Estimation of snow depth in the UK using the HUT snow emission model

Abstract

Snow cover characteristics have significant effects on upwelling naturally emitted microwave radiation through processes of forward scattering. This study simulates numerically the electromagnetic responses from snow in the UK using the radiative transfer‐based semiempirical model developed at the Helsinki University of Technology (HUT), which takes into account the influence of soil surface, forest canopy and atmosphere on space‐borne observed brightness temperature by using empirical and semiempirical formulas. A sensitivity analysis of the HUT model was conducted to determine the most sensitive parameter affecting upwelling radiation from snow in the UK. The model‐based results were compared with observed Special Sensor Microwave Imager (SSM/I) brightness temperatures to better understand the SSM/I response to snow. The available ensemble of data required for input to the HUT model comprise surface physical temperature, ground level pressure and water vapour content, forest stem volume and land cover water fraction. Based on the sensitivity analyses, numerical parameters representing physical snow pack quantities (e.g. snow grain size, snow moisture and snow depth (SD)) were varied and the method of root mean square error (RMSE) minimization was used to invert the SD. The HUT model was applied to different days in 3 months (23–31 January, 1–5 and 26–27 February and 1–7 March 1995) of records of daily SD and SSM/I observations. The results show that the HUT model both underestimates and overestimates SD prediction. For the month of January 1995, the HUT model underestimated SD with a bias of −0.59 cm, whereas for February and March 1995 the HUT model overestimated the SD with a bias of 1.89 cm and 1.64 cm, respectively. This study demonstrates that microwave remote sensing of snow can be used successfully in the UK, where most research on snow cover is conducted by using a visible and infrared radiometer. It is also evident from this work that application of algorithms to snow pack monitoring needs local calibration for effective and reasonable results.