How well do the reanalysis datasets capture hot and cold extremes and their trends in India?

Abstract

Much of the Earth's surface lacks long-term in-situ measurement of essential meteorological variables. Climate reanalysis datasets provide an alternative in data-sparse regions, sometimes replacing gauge-based observations for climatological studies, however, they have inherent biases. Reanalysis is now available at finer spatial and temporal resolutions, that can be considered for hydrological and climatological studies. Although the assessment of reanalysis datasets is common at a daily, monthly, or seasonal scale, how the recent generation reanalysis captures the spatial pattern of extreme temperature events, and their trends remains an open question.

In this study, two regional (IMDAA and EARS) and five global (ERA5-Land, ERA5, MERRA2, CFSR, and JRA3Q) reanalysis datasets are evaluated with a gauge-based gridded temperature dataset from the India Meteorological Department (IMD) to assess their suitability for studying extreme temperature events and their trends over India. Fifteen hot and cold extremes indices are identified to characterize extremes covering frequency, intensity, and duration of extreme temperature events.

The study finds that no single reanalysis outperforms others for all the extreme indices when compared to the IMD gridded data, however, a select few (e.g., ERA5, ERA5L, MERRA2, and JRA3Q) better perform in reproducing the observed spatial pattern of extreme events and their changes across different regions of India. It is also noted that in response to global warming, the frequency, duration, and magnitude of extreme hot events are rising, and cold events are decreasing in India which is also captured by most of these reanalyses. Overall, the increase in hot extremes is more prominent in the south of the tropics and the decline in cold extremes is more evident in the north. However, the trend areas and magnitudes of the reanalysis datasets were not similar in comparison to trends from a regional station-based gridded dataset. Thus, care should be taken when selecting datasets for such applications and interpreting their trends.