In recent years, the absence of nighttime variegated haze image benchmarks results in poor performance of the existing image restoration or other high level vision algorithms under adverse nighttime weather conditions. To address this problem, we present an improved atmospheric scattering model for depicting complex environments in nighttime variegated haze scenes. Based on this model, we propose a synthesis algorithm to generate the Nighttime Variegated Hazy Dataset (NVHD). We further introduce both full-reference and non-reference strategies to evaluate the performance of the state-of-the-art dehazing algorithm, along with utilizing advanced visual tasks to assess the effectiveness of the proposed dataset. Extensive experiments show that the authenticity and practicality of our dataset on nighttime variegated haze weather conditions.