RainRunner – A Multi-Satellite Rainfall Retrieval Algorithm for Northern Ghana Based on Deep Learning and Earth Observation and Supported by Citizen Science

Abstract:

With West African economy mainly sustained on agriculture and most crops being rain-fed, economic and food security in this region are highly dependent on rainfall and the knowledge of rainfall patterns. However, a poor rain gauge network and data transmission challenges make rainfall analysis difficult in West Africa.

The aim of this work is to produce reliable rainfall estimations for northern Ghana following a fully data-driven approach, using the Open Data Cube (ODC) and Convolutional Neural Networks, as an alternative to existing but regionally poor-performing satellite rainfall retrieval products.

In recent years Earth Observation (EO) data cubes are being successfully deployed as an effective way to organize and analyze the growing amounts of satellite data. At the same time, Deep Learning is being increasingly applied to exploit the information contained in EO data, often outperforming physical models in representing complex processes. However, the scarce ground data for training and validation is still a challenge, particularly in the Global South.

Based on these advances, a rainfall retrieval algorithm is being developed within the Schools and Satellites (SaS) project: RainRunner. SaS is being funded by the European Space Agency as one of the pilot projects of the Citizen Science and Earth Observation Lab.

To capture the local characteristics involved in the rain process, RainRunner will use data from diverse sensors onboard ESA’s Sentinel satellites (S1, S2, S3 and S5P), as well as MSG’s Aviris and a DEM that contain information such as cloud top temperature, cloud amount, atmospheric aerosols, soil moisture and land surface temperature. These data have been organized over northern Ghana in a data cube, from which RainRunner will extract smaller cubes and estimate daily precipitation for the central cell of each cube in two stages: Binary rain/no rain classification followed by rainfall amount estimation.

Citizen Science complements rain gauge data to form the training and validation dataset. Concretely, Ghanaian schoolchildren and local farmers are taking daily rainfall measurements in several locations in northern Ghana.

Once operative, RainRunner will guide agricultural extension agents, support crop insurance and ultimately contribute to economic growth and food security in the Global South.