The symbiotic network formed by the interaction between flowering plants and pollinators in a community is known as a pollination network. As a dynamic entity, species and their interactions change over time. Studying the dynamics of pollination networks is crucial for understanding community construction and the mechanisms that maintain ecosystem stability. However, there are few studies on the seasonal dynamics of pollination networks in desert communities. The Xinjiang Kalamaili Mountain Ungulate Nature Reserve is a representative temperate desert ecosystem in northern China. The region is characterized by long, cold winters, late springs, and rapid temperature increases over short durations. Consequently, most desert plants grow and bloom quickly in this season to avoid the summer heat. During this time, pollinating insects are numerous and active, forming the most complex networks. It is hypothesized that there are significant changes in the pollination network during the spring in this region. To explore this, the study constructed a pollination network over six springs through field observations with weekly analysis granularity and examined the spring dynamics of the interaction network in terms of composition, connectivity, and structure. The results showed that: (1) The ambient temperature in the gravel desert rose by 15.5°C during the spring. (2) The composition, connectivity, and structure of the pollination network exhibited strong spring dynamics. From the first week of observation (late April) to the fifth week (late May), the number of flowering plants and pollinator species, the relative abundance of pollinators, connectance, and network specialization all initially increased and then decreased. The number of flowering plant and pollinator species increased from 3 and 9 to 25 and 58, respectively, with pollinator species always outnumbering flowering plants. The relative abundance of pollinators rose from 0.028 to 0.303, connectance from 13 to 157, and network specialization from 0.45 to 0.65. However, connectivity, nestedness, Niche overlap HL, and Niche overlap LL showed opposite trends. Connectivity decreased from 0.15 to 0.05, nestedness from -3.44 to -30.35, and Niche overlap HL and LL from 0.325 and 0.473 to 0.115 and 0.114, respectively. Over the observation period from late April to the end of May, the relative abundance of flowering plants and modularity increased continuously, with the relative abundance of flowering plants rising from 0.002 to 0.520 and modularity from 0.16 to 0.63. However, the overall robustness of the network declined from 0.66 to 0.15. There were significant correlations among network parameters. For example, the number of plant species was positively correlated with the number of pollinator species, connectance, modularity, and network specialization, but negatively correlated with connectivity, nestedness, Niche overlap HL, and Niche overlap LL. (3) The partial least squares path model indicated that temperature changes in spring drove changes in the number of flowering plants and pollinator species (P<0.05), which likely caused changes in their interactions, leading to rapid changes in network connection (connectance and connectivity) (P<0.001). These changes in network connection subsequently affected network structure (nestedness, modularity, robustness, Niche overlap HL and LL, and network specialization) (P<0.05), resulting in significant dynamic changes in the interaction network during spring. These findings suggest that the environmental factors of the gravel desert and the composition, connectivity, and structure of the interaction network undergo different dynamic changes, demonstrating the strong plasticity of the interaction network. This adaptability might be a mechanism for the gravel desert community to cope with harsh environmental conditions, providing important theoretical insights into the dynamics and formation mechanisms of the pollination network in the gravel desert of northern Xinjiang.