This study simulates and predicts nitrogen (N) and phosphorus (P) losses in paddy fields under different irrigation modes and rice varieties in the rice-growing regions of southern China. A field experiment was conducted with two rice varieties (Xiushui and Yongyou) and four irrigation modes: conventional flooding (W0), shallow rain-catching irrigation (W1), improved shallow rain-catching irrigation (W2), and furrow-basin rainfall-adapted irrigation (W3). A model integrating water balance and nutrient concentration dynamics was developed to simulate drainage volume and N/P losses, with parameters calibrated and validated using field-measured data. Results showed that irrigation mode significantly influenced N and P losses. The W0 mode exhibited the highest losses, with total nitrogen (TN) and total phosphorus (TP) losses reaching 8.52 kg N·hm^-2 and 0.72 kg P·hm^-2, respectively. In contrast, the W3 mode reduced TN and TP losses by approximately 39% and 43%, respectively. The exponential decay-based concentration model demonstrated strong performance, with determination coefficients (R2) of 0.68 for TN and 0.62 for TP. Spatial analysis identified eastern Xucun Town and northern Huangwan Town as critical source areas for N and P losses. The total estimated TN and TP losses across the irrigation district were 21.43 tons and 3.69 tons, respectively, with loss coefficients ranging between 0.85%-3.32% for TN and 0.49%-1.37% for TP. The study concludes that optimized irrigation practices, particularly W3, effectively mitigate nutrient losses while maintaining crop water requirements. The proposed model offers a reliable tool for simulating and predicting non-point source pollution in paddy systems, supporting the development of targeted pollution control strategies in agricultural watersheds.