Electronic supplementarymaterial1

Predicting canopy cover changeinTasmanianeucalypt forests using dynamically downscaled regional climateprojections.

GRANT J. WILLIAMSON1*,LYNDA D. PRIOR1, MICHAELR. GROSE2,3, REBECCA M. B. HARRIS2ANDDAVIDM.J.S. BOWMAN1

1School of Plant Science, Universityof Tasmania, Private Bag55,Hobart,Tasmania 7001 ()

2AntarcticClimate and Ecosystems CooperativeResearch Centre(ACE CRC), Private Bag

80 Hobart TAS7000 Australia

3. CSIRO Marineand AtmosphericResearch,107-121 Station Street, Aspendale VIC 3195

*Correspondingauthor

Appendix 1

Figure S1. The distribution of a range of Tasmanian climate variables (black), in comparison to the distribution across the rest of south-eastern Australia (white).

Appendix2.Preliminarymodellingusedto identifythe best-supported variables forthe

Random Forest model.

Preliminarymodellingbased on multi-model inference,generalized linearmodellingand AIC model selection (Burnham and Anderson 2002)was used to select a limited, biologically- relevant suite ofexplanatoryvariables to explaincanopycoverof eucalypt forest, including annual averages and the average July minimum (winter) and average January (summer) maximum temperatures to capture seasonality. Preliminary analyses show that these months are the modal for minimum and maximum temperatures respectively, for both current and projected future climates. Additional variables to capture seasonal shifts in temperature and precipitation were based on the “bioclim” variable suite. We included Bio18 (precipitation of warmest quarter, intended to capture any shift towards or away from summer rainfall), Bio15 (precipitation seasonality), and Bio4 (temperature seasonality). The first suiteof models aimed to identifythe best suiteofclimatevariables to explaincanopycover percentageusingAIC model selection; thesecond identified the best topographic variable in combination with thebest climaticvariables; and thethird thebest water balancevariables. Finally,asuite ofmodelswas constructed combiningvariables from the best climatic, topographic and waterbalancemodels, and the best combined model was selected. Generalized linear modellingwas performed in R2.15.1 (R DevelopmentCoreTeam 2012), with a Gaussian distribution and an identitylink function.

This showed the best-supported set of climatic variables to explainEucalyptus cover in south-eastern AustraliawasMean Annual Precipitation +Mean Maximum January Temperature. Therewasstronger statistical support for addingthe topographic variable Slope than Elevation to the climatic model, and the best-supported water balancevariablewas Annual WaterBalance (Mean Annual Precipitation– Annual Pan Evaporation) (Table S1). Thebest overallmodel was the onethat combinedallfour explanatoryvariables (TableS1).

Table S1Best generalized linear modelsto explain canopycoverfor eachmodel suite, and correspondingAIC values andpercent deviance explained. Lowest AIC values indicate the greatest statistical support.

Model TypeModel VariablesAICDeviance

Explained (%)

ClimateRain_Yr+MaxT_Jan83548.467.8

Climate + Topographic

Rain_Yr+MaxT_Jan +Slope82996.769.3

Water BalanceWB_Yr82971.569.9

CombinedRain_YR +MaxT_Jan +Slope

+WB_Yr

82588.971.0

Appendix 3.Model evaluation

Fig.S2. Actual versus predicted canopycover (%), based on the RandomForest model (R2=

0.806, n = 1,041,030; P < 0.0001).

Fig.S3. Maps showing actual (satellite)tree canopycover, andpredicted cover based on the Random Forest model.There wasgood spatialagreement between theRandom Forest model predictionsand satellite measurements of existingcanopycover forTasmanian eucalypt forest, althoughwith some over-prediction around the boundaryofthe agricultural Midlands region

Appendix 4

Monthly mean maximum and minimum temperatures were used in constructing the eucalypt cover change models, rather than 3-month seasonal climate variables. However, monthly values were highly correlated with the equivalent seasonal mean, as demonstrated in figures S3 and S4.

.

Figure S4. Correlation between July minimum temperature and winter minimum temperature.

Figure S5. Correlation between January maximum temperature and summer maximum temperature.

Appendix 5

Fig S6.Interim Biogeographic Regionalisation for Australia bioregions for Tasmania.