Driven by the carbon neutrality strategy, high-precision State of Charge (SOC) estimation for lithium batteries has become a core technical challenge in battery management systems. However, most current lithium battery SOC estimation methods are limited to constant ambient temperatures and ignore the effects of environmental temperature. Therefore, this paper proposes a SOC estimation method using an Environmental Temperature-Aware Dual-Stream gated recurrent unit (GRU) Fusion Network (ETA-GRU). The model employs a dual-stream GRU architecture to separately encode primary battery variables (voltage, current, and surface temperature) and ambient temperature covariates, while integrating a dominant-covariate cross-attention (DCCA) mechanism to dynamically capture the impact of temperature on electrochemical states. Experimental results based on a public datasets demonstrate that under dynamic ambient temperature conditions (10 ~ 25 ℃) and dynamic load scenarios, the proposed ETA-GRU achieves an average absolute error (MAE) of 0.54% and a root mean square error (RMSE) of 0.65% for SOC estimation, outperforming mainstream models (e.g., PA-LSTM, DA-GRU, and Transformer) by 53.8~60.0%. Ablation experiments verified the effectiveness of each component, and the experimental results indicated that the environmental temperature flow and DCCA contributed to performance improvements of 67.8 and 32.2%, respectively. This study provides a solution that integrates theoretical innovation and engineering practicality for high-precision SOC estimation under complex operating conditions, demonstrating significant potential for real-world engineering applications.