How to Apply Process Information to Improve Prediction Jim McNamara Boise State University Idaho, USA Precipitation • Improved prediction and improved process understanding are mutually reliant flow time time Modified from Mukesh Kumar Lumped Model Q(t ) f ( P(t ), A, C ) Distributed Model, Semi-Distributed Model, Conceptual REW 2 p Physics based .(U ) .( ) Qss t REW 3 REW 4 REW 1 REW 5 REW 7 REW 6 q q q Process Representation: Parametric Physics-Based Predicted States Resolution: Coarser Fine Data Requirement: Small Large Computational Requirement: Perceived Intellectual Value: Small Large Modified from Mukesh Kumar Lumped Model Q(t ) f ( P(t ), A, C ) Distributed Model, Semi-Distributed Model, Conceptual REW 2 p Physics based .(U ) .( ) Qss t REW 3 REW 4 REW 1 REW 5 REW 7 REW 6 q q q Outcome: Right for Wrong Reasons Wrong for Right Reasons History: Mathematical Lumping Process Understanding ? Process Understanding Future: In Defense of Hydrologic Reductionism … an approach to understand the nature of complex things by reducing them to the interactions of their parts… …a philosophical position that a complex system is nothing but the sum of its parts, and that an account of it can be reduced to accounts of individual constituents … My Past Berkely Catchment Science Symposium 2009 My Past: In Defense of Reductionism • Newton was right • Model failures result from poor characterization of heterogeneous landscapes leads to – No emergent properties • Our community struggles to identify grand, overarching questions because…there are no grand unknowns • Hydrology is a local science The Response Ciaran Harman, Catchment Science Symposium, EGU 2011 The Response Ciaran Harman, Catchment Science Symposium, EGU 2011 Catchments Lump Processes Emergent Behavior Decades of case studies have documented the many ways that water moves downhill Recent work has identified many Physically Lumped Properties that are manifestations of the system of states and fluxes -A physical basis for lumped parameter modeling Physically Lumped Properties (emergent behavior) • Connectivity • Thresholds • Residence Time 600 Total Precipitation Water Input 400 Evapotranspiration 0.2 7/2 15 cm 8/31 10/30 30 cm 65 cm Bedrock Flow 12/29 2/27 4/28 6/27 30 20 0.1 10 0 60 7/2 0 Streamflow 8/31 10/30 12/29 2/27 4/28 6/27 8 dissolved solids 40 20 0 5 7/2 8/31 10/30 12/29 2/27 4/28 6/27 (mg/L) Soil Moisture 0 Bedrock Flow (mm) 200 Streamflow (liters/min) • Connectivity • Thresholds • Residence Time Depth (mm) Physically Lumped Properties Physically Lumped Properties • Connectivity • Thresholds • Residence Time Modified from Mukesh Kumar Lumped Model Q(t ) f ( P(t ), A, C ) Distributed Model, Semi-Distributed Model, Conceptual REW 2 p Physics based .(U ) .( ) Qss t REW 3 REW 4 REW 1 REW 5 REW 7 REW 6 q History: Future: q q Mathematical Lumping Process Understanding ? Process Understanding Modified from Mukesh Kumar Distributed Model, Physics based Physically Lumped Model .(U ) .( ) Qss t Q(t ) f ( P(t ), A, C ) Physically lumped properties q History: Future: Mathematical Lumping Process Understanding Process Understanding Physically lumped properties How to Apply Process Information to Improve Prediction • Retain the computationally efficiency and lumped philosophy of systems models • Observe how catchments create physically lumped properties • Replace mathematical lumping approaches with physically lumped properties – Use as validation targets – Build into new model structures What do we do with this awareness? Connectivity Thresholds