resulted in a close approximation to DJF hydroclimate variables with Precipitation = +2% of PRISM (climate), wind/lee distribution 3.0 (2.9 in PRISM), DJF Pearson correlation of 0.77 Snow Cover = +6% of MODIS (climate), wind/lee distribution 1.1 (1.2 in MODIS), DJF Pearson correlation of 0.80 SWE = 5% of Landsat Era (climate), wind/lee distribution 2.0 (2.2 in Landsat Era), DJF Pearson correlation of 0.69 Surface Temperature = 2.9 K (climate) needs to be addressed, DJF Pearson correlation of 0.93 More work is needed to understand the snowpack bias sensitivities to grid resolution and sub grid scale microphysics 18
Future Work http://climate.ucdavis.edu/hyperion Case Study Regions (3 water managers from each): 1) Colorado River Headwaters (Colorado) 2) Kissimmee River and South Florida (Florida) 3) Sacramento San Joaquin Watershed (California) 4) Susquehanna River (Pennsylvania) 19
Future Work http://climate.ucdavis.edu/hyperion Case Study Regions (3 water managers from each): 1) Colorado River Headwaters (Colorado) 2) Kissimmee River and South Florida (Florida) 3) Sacramento San Joaquin Watershed (California) 4) Susquehanna River (Pennsylvania) 20
Acknowledgements Lawrence Berkeley National Laboratory Lab Directed Research and Development (LDRD) Program ``Modeling the Earth's Hydrological Cycle from Watershed to Global Scales'' project Office of Science, Department of Energy ``An Integrated Evaluation of the Simulated Hydroclimate System of the Continental US'' project (award no. DE SC0016605) ``Multiscale Methods for Accurate, Efficient, and Scale Aware Models of the Earth System'' project (contract no. DE AC02 05CH11231) NSF Climate Change, Water, and Society IGERT at UC Davis NSF Award Number:1069333 Leland Roy Saxon and Georgia Wood Saxon Fellowship 21
Effects of Resolution and Microphysics on DJF Mountain Elevation Profile Benefits associated with proper amounts of total precipitation and windward/leeward deposition location are apparent in MG2 mountain elevation profiles for precipitation and SWE However, a systemic cold bias that worsens with elevation is apparent Precipitation Difference (mm/day) SWE Difference (mm) Surface Temperature Difference (K) DJF seasonal variability in observed A more complete analysis of the processes that impact mountain lapse rates need to be conducted 22
Effects of a Nonhydrostatic DyCore on Mountain Hydroclimatology DJF climate averages for WRF_VR CESM MG1 simulations still show a clear windward/leeward mountain precipitation bias. Snowpack bias is also maintained. Surf. temp. cold bias diminished, but still present in high elevations in mountainous regions. 28 km 14 km 7 km Difference from Observed/Reanalysis Precipitation (mm/day) WRF simulations conducted by Zexuan Xu Snow Cover (%) SWE (mm) Surface Temperature (K) DJF seasonal Pearson Pattern Correlations in WRF_VR CESM MG1 were 0.45 to 0.60 across precip., snow cover, and SWE and <0.9 for surf. temp. 23
Effects of a Nonhydrostatic DyCore on Mountain Hydroclimatology DJF climate averages for WRF_VR CESM MG1 simulations still show a clear windward/leeward mountain precipitation bias. Snowpack bias is also maintained. Surf. temp. cold bias diminished, but still present in high elevations in mountainous regions. 28 km 14 km 7 km Difference Between WRF/VR CESM Precipitation (mm/day) WRF simulations conducted by Zexuan Xu Snow Cover (%) SWE (mm) Surface Temperature (K) DJF seasonal Pearson Pattern Correlations in WRF_VR CESM MG1 were 0.45 to 0.60 across precip., snow cover, and SWE and <0.9 for surf. temp. 24
Effects of a Nonhydrostatic DyCore on Mountain Hydroclimatology DJF climate averages for WRF_VR CESM MG1 simulations still show a clear windward/leeward mountain precipitation bias. Snowpack bias is also maintained. Surf. temp. cold bias diminished, but still present in high elevations in mountainous regions. 28 km 14 km 7 km Differences Precipitation (mm/day) WRF simulations conducted by Zexuan Xu DJF climate difference in WRF_VR CESM MG1 vs VR CESM MG1 Highlights that WRF generally Snow Cover (%) Increases precipitation bias Increases average SWE/snow cover Warms surface temperatures (save for SWE portions (mm) of southern Sierra) Surface Temperature (mm) DJF seasonal Pearson Pattern Correlations in WRF_VR CESM MG1 were 0.45 to 0.60 across precip., snow cover, and SWE and <0.9 for surf. temp. 25
Daily Climate Lifecycles and Mountain Windward/Leeward Distributions with WRF forced by VR-CESM Accumulated precipitation similar to those seen in VR CESM results, positive (negative) windward (leeward) bias Positive accumulated SWE bias worsened from VR CESM with increase in resolution Too Wet Too Dry Observed Windward/Leeward Ratios Ensemble Average WRF_VR CESM MG1 Ensemble Average WRF_NCEP PRECT 2.9 10.3 6.1 Surface temperature cold bias halved in DJF, analysis underway to understand why WRF simulations conducted by Zexuan Xu 26
Back of the Envelop Calculation for MG1 diagnostic physics At what horizontal resolution would MG1 microphysics theoretical assumptions break? Lateral Transport Time = dx / lateral transport velocity Drop Time = drop distance / drop velocity when lateral transport time = drop time, then... DX = (drop distance / drop velocity)*lateral transport velocity Rainfall DX Assume Lateral transport vertical integration = lower 10 levels in atmosphere drop velocity = maximum, 1.2 m/s (snow) and 9.1 m/s (rain) drop distance = 5000 m (Morrison and Gettelman) Snowfall DX Min = 1 km, Max = 6 km Lower 10 Level Rainfall dx where lateral transport time = drop time Min = 11 km, Max = 47 km Lower 10 Level Snowfall dx where lateral transport time = drop time it t 27
Rhoades et al., (2016) "Characterizing Sierra Nevada Snowpack Using Variable Resolution CESM" JAMC doi: 10.1175/JAMC D 15 0156.1 Historical analysis to evaluate efficacy of VR CESM as a regional downscaling tool at 28km and 14km 14 km Datasets Rhoades et al., (2017) "Projecting 21st Century Snowpack Trends in Western USA Mountains Using Variable Resolution CESM" Clim. Dyn. doi: 10.1007/s00382 017 3606 0 VR CESM at 28km to evaluate RCP8.5 signal on hydroclimate metrics with comparison to other widely used climate change datasets RCP8.5 by 2025 2050 Pilot Project (2017) with the Global Change Center at the Pontificia Universidad Católica de Chile VR CESM CORDEX style grids to evaluate TSA and PRECT over South America/Chile at 28km and 14km RCP8.5 by 2075 2100 10 Region Analysis 28 km Datasets Snowfall decreases with elevation (<2000 m) SWE decreases with elevation (all elevations) SNOTEL Peak Accumulation Timing (Day 180) Models have Negative Peak Accumulation Bias Freezing line moves upslope by 1000 m 28