Now we have more numbers I tried to make an approximation by a computational model. The model assumes a linear dependency of range on temperature, and a dependency of range on speed that is a combination of linear and quadratic. For the latter part of the model, for higher speeds, such as 110 km/h, the quadratic part dominates by a factor 4 to 5 over the linear part, for lower speeds such as 50 km/h this factor is 1.5 to 2.
As our available population of cases shows a lot of variation, the model naturally deviates from these individual cases. For example, in some of the high temperature cases concerning the Jeju Island rally, a very good hypermiler was selected as driver, whereas in Bjorn Nyland's Winter Tests he did not try to drive in an economic way, and sometimes used the sporty mode, and besides, he had to fight very bad weather conditions in the second part of his trip.
Due to this variation over the different cases, the average deviation between model outcomes and reported data over all cases is about 20 km. Most of this average is determined by the most extreme cases. If two or three most extreme cases are left out (by considering them as socalled outliers), the average deviation between model outcomes and reported data becomes about 10 km, which seems not too bad. Below a comparison between reported data and model outcomes is shown. As you can see for 10 of the 13 cells the difference is around 10 km or even less, but for 3 of them (in red) the difference is more than 20 km. Still keep in mind that the model describes average drivers and circumstances, and extreme cases will not be approximated well.