Hydrologic Predictability in the Mississippi River Basin An Investigation of Continental-Scale Runoff Variability and Persistence June 6, 2002 Ed Maurer

Embed Size (px)

DESCRIPTION

How We Experience Hydrologic Variability Droughts Floods Extreme Events Source: NOAA, Hydrologic Info. Ctr.

Citation preview

Hydrologic Predictability in the Mississippi River Basin An Investigation of Continental-Scale Runoff Variability and Persistence June 6, 2002 Ed Maurer University of Washington Source: NASA Motivation 1)Better observation and prediction of climate signals and their teleconnections to land areas 2)Improved understanding of continental-scale hydrologic variability through data collection and modeling 3)Potential benefits of improved long-lead prediction How We Experience Hydrologic Variability Droughts Floods Extreme Events Source: NOAA, Hydrologic Info. Ctr. Current Water Scarcity Relative Water Scarcity Ref: Vrsmarty et al., Science, 2000 Increasing Water Scarcity Qualitative index provides indication of high water scarcity High variability exacerbates water shortages Increasing population and irrigation add greater stress Accommodating Hydrologic Variability 3) Improved Water Management New Data Sources Improved Models Better PredictionsBetter Predictions 2) Water Reallocation 1) Water Resources Development Science Questions 1)How can relative influences of land surface and climate on continental scale runoff variability be determined? 2)What is the potential predictability of runoff at seasonal to annual lead times? 3)Where might the greatest improvement in seasonal forecasts be expected with improved observations? Specific Objectives 1)Describe Variability of Land Surface Water Budget Observations Modeling 2)Determine Long-Lead Hydrologic Predictability attributable to: Land Surface States with Long Memory (soil moisture) Slowly Varying Remote Climate Signals 3)Identify where and when the greatest improvement in seasonal forecasts might be expected with improved observations. The Land-Surface Water Budget EP W Q (Near Surface) Water Balance Equation Examine variability in water budget components Need long records of observations to define variability and predictability Precipitation Observations Precipitation (Daily) appears well defined, generally since 1948 Station density within the continental U.S.: 1 per 700 km 2 Station data can be supplemented by radar (WSR88-D stations, beginning ) Evapotranspiration Observations Ameriflux program (flux towers) provides some direct measurements of E, since mid 1990s Approximately Station Density (continental U.S.): 1 per 130,000 km 2 Scattered Field Campaigns provide some additional data Snow Water Content Observations About 600 SNOTEL sites in western US Snow water content measured since 1977 Used in Water Supply Outlooks Can be supplemented by satellite-derived products from NOHRSC, since 1990 Only snow areal extent New instruments show better characterization Soil Moisture Observations Spatial coverage poor at continental scale Only IL covers more than 10 years Observational database is expanding Source: A. Robock, Rutgers U. Runoff (Streamflow) Observations Streamflow in the U.S. measured at roughly 7,000 active gauging stations. Stations can represent regulated flow conditions Streamflow is a spatially integrated quantity Determining Runoff Variability Returning to water balance equation: Q observed as streamflow P is the well defined for appropriate averaging period E is virtually unobserved W (Soil moisture and snow water) not well known We cannot close the water budget with observations! The Need for Hydrologic Modeling To derive W and E, use P (and T) to drive a hydrologic model Reproduce observed Q By water balance, E must be close Physically-based land surface representation can provide information on variability Source: Wilks, 1995 Hydrologic Model VIC Model Features: Developed over 10 years Energy and water budget closure at each time step Multiple vegetation classes in each cell Sub-grid elevation band definition (for snow) Subgrid infiltration/runoff variability 3 soil layers used Non-linear baseflow generation Modeling Domain Domain is North America between 25 and 53 N Resolution 1/8 77,000 grid cells through domain (56,000 in Continental U.S.) Model developed for 15 sub-regions Model forcing data ( ) derived from observations Run at a 3-hour time step Gridding Temperature and Precipitation Data Within the U.S.: Precipitation adjusted for time-of- observation Precipitation re-scaled to match PRISM