How to design a study to estimate responses along an ecological gradient or two

We’ll discuss Kreyling et al (2018) and Chalcraft’s (2019) response.  The context is to understand how some ecological response varies along gradients of environmental drivers, especially when the response might be non-linear.  The classical approach is to choose two or a few levels of the environmental driver and apply them multiple times (replicates).  Kreyling argues for the opposite: choosing many levels with no replication.  Chalcraft contests that. What do you think?  Are there additional issues that haven’t been considered?