Using commercial microwave links and SEVIRI observations for rainfall estimation in Zambia

In Zambia, like in many African countries, the dedicated rainfall observation network is sparse, whereas accurate information about rainfall is crucially needed. In such a data-poor country, opportunistic sensors like commercial microwave links (CMLs) can be very beneficial. However, the irregular spatial distribution and the fact that many CMLs are very long and operate at low frequencies are common characteristics for rural areas in Africa which make rainfall retrieval with CMLs challenging. In addition, the lack of reference data complicates the adoption and adjustment of existing CML processing methods. In particular, the detection of rain events in noisy CML data, which can have a significant effect on the resulting estimated rainfall amounts, requires special attention as the long low-frequency CMLs provide comparatively noisy data. One option to support CML data processing is the usage of satellite data. We use level 1.5 data from Meteosat Second Generation (MSG) SEVIRI to generate a precipitation probability (PC) product, similar to the PC products from NWC SAF. Our PC product is generated by a convolutional neural network (CNN) which was trained with SEVIRI and high-resolution radar data in Germany and which was validated with station data in Burkina Faso. We use this PC product to improve the rain event detection during the data processing of almost 1000 CMLs with 15-minute min-max data over several months of the rainy season 2021/2022. In addition, we use two other rain event detection methods, the Python implementation of the nearby-link approach from RAINLINK and the simple rolling standard-deviation method. From the processed CML rainfall estimates, we produce interpolated rainfall maps which we then validate with rain gauge data. Preliminary results show that the nearby-link and rolling standard-deviation method produce satisfactory results in urban regions where CML density is high and CML frequencies are larger than 10 GHz. The application of the SEVIRI-based PC product for improved CML data processing, in particular for the long low-frequency CMLs, is currently being investigated and we will present first results to analyze its potential and limitations.