Artificial Neural Network Models to Predict Katabatic Winds over Ross Ice Shelf, Antarctica
Abstract
Strong Katabatic winds in Antarctica are responsible for blowing snow and wind-scoured blue ice areas, polynya and sea ice formation, overturning ocean circulation, and local-scale variations in Surface Mass Balance. Katabatic wind patterns are controlled by temperature and pressure gradients and have high diurnal and seasonal variability. In this project, we use data from 11 Automatic Weather Stations (AWS) on the Ross Ice Shelf and optimally structured neural networks to quantify the accuracy of near-surface wind data from the European Centre for Medium-Range Weather Forecasts (ECMWF). It is well known that ECMWF temperature, pressure and wind speed values have systematic biases. Our results show that ECMWF Pressure is biased at an average 7.9 2.8 hPa higher than AWS pressure, and ECMWF temperature has seasonal bias which varies by station. For example in the case of Schwerdtfeger AWS station, in the winter months, the ECMWF monthly average temperature is 1-2 °C lower than the AWS monthly average temperature with no overlap between the 99.7% confidence intervals surrounding these means. Using the controlling mechanisms of the Katabatic winds as inputs, optimally structured neural networks are trained and used to fill in the sizeable data gaps in the AWS wind speed measurements. Our results show from 1985 to present the ECMWF wind speed correlates to the AWS wind speed data with r = 0.74. Our trained neural networks used to fill in gaps in AWS wind speed data achieves a correlation of r = 0.69 with ECMWF data.
These preliminary results demonstrate that artificial neural networks can be used to predict missing values and reduce biases in large datasets which has implications for regional climate models that use ECMWF as boundary conditions.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFM.C43E1851L
- Keywords:
-
- 0726 Ice sheets;
- CRYOSPHEREDE: 0728 Ice shelves;
- CRYOSPHEREDE: 0762 Mass balance;
- CRYOSPHEREDE: 0776 Glaciology;
- CRYOSPHERE