Answers are grounded in this paper and cite their sources.
Flowchart of streamflow gap filling using four methods across six gap scenarios, with statistical and hydrological e valuation. The graphical abstract summarizes a multivariate framework for reconstructing missing daily streamflow at Thumpa mon Station, Kerala, India. Observed streamflow is combined with rainfall, temperature, soil moisture, and other hydrological v ariables. Artificial missing data are generated under six scenarios: long, short, continuous, discontinuous, high flow, and low flo w gaps. Four methods are evaluated: Predictive Mean Matching, k-Nearest Neighbour, Random Forest, and Facebook Prophet. Reconstructions are assessed using RMSE, NRMSE, Kling-Gupta Efficiency, and Pearson’s correlation, together with hydrolog ical diagnostics. Random Forest shows strong overall performance across several gap structures, while Predictive Mean Match ing performs comparatively well for high- and low-flow conditions. Reliable streamflow records are essential for hydrological analysis, water resource management, and flood risk assessment. However, these datasets often contain missing values due to equipment failures, transmission errors, or environmental constraints, reducing reliability and leading to inaccurate results. This study evaluates the performance of several data-driven techniques for infilling missing daily streamflow data under different missing-data scenarios. Four approaches: k-Nearest Neighbour (k-NN), Predictive Mean Matching (PMM), Random Forest (RF), and Facebook Prophet (FB Prophet), were applied using a multivariate imputation framework. Missing-data patterns considered include long gaps, short gaps, continuous gaps, discontinuous gaps, and missing values associated with high-flow and low-flow conditions. Daily streamflow data from the Thumpamon station in Kerala, India, were used for model evaluation. Model performance was assessed using Root Mean Square Error (RMSE), Normalised Root Mean Square Error (NRMSE), Kling–Gupta Efficiency (KGE), and Pearson's correlation coefficient. Random Forest achieved Pearson's correlation coefficients up to 0.91 and Kling–Gupta Efficiency values up to 0.81, while Predictive Mean Matching showed comparatively stronger performance under simulated high-flow and low-flow missing-data scenarios. Predictive Mean Matching showed comparatively better statistical performance under the simulated high-flow and low-flow missing-data conditions, while k-NN demonstrates moderate performance across all scenarios. The findings highlight the potential of machine learning approaches to improve the statistical reconstruction of missing streamflow observations under different missing-data conditions.
No takes yet. Share an insight, caveat, or question.
Adarsh et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: