James Luffman’s Post

High-resolution irradiance data 🌤 is a result of combining various data sources, complex algorithms, and years of development. Satellite imagery is a crucial element, but just one piece of the puzzle.

Satellite imagery isn’t the only input to our irradiance dataset, but it drives a lot of our cloud detection and check out how they line up! Our team uses millions of data points to create high resolution live actuals. Converting satellite imagery into GHI data means including irradiance-impacting factors like water vapor, aerosols, terrain, snow soiling, and more. What could your team be doing with live actuals that are that this realistic? To check out what this data looks like for your assets, create a commercial toolkit account and start evaluating Solcast data for free! #solar #datascience #visualisation #innovation Will Hobbs Mark Mikofski Mucun Sun Tom Bowcutt

New Insights into Global Aerosol Models from 3 Decades of AERONET Data! In our latest work, our team have analyzed three decades of AERONET data, revealing key trends in fine and coarse mode aerosols across the globe. While not using satellite data directly, this work enhances satellite-based aerosol remote sensing, especially for single-view spectroradiometers. Check out the full study on arXiv!

ICEYE Expands High-resolution SAR Data Products with Dwell Fine Imaging Mode

Reimagining Radar in the Form of High-Resolution 4D Sensing Systems (by Clive (Max) Maxfield) https://lnkd.in/eV8VCUuX

In remote sensing, there's a #SAR superpower that many people are unfamiliar with: natural footprint. This lets you "unlock" important parts of an image that you may have missed (no more "damn, it's just out of the frame"). Here's yesterday's St Petersburg shot as an example. But how? How is this possible? When you image with an electro-optical sensor, the edge of the frame is defined by your lens and your focal plane. For SAR, it's defined by the edge of your beam, which has a gradual roll-off from its peak, unlike the sharp cutoff for optical. When we sell an image, we optimize the radar parameters for the area size that a customer wants to collect. But beyond that delivered image area, more data is still collected - which means there is often a larger image area to exploit. Images can be reprocessed to a large footprint, even after collection, to expose this additional data as an image product. So, while the defined footprint of the radar beam is the part that meets specifications, the "natural footprint" is the total extent of the beam that extends beyond those parameters. In this example, we've taken what was planned as a 5x5 image and reprocessed it to 10x10km - *4x as much data!* There are a few caveats for this process: the usable size is dependent on the geometry of the collection, the signal strength is reduced, and radar ambiguities may creep in (so quality is lower, especially towards the edges of the frame). But when the thing you want to see turns out to be just outside the original frame you planned for, extended scene processing provides enormous benefit to customers to solve some of their biggest challenges.

Ground Sampling Distance (GSD), Ground Resolution and Spatial Resolution are terms used when specifying remote sensing system resolution. They are all used to define the smallest feature a remote sensing system can resolve on the ground. The simplest definition of resolution is the size of one camera pixel on the ground.  This is dependent on the altitude and focal length of the lens being used in the system. By similar triangles, we see that GSD / H = w / F, where GSD is the width of a pixel on the ground, H is the altitude, w is the width of a camera pixel and F is the focal length of the lens. A better metric for resolution includes the effects of diffraction and aberrations of the lens used in the remote sensing system. Both effects are included in the Point Spread Function (PSF) of the lens, which represents the image of a point on the ground. If the PSF of a remote sensing lens is less than the width of one pixel on the camera, then the spatial resolution will be limited by the pixel size and the simple definition is accurate. However, if the PSF is larger than one camera pixel, then the resolution is limited by diffraction from the lens aperture or the aberrations of the lens. For more information on this topic, please see the extended discussion on our website: https://lnkd.in/g9da_-C3

Ground Sampling Distance (GSD), Ground Resolution and Spatial Resolution are terms used when specifying remote sensing system resolution. They are all used to define the smallest feature a remote sensing system can resolve on the ground. The simplest definition of resolution is the size of one camera pixel on the ground.  This is dependent on the altitude and focal length of the lens being used in the system. By similar triangles, we see that GSD / H = w / F, where GSD is the width of a pixel on the ground, H is the altitude, w is the width of a camera pixel and F is the focal length of the lens. A better metric for resolution includes the effects of diffraction and aberrations of the lens used in the remote sensing system. Both effects are included in the Point Spread Function (PSF) of the lens, which represents the image of a point on the ground. If the PSF of a remote sensing lens is less than the width of one pixel on the camera, then the spatial resolution will be limited by the pixel size and the simple definition is accurate. However, if the PSF is larger than one camera pixel, then the resolution is limited by diffraction from the lens aperture or the aberrations of the lens. For more information on this topic, please see the extended discussion on our website: https://lnkd.in/gYQjZewv

❓What about georeferenced digital inspection❓ 🚀Now you can incorporate precise spatial data by uploading georeferenced models. Simply define the Spatial Reference System (SRS), such as EPSG 32633 (UTM Zone 33N), and export your model accordingly. If your images contain GNSS data, Twinspect can automatically generate a valid SRS, even if the model wasn’t initially georeferenced. 🗺️ The new Project Map View allows you to visualize your model's exact location along with the images used to create it on an interactive map. 💡This feature also displays annotations and AI detection results directly on the map, offering a comprehensive overview to identify issues or gaps. Additionally, Twinspect now includes a satellite map mode, enabling you to overlay your data on satellite images for more detailed spatial analysis. The high-resolution reality capture data of the pictured building is from our partner Haviq Inspect #twinspect #digitalinspection #innovation #infrastructure #droneinspection #engineering #safety #technology

While multimodal foundation models are already common in some domains, in others their utility is still being explored. Zhitong Xiong et al. developed a unified foundation model for remote sensing to accommodate a variety of types of satellite sensors. Their model accepts RGB, RGB+NIR, multispectral, SAR, and even hyperspectral imagery at different spatial resolutions using a shared backbone. Most previous models have used an independent backbone for each modality. With One-For-All Network (OFA-Net), each modality uses a unique patch embedding before the shared transformer backbone. The network is trained using masked image modeling. In validating on 12 downstream tasks, OFA-Net outperformed alternatives like SatMAE, but a more extensive validation is still to be done. https://lnkd.in/e4fQ7VdY #RemoteSensing #EarthObservation #MachineLearning #DeepLearning #ComputerVision __________________ Enjoyed this post? Like 👍, comment 💬, or re-post 🔄 to share with others. Click "View my newsletter" under my name ⬆️ to join 1500+ readers.