Harnessing GNSS-IR and Deep Learning for Soil Moisture Estimation in Malawi

Soil moisture estimation is of utmost importance in various applications such as agriculture, hydrology, and climate modelling. Traditional approaches for soil moisture estimation often rely on remote sensing or sensor network dataset. However, these methods have limitations in terms of coverage, resolution, and cost-effectiveness. The emergence of Global Navigation Satellite System Interferometric Reflectometry (GNSS-R) as an innovative technique utilizing satellite observations has opened up new possibilities for both proximal and remote sensing in this domain. In recent years, deep learning has emerged as another approach for accurate and widespread soil moisture estimation. This study presents an investigation into the use of GNSS-IR in conjunction with deep learning for soil moisture estimation in Malawi. This is achieved by utilising datasets from the Malawi Continuously Operating Reference Stations (CORS) and geodetic measurements collected from various locations across the country. The datasets are refined through preprocessing to enhance their quality and relevance in the deep learning model. Both datasets are used to estimate the water content in the soil, and the deep learning model is used to learn the complex relationship between the geodetic observations and soil moisture content. A comparative analysis is conducted, comparing the results obtained from CORS and GNSS campaigns, taking into account factors such as the supported satellite system, signals, and terrain characteristics. Notably, the findings highlight variations in outcomes based on terrain conditions and soil roughness in the vicinity of the geodetic monuments. Further comparative analysis with traditional methods highlights the superior performance and wider coverage achieved by the GNSS-IR and deep learning combined approach. Generally, this research opens up avenues for further exploration in the field of soil moisture estimation. Future work could focus on optimizing deep learning architectures, incorporating multi-sensor data fusion, and investigating transferability across different soil types and regions in Malawi.


Keywords: GNSS-IR, Deep learning, Soil moisture estimation, CORS, Geodetic observations, Malawi