Residence Time Lump the lumps It’s about Storage P-ET-Q =dS/dt Storage Connectivity Thresholds Residence Time A Tale of Two Catchments Runoff Ratio % Old Water Flow (cms) Thaw Depth (cm) A Natural Storage Experiment 60 Storage Capacity 40 20 0 6 4 2 0 85 13-Jun 3-Jul 23-Jul 12-Aug 75 65 0.4 13-Jun 13-Jul 12-Aug 0.2 0.0 13-Jun 13-Jul 12-Aug • Storage regulates fluxes (ET, Recharge, Streamflow) • Storage is responsible for emergent behavior such as connectivity, thresholds, and residence time % Old Water • The mechanisms by which catchments STORE water ultimately characterize the hydrologic SYSTEM Runoff Ratio P-ET-Q =dS/dt Flow (cms) Thaw Depth (cm) The Case for Storage 60 Storage Capacity 40 20 0 6 4 2 0 85 13-Jun 3-Jul 23-Jul 12-Aug 75 65 0.4 13-Jun 13-Jul 12-Aug 0.2 0.0 13-Jun 13-Jul 12-Aug • We should concern ourselves with how catchments Retain Water in addition to how they release water % Old Water • We should focus on Runoff Prevention mechanisms in addition to runoff generation mechanisms Runoff Ratio P-ET-Q =dS/dt Flow (cms) Thaw Depth (cm) A Natural Storage Experiment 60 Storage Capacity 40 20 0 6 4 2 0 85 13-Jun 3-Jul 23-Jul 12-Aug 75 65 0.4 13-Jun 13-Jul 12-Aug 0.2 0.0 13-Jun 13-Jul 12-Aug The Storage Problem • Storage is not commonly measured • Storage is often estimated as the residual of a water balance • Storage is treated as a secondary model calibration target CUAHSI Catchment Comparison Exercise Girnock, Scotland, Rain, humid Dry Creek, Idaho, USA Snowy, semi-arid, ephemeral Panola, Georgia, USA Rain, humid, perennial Reynolds Creek, Idaho, USA Snowy, semi-arid, perennial Gårdsjön, Sweden, Snow, ephemeral Storage Zones Snow Vegetation Surface Soils Bedrock S t Storage Time Series by Direct Measurement Main and Seep Channels Soil Moisture Sample Locations Weir Weather Station Precipitation Gage 52 51 50 North 55 45 46 47 44 43 42 54 1654 56 30 1646 29 1638 31 32 28 27 26 48 250 41 40 39 38 33 34 25 35 36 37 Treeline Study Site 24 22a 23 1630 16 17 18 11 10 9 1606 20 8 12 13 1590 snow 21 7 6 1598 Meters 200 22b 19 1614 15 5 3 4 2 1 14 Boise 0 50 100 150 soil moisture Dry Creek Experimental Watershed 200 Meters Stored Water (mm) 58 57 1622 Total 150 100 50 0 10/1 11/30 1/29 3/29 2003-2004 5/28 7/27 9/25 Storage-Discharge Signature of geology? Storage (mm) 10000 Dry Creek Gårdsjön Girnock Reynolds Panola 1000 S = 786.Q0.32 S = 253Q0.11 S = 251Q0.05 S = 219Q0.06 100 S = 87Q0.05 10 0.0 0.1 1.0 10.0 100.0 Streamflow (mm) McNamara et al., 2011 Storage-Discharge Storage (mm) 10000 Similar dynamic range? Dry Creek Gårdsjön Girnock Reynolds Panola 1000 S = 786.Q0.32 S = 253Q0.11 S = 251Q0.05 S = 219Q0.06 100 S = 87Q0.05 10 0.0 0.1 1.0 10.0 100.0 Streamflow (mm) McNamara et al., 2011 Relative Storage-Discharge 1.2 Relative Storage S/Smax 1 0.8 0.6 0.4 Dry Creek Gardsjon Girnock Reynolds Panola 0.2 0 0.0 2.0 4.0 6.0 8.0 10.0 Streamflow (mm) 12.0 14.0 16.0 McNamara et al., 2011 Relative Storage-Discharge 1.2 Relative Storage S/Smax 1 Course soils drain rapidly No deep groundwater Active storage uses limited range 0.8 0.6 0.4 Dry Creek 0.2 0 0.0 2.0 4.0 6.0 8.0 10.0 Streamflow (mm) 12.0 14.0 16.0 McNamara et al., 2011 Relative Storage-Discharge 1.2 Swedish catchment uses larger range at a wetter state Relative Storage S/Smax 1 0.8 0.6 0.4 Dry Creek Gardsjon 0.2 0 0.0 2.0 4.0 6.0 8.0 10.0 Streamflow (mm) 12.0 14.0 16.0 McNamara et al., 2011 Relative Storage-Discharge 1.2 Southeast US Highly seasonal Connected to deep groundwater Uses large range of storage Relative Storage S/Smax 1 0.8 0.6 