The problem could be approached as a regression instead of a classification if If we use crop yield as target variable. A regression to predict a yield value would add some flexibility, because the threshold that triggers an action could be decided by the professional. In our experience, regressions have often returned inaccurate predictions, but such approach should be investigated.
Now you might think "hey, this might work for humans too!" Indeed, there is no reason why it won't.
All computations done for this article can be found in a R Jupyter notebook on GitLab.