Drought is a critical climatic hazard in Iran, posing severe risks to water and food security. While global Teleconnection indices such as the El Ni & ntilde;o-Southern Oscillation (ENSO) and North Atlantic Oscillation (NAO) are known drivers of Iran's climate variability, their relationship with drought is complex, nonlinear, and spatiotemporally unstable. This study overcomes the limitations of traditional linear models by employing eXplainable Artificial Intelligence (XAI) to systematically decode the lagged and nonlinear impacts of a comprehensive suite of 36 teleconnection indices on monthly meteorological droughts across Iran. Based on 30 years of data (1993-2022) from 96 stations, we developed multiple machine learning models to predict the Standardized Precipitation Index (SPI). The Random Forest (RF) model demonstrated superior predictive performance, achieving a coefficient of determination (R2) of 0.40. While modest, this value is robust for predicting chaotic climate phenomena and confirms the model's ability to capture significant underlying signals. We then applied Shapley Additive Explanations (SHAP) to the trained RF model to unlock its "black box". The SHAP analysis quantified the importance, directional influence, and spatiotemporal heterogeneity of all 36 teleconnections at lags of 0-3 months. Our results confirm the predominantly nonlinear nature of these relationships and reveal significant spatial instability, with no single pattern dominating the entire country. Crucially, beyond the well-documented influence of ENSO and NAO proxies, our findings elevate the importance of lesser-known patterns, including the Atlantic Meridional Mode (AMM), the Zonal Wind at 200 hPa (ZWNDz200), the Western Hemisphere Warm Pool (WHWP), and the Western Pacific 850 hPa Zonal Wind index (WPAC850), as key modulators of Iran's drought variability. By identifying region-specific drivers and their critical time lags, this work provides a scientifically robust foundation for developing more accurate operational drought forecasting systems. These physically interpretable insights can directly enhance climate adaptation and water management strategies, demonstrating a critical move beyond linear assumptions to build resilience in vulnerable arid and semi-arid regions.Graphical AbstractThis graphical abstract illustrates the research workflow designed to decode the complex relationships between global teleconnection indices and monthly droughts in Iran using an eXplainable Artificial Intelligence (XAI) approach. The process is depicted in two main tiers: Top Tier (Modeling & Evaluation): This sequence begins with the foundational Data: monthly precipitation records from 96 meteorological stations across Iran (1993-2022) and 36 global teleconnection indices. These inputs feed into a suite of Machine Learning Models (ANN, SVM, XGBoost, RF). An Analyses phase follows, where the models' performance is evaluated using key metrics (MSE, MAE, R2). Based on this evaluation, the Random Forest is chosen as the optimal Model due to its superior predictive accuracy. Bottom Tier (Interpretability & Results): This tier focuses on the study's core innovation: "Opening the black box" of the Random Forest model using Shapley Additive Explanations (SHAP). This powerful Analyses technique allows for the extraction of key Results: (1) A ranked summary of the most effective teleconnection indices, highlighting not only established drivers but also lesser-known influencers (e.g. , ZWNDz200, WPAC850); and (2) A series of spatial distribution maps revealing how the impact of these patterns varies significantly with geographical location and at different time lags (0-3 months). The overarching Conclusion derived from this process is that the XAI framework provides critical physical insights, paving the way for improved drought forecasting and targeted risk management across Iran's diverse regions.