We explore the use of distributed PV power measurements for real-time short-term forecasting of the maximum potential power output of a utility-scale PV power plant, to support future incorporation of PV plants in automatic generation control (AGC) systems. PV plants operating under AGC may run in a curtailed state but must be capable of accurately forecasting their maximum potential non-curtailed power output, or potential high limit (PHL), 10-20 seconds in advance – despite variable environmental conditions. One forecast approach estimates PHL from a subset of inverters, designated reference inverters, which are never curtailed. We propose an alternative approach using in-situ I V tracers to measure maximum power point of reference modules distributed throughout a PV plant. This would allow for greater sampling of variations in irradiance, module temperature, soiling, and albedo, permitting better PHL prediction accuracy. To study potential benefits of this method, we analyzed a one-year data set from an approximately 100 MW PV power plant in the US Southeast, using non-curtailed string combiner outputs as a proxy for distributed module-level power measurements. We compare the reference inverter and distributed power measurement approaches. Results indicate that for highly flexible operation, substantial improvement in prediction accuracy can be achieved using distributed power measurements compared to using a low fraction of reference inverters. We review the study methodology, conclusions, and shortcomings and discuss prospective future work.