Multi-Sensor Assessment and Uncertainty Quantification of Integrated Hydrological Components in a Tropical Experimental Catchment.

Comprehensive in-situ hydrological measurements in tropical environments face significant data challenges. Multi-sensor deployments often result in incomplete temporal coverage due to installations in phases, sensor malfunctions, and maintenance requirements. These data gaps, combined with sensor-specific calibration uncertainties and measurement noise, introduce substantial uncertainties that are rarely quantified, particularly in the tropics where such datasets are scarce.

Data quality control and uncertainty quantification become critical when integrating measurements from diverse sensor types that measure different areas and have different types of errors. Raw sensor data require cleaning protocols to identify and address outliers and systematic biases. Furthermore, translating single-point measurements into catchment-scale estimates introduces scaling challenges that add to the existing uncertainties of each sensor.

This study looks at these challenges using data from 2022–2025 in the Kent Ridge experimental catchment in Singapore characterized by different land types (grass, forest, built-up areas). Our integrated sensor network combines pressure transducers for surface water level (used to derive flow rates) and groundwater table monitoring, drainage lysimeters, plant physiological sensors (sap flow meters, dendrometers, leaf wetness sensors), multi-depth soil moisture monitoring, soil temperature sensors, PAR sensors (Photosynthetically Active Radiation), and weather data including rainfall, wind speed and direction, air temperature, vapor pressure deficit, and solar radiation.

We apply data cleaning procedures and employ different uncertainty quantification methods to interpret sensor-specific outputs and evaluate how data gaps affect the overall measurement uncertainty. Results quantify typical measurement errors in urban tropical catchment and demonstrate practical approaches for handling imperfect multi-sensor datasets in real world environments.