It is certainly the global irradiance on the plane of interest, which combines direct radiation plus diffuse radiation either from the sky or from the ground (this component may be high if there is a high albedo like for example that of snow or sand).
The total incident solar radiation on a surface is the sum of three components: beam radiation, radiation reflected by the ground (surrounding area) and sky diffuse radiation. having these data, you can use the isotropic model with the Lui and Jordan formula.
Many models were developed in this subject in the aim of representing the realistic total radiation reached by the panels and then the output power. The radiation, as mentioned above, is the sum of three components. You have to choose between models and with regard of the site climatic condutions, you can compute the total radiation reached by the panel. Besides isotropic model, the anisotropic model was elaborated by introducing corrections on the isotropic model. It is often used by scientis. Please take a look on my contributions for further information.
if you use software tools (such as PVsyst) to perform the analysis of a PV system.
you need 12 monthly daily values for each variable: 12 global solar radiation on horizontal plane values and 12 ambient temperature values. if available you can use also 12 diffuse irradiance values and 12 wind speed values.
Definitely Global Solar Radiation is better to be considered here. However the DNI has higher contribution to the estimated or predicted power output of the PV panel.