Search

Air temperature is an important baseline parameter for terrestrial Antarctica in the context of patterns and processes in climatology, hydrology or ecology. There are still large uncertainties on how the Antarctic system responds to spatio-temporal variability of temperature. This can partly be attributed to the lack of high resolution datasets. In this paper, we present AntAir, a new dataset of gridded air temperatures in 1 km spatial and daily temporal resolution that is available since 2003. AntAir was created by modelling daily air temperature from MODIS land surface temperature using machine learning algorithms. Data from 70 weather stations was used as a reference. Daily temperatures could be estimated with a R² of 0.91 and a RMSE of 5.07 °C validated on independent years. The performance to estimate the time series of a new spatial location was R² = 0.78 and RMSE = 5.83 °C. Hence the high spatial and temporal resolution of the dataset as well as the high accuracy make AntAir an important baseline dataset for a wide range of applications in environmental science of Antarctica. [ABSTRACT FROM AUTHOR]

Spatial predictions of near-surface air temperature (Tair) in Antarctica are required as baseline information for a variety of research disciplines. Since the network of weather stations in Antarctica is sparse, remote sensing methods have large potential due to their capabilities and accessibility. Based on the MODIS land surface temperature (LST) data, Tair at the exact time of satellite overpass was modelled at a spatial resolution of 1 km using data from 32 weather stations. The performance of a simple linear regression model to predict Tair from LST was compared to the performance of three machine learning algorithms: Random Forest (RF), generalized boosted regression models (GBM) and Cubist. In addition to LST, auxiliary predictor variables were tested in these models. Their relevance was evaluated by a Cubist-based forward feature selection in conjunction with leave-one-station-out cross-validation to reduce the impact of spatial overfitting. GBM performed best to predict Tair using LST and the month of the year as predictor variables. Using the trained model, Tair could be estimated with a leave-one-station-out cross-validated R² of 0.71 and a RMSE of 10.51 °C. However, the machine learning approaches only slightly outperformed the simple linear estimation of Tair from LST (R² of 0.64, RMSE of 11.02 °C). Using the trained model allowed creating time series of Tair over Antarctica for 2013. Extending the training data by including more years will allow developing time series of Tair from 2000 on. [ABSTRACT FROM AUTHOR]