Lidars are commonly used on autonomous vehicles, but their performance can be significantly affected by adverse weather. A number of studies have been devoted to analyzing and improving lidars' performance in rain, fog, and snow. Yet, relatively little attention has been paid to road spray which occurs when vehicles travel on wet surfaces at high speed. Road spray produced by the vehicles causes false positive point measurements in lidar scans. To evaluate the performance of lidar perception systems or train learning-based perception models, a large amount of data collected in various weather conditions are needed. Unfortunately, collecting spray data is challenging due to the requirements for certain weather conditions (e.g., heavy rain), and vehicles with high speed. In this letter, we propose the first data-driven method combined with simulation to reconstruct and synthesize spray data. The proposed pipeline can be applied to data augmentation by adding spray effects to existing lidar data collected under good weather conditions. We compare the performance of vehicle detection models trained with and without augmented data. The model trained with augmented data achieve significant performance improvement given real-world spray-affected point cloud data.