0.4 Dry Creek Gardsjon Panola 0.2 0 0.0 2.0 4.0 6.0 8.0 10.0 Streamflow (mm) 12.0 14.0 16.0 McNamara et al., 2011 Relative Storage-Discharge 1.2 Relative Storage S/Smax 1 0.8 0.6 0.4 Dry Creek Gardsjon Girnock Reynolds Panola 0.2 0 0.0 2.0 4.0 6.0 8.0 10.0 Streamflow (mm) 12.0 14.0 16.0 McNamara et al., 2011 Improved storage characterization will lead to improved prediction Reynolds Creek Dry Creek Precipitation Boise Front annual precipitation 1999-2009 1200 precipitation (mm) 1100 1000 900 2000 2002 2004 2006 2008 2001 2003 2005 2007 2009 800 700 600 500 400 300 200 100 800 1000 1200 1400 1600 elevation (m) 1800 2000 Precipitation (mm) precipitation precipitation (mm)(mm) 200 150 963 mm 77% Snow 100 50 0 october january april 200 july 335 mm 32% Snow Rain 150 Snow 100 50 0 october january april 2008 Water Year july Distributed Soil Moisture Measurements - Aspect Moisture and Aspect Volumetric Moisture Content 0.30 0.20 0.10 0.00 S O N D J F M A M J J A S O S O N D J F M A M J J A S O 0.30 S O N D 0.20 0.10 0.00 J F M A M J J A S O S O N D J F M A M J J A S O 0.30 0.20 0.10 0.00 0.30 0.20 0.10 0.00 2008 Water Year Smith et al., in review Soils properties vary with aspect 140 South aspect Soil Depth (cm) North aspect 120 100 80 60 40 20 0 0 Smith, T., in progress, MS Thesis 1 2 3 Smith et al., in review • North facing aspects retain more water than south facing aspects Both Aspects 1000 900 800 Tension (cm) 700 600 500 400 300 200 100 0 0 0.2 0.4 0.6 3 3 Volumetric Water Content (cm /cm ) Geroy et al., in review Soil Moisture Storage (mm) More Storage on North aspects Except at high elevation 300 S=(soil depth)(moisture content) 200 -Higher moisture content 100 -Higher moisture retention 0 -Deeper soils 300 7/ 7/ 8/ 9/ 10 11 12 1/ 2/ 3/ 4/ 5/ 6/ 7/ 8/ 9/ -Finer grained soils 200 1 31 30 29 /2 /2 /2 27 26 28 27 27 26 26 25 24 -Lower insolation 100 9 8 8 0 7/ 7/ 8/ 9/ 10 11 12 1/ 2/ 3/ 4/ 5/ 6/ 7/ 8/ 9/ 1 31 30 29 /2 /2 /2 27 26 28 27 27 26 26 25 24 9 8 8 300 200 100 0 July Sep 1-Jul 30- Nov 29- Jan 28- Mar 26- May 27- Jul 26- Sep 25- Aspect-Insolation-Soil Carbon • Soil organic carbon content increases with aspect and elevation S N S N S Up Hill S Kunkel et al., in review Geomorphology and Aspect North facing slopes are steeper and shorter N (ish) Poulos et al., in review Storage Capacity • Rooted in the co-evolution of landscape form and hydrologic processes Both Aspects 1000 900 800 Tension (cm) 700 600 500 400 300 200 100 0 0 0.2 0.4 0.6 3 3 Volumetric Water Content (cm /cm ) • Responsible for catchment-scale emergent behavior – Physical Lumped Properties – Connectivity, Thresholds, Residence time Our Modeling Experience • Soil Water Assessment Tool (SWAT) Hydrograph “Right” Storage Wrong 3.5 250 SWAT Soil Moisture Observed Average Soil Moisture BFU and BFL Date Oct-07 Jul-07 Apr-07 Oct-06 Jul-06 0 Apr-06 Jan-05 Oct-04 Jul-04 Apr-04 Jan-04 Oct-03 Jul-03 Apr-03 Jan-03 Oct-02 Jul-02 Apr-02 0.0 50 Jan-06 0.5 100 Oct-05 1.0 Jul-05 1.5 150 Apr-05 2.0 Jan-02 Flow (m^3/s) 2.5 200 Jan-05 Soil Moisture of Soil Colmn (mm) Predicted Jan-07 3.0 Date Stratton et al., 2009 VERY Physically-Based 1D Kelleners et al., 2010 VERY Physically-Based 1D • Simulates Snow accumulation, melt, infiltration, and Bedrock infiltration with “NO” shortcuts – Over 70 equations just for 1 dimension • NOT PRACTICAL • Allows us to determine the relative importance of physical controls • • • • • • • Wavelength-dependent solar radiation Iterative canopy energy balance => Tleaf Iterative surface energy balance => Tsurface Snow