-------------------------------------------------------------------------------------------------- This file describes the 1/8th degree forcing file datasets and their generation, as well as the associated VIC soil file developed for the North Pacific Rim (NPR) hydrologic study. Original: 12 Jan 2011 Last Revised: 19 March 2012 -------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------- 1. DATA FORMAT AND ORGANIZATION ----------------- (a) Directory listing ----------------- Following is a list of the data that can be found in this directory: daily_forcings_0.125deg -- directory containing zipped files (described below) containing the daily forcings developed for this project, subsampled to 1/8th degree vic_soil_files -- VIC soil parameter files, calibrated over the test basins. Included are the original 1/2 degree file, as well as the same file subsampled to 1/8 degree. vic_calibrations_final -- directory containing plots comparing guage observations over the four test basins (Yukon, Skeena, Fraser, and Sacramento) to VIC simulated runoff using the basis 1/2 degree dataset. monthly_summaries_forcings -- directory containing plots of the 1/2 degree forcing dataset for historical and each downscaling scenario. Each plot displays the record-mean (e.g., 1948-2000) values for each month for each variable (tmax, tmin, precip). Also included are zipped files containing ArcInfo ascii-grid files of the above monthly summaries at 1/2 and 1/8th degree. monthly_summaries_hydrology -- directory containing zipped files, each of which contains ArcInfo ascii-grid files of the results of each 1/2 degree VIC run. These VIC runs were used for diagnostic purposes and to generate maps of expected changes across the domain. As above, As above, these are monthly summary files. ex_plots_of_previous_dataset_errors -- directory containing plots showing examples of the errors in the previous versions of this dataset. ----------------- (b) Zipped forcing files ----------------- The forcing file datasets are available for download as .tgz files (tarred and zipped) as follows: 1. npr_historical_0.125deg_subsmpld.tgz -- historical forcings (1948-2000) "Spatial Delta" downscalings*: 2. npr_SD_A1B_2040s_0.125deg_subsmpld.tgz -- A1B 2040s forcings 3. npr_SD_A1B_2080s_0.125deg_subsmpld.tgz -- A1B 2080s forcings 4. npr_SD_B1_2040s_0.125deg_subsmpld.tgz -- B1 2040s forcings 5. npr_SD_B1_2080s_0.125deg_subsmpld.tgz -- B1 2080s forcings BCSD downscalings* (available soon): 6. npr_BCSD_cgcm3.1_t47_A1B_0.125deg_subsmpld.tgz -- cgcm3.1_t47 forcings, A1B scenario 7. npr_BCSD_cgcm3.1_t47_B1_0.125deg_subsmpld.tgz -- cgcm3.1_t47 forcings, B1 scenario 8. npr_BCSD_cgcm3.1_t63_A1B_0.125deg_subsmpld.tgz -- cgcm3.1_t63 forcings, A1B scenario 9. npr_BCSD_cgcm3.1_t63_B1_0.125deg_subsmpld.tgz -- cgcm3.1_t63 forcings, B1 scenario 10. npr_BCSD_echam5_A1B_0.125deg_subsmpld.tgz -- echam5 forcings, A1B scenario 11. npr_BCSD_echam5_B1_0.125deg_subsmpld.tgz -- echam5 forcings, B1 scenario 12. npr_BCSD_gfdl_cm2_1_A1B_0.125deg_subsmpld.tgz -- gfdl_cm2_1 forcings, A1B scenario 13. npr_BCSD_gfdl_cm2_1_B1_0.125deg_subsmpld.tgz -- gfdl_cm2_1 forcings, B1 scenario 14. npr_BCSD_hadcm_A1B_0.125deg_subsmpld.tgz -- hadcm forcings, A1B scenario 15. npr_BCSD_hadcm_B1_0.125deg_subsmpld.tgz -- hadcm forcings, B1 scenario 16. npr_BCSD_miroc_3.2_A1B_0.125deg_subsmpld.tgz -- miroc_3.2 forcings, A1B scenario 17. npr_BCSD_miroc_3.2_B1_0.125deg_subsmpld.tgz -- miroc_3.2 forcings, B1 scenario *note: downscaling methods are described below Each historical and spatial delta file is about 16GB in size and about 22GB once unzipped, and each unzips all forcing files to a directory entitled "forcings". The BCSD files are about 46GB in size and about 62GB once unzipped. ----------------- (c) Forcing file organization and naming ----------------- Since VIC calculates water balance variables separately for each grid cell, the input forcing dataset is stored in a different file for each grid cell, with the following naming convention: data_LAT.XXXXX_LON.XXXXX (where "LAT.XXXXX" and "LON.XXXXX" refer to the lat/lon of the grid cell, respectively, specified to the 5th decimal place) At the 1/8th degree resolution used for this project, there