The ModelChain cell temperature methods (sapm_celltemp, pvsyst_celltemp and faiman_celltemp) expect that 'poa_global' is present in ModelChain.total_irrad. When running a model chain from e.g. effective irradiance, it would be good for the cell temperature models to fall back to using effective irradiance rather than asking for 'poa_global' to also be specified.
Related to #943