<b>Dataset used for </b><b>Next-Generation Drought Forecasting: Hybrid AI Models for Climate Resilience</b>

Droughts threaten ecological, agricultural, and socio-economic systems, particularly in climate-sensitive regions such as China's Inner Mongolia section of the Yellow River Watershed. This study presents a hybrid AI-based drought forecasting framework that integrates machine learning (ML) and deep learning (DL) techniques with high-resolution climate projections to enhance predictive accuracy and climate resilience. The approach leverages observed data from the TerraClimate archive (1985–2014) and bias-corrected CMIP6 projections (2030–2050) under SSP245 and SSP585 scenarios, sourced from NASA's GDDP v2 dataset. A rigorous preprocessing pipeline—including quantile mapping bias correction and climate-informed feature engineering—was implemented to ensure robust model inputs. Several ML algorithms (RF, XGBoost, GBR, SVR) were benchmarked, and Random Forest was selected based on its predictive strength and interpretability.干旱威胁着生态、农业与社会经济系统，在气候敏感区域尤为显著——例如中国黄河流域内蒙古段。本研究提出了一种基于人工智能的混合干旱预测框架，该框架融合机器学习 (Machine Learning，ML) 与深度学习 (Deep Learning，DL) 技术，并结合高分辨率气候预估数据，以提升预测精度与气候韧性。本研究采用了TerraClimate数据集档案（1985–2014年）的观测数据，以及源自美国国家航空航天局（NASA）GDDP v2数据集的、共享社会经济路径（Shared Socioeconomic Pathways，SSP）245与SSP585情景下经过偏差校正的耦合模式比较计划第六阶段（CMIP6）预估数据（2030–2050年）。同时，研究构建了一套严格的预处理流程，涵盖分位数映射偏差校正与融合气候先验知识的特征工程环节，以确保模型输入数据的鲁棒性。此外，研究对多种机器学习算法（RF、XGBoost、GBR、SVR）开展了基准对比测试，并基于预测性能与可解释性优势，最终选用了随机森林（Random Forest）算法。