Repetition of large-scale surveys within a 24-hour period is made possible by the critical role that satellite usage plays in the process. In broad, uniform landscapes, high-resolution satellite photography has been shown to be useful for identifying and counting various animal species. Still, the main problem is to use deep learning and high-quality satellite data to spot elephants from space or identify animals in complicated situations. With the use of a convolutional neural network model, our study aims to automate the recognition and counting of African elephants in South Africa’s complex woodland savanna ecology. Using the highest commercially available resolutions from WorldView-2/3 satellite data, 15 photos from 2015 to 2020 are used to train and test the algorithm. Subsequently, the model’s generalizability is evaluated by comparing its performance accuracy to human accuracy and utilizing a lower resolution satellite picture (GeoEye-1) taken in Kenya without any further training data. The results show that Convolutional Neural Network (CNN) based Yolov7 applied on a custom dataset, as well as a human detector, in terms of accuracy. The detection accuracy is 0.744 in homogeneous regions and 0.777 in heterogeneous areas. Most remarkably, the model shows that it can generalize and identify elephants from a low-resolution satellite image and in a different geographic region. These results provide guidance for identifying the distant data required for animal detection tasks, enabling ecologists and farmers to use more readily available low-resolution photography with confidence.