mean for Precipitation and Temperature from gauge observations gridded to 1/8 o Avg. Station density: AreaKm 2 /station U.S Canada2500 Mexico6000 Derived Meteorological Variables Certain meteorological variables not well observed Use parameterizations to derive them from better know variables (T min, T max, P) Humidity (Vapor Pressure) MTCLIM - T dew estimated from T min (with aridity index based on P annual and R solar ) Downward Solar Radiation transmissivity estimated from T dew, T max T min Downward Longwave Radiation estimated from T average, humidity, atm. transmissivity Wind daily wind speed from NCEP/NCAR reanalysis Soil Characterization Soil parameters: derived from Penn State State STATSGO in the U.S., FAO global soil map elsewhere. Land Cover Determination Land Cover/Vegetation: from the University of Maryland 1-km Global Land Cover product (derived from AVHRR) 14 Land Cover Classes Published Values for each Land Class: Rooting Depth Roughness Height Min. Stomatal Resistance Leaf Area Index from remotely-sensed global product of Myneni Model Simulation for 50-Years Using VIC model, simulation run for 50 years at 3-hour time step Input Time series of spatial data One terabyte of output archived Evaluation of Energy Parameterization Comparison with 4 SURFRAD Sites 3-minute observations aggregated to 3-hour Average Diurnal Cycle is for June, July, August Peak underestimated 3- 15% at each site (avg. 10% for all sites) Daily average within 10%, (avg. 2%) Obs VIC Comparison of Simulated and Observed Runoff Daily grid cell runoff routed to edge of grid cell Daily grid cell outflow routed through river network to observation point. Monthly hydrographs of routed daily flows are compared to observations 6 Sample Hydrographs Good agreement of Seasonal cycle Low Flows Peak Flows Model Obs. Comparison with Illinois Soil Moisture 19 observing stations are compared to the 17 1/8 modeled grid cells that contain the observation points. Persistence Moisture Level Moisture Flux Variability Obs. Model Soil Moisture - Active Range 50-Year Soil Moisture Range Scaled by Annual Precipitation Long term spatial data set allows characterization of variability Dynamic range of the soil column Degree to which source of variability (P) is buffered by soil column Level of hydrologic interaction of soil column Case Study - Estimation of Runoff Predictability Mississippi River Basin Climatic Variability Land Cover Variability Aggregation of Grid Cells Precipitation and Runoff - Mississippi River Basin High P in SE in Winter and Spring Summer has broader patterns of P Winter runoff high in SE High Winter P, Summer runoff in NW Scaled Runoff Magnitude Seasonal average runoff unitless High Values Indicate Importance of Runoff Predictability Factors Exhibiting Persistence Climate Signal ENSO, AO Precipitation and Temperature (and Evaporation) Anomalies Land Surface State Snow Soil Moisture Heat and Moisture Drainage and Surface Runoff Melt Unpredictable Noise Seasonal Average Runoff Persistence tendency to retain memory of earlier state Two sources of seasonal persistence in runoff: Land Surface Climate Signal Unpredictable Noise Climatic Sources of Runoff Predictability ENSO Predictable SST up to 1 year in advance Events may persist up to 2 years Arctic Oscillation Affects winter climate in the Northern Hemisphere. Varies on many time scales. Persistence through 2 months Source: NOAA Source: JISAO, U. Washington Monthly Variation of Climate Indices SOI Exhibits Long-term Persistence AO Fluctuates Rapidly AO Includes some slowly-varying signal Methods for Determining Predictability Indices Characterizing Sources of Predictability: SOI An index identifying ENSO phase AO An index of phase of the Arctic Oscillation SM Soil moisture level (normalized) SWE Snow water equivalent (normalized) Varying Lead Times between IC and Forecast Runoff Forecast Season DJF Initialization Dates for DJF Forecast Dec 1 Mar 1Jun 1Sep 1 Lead-0Lead-4Lead-3Lead -2Lead-1 DJ FMA M JJ A SO N Climate Land Predictability: due to linear relationships of runoff with indices at some initialization time Multiple linear regression used between IC and runoff Variance explained (r 2 ) indicates level of predictability Variables introduced in order of how well indices represent current knowledge of state: 1.SOI/AO 2.SWE 3.SM Incremental predictability Methods 2 r 2 SOI r 2 SM Runoff SOI SM Methods 3 Test for Significant Predictability (r 2 ) in 2 steps Local Significance: Tested at each grid cell Effective time between samples