Photovoltaic (PV) monitoring is commonly performed through individual system indicators such as Performance

Ratio (PR). Although useful, these indicators often identify faults only after measurable performance degradation has

already occurred. Drone-based visual and thermal inspection can provide physical diagnosis but it is periodic not

continuous. We propose an information-theoretic framework for early-warning monitoring of co-located PV systems

using Transfer Entropy (TE). Instead of treating each PV system as an isolated unit, the proposed approach models a

solar facility as a directed network of interacting time series. Three years of 5-minute resolution data from the Desert

Knowledge Australia Solar Centre (DKASC) in Alice Springs were analyzed which covers 12 co-located PV systems

and 43,044 clean daytime observations from 2019 to 2021. TE was compared with Pearson correlation and Mutual

Information (MI) as symmetric baseline measures. Results show that all 132 directed system pairs exhibit significant

TE, with values ranging from 0.029 to 0.139 bits and a 4.7× spread in information-flow strength. The TE network

reveals identifiable source-sink behavior, with M1_A, M7_A, and M2_C acting as dominant sources and M4_A and

M5_A acting as dominant sinks. Fault-adjacent periods show significantly higher TE than normal periods, with mean

TE increasing from 0.0772 to 0.0945 bits (approximately 22%). Welch’s t-test and the Kolmogorov-Smirnov (KS)

test confirm statistical significance. Early-warning case studies further suggest that TE may rise before visible PR

degradation, supporting its potential for risk-aware PV monitoring, predictive maintenance, and inspection planning.