I'm a cross-over scientist, half fieldworker and half modeller. I've done enough fieldwork to understand that field experience is essential for real understanding of ecosystems, but I've also done enough modelling to understand that there is critical understanding of large-scale events that can only come from looking at the wider scale.
   Progress requires expertise and input from both camps: fieldworkers need to collect not just data, but model-appropriate data (all data is good, yes, but prediction models are more sensitive to some variables than others so why not let that drive the data collection agenda?) and modellers need not only to have code that works, but code that actually reflects the dynamics of real systems (models need simplifications, of course, but some simplifications are more ecologically reasonable than others and too often simplifications are chosen to optimise runtime rather than realism).
   Models like JULES are very widely-used nowadays and underpin predictions referred to and used by policymakers (e.g. through the IPCC). I am involved in the work of ensuring that we get the very best predictions out of these models that we can. Effectively, I don't say everyone should use models: I say that given that they are now being used in ways that affect policy and therefore all of our lives, I feel it is important to ensure that we get out of them the best results and predictions that we possibly can.