Correcting Precipitation Undercatch Caused by the Wind Using Improved Instrumentation and Machine Learning

Precipitation measurements offer historical and near real-time data for meteorological services, serving as ground truth references for modelling and forecasting.

However, current precipitation measurement solutions, such as tipping-bucket rain gauges (TBRG) are affected by the well-known issue of undercatch, caused by factors like wind effects on the gauge, out-splash, evaporation, and internal tipping bucket ('counting') errors, leading to water-balance inaccuracies for hydrologists [Sevruk, Pollock].

While effective aerodynamic rain gauge design and proper placement can help minimize the impact of these issues, they cannot eliminate them; ideally, gauges should be installed out of the effects of the wind, such as World Meteorological Organization (WMO) approved pits, although this is seldom practised, typically occurring only at select high-profile meteorological sites [Burt].

Research has focused on identifying the optimal aerodynamic shape for a rain gauge to reduce out-splash and enhance catch efficiency [Strangeways]. Field comparisons and computational fluid dynamics (CFD) studies were conducted to assess various designs, including standard straight-sided, ‘chimney’ shaped, aerodynamic, and pit-installed (wind-protected) gauges [Pollock, Colli].

This research suggests that undercatch could be quantified by determining wind speed from the rain gauge rim, along with information on the size, velocity, and distribution (number of droplets over time) of droplets and applied using to wind correction algorithms.

On this matter, various algorithms have been developed to tackle the challenge of undercatch in precipitation measurements. The accuracy of these algorithms hinges on the ability to measure and understand how these factors influence the collection of precipitation, as any deviation in these variables can significantly impact the correction process [Cauteruccio].

With regards to drop analysis, current optical distrometers can detect and classify droplets, but they come at a high price and often underestimate precipitation totals [Johannsen]. They can also be bulky, hindering the precipitation they are designed to measure [Chinchella].

This abstract summarises our innovative and cost-effective method using a combination of ultrasonic sensors, and a laser-based optical analyser tailored for seamless integration with a rain gauge funnel. This instrument system is designed to measure the wind speed, drop size, velocity, and distribution of the droplets passing through a (dual) laser beam spanning across the funnel of a standard aerodynamic TBRG.

To enhance the instruments capabilities, a detection algorithm based on machine learning (ML)/neural networks (NNs) has been integrated, which enables accurate prediction of droplet characteristics (size and number of droplets).