Gap filling of water level time series with water area using remote sensing data: a comparative performance analysis of polynomial functions, XGBoost, Random Forest and Support Vector Machine

OpenAlex · English

Freshwater resources are crucial in many fields, such as food and transportation; hence, monitoring and interpretation of inland waters are necessary. Nowadays, remote sensing enables the detection of both water level and surface area of inland waters. In this study, Rathbun and Murray lakes in the USA were selected to fill gaps in the time series of water levels. Three polynomial functions (first, second, and third order) and eXtreme Gradient Boosting (XGBoost), random forest, and support vector machine regression methods were used to fill the gaps. The results showed that the XGBoost method outperformed other methods for both lakes. The water level in Lake Murray was determined with an Root Mean Square Error (RMSE) = 0.31 m and Coefficient of determination (R2) = 0.92, while in Lake Rathbun, it was RMSE = 0.30 m and R2 = 0.95. Based on these results, the XGBoost method could be an important approach for filling gaps in the water level time series.