Thermophoresis is an efficient process for the manipulation of molecules and nanoparticles due to the strong force it generates on the nanoscale. Thermophoresis is characterized by the Soret coefficient. Conventionally, the Soret coefficient of nanosized species is obtained by fitting the concentration profile under a temperature gradient at the steady state to a continuous phase model. However, when the number density of the target is ultralow and the dispersed species cannot be treated as a continuous phase, the bulk concentration fluctuates spatially, preventing extraction of temperature-gradient induced concentration profile. The present work demonstrates a strategy to tackle this problem by superimposing snapshots of nanoparticle distribution. The resulting image is suitable for the extraction of the Soret coefficient through the conventional data fitting method. The strategy is first tested through a discrete phase model that illustrates the spatial fluctuation of the nanoparticle concentration in a dilute suspension in response to the temperature gradient. By superimposing snapshots of the stochastic distribution, a thermophoretic depletion profile with low standard error is constructed, indicative of the Soret coefficient. Next, confocal analysis of nanoparticle distribution in response to a temperature gradient is performed using polystyrene nanobeads down to 1e-5% (v/v). The experimental results also reveal that superimposing enhances the accuracy of extracted Soret coefficient. The critical particle number density in the superimposed image for predicting the Soret coefficient is hypothesized to depend on the spatial resolution of the image. This study also demonstrates that the discrete phase model is an effective tool to study particle migration under thermophoresis in the liquid phase.