Improving CYGNSS-based Soil Moisture Coverage through Autocorrelation and Machine Learning-Aided Method

Global Navigation System-Reflectometry (GNSS-R) is a microwave remote sensing technology that can estimate soil moisture (SM) by receiving Earth’s surface reflected signals from GNSS satellites. The Cyclone GNSS (CYGNSS) constellation, with its high temporal resolution Earth observation data, is a commonly used data source for soil moisture. However, due to the continual movement of GNSS transmitters and GNSS-R satellites, observations of the Earth’s surface can be chaotic and random when compared to the regular paths followed by other remote sensing satellites. Moreover, there are numerous gaps in the observations that are unevenly distributed. In this paper, a gapfilling method based on spatial autocorrelation is proposed to interpolate the gaps within these uneven observation datasets, with SM being estimated post-interpolation. The sample set for the model comprises points surrounding the interpolation target, with modeling conducted considering factors of spatial weighting to estimate values at the interpolation target. Different autocorrelation-based gap-filling methods using CYGNSS data can achieve good estimation accuracy, and the data coverage after interpolation is on average 1.8 times greater than before interpolation. The gap-filling method using XGBoost achieves the best performance and offers the highest accuracy in SM estimation, with an average correlation coefficient of 0.8445, and an average RMSE of 0.0457 m³/m³. The gap-filling approach can significantly enhance data coverage and facilitate the filling of daily gaps in CYGNSS data with all maintaining high SM estimation accuracy. The estimation of daily missing values using CYGNSS data can fully exploit the embedded surface features in the data’s fine resolution and can provide high-resolution SM retrieval, as well as for the warning and monitoring of floods, droughts, and changes in water bodies.