Bias Estimation and Correction for In-situ Observations over Complex Terrain using an Ensemble Kalman Filter with the WRF model

Assimilating near-surface in-situ observations over complex terrain is challenging for several reasons. One is that observation representativeness errors can be fundamentally different, and greater than, many other observing platforms. An example is an anemometer sited on a slope that cannot be properly represented within discretized numerical weather prediction (NWP) model equations, or affected by surrounding features such as buildings or vegetation that a model cannot represent. Observation errors can be both random and systematic. In the data assimilation process, systematic observation representativeness errors can lead to systematic errors in initial conditions for a prediction.

To mitigate systematic errors in data assimilation over complex terrain, this work presents a methodology to correct and estimate biases of individual in-situ observations. The approach is based on parameter estimation using an augmented state in an ensemble adjustment Kalman filter. This methodology is applied by coupling the Weather Research and Forecasting (WRF) model with the Data Assimilation Research Testbed (DART). The domain encompasses a region with complex terrain over the western U.S. where the MATERHORN field campaign was carried out during fall 2012 (20 September to 25 October). The data used for assimilation include radiosonde, aircraft, satellite winds, METARs, and surface mesonets. Observations are assimilated every 3 hours using a 96-member ensemble on a 30-km domain. Biases are estimated for a mesonet affected by a representativeness error characterized by height differences between model and observations larger than 100 m. For this mesonet, biases for wind components at 10 m and temperature at 2 m are estimated.