accounts for temporal autocorrelation in series Degrees of freedom derived from this 95% confidence level estimated Field Significance (Livezey and Chen, 1983): Tests area showing local significance over entire basin Accounts for limited sample size, spatial correlation in both predictors and predictand Significance levels determined using Monte Carlo simulations 95% confidence for field significance Total Runoff Predictability Lead, months Uses all 4 indices to predict runoff X no field significance Predictability deteriorates with time Predictability due to Climate Signals Moderate levels of r 2 Predictors currently available Greater influence in winter, in area and lead time Difficulty in long- lead persistence prediction with climate signals Predictability Due to Snow r 2 represents incremental increase Focus at 1 or more season lead is in Rocky mountains At level of Mississippi basin, predictability limited to 1-2 seasons Analysis by sub-area could reveal greater predictability Predictability due to Soil Moisture Widespread predictability at 0 lead (1 month) Winter Runoff: little predictability where runoff is high Summer Runoff: limited predictability to 3 seasons 2 season lead:Important predictability limited Snowmelt signal in W mountains 1 season lead:mild climate influence in SE snow influence on summer runoff in W 3 season lead: climate signal provides predictability of winter runoff in SE 0 season lead:soil moisture signal dominant snow signal dominant in W in summer climate signal strong in SE in winter Scaled Runoff Predictability Conclusions At a lead-0 (1.5 month), soil moisture is dominant for predictive capability of runoff At lead times over 1 season, limited potential forecast skill from snow in west and climate signal in east Important runoff forecast skill at long lead times is limited, and is due to modest predictive skill in areas with high runoff Implications of Added Predictability How can the potential added predictability affect water management? Missouri River Basin Predictability study provides context for: Focus on Missouri River Basin Determine utility of changes in predictability Assess recent changes in remote sensing Aside: Climate Model Water Budget Assessment Derived Pseudo- observations used in coupled model comparison Used as benchmark for water budget simulations Helped Assess: oSeasonal runoff cycle oSoil Moisture Variability oWater Budget Closure Improvement in Snow Characterization MODIS Gridded Observations NOHRSC April 4, 2000 MODIS: images since early 2000 NOHRSC: current operational product Recent study shows improvement THE END? Second Topic Identifying Appropriate Temporal and Spatial Scales for Detecting Global Water Cycle Intensification Adapted from AGU presentation, May 29, 2001 by: Justin Sheffield 1, Alan Ziegler 1, Edwin P. Maurer 2, Bart Nijssen 2, Eric F. Wood 1, Dennis P. Lettenmaier 2, Anthony Broccoli 3 1 Princeton University 2 University of Washington 3 NOAA/GFDL Is the global water cycle intensifying? How to detect intensification? Trends in water fluxes: P, E, Q, ds/dt What is the required time to detect a significant trend? Where and to what extent should monitoring be done? Questions 1.Determine Continental-scale variability in P, E, Q (from VIC simulation) 2.Determine predicted trend for each continent for P, E, Q (from GCM run) 3.Estimate minimum time required to detect warming-induced trends Methods Variable Infiltration Capacity LSM 14-year ( ) Water balance variables Global: six continents 2 grid cells Variance derived for 14-year period (summarized in Nijssen et al., J. Climate 2001) GLOBAL VIC SIMULATION Datasets: variance Trends determined from PCM run B IPCC A2 scenario: worst plausible case A2 (IPCC, 2001) Dataset: trends Period: Period: Mann-Kendall trend Mann-Kendall trend Global Trends P: 0.6 mm y -1 P: 0.6 mm y -1 E: 0.4 mm y -1 E: 0.4 mm y -1 Q * : 0.2 mm y -1 Q * : 0.2 mm y -1 Predicted Trends PT min E Q = trend mm y -1 S = std. deviation of annual total, mm Trends and Variability Minimum number of years ( n min ) required to detect a trend Time series variance (VIC) Trend magnitude (GCM) = probability of NOT detecting intensification when it is actually occurring = probability of detecting intensification when there is none Trend Detection Low Risk/High Detectability: = 0.05; = 0.10 Results: Minimum years for detection units: years P only Levels of Risk Preliminary results indicate decades may be required to detect trends in P, E, Q Detectability depends on risk acceptance Conclusions