See vertical_skill notebook for plots examples, implementation ideas etc.

Summary of qustionnaire ~40 answers

• People use sigma-layers (some use Mike Wave but I guess it also work with sigma)
• Most use fixed profile location in time
• Some use Matlab and excel but most use python and Mike Zero Probably P or M in combination with Mike Zero
• Most people interpolate model depth to exact obs pos. However, ~1/3 binned the obs to model cells (with or without aggregate)
• Clear, ready to use visualizations are highly valued: observed vs modelled vertical profiles
• Vertical profile analysis is a key gap: e.g. depth range stats, and easier profile to profile comparisons.
• Advanced diagnostic plots are requested: e.g. Hovmöller diagrams
• Users have example plots and scripts

What does this mean for Modelskill dev:

1. Focus on sigma-z layers
2. Start with x,y fixed profiles
3. Options to interpolate or bin observations
4. Need dedicated vertical plots
5. Vertical profiles stats, e.g. vertical_grid_stats

Features needed for minimal version

• Support vertical observations with fixed position and depth coordinates. Time coordinate should support non-equidistant timsteps.
• Match 3D sigma model to these observations with binning and interpolation. Get a comparerer.
• Calculate statistics on vertical grids (e.g. rmse with depth)
• Slice Comparer in vertical (results similar as comaprer with pointObs)
• Aggregate Comparerer in vertical (results similar as comaprer with pointObs)
• Plots