are a total of 145954 grid cells in the full NPR domain. As a result, there will be 145954 forcing files in each unzipped "forcings" directory, each following the naming convention outlined above. Note: the naming convention is identical for historical and future scenarios, meaning that care must be taken to avoid confusing one dataset with another. ----------------- (d) Forcing file format ----------------- In order to minimize memory usage, data are stored in binary format as short integers, with one entry per day for the 53 years (1948-2000, corresponding to a total of 19359 days). Each entry contains a value, in the following order, for precipitation (precip, mm), maximum daily temperature (tmax, degC), minimum daily temperature (tmin, degC), and wind (m/s). Conversion from floating point to short integers is implemented as follows: FORCE_TYPE PREC UNSIGNED 40 FORCE_TYPE TMAX SIGNED 100 FORCE_TYPE TMIN SIGNED 100 FORCE_TYPE WIND SIGNED 100 notes: - this is the specification that can be used in the VIC global parameter file - 'signed' / 'unsigned' refers to the range used for conversion to short integer (-32768/32767 vs 0/65535) - A scaling factor of 40 was chosen for precipitation in order to be sure that maximum values in precipitation did not exceed the maximum value that can be stored as an unsigned short integer (65535) With 4 entries per day (precip, tmax, tmin, wind) at 2 bytes per entry, and a total of 19359 entries, each forcing file has a size of 154872 bytes, or approximately 152Kb. -------------------------------------------------------------------------------------------------- 2. DEVELOPMENT OF THE HISTORICAL FORCING DATASET ----------------- (a) basis dataset ----------------- The basis meteorological dataset is a 0.5 degree gridded daily time series in precipitation (precip, mm), daily maximum temperature (tmax, degC), daily minimum temperature (tmin, degC), and wind (m/s), generated following the description of Nijssen et al. (2001) and corrected for undercatch following Adam and Lettenmaier (2003). In brief, the dataset was created as follows: - construct daily gridded time series in temp/precip using global surface observations - fill gaps in data using stochastic model - scale monthly time series in temperature to match CRU dataset (Jones et al., 1994) - scale monthly time series in precipitation to match the GPCP (Huffman et al., 1997) and Hulme (1995) datasets - correct precipitation for undercatch (Adam and Lettenmaier, 2003). - obtain wind estimates by linearly interpolating daily NCEP reanalysis winds (Kalnay et al., 1996) - subsample 1/2 degree meteorological data to 1/8th degree resolution. Two steps (described below) were taken to ensure that this dataset represents a best estimate of the actual meteorological conditions over the NPR domain: 1) a precipitation calibration intended to account for unresolved changes in precipiation with elevation over the data sparse regions of Northern North America, and 2) corrections based on high-quality (and high resolution) reference datasets obtained for Alaska, British Columbia, the Columbia River Basin, and the Western US. ----------------- (b) precipitation calibration ----------------- Precipitation over the data sparse northern portion of North America was calibrated by comparing historical streamflow simulations with observations. Based on the comparison, it was determined that the basis dataset under-resolved the elevation dependence of streamflow. Precipitation was thus adjusted for all grid cells in North America north of 55N, using the following formula: scale factor = 1 + 0.002 * elevation (where elevation is in meters) The result is a scaling of 1 (i.e., no change) at zero elevation, and 6 at 3000 m. These adjustments were only applied to winter months (Oct-Apr), and applied at 1/2 degree resolution, before subsampling to 1/8th degree. ----------------- (c) reference meteorological datasets ----------------- Four reference datasets were used to adjust the climatology of the 1/8th degree forcing dataset: - 1/16th degree precip/tmax/tmin for the Columbia River Basin and coastal drainages (Elsner et al., 2010) - 1/24th degree precip/tmax/tmin for the western US (PRISM, Daly