Development of non-stationary temperature duration frequency curves for Indian mainland

Extreme temperature events are one of the most serious threats resulting from climatic change across the globe. Quantifying the intensity, duration, and frequency of temperature extremes is of huge societal and scientific interest. Recent evidence in climate change challenges the stationarity assumption conventionally followed while performing temperature-duration-frequency (TDF) analysis. India has distinct climate zones and topography, distributing climate threats unevenly. In this study, stationary (S) and non-stationary (NS) TDF analysis for India and its seven temperature homogenous areas is performed using a gridded (1 degrees x 1 degrees) daily maximum temperature dataset from 1951 to 2019. Time is employed as a covariate to incorporate linear, quadratic, and exponential trends in the location and/or scale parameters of the generalized extreme value distribution to demonstrate the impact of non-stationarity in developing TDF curves. According to the findings, NS TDF models provide a better fit to the dataset when compared to the S TDF model. More than 55% of the grid points have NS Model-1, viz., location parameter linearly varying with time, as the best-fit model. In contrast to their stationary counterparts, NS temperature return levels were consistently higher across all return periods. Furthermore, temperature homogenous zones in the North-West, North Central, and Interior Peninsula are more susceptible to temperature rises beyond 45 degrees C. While envisioning long-term solutions in a changing climate scenario, considering non-stationarity significantly improved the accuracy of TDF curves. This will indeed support more robust predictions, which will ultimately aid in the mitigation of future extreme temperature events.