You can use principal component analysis or pan sharpening. You might find this article on increasing the resolution of your images using deep learning techniques useful. It may work effectively for your case.

Conference Paper Lecture Notes in Computer Science

You may explore geostatistical approaches due to their appealing property of perfect coherence. Since you didn't mention anything about your dataset (i.e., if you have any ancilliary data to assist the downscaling), my recommendation is to do a little bit of research online about this topic (downscaling) and come back with a more focused question.

To increase the resolution of Landsat images to 5 meters, a procedure known as “image fusion” . This process involves a combination of high spatial resolution data with high spectral resolution data. Although Landsat images have a spatial resolution of 30 meters, techniques exist to increase the apparent resolution of images using data from other high-resolution sources or super-resolution algorithms.

Here are some possible approaches: 1. Use of High Resolution Data Fusion with High Resolution Images: If high resolution satellite images are available (such as those from the Sentinel-2 satellite, which has a spatial resolution of 10 meters, or from commercial satellites such as WorldView-3), these can be fused with Landsat images to improve resolution. Procedures: Image Pan-sharpening: Pan-sharpening techniques can be used to increase spatial resolution. A common example is the use of the high-resolution panchromatic band together with the lower-resolution multispectral bands. In the case of Landsat, this can be done using the panchromatic band (band 8) which has a resolution of 15 meters, combined with other multispectral bands. Multispectral Fusion Methods: There are several algorithms such as Brovey, IHS (Intensity, Hue, Saturation), and PCA (Principal Component Analysis) that can be used for this purpose.

2. Super-Resolution Algorithms Machine Learning-Based Super-Resolution: Using convolutional neural networks (CNNs) to improve image resolution. Deep learning-based super-resolution techniques can be trained to increase the spatial resolution of Landsat images. Tools and Libraries: Google Earth Engine: Offers several tools for image processing and can be used to implement super-resolution algorithms. TensorFlow and PyTorch: These libraries can be used to develop and train super-resolution models.

3. Interpolation Methods Spatial Interpolation: Interpolation methods, such as bicubic or spline, can be used to increase resolution. However, these methods generally do not add new details and may not be suitable for all types of analysis.

4.Step by Step for Image Fusion Image Acquisition: Download Landsat images from USGS Earth Explorer. If available, obtain high spatial resolution imagery from Sentinel-2 or commercial satellites. Pre-processing: Geometrically correct the images. Align images from different sources. Application of Fusion Technique: If using pan-sharpening, use Landsat's panchromatic band to increase the resolution of the multispectral bands. If using data from another source, use algorithms such as IHS or Brovey to combine the images. For super-resolution, train a suitable model and apply it to Landsat images. Validation: Compare resulting images with reference or field data to ensure accuracy. Analysis and Usage: Use high-resolution images for urban planning, such as land use analysis, green space monitoring, and infrastructure planning.