Dialect identification has been explored profusely in major languages, such as Arabic, Chinese and Spanish. This paper presents an automatic dialect identification system in the Ao language using prosodic features. Ao is a low-resource Tibeto-Burman tonal language spoken in Nagaland in the North-Eastern part of India. It consists of three distinct dialects: Chungli, Mongsen and Changki. Prosodic characteristics are believed to have an essential role in tonal languages. In this direction, the current work focuses to investigate the prosodic characteristics to build a discriminative system in identifying the three Ao dialects. The statistical and Low-Level Descriptors (LLD) of prosodic features are used in this work. The prosodic features such as F0, loudness, shimmer, jitter, voiced and unvoiced segment length, etc., are utilized in this study. The experiments are conducted using SVM and attention-based Bi-GRU classifiers in trisyllabic words and passage-level datasets, respectively. The combination of prosodic features outperforms the MFCC (baseline) feature. The Voice Quality and Temporal (VQT) feature set is the best performing prosodic feature. The statistical analysis also shows that the VQT features are statistically significant. The performances of SVM and attention-based Bi-GRU classifiers indicate the significance of prosodic information in classifying the three Ao dialects.