Precipitation event detection based on air temperature over the Equatorial Indian Ocean

Abstract

Air temperature (AT) and precipitation observations obtained from RAMA (Research Moored Array for African-Asian-Australian Monsoon Analysis and prediction) buoy at 0° N, 90° E from July 2009 to June 2017 are used to identify rainfall events. Based on the Random forest method, which consists of classification and regression based on decision trees, an algorithm is developed to identify the rainfall events from the change in AT data with high accuracy. During the study period, a total of 22461 abrupt drops in air-temperature events were identified by the algorithm. Around 75 % of these events were used to train and develop the clustering algorithm, and the rest of the events were used for validation with the precipitation data available from the buoy. The algorithm can identify more than 94 % of rain events accurately when the classification is binary. When the rain events are classified similar to the India Meteorological Department's classification, the algorithm is still able to identify the rain events; however, the performance degrades to ~ 84 % accuracy.