[Objective] Traditional evaporation models are limited in floating ball coverage scenarios by complex parameterization, reliance on sensible heat flux data, and cumbersome aerodynamic calculations. This study aims to develop a highly accurate evaporation prediction formula with simplified parameters, adaptable to varying coverage ratios, to provide a reference for water resource quantification. [Method] Based on the Priestley-Taylor model, a modified model was developed by introducing correction term g(m) to address evaporation prediction under floating ball coverage. By analyzing the relationship between coverage ratio and latent heat flux, exponential (for high-temperature seasons) and linear (for transition seasons) correction forms were determined. Robust regression was employed to minimize the impact of outliers. The optimal model was selected through model comparison and cross-validation. [Result] (1) The response relationship between latent heat flux and floating ball coverage ratio exhibited significant monthly variations. At high-temperatures during high-radiation season (June-August), the exponential correction yielded the best fit (R²≥0.985), whereas the linear correction performed better during the transition seasons (March-May and September-October) (R²≥0.986). (2) The exponential model demonstrated superior adaptability and predictive advantages for floating ball evaporation. Over the entire experimental period, it achieved a Willmott’s index of agreement (D) of 0.986, a Nash-Sutcliffe efficiency (NSE) of 0.947, a root mean square error (RMSE) of 0.54 mm/d, and a mean absolute error (MAE) of 0.43 mm/d, with strong parameter stability and no systematic bias. (3) The model also showed good adaptability during external validation under high-coverage conditions (NSE=0.734, D=0.890). [Conclusion] The proposed modified model establishes a unified prediction framework based on basic meteorological parameters for varying floating ball coverage ratios. It enables precise estimation of evaporation processes across different coverage levels, providing an efficient and practical computational method for water resource management in arid regions.