water flow Soil water flow Snow-soil-bedrock heat transport Snow-soil water phase change Kelleners et al., 2010 0.4 50 measured calculated 5 cm 0.2 0.1 0 Jun-03 Oct-03 Jan-04 Apr-04 Aug-04 Nov-04 soil temperature (Celsius) 0.3 measured calculated 40 5 cm 30 20 10 0 -10 Jun-03 Oct-03 Jan-04 Apr-04 Aug-04 Nov-04 0.4 50 60 cm 0.3 0.2 0.1 0 Jun-03 Oct-03 Jan-04 Apr-04 Aug-04 Nov-04 Root Mean Square Error: 20-38 % Modeling Efficiency: 0.65-0.86 soil temperature (Celsius) volumetric soil water content volumetric soil water content VERY Physically-Based 1D 40 60 cm 30 20 10 0 -10 Jun-03 Oct-03 Jan-04 Apr-04 Aug-04 Nov-04 Root Mean Square Error: 11-28 % Modeling Efficiency: 0.88-0.95 Very Physically Based – 3D Kelleners et al., 2010 Very Physically Based – 3D • Catchment is divided into 141 grid cells (1010 m) s u r fa ce ru n of f vertical soil water flow perched watertable soil depth H bedrock lateral subsurf ace flo w qdp * j Qstreami Qstreami j * i 2 2 i* q dp Kelleners et al., 2010 H D k sr D Very Physically Based – 3D • Decent simulation of soil moisture, unsatisfactory simulation of streamflow – Wrong for the right reasons Distributed Hydrology Soil Vegetation Model (DHSVM) Evaluate the impact of “improved” soil depth information on streamflow and soil moisture simulations SSURGO Soil Depth Modeled Soil Depth Soil Depth: Field Data Soil depth measurement - 2.2 m rod - 1.27 cm diameter - Pounded to refusal - 2 or 3 repeats 819 points (calibration) - 8 subwatersheds - 130 random points (testing) - During Spring when soil was moist Fence Post Pounder Copper Coated Steel Rod 52 Predictor Variables Symbol Description sca Specific catchment area from the D method. This is contributing area divided by the grid cell size modcurv Curvature modeled based on field observed curvature. ang The D flow direction: the direction of the steepest outwards slope reported as the angle in radians counter-clockwise from east avr Average D vertical rise to ridge lspv Longest vertical slope position lvs Longest D vertical drop to stream slpg Magnitude of topographic slope computed using finite differences on a 3x3 grid cell window sd8a Slope averaged over a 100 m path traced downslope along D8 flow directions elv Elevation above sea level plncurv Plan curvature: the curvature of the surface perpendicular to the direction of the maximum slope pc1 First principal component from ERDAS IMAGINE Newly derived variables Land cover variable 53 Measured vs Modeled Soil Depth Testing Training and Testing NSE = 0.58 95% Confidence interval NSE = 0.47 1:1 Tesfa et al., 2009 54 Predicted Soil Depth Map Soil Depth Map of Dry Creek Experimental Watershed Soil Depth (cm) Value High : 374.366 Contour Low : 16.7556 DHSVM 4 Recorded and Simulated stream flow (2002-2003) Recorded MODSD Simulated SUSD Simulated 1 2 Calibration to streamflow obliterates effect of soil information 0 Stream flow (m3/s) 3 Insignificant difference with improved soil depth information Oct Dec Feb Apr Jun Aug Oct Months Dec Feb Apr Jun Aug Oct Soil Capacitance Model (Reynolds Creek) Snow Water Input (ISNOBAL) SWI Get the inputs right (accumulation, STORAGE,and ablation of snow) Get the 1D soil water storage right Ignore all lateral movement No calibration to streamflow See what happens Throughflow Seyfried et al., 2009, Hydrological Processes Soil Capacitance Model (Reynolds Creek) SWI • Throughflow occurs when soil column water holding capacity is exceeded • Soil water storage parameterized by field capacity, plant extraction limit, soil depth S i numlayer T i 1 