Photovoltaic (PV) output exhibits strong randomness and multi-scale temporal volatility. In long-term system planning, scenario generation and reduction are required to balance scenario diversity and computational complexity. However, existing methods primarily depend on distance or mean error criteria, struggling to systematically uncover the intrinsic relationship between reduction strategies and scenario characteristics. This paper proposes an evaluation method for PV scenario reduction based on multiple discrepancy metrics and Spearman correlation. First, the K-means clustering is applied to compress PV output scenarios into diverse reduced sets, which are then extrapolated into annual time series via a Markov chain model. Subsequently, the optimal sizing of PV and energy storage systems is executed across these scenarios. Finally, the Spearman rank correlation coefficient is used to quantify the correlation between scenario discrepancy metrics and optimization results, thereby establishing a comprehensive evaluation model weighted by these coefficients. The results show that specific configuration outcomes fluctuate by over 5% across different reduced and reconstructed scenarios, proving that scenario discrepancy features dictate the sensitivity and stability of optimization results. This paper offers a quantitative basis for selecting and assessing scenario reduction techniques in renewable energy planning.