Machine learning and dynamics based error‐index method for the detection of monsoon onset vortex over the Arabian Sea: Climatology and composite structures

Abstract

AbstractMonsoon onset vortex (MOV) forms over the Arabian Sea near the northern flank of the low‐level jet during the monsoon onset over Kerala (MOK). The study concerns the development and evaluation of an algorithm for detecting and tracking MOVs in regional/global analyses. The first step involves preparing the first‐guess database of MOV locations based on geopotential height, surface and 850 hPa wind magnitude and circulation from ERA5 reanalysis for 1982–2020. Three different approaches: (a) error‐index of MOV, (b) machine‐learning (ML), and (c) combination of error‐index and ML models, are employed to detect MOV. The error‐index method, in which the detected vortex is compared with the idealized vortex, achieves an accuracy of 0.6 with a 0.95 true‐positive‐rate and 0.55 false‐positive‐rate. The best ML models can identify the MOVs in the training samples with maximum accuracy of 0.99. However, their accuracy is limited in tracking the MOVs continuously in the global analyses as they are not trained with wind circulation. The combined error‐index and ML models could detect all the 27 observed MOVs in the ERA5 reanalysis with 5 false‐positives. This approach is tested on IMDAA reanalysis, and the success rate is 0.79 with 6 false‐positives. Temporal analyses show that ∼95% of the MOVs occur during −10 to +20 days from MOK. Composite structures indicate that the MOVs exhibit higher sea‐surface temperatures (>0.3 °C) in the forward sector with 85% cloud cover in the left‐rear sector. Rainfall of 4–5 mm·hr−1 is seen in the left sector. Upper‐level (700–200 hPa) warm core (>3.5°C) and lower‐level (1000–700 hPa) cold‐core (<1°C) is evident. The composite structures of MOVs are almost similar to that of monsoon depressions with higher asymmetry in the forward‐rear sectors. This study may help explore future projections of MOV activity from climate models and its relationship with monsoon rainfall activity.