The discovery of Land-Use and Land-Cover (LULC) mapping is vital task in remote sensing, supporting critical applications in environmental monitoring, urban planning, agriculture, and in disaster management. Traditionally, LULC classification has relied on supervised learning techniques that demand large volumes of labelled data, which are often expensive and difficult to procure. With the emergent deep learning techniques, mainly Generative Adversarial Networks (GANs), a paradigm shift has emerged in how satellite imagery is analyzed. GANs offer a powerful framework for unsupervised and semi-supervised LULC classification by learning complex spatial patterns without labelled data. This survey presents a detailed review of recent developments in applying GANs for LULC mapping. We discuss fundamental GAN architectures such as DCGAN, Pix2Pix, CycleGAN, and Self-Attention GAN, and their adaptations for remote sensing. Important feature reviewed include image-to-image translation, data augmentation, super-resolution as well as unsupervised clustering. This survey also highlights performance comparisons with traditional methods outlines future directions including hybrid GAN models, temporal analysis, and integration with self-attention GANs and accurate approaches to land-cover analysis using generative models.