Abstract: This study aims to explore the real water-saving potential in the Hetao Irrigation Area of Inner Mongolia of China. The distributed water cycle and salinity models were constructed using machine learning. A series of water-saving schemes were established to quantitatively analyze the water intake, water consumption, groundwater depth, and salt accumulation. The results were as follows: 1) The distributed water cycle model was used to calibrate and verify some indicators, such as evaporation, runoff, and groundwater depth in the water cycle process. The Nash coefficients of water surface evaporation, drainage processes, and groundwater depth were no less than 0.654, 0.600, and 0.628, respectively, where the absolute relative errors were no higher than 4.82%, 5.11%, and 5.12%, respectively, and the correlation values were 0.88, 0.82, and 0.86, respectively, fully meeting the accuracy requirements. Three algorithms of machine learning were selected to construct the soil salinity model. The Nash coefficients of soil salt accumulation were no less than 0.76, compared with the measurement. 2) Canal and field water-saving measures were optimized to screen the crop structure adjustment. Seven water-saving schemes and combinations were constructed for the main water-saving measures. In Scheme water saving of channel lining (S1), the water utilization coefficient of the canal increased to 0.60, with a water-saving amount of 293 million cubic meters. Scheme field water-saving control (S2) was used to implement the field water-saving regulation, particularly with a water-saving amount of 302 million cubic meters. In Scheme planting structure adjustment (S3), the crop structure was adjusted without the reduction in the amount of water that diverted from the Yellow River, where the water-saving amount of 254 million cubic meters was obtained. Among different scheme combinations, Scheme S1+S2+S3 shared the highest water-saving amount of 911 million cubic meters, followed by Schemes S2+S3 and S1+S3 with a water-saving amount of 569 million and 557 million cubic meters, respectively. 3) An increase in the canal water utilization coefficient led to a decline in groundwater levels, which was unfavorable for salt drainage. In Scheme S1, the proportion of areas with a groundwater depth greater than 3 meters increased by 7.59%, compared with the baseline scheme, whereas, the proportion with a groundwater depth between 2.5 and 3.0 meters increased by 4.44%, which was not conducive to the salt drainage. The field engineering measures reduced the infiltration rate of farmland, leading to a decline in the groundwater levels, which was beneficial for the salt drainage. In Scheme S2, the groundwater recharge in the irrigation area decreased by 257 million cubic meters, compared with the baseline scheme, indicating a more significant decline in groundwater level. Scheme S2 was favorable for the salt drainage. In Scheme S3, the groundwater recharge slightly decreased with the relatively stable groundwater level, which was also favorable for salt drainage. Among them, the S1+S2 and S1+S2+S3 combination shared a greater impact on the groundwater depth, especially S1+S2+S3 in the northwest part of the irrigation area. A high-value area of continuous buried depth was formed to dominate the growth of crops and vegetation during the growing season. There was a 5.46% increase in the proportion of naturally vegetated areas with an average buried depth exceeding 2.5 meters in Scheme S1+S2, compared with the baseline scheme. Therefore, the scheme S2+S3 was recommended with the largest water-saving amount, where the suitable water-saving potential of the irrigation area was 569 million cubic meters, considering the constraints of the ecological environment. Consequently, this scheme can be the most favorable for salt drainage in the irrigation area.