i i S FC i numlayer i 1 T fci i 0.3 Throughflow Soil Moisture Ө0.2 fc 0.2 65 cm 0.1 Өpel 15 cm 0.1 30 cm 0 0 7/2 Seyfried et al., 2009, Hydrological Processes 8/31 10/30 12/29 2/27 4/28 6/27 Good 1D Performance Distributed Model Distributed energy balance forcing Distributed soil properties by similarity classes No lateral flow simulated Simulated storage excess agrees with streamflow Connectivity Index SN S S PEL S FC S PEL How do we Apply Process Understanding to Improve Prediction? • Revisit the lumped philosophy of systems models • Recognize catchments create physically lumped properties • Replace mathematical lumping approaches with physically lumped properties – Use as validation targets – Build into new model structures • The mechanisms by which catchments STORE water characterize the catchment system • We should concern ourselves with how catchments Retain Water in addition to how they release water – Get storage right, and everything else will work out How do we Apply Process Understanding to Improve Prediction? • …ecohydrology should synthesize Newtonian and Darwinian approaches to science…combining Newtonian principles of simplification, ideal systems, and predictive understanding (often, but not solely embraced by hydrologists) with Darwinian principles of complexity, contingency, and interdependence (often, but not solely embraced by ecologists)…offers the potential for profound and more rapid advances in our understanding of environmental process… Brent Newmann However… • …ecohydrology should synthesize Newtonian and Darwinian approaches to science…combining Newtonian principles of simplification, ideal systems, and predictive understanding (often, but not solely embraced by hydrologists) with Darwinian principles of complexity, contingency, and interdependence (often, but not solely embraced by ecologists)…offers the potential for profound and more rapid advances in our understanding of environmental process… Brent Newmann How do we compare catchments? • M. Robinson (1992) – 20 papers from western and central Europe – Key Conclusion: Intercomparison is difficult How do we compare basins? • Jones and Swanson (2001) – A basin’s streamflow may be predicted by characterizing basin storage capacities in vegetation, soil, and snow… How do we compare basins? • McDonnell and Woods (2004) – Governing principles are known • Heterogeneity rules the day – Possible classification metrics include • Response time of dominant storage How do we compare basins? • Wagener et al. (2007) • Classification is a rigorous scientific inquiry into the causes of similarities and relationships between catchments. How do we compare basins? • Wagener et al. (2007) Runoff generation commonly studied How do we compare basins? • Wagener et al. (2007) Storage uncommonly studied How do we compare basins? • Landscape structure moderates transit times How to Apply Process Information to Improve Prediction • Recognize that the existence of true physically-based models is a myth • Identify physically lumped properties • Build conceptual models based on the ways catchments lump properties, not mathematical – Systems approaches using “essential” parameters How to Apply Process Information to Improve Prediction? • Recognize that the existence of true physically-based models is a myth • Identify hydrologically relevant processes or properties for hydrogeographic regions – Classification • Build models that target relevant hydrologic processes or properties – Systems approaches using “essential” parameters
© Copyright 2024