The article discusses creating optimal designs for a non-linear regression function called Mitscherlich's function. The researchers developed designs for both discrete and continuous responses, not just normally distributed ones. They found that these designs can be the same as or different from optimal designs for variance weighted linear regression.