From skin to bulk: An adjustment technique for assimilation of satellite-derived temperature observations in numerical models of small inland water bodies

Abstract

Data Assimilation (DA) has been proposed for multiple water resources studies that require rapid employment of incoming observations to update and improve accuracy of operational prediction models. The usefulness of DA approaches in assimilating water temperature observations from different types of monitoring technologies (e.g., remote sensing and in-situ sensors) into numerical models of in-land water bodies (e.g., lakes and reservoirs) has, however, received limited attention. In contrast to in-situ temperature sensors, remote sensing technologies (e.g., satellites) provide the benefit of collecting measurements with better X-Y spatial coverage. However, assimilating water temperature measurements from satellites can introduce biases in the updated numerical model of water bodies because the physical region represented by these measurements do not directly correspond with the numerical model's representation of the water column. This study proposes a novel approach to address this representation challenge by coupling a skin temperature adjustment technique based on available air and in-situ water temperature observations, with an ensemble Kalman filter based data assimilation technique. Additionally, the proposed approach used in this study for four-dimensional analysis of a reservoir provides reasonably accurate surface layer and water column temperature forecasts, in spite of the use of a fairly small ensemble. Application of the methodology on a test site - Eagle Creek Reservoir - in Central Indiana demonstrated that assimilation of remotely sensed skin temperature data using the proposed approach improved the overall root mean square difference between modeled surface layer temperatures and the adjusted remotely sensed skin temperature observations from 5.6°C to 0.51°C (i.e., 91% improvement). In addition, the overall error in the water column temperature predictions when compared with in-situ observations also decreased from 1.95°C (before assimilation) to 1.42°C (after assimilation), thereby, giving a 27% improvement in errors. In contrast, doing data assimilation without the proposed temperature adjustment would have increased this error to 1.98°C (i.e., 1.5% deterioration).