Ecography. Supplementary material

Transcription

1 Ecography E7805 Platts, P. J., Gereau, R. E., Burgess, N. D. and Marchant, R Research article: Spatial heterogeneity of climate change in an Afromontane centre of endemism. Ecography 35: xxx xxx. Supplementary material

2 Supplementary material Spatial heterogeneity of climate change in an Afromontane centre of endemism Philip J. Platts, Roy. E. Gereau, Neil D. Burgess, Rob Marchant Appendix 1. Improving normality for statistical regression For use in multivariate regression (generalised additive models), we attempted to improve the normality of climatic predictors using square root, logarithmic (bases 2, 10 and e), inverse and power transformations. For mean annual temperature and temperature seasonality, no improvement was gained using these methods, but departures from normality were not considered serious (Pearson s correlation coefficient r > 0.98, ordered response vs. normal order statistic medians). For moisture predictors, Box-Cox power transforms were applied. Power transformations work by compressing the right side of the distribution more than the left. For negatively skewed data, we reflected the distribution, adding a constant such that all values were greater than zero, applied the Box-Cox transform [T(X) = (X λ 1) / λ] and then reflected once more to restore the original order of the variable. The chosen λ was that which optimised linear dependence between the (transformed) ordered response and normal order statistic medians (probability plot correlation coefficient; PPCC plot). Annual rainfall Positive skew; Box-Cox, λ = 0.08 Pearson s correlation with normal order statistic medians, r =

3 Annual moisture index Positive skew; Box-Cox, λ = 0.14 Pearson s correlation with normal order statistic medians, r = Dry season water stress Negative skew (data reflected, constant = 91). Box-Cox, λ = 0.13 Pearson s correlation with normal order statistic medians, r =

4 Mean annual temperature and temperature seasonality No transformation improved normality Pearson s correlation with normal order statistic medians, r = (annual mean) and r = (annual range)

5 Appendix st century climate anomalies Future climate anomalies summarised across all EAM forests, and individually by mountain bloc. Forecasts are according to IPCC-AR4 emissions scenarios B1 (white) and A1B (grey), downscaled from a regional climate model (55 km, 0.5, ECHAM5 boundary data) using WorldClim (temperature) and TRMM (rainfall) surfaces as present-day baselines (1 km, ). Whiskers extend up to 150% of the interquartile range from each box. Filled and open circles show means ± standard deviations, respectively. All climates are significantly different to the present day (p < 0.05; Wilcoxon Signed-Rank), unless denoted ns in the plots (e.g., temperature seasonality, B1 2090).

6

7

8

9 Appendix 3. Climates lost and gained Climatic conditions lost and gained in the EAMs, relative to present-day conditions, by the years 2055 and Due to the dispersal limitations of narrow-ranged mountain plants, the search for climate analogues is restricted to individual mountain blocs. Climate lost is the proportional distance of present-day conditions beyond future extremes with respect to, and summed over: mean annual temperature (TM), temperature seasonality (TS), mean annual rainfall (RF) and dry season water stress (WS). Climate gain sums the proportional distance of future conditions beyond present-day extremes. Scatter plots compare no-zero loss and gain (mean across known endemism sites) with endemic plant richness. Regression lines and corresponding p-values (F-tests) show significant negative relationships between the degree of climate loss and endemic richness for both 2055 and 2090 and under both emissions scenarios. Climate gain, less directly associated with species persistence (because familiar conditions may be accessible in neighbouring cells), is significant only for A1B 2090.

10

11

12 Appendix 4. Multivariate species models Of the 56 endemic plant species eligible for multivariate regression (generalised additive modelling with ten or more presence records), 34 produced models with an average cross-validated (five-fold) AUC of at least 0.7. This level of model accuracy, balancing type I and type II error rates on unseen data, was considered sufficiently robust to provide a good indication of the relative importance of climatic gradients in explaining observed distributions. It is worth noting that background (pseudoabsence) data outnumbered presence data (5:1) and were biased towards sites close to known occurrences, providing particularly constrained models. That is, commission error may be lower than the AUC suggests. For these 34 species, we summarise the relative importance of model predictors and directions of change, with respect to range size and altitude, under future climate scenarios.

13 Relative important of predictor variables Selection rate was calculated over 20 model runs: forward-backward and backward-forward stepwise procedures for ten independent realisations of background data. Contribution is the average drop in explained deviance when a variable was removed from the final model. Selection rate (%) Contribution (%) Species Growth form TM TS RF WS TM TS RF WS Allanblackia ulugurensis Tree Allophylus melliodorus Tree Alsodeiopsis schumannii Tree Casearia engleri Tree Chassalia albiflora Shrub Chassalia zimmermannii Shrub Cola stelechantha Tree Craterispermum longipedunculatum Tree Danais xanthorrhoea Herb Diplazium pseudoporrectum Herb Dissotis polyantha Shrub Eugenia toxanatolica Tree Gravesia pulchra Shrub Isolona linearis Tree Lasianthus cereiflorus Tree Lasianthus glomeruliflorus Shrub Lasianthus macrocalyx Shrub Lasianthus pedunculatus Tree Lobelia longisepala Herb Memecylon cogniauxii Shrub Pauridiantha coalescens Shrub Pavetta amaniensis Shrub Pavetta holstii Shrub Plectranthus triangularis Herb Polysphaeria macrantha Tree Psychotria brucei Tree Psychotria megalopus Tree Psychotria pandurata Herb Psychotria porphyroclada Shrub Psychotria triclada Shrub Sorindeia calantha Tree Tarenna roseicosta Shrub Tricalysia aciculiflora Shrub Vangueria fuscosetulosa Shrub All 34 species Mean s.d. ± 32 ± 37 ± 36 ± 40 ± 8 ± 9 ± 10 ± 9 TM, mean annual temperature; TS, temperature seasonality RF, annual rainfall (moisture index [RF/PET] was preferred in 46% of model runs) WS, dry season water stress

