Development of a Gridded Observation-based Ensemble Precipitation Product for India

Knowledge of precipitation is important for understanding the water cycle, weather and climate conditions, and natural disasters such as flood. Available gridded products over India are highly uncertain due to several factors such as less density of station network, effect of complex terrain, and measurement errors. Existing gridded datasets for India were made using a similar method, which included a multi-stage quality check followed by algorithms like Shepard's Interpolation method and probabilistic interpolation method. In addition, these datasets do not provide uncertainty estimates for any of the calculated fields and are, in fact, deterministic. Using a large network of 2186 precipitation stations over India, we generate an ensemble precipitation product for the Indian region, that explicitly account for the terrain complexity as well as errors in the precipitation estimates. The EnsIND (Ensemble precipitation product for India) generates a daily ensemble precipitation product for the given grid using the gauge station measurements as input, along with spatial attributes such as latitude, longitude, elevation, and slope. Thirty different ensemble members are generated daily at a resolution of 0.25 degree and 0.1 degree. The mean and median of the generated ensemble members are compared with the IMD (Indian Meteorological Department) and the CHIRPS gridded precipitation data. To validate the results, metrics such as mean absolute error, root mean square error, pearson correlation etc are calculated. The pearson correlation between 0.25-degree ensemble mean and IMD gridded precipitation shows high correlation in the Northeast region and Northern India. This research contributes to a more realistic depiction of precipitation over the Indian subcontinent. Estimates of the mean, median and standard deviation allows us to assess the spatial distribution of precipitation across the Indian region. Estimations of uncertainty as an integrated part of developed datasets will help us to understand and describe forcing uncertainties better.