Eliminating calibration periods in rainfall estimation through soil moisture using Growing Neural Gas clustering

Rainfall is essential for sustaining life on Earth, as it plays a crucial role in the water cycle, influencing the amount of water stored in the soil. This relationship significantly impacts agriculture and water resource management, highlighting the importance of understanding and preserving natural rainfall patterns. Considering the importance of these two variables and their impact on the water cycle, it is essential to understand the profound connection between rainfall and soil moisture (SM). Due to the immediate response of soil to rainfall events, the amount of water stored in the soil can be regarded as a reliable signal for estimating rainfall. Considering this context, we explore the Soil Moisture to Rainfall (SM2RAIN) algorithm with Net Water Flux (NWF) through the International Soil Moisture Network component. By doing this, we can consider a measure of drainage to formulate a new analytical approach for the previous SM2RAIN algorithm. However, NWF enhanced SM2RAIN algorithm (hereafter SM2RAIN-NWF) highly dependent on historical data and a calibration period, and this dependency significantly limits its applicability in regions lacking sufficient historical data. To address this issue, we introduce and evaluate a novel technique that eliminates the need for a calibration phase in the SM2RAIN-NWF. Using the Growing Neural Gas algorithm and Bayesian optimization, we categorize patterns based on ERA5 land datasets. The proposed method eliminates the heavy reliance on historical data sets, which are often challenging and costly to collect, and enhances our understanding of the global-scale water cycle and land-atmosphere interaction.