Cross-Climate Comparison of Rainfall Estimates from Commercial Microwave Links with Spaceborne Radiometer and Radar Observations

Rainfall estimates from commercial microwave links (CMLs) have recently emerged as a novel ‘opportunistic’ source of hydrological information, complementing estimates from dedicated rainfall sensors, both ground-based (rain gauges, weather radars) and spaceborne (low earth orbit and geostationary satellites). CML-based rainfall estimates are particularly promising for hydrological applications in the Global South, where weather radars are virtually absent and rain gauge networks are typically sparse and not always well maintained. However, as CML networks have been designed to facilitate cellular communication rather than for high-resolution space-time rainfall monitoring, their rainfall estimates suffer from a number of errors and uncertainties. In addition, they do not provide coverage over larger water bodies, such as lakes and coastal seas, a potentially serious shortcoming for short-term rainfall forecasting and flood early warning. This is where spaceborne rainfall estimates could come to the rescue.

The objective of this work is to systematically compare and evaluate the potential of complementing rainfall estimates from CML networks with those from spaceborne radiometers, and vice versa, across various climates. Specifically, we focus on the radiometer on board the GPM core satellite (i.e. the GMI), which provides simultaneous observations with the GPM radar (i.e. the DPR), which is used as reference. We use low earth orbit observations because the cloud-penetrating capabilities of both the GMI and DPR allow for more direct rainfall estimates at the land surface than the cloud-top observations from geostationary satellites.

We consider coinciding CML and GPM observations for rainfall events over three countries with contrasting climates, namely two in the Global South (Sri Lanka and Nigeria) and one in the Global North (the Netherlands). A total of 19 cases was selected for Sri Lanka (between September 2019 and December 2020), 14 cases for Nigeria (between May and October 2020), and 28 cases for the Netherlands (between March 2014 and March 2015). Comparisons are made in the form of scatter plots of instantaneous rainfall estimates at the locations of individual CMLs (including a number of statistics) as well as in the form of rainfall maps for the three countries concerned.

Preliminary results indicate significant discrepancies between the rainfall estimates from CMLs on the one hand and those from GMI and DPR on the other. These discrepancies can be attributed to differences in the nature of the measurements, to differences in spatial and temporal sampling as well as to the space-time variability of rainfall. Furthermore, our analyses indicate that the default wet antenna attenuation correction applied to the CMLs may be too high, resulting in low-intensity precipitation going undetected. Additionally, the relatively large distances between CMLs can result in misclassification of rainy conditions as dry moments when the ‘nearest’ CML is too far away to detect the same event. In both cases, radiometer overpasses could assist in improving the wet-dry classification. Conversely, CMLs can improve the retrieval of high-intensity precipitation, which radiometers tend to underestimate. This is particularly valuable in the tropics, where dedicated measurements are sparse while CMLs are generally available.

These early analyses provide a first step towards the development of a merging scheme, taking the strengths and weaknesses of the various sensors into account. Our ultimate objective is the development of an algorithm that produces seamless surface rainfall products for those countries in the world lacking dense ground-based rainfall observation networks.