The analysis of the frequency response of integrated transmission-distribution networks with deep penetration of solar photovoltaic (PV) generation faces major challenges due to the complexity emerging from the dynamic models of the numerous and diverse PV units involved. This article proposes converter-based dynamic equivalent models for both distributed (distribution network-connected) and large-scale (transmission network-connected) PV units, which take into account practical issues, such as measurement and coordination delays. Different from the previous work that adopted an open-loop identification, the unknown model parameters are identified here through a novel closed-loop identification process based on the least-square minimization. This allows capturing the continuous interaction between system and PV responses, thus improving the outcome of the overall frequency response model. The proposed models are validated with the real data from the August 2018 separation event in Australia. The results demonstrate the excellent performance of the proposed models in determining the frequency response from PV in both transmission and distribution networks, hence paving the way to its adoption in the frequency stability analysis in low-carbon grids dominated by frequency-responsive renewables.