Abstract: Efficient water resource use is often required for precision irrigation in sustainable agriculture. In this study, a decision-making approach was proposed for real-time irrigation to integrate crop growth dynamics with environmental conditions. Irrigation strategies were optimized for water use efficiency in the field. Specifically, a real-time irrigation decision model was developed to integrate the SWAP (soil-water-atmosphere-plant) model, the iterative ensemble smoother (IES) data assimilation, and the multi-objective particle swarm optimization (MOPSO). Field tests were conducted in Yongning County, Ningxia, China, during 2019-2020. SWAP model parameters were calibrated to examine the effects of soil texture on real-time irrigation decision-making. Common loam and sandy loam soil textures were measured using the Hydrus-1D parameter set. Soil water content at 10 cm (SW10) and leaf area index (LAI) were selected as the observation variables, and crop water stress (w_s) was used as the optimization objective to generate reference irrigation schedules. A systematic evaluation was conducted to explore the effects of soil type, assimilation interval, and measurement error on model performance. The results showed that: 1) The reference irrigation schedules represented the differences in water movement and water-holding capacity between loam and sandy loam. Under loam, real-time irrigation was concentrated from 40 to 70 days after emergence, with a total irrigation amount of 195 mm and 23 irrigation events. The yield and irrigation water use efficiency (IWUE) were 12737 kg/hm² and 6.53 kg/m³, respectively. Under sandy loam, the irrigation started earlier and continued until the end of the growth period, because of weak water retention and rapid infiltration and drainage; The total irrigation amount and events increased to 384 mm and 30, respectively, while the yield and IWUE were 13199 kg/hm² and 3.44 kg/m³, respectively. The loam maintained a higher water productivity with less irrigation input, whereas sandy loam required more frequent water replenishment to maintain crop water status. 2) The soil type of assimilated observation shared the influence on accuracy and irrigation decisions. The SW10 model more accurately reproduced irrigation amount and irrigation frequency, compared with the LAI. The relative error (RE) values of irrigation amount and events were 1.16% and 13.13% under loam and 13.44% and 13.93% under sandy loam, respectively, when SW10 was assimilated, whereas those values were 11.49% and 17.74% under loam and 23.20% and 19.30% under sandy loam, respectively, when LAI was assimilated. Otherwise, LAI alone improved the accuracy after simulation, with yield RE values of 1.04% and 1.30% under loam and sandy loam, respectively, which were lower than those in the SW10 model. The soil water stress was also triggered during irrigation, while LAI provided stronger constraints on canopy development, biomass accumulation, and final yield. SW10 and LAI assimilation were combined to simultaneously constrain soil water and crop canopy growth, indicating the best performance under both soil conditions. Under loam, the RE values of irrigation amount, irrigation events, yield, and IWUE were 1.05%, 9.74%, 0.41%, and 1.53%, respectively; Under sandy loam, they were 4.75%, 6.53%, 0.74%, and 6.69%, respectively. Therefore, joint assimilation was more efficient to synchronously evaluate the irrigation schedule, crop growth, and IWUE. 3) Sensitivity to the assimilation interval differed between soil types. There were weak responses to interval changes; Once the interval was extended from 3 to 10 days, the RE of irrigation amount increased only from 1.53% to 2.48%, and the yield RE remained below 2%, indicating that the stable soil water process in loam buffered error propagation within the assimilation cycle. There was more sensitive to the assimilation interval under sandy loam. When the interval increased from 3 to 5 days, the irrigation amount RE increased from 2.44% to 8.53%, but the yield RE remained low at 0.75%. Once the interval reached 10 days, the RE values of irrigation amount, number of irrigation events, and IWUE increased to 17.93%, 9.27%, and 16.57%, respectively. The rapid infiltration, drainage, and depletion in sandy loam shorten the effective time for correction. Therefore, the assimilation interval was kept within 5 d to maintain model reliability. 4) Observation errors markedly reduced the accuracy and decision stability. When observation errors increased from 10% to 50%, the RE of irrigation amount under loam increased from 1.16% to 3.63%, while the yield root mean square error (RMSE) increased from 409.25 to 1045.10 kg/hm². Meanwhile, the IWUE RMSE increased from 0.52 to 1.04 kg/m³. The RE of irrigation amount increased from 6.29% to 24.54% under sandy loam, while the RMSE of irrigation amount increased from 50.17 to 191.26 mm, the yield RMSE increased from 209.87 to 922.92 kg/hm², and the RE and RMSE of IWUE increased to 18.31% and 1.12 kg/m³, respectively. The high-quality observations were attributed to the stronger error amplification under sandy loam, particularly in soil with rapid water movement. Overall, this finding can provide technical guidance for real-time irrigation decision-making under different soil conditions in precision agriculture.