et al., 2002) - 1/16th degree precip/tmax/tmin for British Columbia (PCIC, Werner, 2011) - 2 km precip/tavg for Alaska (Scenarios Network for Alaska Planning, http://www.snap.uaf.edu) Each dataset was regridded to match the 1/8th degree grid and compared to the basis dataset using the 1950-2000 climatology. Corrections were applied to the basis climatology based on the difference (tmin, tmax) or the ratio (precip) of the reference climatology to the climatology of the basis dataset. ----------------- (d) smoothing of boundaries ----------------- In order to minimize discontinuities in the final dataset, all of the above corrections were smoothed along boundaries. The smoothing was applied to the 1/2 degree historical dataset. On boundaries between reference datasets, any point within 2 grid cells of the boundary was smoothed using a 3 by 3 point box filter (average of all grid cells in a 3 by 3 box centered on the grid cell in question). For regions where there is no reference dataset, grid cell corrections were smoothed if they were within about 2 degrees of a reference dataset ----------------- Note that: 1. The above corrections were only applied to the North American portion of the dataset, and 2. None of the above corrections impact the time series at a grid cell: the corrections only impact the mean (i.e., monthly climatology) at each grid cell. ----------------- -------------------------------------------------------------------------------------------------- 3. FUTURE CLIMATE PROJECTIONS ----------------- (a) Scenarios, climate models, and downscaling ----------------- Projections of future climate are obtained from global climate model (GCM) simulations, archived for the 4th assessment of the Intergovernmental Panel on Climate Change (IPCC AR4). Simulations are forced following the A1B and B1 emissions scenario. The former is a middle-of-the-road projection of future emissions, while the latter is more optimistic about mitigation of future greenhouse gas emissions. In the present work, we used data from the following 6 GCMs: - cgcm3.1_t47 - cgcm3.1_t63 - echam5 - hadcm - gfdl_cm2.1 - miroc_3.2 GCM simulations are run at low spatial resolution (GCM gridboxes are generally about 2 deg x 2 deg) and archived at monthly temporal resolution. We use two "downscaling" techniques to relate the relatively coarse GCM output to the final spatial and temporal resolution required for the present study (daily, 1/2 degree). These are the "spatial delta" and "BCSD" methods, and are briefly described in the following two sections. These and other downscaling techniques are discussed in much more detail by Hamlet et al. (2010, linked below). All future scenarios are generated at the base resolution of the historical dataset: 1/2 degree. These are then subsampled to 1/8th degree to provide the final forcings dataset. ----------------- (b) Spatial Deltas ----------------- Monthy projected changes in precipitation and temperature at 1/2 degree are obtained as follows: - obtain model output for emissions scenarios A1B and B1 - interpolate global climate model (GCM) output to 1/2 degree grid - calculate mean precip/temperature for 1948-2000 (historical), 2030-2059 (2040s), and 2070-2099 (2080s) - calculate spatially resolved temperature "deltas" by subtracting historical from future period - calculate spatially resolved precipitation "deltas" by dividing future period by historical The result is a delta for each grid cell, month, and variable (temperature, precipitation), which is applied to the historical forcings to obtain estimated future forcings for each of the selected future decades (2040s, 2080s). 