<p>Recently, deep neural networks have achieved state-of-the-art performance in remote sensing scene classification. Traditionally, neural networks are trained on a predefined number of classes, expecting that all training and test data are identically distributed. But, when the classifier is deployed into the operating environment, it is likely to be exposed to data from a different distribution, i.e., out-of-distribution (OOD) data. In remote sensing scene classification, OOD images can come from different geographic areas, sensors, and imaging conditions or can contain scene classes not present in the training dataset. Methods for detecting OOD images can benefit from outlier exposure, which involves training a classifier using a library of OOD images in addition to in-distribution training images. In the present paper, we propose a training-free outlier exposure method for improving OOD detection in remote sensing scene classification. The method is based on computing ratios of distances between the embeddings of test and training images, as well as outlier exposure samples. The experimental results show that, for small outlier exposure libraries and without updating the basic classification model, the proposed method achieves performance improvements comparable or better than the approaches from the literature, making it an attractive baseline for OOD detection with outlier exposure.</p>