14 Taita Hills North Pare South Pare West Usambara East Usambara Nguu Nguru Ukaguru Uluguru Malundwe Rubeho Udzungwa Mahenge Direction of change in climatic suitability Increase ( ) or decrease ( ) in modelled climatic suitability across mountain blocs within each species range. Emissions scenarios B1 and A1B. Pairs of arrows correspond to the years 2055 (left) and 2090 (right). All EAM Allanblackia ulugurensis B A1B Allophylus melliodorus B A1B Alsodeiopsis schumannii B A1B Casearia engleri B A1B Chassalia albiflora B A1B Chassalia zimmermannii B A1B Cola stelechantha B A1B Craterispermum longipedunculatum B A1B Danais xanthorrhoea B A1B Diplazium pseudoporrectum B A1B Dissotis polyantha B A1B Eugenia toxanatolica B A1B Gravesia pulchra B A1B Isolona linearis B A1B

15 Taita Hills North Pare South Pare West Usambara East Usambara Nguu Nguru Ukaguru Uluguru Malundwe Rubeho Udzungwa Mahenge [climatic suitability continued] All EAM Lasianthus cereiflorus B A1B Lasianthus glomeruliflorus B A1B Lasianthus macrocalyx B A1B Lasianthus pedunculatus B A1B Lobelia longisepala B A1B Memecylon cogniauxii B A1B Pauridiantha coalescens B A1B Pavetta amaniensis B A1B Pavetta holstii B A1B Plectranthus triangularis B A1B Polysphaeria macrantha B A1B Psychotria brucei B A1B Psychotria megalopus B A1B Psychotria pandurata B A1B Psychotria porphyroclada B A1B Psychotria triclada B A1B

16 Taita Hills North Pare South Pare West Usambara East Usambara Nguu Nguru Ukaguru Uluguru Malundwe Rubeho Udzungwa Mahenge [climatic suitability continued] All EAM Sorindeia calantha B A1B Tarenna roseicosta B A1B Tricalysia aciculiflora B A1B Vangueria fuscosetulosa B A1B

17 Taita Hills North Pare South Pare West Usambara East Usambara Nguu Nguru Ukaguru Uluguru Malundwe Rubeho Udzungwa Mahenge Direction of change in altitudinal optima Increase ( ), decrease ( ), or stasis (o) in the altitudinal optima of species ranges under future compared with present-day climatic conditions, calculated by summing climatic suitability within 100 m altitudinal bands. Emissions scenarios B1 and A1B. Pairs of arrows correspond to the years 2055 (left) and 2090 (right). All EAM Allanblackia ulugurensis B oo - oo A1B oo - oo Allophylus melliodorus B o A1B Alsodeiopsis schumannii B oo A1B oo Casearia engleri B1 - - oo oo oo A1B - - oo oo o Chassalia albiflora B1 - - o o - o oo A1B - - o o Chassalia zimmermannii B oo oo oo oo A1B oo o oo oo Cola stelechantha B oo o - A1B oo Craterispermum longipedunculatum B o oo A1B o oo Danais xanthorrhoea B1 - - o o A1B - - o o Diplazium pseudoporrectum B A1B Dissotis polyantha B oo - o o - o A1B oo o oo - oo o - Eugenia toxanatolica B1 - - oo oo oo oo A1B - - oo oo oo oo Gravesia pulchra B oo - o o A1B oo - oo o

18 Taita Hills North Pare South Pare West Usambara East Usambara Nguu Nguru Ukaguru Uluguru Malundwe Rubeho Udzungwa Mahenge [altitudinal optima continued] All EAM Isolona linearis B oo o - o A1B oo Lasianthus cereiflorus B A1B Lasianthus glomeruliflorus B oo oo A1B o o Lasianthus macrocalyx B A1B Lasianthus pedunculatus B oo oo - oo oo - oo A1B oo oo o - oo oo - oo Lobelia longisepala B oo - oo - - oo oo oo A1B oo - oo - - oo oo oo Memecylon cogniauxii B oo o A1B - - o o - - oo oo Pauridiantha coalescens B oo - oo A1B Pavetta amaniensis B oo A1B oo Pavetta holstii B1 - - oo oo - oo - oo o A1B - - o o o - o - oo oo Plectranthus triangularis B1 oo - oo - - oo - oo oo A1B oo - oo - - oo - oo o Polysphaeria macrantha B1 - - oo - oo oo - oo A1B - - oo - oo oo - oo Psychotria brucei B1 - - oo oo - oo A1B - - oo oo - oo Psychotria megalopus B oo A1B oo Psychotria pandurata B oo oo oo A1B oo oo oo Psychotria porphyroclada B A1B

19 Taita Hills North Pare South Pare West Usambara East Usambara Nguu Nguru Ukaguru Uluguru Malundwe Rubeho Udzungwa Mahenge [altitudinal optima continued] All EAM Psychotria triclada B oo - oo A1B o o - o o Sorindeia calantha B1 - oo o A1B oo - oo Tarenna roseicosta B o - A1B oo - Tricalysia aciculiflora B A1B Vangueria fuscosetulosa B A1B