Removal of Systematic Model Bias on a Model Grid

Abstract

Virtually all numerical forecast models possess systematic biases. Although attempts to reduce such biases at individual stations using simple statistical corrections have met with some success, there is an acute need for bias reduction on the entire model grid. Such a method should be viable in complex terrain, for locations where gridded high-resolution analyses are not available, and where long climatological records or long-term model forecast grid archives do not exist. This paper describes a systematic bias removal scheme for forecast grids at the surface that is applicable to a wide range of regions and parameters.Using observational data and model forecasts over the Pacific Northwest, a method was developed to reduce the biases in gridded 2-m temperature, 2-m dewpoint temperature, and 12-h precipitation forecasts. The method first estimates bias at observing locations using errors from forecasts that are similar to the current forecast. These observed biases are then used to estimate bias on the model grid by pairing model grid points with stations that have similar elevation and/or land-use characteristics.Results show that this approach reduces bias substantially, particularly for periods when biases are large. Adaptations to weather regime changes are made within a short period, and the method essentially "shuts off" when model biases are small. With modest modifications, this approach can be extended to additional variables.

Locally Calibrated Probabilistic Temperature Forecasting Using Geostatistical Model Averaging and Local Bayesian Model Averaging

“…Errors are used from observing locations that are within a predefined radius of s 0 , with observing locations that are closer to s 0 given preference over those that are further away. Observing locations also must have a similar land use category and elevation to s 0 ; the 24 distinct land use types are grouped into 9 groups sharing similar characteristics, described in Mass et al (2008). To mitigate the effects of change of meteorological regime, only errors from (ensemble mean) forecasts that are within a set tolerance of the current (ensemble mean) forecast value at s 0 are used.…”

“…describe an approach to gridding MOS predictions that accounts for elevation and the distinction between water and land based model grid points. Mass et al (2008) describe an approach to gridded bias correction that is sensitive to features which affect model bias, such as elevation, land use type, and forecast value. It is based on the following interpolation scheme, which we refer to as the Mass-Baars interpolation method.…”

“…Land use types are split into nine categories, defined in Mass et al (2008). In order to accommodate regime changes, the forecast errors must arise from a similar forecast value; for instance if the forecast at s 0 is 70 • F, then the only errors considered come from forecasts between, say, 65…”

Evaluation of NWP results for wintertime nocturnal boundary‐layer temperatures over Europe and Finland

“…Järvenoja, 2005;Maas et al, 2008;Tastula and Vihma, 2011). The largest errors typically appear in the forecast of T2m in wintertime when the temperature is lowest and stratification is strongest.…”

Use of Data Denial Experiments to Evaluate ESA Forecast Sensitivity Patterns

“…One technique developed by Janjić and Cohn (2006) specifically focuses on bias corrections due to unresolved scales using a Kalman filter correction technique. A second more general technique developed by Mass et al (2008) directly makes corrections to observations while taking into account model bias for a number of different observed variables. These techniques highlight the need to assess model bias during the assimilation process to produce a more representative set of atmospheric initial conditions and lead to a more accurate forecast.…”