Impact model: BioScen1.5-MEM-GBM

The BioScen1.5-MEM-GBM models are macroecological models of terrestrial vertebrate richness (amphibians, birds, mammals) using Generalized Boosted Models (GBMs) as model algorithm. BioScen1.5-MEM-GBM are statistical models, which were calibrated using a 30-year average (1980-2009) of the EWEMBI observed climate input data set and species richness maps derived from IUCN and BirdLife range maps of 15496 terrestrial vertebrate species (2964 amphibians, 8493 birds and 4039 mammals). The models are then projected to simulated historical, current and future 30-year average climate data from ISIMIP2b, resulting in species richness estimates for the three different vertebrate taxa.

Sector
Terrestrial biodiversity
Region
global
Contact Person

Information for the model BioScen1.5-MEM-GBM is provided for the simulation rounds shown in the tabs below. Click on the appropriate tab to get the information for the simulation round you are interested in.

Basic information
Reference Paper: Main Reference: Biber M, Voskamp A, Niamir A, Hickler T, Hof C et al. A comparison of macroecological and stacked species distribution models to predict future global terrestrial vertebrate richness. Journal of Biogeography,None,,2019
Resolution
Spatial Aggregation: regular grid
Spatial Resolution: 0.5°x0.5°
Input data sets used
Simulated Atmospheric Climate Data Sets Used: IPSL-CM5A-LR, HadGEM2-ES, GFDL-ESM2M, MIROC5
Observed Atmospheric Climate Data Sets Used: EWEMBI
Other Data Sets Used: Land-sea mask
Climate Variables: tasmax, tasmin, pr
Additional Information About Input Variables: 30-year monthly means of minimum temperature (tasmin), maximum temperature (tasmax) and total precipitation (pr) were calculated and used to derive 19 bioclimatic variables (see Hijmans et al. 2005, International Journal of Climatology). The best combination of 4 variables was selected for each taxon (amphibians, birds, mammals) by comparing the performance of all uncorrelated combinations of 4 variables. As a result, Bio4 (temperature seasonality), Bio5 (max temperature of warmest month), Bio12 (annual precipitation) and Bio15 (precipitation seasonality), were used for modelling birds and mammals and Bio4, Bio5, Bio18 (precipitation of warmest quarter) and Bio19 (precipitation of coldest quarter) were used for modelling amphibians.
Additional Input Data Sets: Species richness data for amphibians, terrestrial mammals and terrestrial birds were obtained by combining expert range maps provided by the International Union for Conservation of Nature (2011) and BirdLife International (2014) gridded to a resolution of 0.5°x0.5°.
Model specifications
Model Algorithm: GBM
Explanatory Variables: Bio4 (temperature seasonality), Bio5 (max temperature of warmest month), Bio12 (annual precipitation) and Bio15 (precipitation seasonality) were used for modelling birds and mammals. Bio4, Bio5, Bio18 (precipitation of warmest quarter) and Bio19 (precipitation of coldest quarter) were used for modelling amphibians.
Response Variable: species richness of taxon
Distribution Of Response Variable: Poisson
Parameters: n.trees=10500, n.minobsinnode = 10, interaction.depth=c(1,2,3), shrinkage=c(0.01, 0.001), bag.fraction=0.5
Additional Information About Parameters: Model performance was evaluated using cross-validation and the model parameters (combination of interaction.depth and shrinkage) that resulted in the lowest cross-validation error were used for the final model. Only models with an AUC of > 0.7 were kept as results.
Software Function: gbm()
Software Package: gbm
Software Program: R
Note of caution
Note of caution about applicability of model use due to modeling concept: Projecting models to pre-industrial and extended future time periods was only performed to allow for comparability with the other modeling sectors and should be used with caution. Historical projections are limited to the species that were modeled and do not include species which went extinct over the last decades or centuries (which we did not model). Historical richness is thus very unlikely to be a realistic estimate of past richness as it is probably highly underestimated. Future and especially extended future projections should also be used with caution, as species interactions and other processes are not considered when running macroecological models even though they do influence species richness and may change in the future, altering the future distribution of species richness. Furthermore, the distribution data used to calibrate the model are range maps drawn by experts based on current and historical distribution records as well as additional information and expert knowledge. Thus, they do not represent the distribution at a specific point in time, but a rough outline of a species current distribution. Therefore, precise projections of individual years are not sensible for this modelling approach.
Basic information
Reference Paper: Main Reference: Biber M, Voskamp A, Niamir A, Hickler T, Hof C et al. A comparison of macroecological and stacked species distribution models to predict future global terrestrial vertebrate richness. Journal of Biogeography,None,,2019
Resolution
Spatial Aggregation: regular grid
Spatial Resolution: 0.5°x0.5°
Input data sets used
Observed Atmospheric Climate Data Sets Used: EWEMBI
Other Data Sets Used: Land-sea mask
Climate Variables: tasmax, tasmin, pr
Additional Information About Input Variables: 30-year monthly means of minimum temperature (tasmin), maximum temperature (tasmax) and total precipitation (pr) were calculated and used to derive 19 bioclimatic variables (see Hijmans et al. 2005, International Journal of Climatology). The best combination of 4 variables was selected for each taxon (amphibians, birds, mammals) by comparing the performance of all uncorrelated combinations of 4 variables. As a result, Bio4 (temperature seasonality), Bio5 (max temperature of warmest month), Bio12 (annual precipitation) and Bio15 (precipitation seasonality), were used for modelling birds and mammals and Bio4, Bio5, Bio18 (precipitation of warmest quarter) and Bio19 (precipitation of coldest quarter) were used for modelling amphibians.
Additional Input Data Sets: Species richness data for amphibians, terrestrial mammals and terrestrial birds were obtained by combining expert range maps provided by the International Union for Conservation of Nature (2011) and BirdLife International (2014) gridded to a resolution of 0.5°x0.5°.
Model specifications
Model Algorithm: GBM
Explanatory Variables: Bio4 (temperature seasonality), Bio5 (max temperature of warmest month), Bio12 (annual precipitation) and Bio15 (precipitation seasonality) were used for modelling birds and mammals. Bio4, Bio5, Bio18 (precipitation of warmest quarter) and Bio19 (precipitation of coldest quarter) were used for modelling amphibians.
Response Variable: species richness of taxon
Distribution Of Response Variable: Poisson
Parameters: n.trees=10500, n.minobsinnode = 10, interaction.depth=c(1,2,3), shrinkage=c(0.01, 0.001), bag.fraction=0.5
Additional Information About Parameters: Model performance was evaluated using cross-validation and the model parameters (combination of interaction.depth and shrinkage) that resulted in the lowest cross-validation error were used for the final model. Only models with an AUC of > 0.7 were kept as results.
Software Function: gbm()
Software Package: gbm
Software Program: R
Note of caution
Note of caution about applicability of model use due to modeling concept: Projecting models to pre-industrial and extended future time periods was only performed to allow for comparability with the other modeling sectors and should be used with caution. Historical projections are limited to the species that were modeled and do not include species which went extinct over the last decades or centuries (which we did not model). Historical richness is thus very unlikely to be a realistic estimate of past richness as it is probably highly underestimated. Future and especially extended future projections should also be used with caution, as species interactions and other processes are not considered when running macroecological models even though they do influence species richness and may change in the future, altering the future distribution of species richness. Furthermore, the distribution data used to calibrate the model are range maps drawn by experts based on current and historical distribution records as well as additional information and expert knowledge. Thus, they do not represent the distribution at a specific point in time, but a rough outline of a species current distribution. Therefore, precise projections of individual years are not sensible for this modelling approach.