Modeling long-term progressive erosion

at the West Valley site

Erosion Working Group modeling team:

Gregory E. Tucker1, Katherine R. Barnhart1, Sandra G. Doty, Rachel Glade2,

Mary C. Hill3 , Matthew Rossi1, Charles M. Shobe1

1 – CIRES and Department of Geological Sciences, University of Colorado, Boulder

2 – INSTAAR and Department of Geological Sciences, University of Colorado, Boulder

3 –Department of Geology, University of Kansas

West Valley Quaterly Public Meeting, February 28, 2018

1

Approach involves six major steps

DEVELOPA SET OFEROSIONMODELS

ANALYZESENSITIVITY

TO INPUTS & PARAMETERS

CALIBRATEBASED ON

GEOLOGIC PAST (13,000 YRS)

VALIDATEUSING

SECONDWATERSHED

IDENTIFYBEST-

PERFORMINGMODELS

PROJECT POTENTIALFUTURE EROSION,WITH QUANTIFIED

UNCERTAINTIES

2

Models simulate long-term erosion at gridded locations in a drainage basin

• Developed 37 process models

• Each model incorporates:– Mass movement

– Hydrology

– Channel/gully erosion

– Material properties

• Grid resolution is 24’

3

SDA

WVDP

Erosion Working Group Study 1 data allow reconstruction of past topography and downcutting history

modern topography

alternative reconstructions of paleo (~13 ka) topography, with post-glacial ravines filled in

3 k

m

Downcutting history at outlet

(data from Erosion Working Group Study 1;R. Young & M. Wilson)

4

Input parameter ranges are informed by results from Erosion Working Group field studies

Soil / till erodibility

Soil infiltration capacity

Channel grainsize

SOURCE: S. Bennett (2017)Report of the West Valley Erosion Working Group Study 2: Recent Erosion and Deposition Processes.

5

Sensitivity analysis shows low sensitivity to downcutting history or paleo-topography

EXAMPLE OF SENSITIVITY TO PARAMETERS, INITIAL CONDITIONS, AND LOWERING HISTORY FOR MODEL “BasicRt” 6

K1 = till erodibilityK2 = rock erodibilityD = soil transport efficiencyWc = contact-zone thickness

Models and parameters are tested by comparing observed and simulated modern topography

MODERN TOPOGRAPHY (CENTER) COMPARED WITH FOUR MODEL RUNS

7

Calibration used to test and rank models and identify best parameters

• At least two possible reasons for a poor fit:1. Poor model2. Great model but wrong parameter choice

• Calibration provides:– Optimal parameter values– Measure of goodness of fit for each model

• Calibration performed on CU’s Summit supercomputer– Project overall required over 1.3 million CPU hours– 34 of 37 successfully calibrated

8

Moderntopography

9

Example Calibration: Basic Model(rank 25 of 34)

10

Observed versus modeled terrain:Basic model

OBSERVED BASIC MODEL (rank 25 of 34)

11

Calibration example #2:Model “BasicChRtTh”(rank 1 of 34)

DRAFT calibration,to be revised.Model BasicChRtTh.Duration 13,000 years.

12

Observed versus modeled terrain:Erosion threshold, nonlinear hillslope law, rock and till

OBSERVED BASIC MODELWITH EROSION THRESHOLD,NONLINEAR HILLSLOPE LAW,AND ROCK AND TILL UNITS

(rank 1 of 34)

DRAFT calibration,Model BasicChRtTh.Duration 13,000 years.

13

SUMMARY OF CALIBRATION RESULTS

LOWER SCORE (VERTICAL AXIS) INDICATES A BETTER-PERFORMING MODEL 14

Models were validated by running on a nearby watershed of similar size and relief

OBSERVED

SIMULATED

15

SDA

WVDP

Models that performed well in calibration also performed well in validation tests

• Top 9 models in calibration and validation selected for erosion projection

• Top-performing models distinguish between glacial sediments & bedrock16

Future projections quantify uncertainty in five main areas:

• Future climate: run three alternative scenarios

• Future downcutting on Buttermilk: run three alternative scenarios

• Terrain modification by humans: run ensemble of simulations with random +/-5’ elevation perturbations

• Model structure: run 9 different models

• Model parameters: propagate calibration uncertainty forward into prediction (seven models only due to compute time limits)

17

Sensitivity tests examine uncertainty from two additional sources:

• Potential for upper Franks capture by gully: run capture-from-southeast scenario

• Potential for rapid Buttermilk widening: run capture-from-east scenario

18

Scenarios for future climate were developed using MACA climate-model downscaling product