30-year averages are necessitated by the large amount of interannual variability: a shorter averaging period would run a greater risk of sample bias. The strength of this method is that it provides a robust, easily-interpreted, realistic time series, and removes any biases from GCMs, taking only the mean monthly change at each grid point. This strength is also the primary weakness: no information about changes to the range, distribution, or sequencing of weather events is retained from the GCM simulations. ----------------- (b) BCSD (Bias Correction and Statistical Downscaling) ----------------- This technique is based on the methods of Wood et al. (2002, 2004), and is implemented as follows: - obtain model output for emissions scenario A1B and B1 - aggregate observations to GCM resolution - use reference historical period (1950-1999*) to bias-correct the monthly GCM time series for the full length of the simulation (e.g., 1900-2099). Bias correct both the mean and the distribution. - interpolate bias-corrected monthly GCM values to the output grid resolution - using reference historical period (1950-1999*), reproduce observed spatial variability by calculating "deltas" between GCM and observed values at each grid point. Apply these to GCM anomalies in order to obtain the final 1/2 degree monthly time series. - temporally disaggregate data from monthly to daily time scale using a random sampling of observed daily variability. Preserve monthly mean values. The result is a transient daily time series for the full length of the GCM simulation, showing the progression from 20th century to late 21st century climate conditions. The primary advantages of this method are that it extracts more information from the GCM simulations, and that it provides a transient time series as opposed to an estimate for a fixed future decade. As above, these strengths are also the main weaknesses of this method: this method is more susceptible to GCM shortcomings, and the transient time series complicates interpretation, since changes resulting from natural (e.g., interannual) variability must be distinguished from any identifiable long term trend. ----------------- *note: the reference period used for the BCSD downscalings (1950-99) is not the same as that use for the spatial delta downscalings (1948-2000). This is because the BCSD downscaling method is sensitive to spurious values in the dataset, a problem that is present for some gridcells in the first two years of the forcing dataset. The spatial delta method, in contrast, is not sensitive to such variations. ----------------- -------------------------------------------------------------------------------------------------- 4. VIC MODEL CALIBRATION VIC model parameters are based on the work by Nijssen et al 2001. The soil parameters used in calibration for this study includes: - the variable infiltration parameter (bi), - the depth of the second soil layer (D2), - the saturated hydraulic conductivity of the first and second layer and their exponents. Based on our experience of model sensitivities, we conducted further calibration of: bi, D2, and in addition D3 (soil depth of third layer) The following summarizes the calibration approach and optimized soil parameters for the Sacramento, Skeena, Yukon, and Fraser basins. Calibration was conducted on a monthly basis comparing error statistics of streamflow on a monthly timestep. The calibration period is 10/1975 to 9/1990. The validation period is 10/1990 to 9/2000. Error statistics used in calibraton include: - Nash Sutcliffe efficiency (NSeff) - higher is better - ln (NSeff) - higher is better - R2 - higher is better - mean ann volume error in KAF (AMVE), computed as abs((sim_annvol/obs_annvol)-1) - lower is better - route mean squared error in KAF (RMSE), computed as RMSE/mean_obs - lower is better - peak difference in KAF (PDiff), computed as abs((sim_peak/obs_peak)-1) - lower is better Here we provide error statistics for results using calibrated soil parameters only. ------------------ (a) Skeena ------------------ Calibration was performed at Skeena River at USK (08EF001). A 1/2 degree routing network was not available for this watershed so summed monthly runoff+baseflow was compared with observed natural flows for calibration. We used the MOCOM-UA autocalibraton scheme (Yapo et al, 1998), modified by scientists at the Pacific Climate Impacts Consortium (PCIC) at the University of Victoria, British Columbia. Error statistics after calibration and validation: CALIBRATION PERIOD VALIDATION PERIOD NSeff 0.807 0.828 ln(NSeff) 0.549 0.670 R2 0.848 0.867 PDiff 0.057 0.012 RMSE 0.418 0.399 AMVE 0.193 0.185 ------------------ (b) Sacramento ------------------ Calibration was performed at Sacramento River at Bend Bridge (A02785). A 1/2 degree