A limitation of this is that the researchers do not have a current analogue for the anticipated temperatures, yet to overcome this the researchers looked at data form the fossil record, using particularly information from the middle Eocene. This study used two climate models, the EC-Earth model and the IPSL-CM5A-LR model. The researchers mapped the current and future sea surface temperatures by using simulations of global oceans for the years 2005-2014, 2050-2060, and 2090-2100. There were 204 fossil occurrences of coral reefs and 8789 current occurrence, to overcome the difference in the amount of data, random sampling occurred from the current occurrences to match the number from the fossil record. The random sample from current data and the data from the fossil record were coupled and used to model the suitability of thermal conditions for the growth of coral reefs. A generalized linear models with binomial distribution was used to gain a curve. Then to evaluate the model the team ran runs with a split of 70% of the data used to model and the remaining 30% used to verify the predictions; this was able to prove the information
A limitation of this is that the researchers do not have a current analogue for the anticipated temperatures, yet to overcome this the researchers looked at data form the fossil record, using particularly information from the middle Eocene. This study used two climate models, the EC-Earth model and the IPSL-CM5A-LR model. The researchers mapped the current and future sea surface temperatures by using simulations of global oceans for the years 2005-2014, 2050-2060, and 2090-2100. There were 204 fossil occurrences of coral reefs and 8789 current occurrence, to overcome the difference in the amount of data, random sampling occurred from the current occurrences to match the number from the fossil record. The random sample from current data and the data from the fossil record were coupled and used to model the suitability of thermal conditions for the growth of coral reefs. A generalized linear models with binomial distribution was used to gain a curve. Then to evaluate the model the team ran runs with a split of 70% of the data used to model and the remaining 30% used to verify the predictions; this was able to prove the information