Data source: Multivariate Adaptive Constructed Analogs (MACA) Datasetshttps://climate.northwestknowledge.net/MACA/

19

Three future climate

scenarios1. Representative Concentration

Pathway (RCP) 8.5: Increase mean wet day totals to 2100, then stabilize

2. Representative Concentration Pathway (RCP) 4.5: Increase mean wet day totals that level off by 2100

3. No change in mean wet day precipitation

20

MEAN ANNUAL PRECIPITATION

MEAN WET-DAY PRECIPITATION

PRECIPITATION FREQUENCY

Three scenarios for future downcutting on Buttermilk Creek

21

Erosion projections plotted for 25 selected points at site

• Time intervals of 100 years

• All model and scenario projection runs store data for every grid location

• Parameter uncertainty runs focus only on the 25 points

22

SDANDA

Dominant source(s) of uncertainty may vary from one location to another, and through time

• Sources include:

– Unknown future climate

– Unknown future rate of lowering in surrounding areas

– Small variations or perturbations in topography

– Parameters in erosion models

– Model structure

• Side-by-side comparison of projections with two different models illustrates model structure uncertainty

23

MODEL “BasicRt”10,000-year run

Lowering scenario 2

MODEL “BasicChRtTh”10,000-year run

Lowering scenario 2

Example of model structure uncertainty

24

Example of uncertainty in initial topography

(representing human modification of landscape)

25

800 802 842

0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0

1300

1325

1350

1375

Model Time (ka)

Ele

vatio

n (

ft)

lowering_future

lowering_future_1

lowering_future_2

lowering_future_3

climate_future

constant_climate

RCP45

RCP85

sew SDA2 IC Uncertainty

MODEL

Example of erosion

projections at a point, with uncertainty arising from

parameter value uncertainty

26

800

0.0 2.5 5.0 7.5 10.0

1320

1340

1360

1380

Model Time (ka)

Ele

vaio

n (

ft)

lowering_future

lowering_future_1

lowering_future_2

lowering_future_3

climate_future

constant_climate

RCP45

RCP85

At-a-point predictions with uncertainty bounds,combining all quantified

uncertainty sources

27

SDA4 SDA5 SDA6 UFrankEdge1 UFrankEdge2

ProcessBLD QuarryEdge SDA1 SDA2 SDA3

NDA1 NDA2 NDA3 NDA4 NDA5

HLWT1 HLWT2 Lagoon2 Lagoon3 LFrankEdge

ErdmanEdge GullyHead1 GullyHead2 GWPlume1 GWPlume2

0.0 2.5 5.0 7.5 10.00.0 2.5 5.0 7.5 10.00.0 2.5 5.0 7.5 10.00.0 2.5 5.0 7.5 10.00.0 2.5 5.0 7.5 10.0

1200

1250

1300

1350

1400

1200

1250

1300

1350

1400

1200

1250

1300

1350

1400

1200

1250

1300

1350

1400

1200

1250

1300

1350

1400

Model Time (ka)

Ele

va

tio

n (

mu

lti−

mod

el expecte

d v

alu

e ±

2σ

, ft

)

Method

Model 842 Only

Nine Model Average:Independent

Nine Model Average:Combined

Model−CalibrationUncertainty Approach

Independent

Combined

Comparison of Three Uncer tainty Quantification Methods

28

Relative contributions of different sources of uncertainty, by location and time

Example of ensemble-based projected erosion maps

29

EXPECTED EROSION UPPER 95% PERCENTILELOWER 95% PERCENTILE

9-M

OD

EL C

OM

PO

SITE

Summary of uncertainty results

• Major sources of uncertainty in future erosion estimates include:– Initial topography / human modification of landscape

– Model structure

– Model parameters

• Other sources are:– Climate

– Downcutting in Buttermilk valley

• Degree of uncertainty and relative importance of different sources varies among locations

30

= Model selection and calibration}

Erosion modeling provides information for further erosion assessment:

• Calculations of potential future erosion at each model grid cell

• Calculations include quantitative estimates of uncertainty in model structure, future climate, initial topography, and future Buttermilk Creek downcutting

• Estimates of uncertainty arising from model parameters are provided 7 models at 25 selected points– Workflow and codes available to perform calculations for other models

and/or locations

• Scenarios also calculated for potential capture of upper Franks Creek by gully erosion to the southeast or Buttermilk valley widening near Heinz Creek fan

• Process model results provide basis for probabilistic modeling of erosion using multiple alternative scenarios

31

QUESTIONS?

32