I have an article under revision where I created a regression model (Random Forests) to estimate a certain X variable. For that model I have reported the coefficient of determination (R2) and the root mean squared error (RMSE). Both R2 and RMSE are calculated from an independant dataset. However, the reviewer is saying that my reported R2 and RMSE are fit statistics only. He suggests that Actual prediction errors for a variable with no in situ observations are much larger. Is he correct ? Given that my R2 and RMSE are from predictions and observations not involved when training the model.