|
Bayesian modelling - Predicting where proteas occur naturallyFor each map the probability of a species presence is mapped on a scale from green (very low) to red (very high). There are 10 shades shown each of about 10%, but sometimes the intervals are lower as the upper limit is the highest level of prediction for the species. Some species have a high prediction (almost 100%), whereas some rarer species are poorly predicted. The highest level of prediction is given below as a percentage. Not shown is the varience (or uncertainty) of the predictions - these are available for each species. Usually the "funny areas" have a high uncertainty. Also, poorly predicted species usually have a high uncertainty as well - in these cases it is only by comparing the probability versus the variability that a reasonable picture of distribution can be achieved. Aulax cancellata 90.5% Aulax pallasia 95.4% Aulax umbellata 99.8% Brabejum stellatifolium 99.8% Diastella fraterna 97.1% Diastalla myrtifolia 61.6% - not very high but a good map! Diastalla thymelaeoides 98.8% Hakea drupacea 98.3% Hakea gibbosa 95.5% Leucadendron album 99.2% Leucadendron arcuatum 98.8% Leucadendron argenteum 99.3% Leucadendron barkarae 87.9% Leucadendron bonum 76.3% - not too good! Leucadendron brunioides 97.1% Leucadendron burchellii 56.0% - not very good! Leucadendron cadens 94.0% Leucadendron conicum 99.7% Leucadendron flexuosum 59.5% - not a good prediction! Leucadendron laureolum 99.92% Leucadendron modestum 99.9% Leucadendron muirii 98.0% Leucadendron nervosum 75.5% Leucadendron rubrum 99.5% Leucadendron salignum 100.0% - top that! Leucadendron xanthoconus 99.8% Protea acaulos 99.8% Protea acuminata 94.2% Protea amplexicaulis 98.8% Protea angustata 94.6% Protea aristata 86.4% Protea aurea 95.0% Protea caespitosa 97.5% Protea canaliculata 98.3% Protea convexa 90.6% Protea cordatum 93.4% Protea coronata 99.6% Protea coronata Comparison of Protea Atlas data and Bayesian modelling Protea coronata Protea Atlas data on its own. Protea coronata A closeup showing Atlas data (black dots) for Pr coronata (coloured dots) and the model predictions. Note the fits and filling in of gaps! Protea compacta 99.0% Protea cryophila 85.3% Protea cynaroides 99.998% - beat that! Protea decurrens 77.8% Protea denticulata% Protea effusa 87.8% Protea eximia 99.4% Protea foliosa 96.4% Protea glabra 99.2% but some leakage into the Karoo Protea holosericea 62% Protea humiflora 97.2% Protea inopina 28.2% - one of the worst! Not much of a prediction! Protea intonsa 94.8% Protea laevis 99.4% Protea lanceolata 99.0% Protea laurifolia 99.91% Protea lepidocarpodendron 99.7% Protea longifolia 99.3% Protea lorea 76.2% - not too good! Map not too bad though. Protea lorifolia 99.8% Protea magnifica 93.9% Protea montana 92.6% - something funny happened here : this cannot be Pr montana! Interestingly all the high altitude Swartberg endemics give the same problem. Protea montana Closeup: From the Protea Atlas data the Bayesian Modelling is all wrong - any ideas? Protea nana 91.7% Protea neriifolia 99.6% Protea nitida 99.86% Protea obtusifolia 99.5% Protea pendula 96.7% Protea piscina 98.8% Protea pruinosa 73.6% - not too good! Another Groot Swartbert endemic. Protea punctata 99.97% Protea recondita 92.9% Protea repens 99.97% Protea restionifolia 84.1% Protea revoluta 96.1% Protea rupicola 89.2% Protea scabra 98.3% Protea scorzonerifolia 76.7% - not too good! Map not too bad though. Protea speciosa 99.5% Protea scolymocephala 94.8% Protea scolopendriifolia 97.5% Protea scabriuscula 97.2% Protea stokoei 56.0% - not very good! Map not too bad though. Protea subulifolia 95.2% Protea sulphurea 97.4% Protea susannae 99.1% Protea tenax 95.8% Protea venusta 77.7% - not too good! Another Groot Swartbert endemic. Protea vogtsiae 93.1% Protea witzenbergiana 95.6% Summarizing species richness. Two maps are provided: 1. The Protea Atlas data showing richness as atlassed: this varies from 0 (no colour) through green ( = 1 species) to red ( > 18 species) to a maximum of 38 species 2. The Bayesian prediction showing richness obtained by summing all the probabilities for the modelled species above: This varies from a minimum of 0.75 species (green ( < 1.75 species) to 17.5 species (red > 15.8 species) Note how the species richness is greatly exaggerated by the modelling - we may have to correct for uncertainty to get a better fit. Back Protea Ecology |