This work reports a method to measure thermal diffusivity of thin disk samples at high temperatures (900 -1150K) using a modified Angstrom's method. Conventionally, samples are heated indirectly from the surroundings to reach high temperatures for such measurements, and this process is time-consuming, typically requiring hours to reach stable temperatures. In this work samples are heated directly in a custom instrument by a concentrated light source and are able to reach high steady-periodic temperatures in 10 mins, thus enabling rapid thermal diffusivity characterization. Further, existing Angstrom's methods for high temperatures use thermocouples for temperature detection that are commonly attached to samples via drilling and welding, which are destructive to samples and introduce thermal anomalies. In this work we use an infrared camera calibrated to 2000 C for non-contact, non-destructive and data-rich temperature measurements. We present an image analysis approach to process the IR data that significantly reduces random noise in temperature measurements. We extract amplitude and phase from processed temperature profiles and demonstrate that these metrics are insensitive to uncertainty in emissivity. Previous studies commonly use regression approaches for parameter estimation that are ill-posed (i.e., non-unique solutions) and lack rigorous characterization of parameter uncertainties. Here, we employ a surrogate-accelerated Bayesian framework and a 'No-U-Turn' sampler for uncertainty quantification. The reported results are validated using graphite and copper disks and exhibit excellent agreement within 5% as compared to reference values obtained by other methods.