routing network was not available for this watershed so summed monthly runoff+baseflow was compared with observed natural flows for calibration. We used the MOCOM-UA autocalibraton scheme (Yapo et al, 1998), modified by scientists at the Pacific Climate Impacts Consortium (PCIC) at the University of Victoria, British Columbia. Error statistics after calibration and validation: CALIBRATION PERIOD VALIDATION PERIOD NSeff 0.794 0.600 ln(NSeff) -5.5 -4.547 R2 0.866 0.693 PDiff 0.008 0.041 RMSE 0.453 0.627 AMVE 0.206 0.270 ------------------ (c) Fraser ------------------ Calibration was performed at Fraser River at Hope (08MF005). A 1/2 degree routing network was available for this site. We used the MOCOM-UA autocalibraton scheme (Yapo et al, 1998), modified by scientists at the Pacific Climate Impacts Consortium (PCIC) at the University of Victoria, British Columbia. Error statistics after calibration and validation: CALIBRATION PERIOD VALIDATION PERIOD NSeff 0.813 0.845 ln(NSeff) 0.706 0.764 R2 0.863 0.879 PDiff 0.007 0.059 RMSE 0.322 0.301 AMVE 0.161 0.127 ------------------ (d) Yukon ------------------ Calibration was performed at Yukon at Pilot Station (USGS 15565447). A 1/2 degree routing network was available for this site. We used the MOCOM-UA autocalibraton scheme (Yapo et al, 1998), modified by scientists at the Pacific Climate Impacts Consortium (PCIC) at the University of Victoria, British Columbia. Error statistics after calibration and validation: CALIBRATION PERIOD VALIDATION PERIOD NSeff 0.737 0.500 ln(NSeff) 0.390 -0.365 R2 0.837 0.506 PDiff 0.160 0.063 RMSE 0.420 0.934 AMVE 0.258 0.006 -------------------------------------------------------------------------------------------------- 5. REFERENCES: Adam, J.C. and D.P. Lettenmaier, 2003: Adjustment of global gridded precipiation for systematic bias. J. Geophys. Res., 108, D9. 4257. Daly, C. et al., 2002: A knowledge-based approach to the statistical mapping of climate. Climate Research, 22 (2), 99-113. Elsner et al., 2010: Implications for 21st century climate change for the hydrology of Washington state. Climatic Change, 102 (1-2), 225-260. Jones, P.D., 1994: Hemispheric surface air temperature variations: A reanalysis and an update to 1993. J. Climate, 7, 1794-1802. Hamlet A.F. et al., 2010: Statistical downscaling techniques for global climate model simulations of temperature and precipitation with application to water resources planning studies. Chapter 4 in Final Project Report for the Columbia Basin Climate Change Scenarios Project, http://www.hydro.washington.edu/2860/report Huffman, G.J. et al., 1997: The Global Precipitation Climatology Project (GPCP) Combined Precipitation Dataset. Bull. Amer. Meteor. Soc., 78, 5-20. Hulme, M., 1995: Estimating global changes in precipitation. Weather, 50, 34-42. Jones, P.D., 1994: Hemispheric surface air temperature variations: A reanalysis and an update to 1993. J. Climate, 7, 1794-1802. Kalnay E. et al., 1996: The NCEP/NCAR 40-year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437-471. Nijssen, B. et al., 2001: Global retrospective estimation of soil moisture using the variable infiltration capacity land surface model, 1980-93. J. Climate, 14, 1790-1808. Werner, A.T., 2011: BCSD Downscaled Transient Climate Projections for Eight Select GCMs over British Columbia, Canada. Pacific Climate Impacts Consortium, University of Victoria, Victoria, BC, 63 pp. http://www.pacificclimate.org/sites/default/files/publications/Werner.HydroModelling.FinalReport1.Apr2011.pdf Wood A.W. et al., 2002: Long range experimental hydrologic forecasting for the eastern U.S. J. Geophys. Res., 107 (D20): 4429 Wood A.W. et al., 2004: Hydrologic implications of dynamical and statistical approaches to downscaling and statistical approaches to downscaling climate model outputs. Climatic Change, 62 (1-3): 189-216 Yapo, P., H. Gupta, and S. Sorooshian, 1998. Multi-objective global optimization for hydrologic models. J. Hydrology, 204(1), 83-97. -------------------------------------------------------------------------------------------------- For additional questions, please contact: Marketa McGuire Elsner (mmcguire at uw dot edu) or Guillaume Mauger (gmauger at uw dot edu)