Abstract: Extreme drought has exacerbated soil erosion in laterite regions. Yet existing research exhibits two limitations: (1) It remains unclear on the relationship between volumetric shrinkage and water retention in laterites under the wide moisture-content ranges; (2) The bimodal prediction models are often required to explicitly integrate the volumetric deformation and multi-scale pore structure into the soil-water characteristic curve (SWCC) framework. In this study, a bimodal SWCC equation was established to integrate the volumetric variation, the multi-scale pore structure (considering inter-/intra-aggregate pore systems), and the Young-Laplace capillarity (pore networks as randomly connected capillary bundles). The desiccation also induced the shrinkage evolution patterns. Parameter sensitivity analysis was used to investigate the physical significance of the model parameters. Wide-suction-range experiments (including both SWCC and parallel shrinkage tests) were conducted on Kunming laterite. The low-suction range (400 kPa or below) was determined using a pressure plate apparatus, while the high-suction range (10¹-10⁵ kPa) was measured using the filter paper method. The experimental procedures involved the soil sample desiccation, 14-day constant-temperature and humidity curing, and moisture-content determination for both filter paper and soil. Shrinkage tests were performed using the air-drying method, where the specimen dimensions and mass-based moisture content were measured at 1-4-hour intervals. The engineering applicability was validated using data from four typical regional soils. Kunming laterite data were corrected for the volumetric variation. The results show that the pore ternary classification and the water-filling critical criterion hypothesis effectively defined the boundary between macropores and micropores. Consequently, the pore distribution function was constructed using multi-scale pore superposition. Segmented van Genuchten functions at the suction intervals were then transformed, superimposed, and integrated with the correction functions. A bimodal adjustment function was extended for the volumetric change-integrated bimodal prediction model. Two-parameter datasets were identified to modulate the curve morphology. Desiccation initiation timing (either advancing or delaying) was regulated to accelerate the post-air-entry suction desiccation and residual moisture. According to Fredlund's classical unimodal model, these parameters were defined as the characteristic parameters to control the air-entry value, desaturation rate, and residual moisture content for the macropores and micropores. Data trends from both the pressure plate and filter papers were consistent in the suction overlap region. The filter paper data points were slightly higher than before. The Kunming laterite SWCC exhibited the bimodality with the two air-entry points and a plateau in the moderate suction range. A critical suction threshold of 34 MPa was identified, where the specimens with the higher initial dry density shared the superior water retention below this threshold, while the SWCC converged beyond it. The void ratio evolution during desiccation exhibited an initial dry-density dependence, followed by a two-stage pattern: an initial rapid decrease followed by gradual stabilization. Consistently, all shrinkage curves tended to stabilize when the gravimetric water content decreased to approximately 20%, regardless of the initial dry density. Volumetric shrinkage had a significant impact on water retention performance. Below a suction of 34 MPa, the shrinkage caused an upward shift in the SWCC curve after correction. Once the suction reached or exceeded 34 MPa, the volume was stabilized to avoid the correction. Energy dissipation analysis showed that the desiccation essentially involved the continuous solid-liquid-gas phase transitions under suction-gradient-induced microstructure. Model validation revealed that there was a three-phase behavior in the volumetric water content versus matric suction relationship: Phase I featured macropore-dominated drainage, Phase III reflected micropore-controlled retention, and Phase II (the transition zone) represented the synergistic interaction of macropores and micropores. Evaluation metrics confirmed that the excellent prediction performance was achieved (R²= 0.987-0.999; root mean square error= 0.3145-1.381). The classical unimodal model was also extended into an engineering application. The finding can also provide the theoretical support for the water retention evolution during desiccation shrinkage in the extremely arid laterite regions, particularly for disaster prevention and control.