1

Comparison of forest-snow process models (SnowMIP2): uncertainty in estimates of snow

water equivalent under forest canopies

Nick Rutter and Richard Essery

Centre for Glaciology

University of Wales Aberystwyth

nick.rutter@aber.ac.uk

IUGG, Perugia, Italy, 10 July 2007

2

Why, what and how?

Current Land Surface Schemes (LSS) in models either neglect or use

highly simplified representations of physical processes controlling the

accumulation and melt of snow in forests

Snow Model Inter-comparison Project 2 (SnowMIP2)

Quantify uncertainty in simulations of forest snow processes

Range of models of varying complexity (not just LSS)

Primarily evaluate the ability of models to estimate SWE

32 models

5 locations: 3 presented herein (Switzerland, Canada, USA)

2 sites per location: forest and clearing (open)

IUGG, Perugia, Italy, 10 July 2007: Why, what and how?

3

Model inputs

IUGG, Perugia, Italy, 10 July 2007: Model inputs

Meteorological driving data :

Precipitation rate (rain and snow)

Incoming SW and LW

Air Temperature

Wind Speed

Relative humidity

Site specific data:

Tree height

Effective LAI

Instrument heights

Snow free albedo

Soil composition

Initialisation data:

Soil temperature profile

Soil moisture

Calibration data:

In-situ snow water equivalent

(SWE) from Year 1, forest sites

4

Model outputs

IUGG, Perugia, Italy, 10 July 2007: Model outputs

Energy fluxes

Radiative, turbulent, conductive, advected and phase changes

Mass fluxes

sublimation, evaporation, transpiration, phase change, infiltration, runoff,

unloading, drip

State variables

mass (solid and liquid), temperature

5

Model complexity

Model complexity is traditionally defined using numerous computational

and snowpack metrics (multi-metric categorisation):

Dimensions, purpose, lines of code, number of layers, process replication

e.g. Snow Modellers Internet Platform (Zong-Liang Yang)

Snow modelling metrics rarely go beyond presence or absence of

canopy / vegetation

Really require metrics describing: gap fraction, LAI (PAI), turbulent

exchange, LW absorption and emittance

IUGG, Perugia, Italy, 10 July 2007: Model complexity

6

Model complexity

Still need to populate a vertical axis describing canopy

representation in models to create a 2-D plot of complexity

Your input at IUGG SnowMIP2 workshop is appreciated!

High complexityLow complexity

e.g. Snow-17e.g. ISBA

& CLASS

e.g. SNOWPACK

& SNOWCAN

Medium complexity

Degree day models

Land surface schemes

Snow-physics models

PLUS 27 OTHER MODELS

S N O W

IUGG, Perugia, Italy, 10 July 2007: Model complexity

7

Alptal, Switzerland (47°N, 8°E)

IUGG, Perugia, Italy, 10 July 2007: Results

OPTIONAL CALIBRATION UNCALIBRATED

UNCALIBRATED ‘CONTROL’

8IUGG, Perugia, Italy, 10 July 2007: Results

BERMS, Saskatchewan, Canada (53°N, 104°W)

9IUGG, Perugia, Italy, 10 July 2007: Results

Fraser, Colorado, USA (39°N, 105°W)

10

Model uncertainties: forest vs open

IUGG, Perugia, Italy, 10 July 2007: Results

11

Model uncertainties: forest vs open

IUGG, Perugia, Italy, 10 July 2007: Results

x 2.16

x 2.42 x 1.35

x 1.59

x 1.17

x 1.53

12

Model uncertainties: forest vs open

IUGG, Perugia, Italy, 10 July 2007: Results

x 1.02

x 5.35

x 3.00

x 1.40

x 1.37

x 1.05

13

Conclusions

IUGG, Perugia, Italy, 10 July 2007: Conclusions

32 models successfully participated in SnowMIP2 – thanks to the hard

work and goodwill of the participants

Low correlation coefficients suggests good model performance at

forest sites do not necessarily mean good model performance at open

sites (and vice versa)

One year of calibration data in forest models is not necessarily good

enough for subsequent years

It is easier to model inter-annual variability at open sites than forest

Uncertainty still exists whether 1) strength of calibration or 2) canopy

complexity makes inter-annual variability of forest snow hard to model