Inputs and Algorithms

Solar-Data-Input-and-Algorithms

We collect the most recent images based on the sample rate of the chosen satellite and load them into our cloud-based storage system. Then, using a statistical ensemble, we compute over 600 million projections every hour, refreshing all of our prediction data with each fresh satellite photo. We call this “Rapid Update” modeling, and we utilize it for all of our solar data services.

We make all of this information accessible in real-time using our ServiceStack API platform, which includes a range of APIs customized to every data product. What follows is a greater overview of our inputs and processes to help you better comprehend how our system works.

Input Data Models

Every 5 to 15 minutes, we integrate updated satellite images from five geostationary satellite systems for our live and forecast data. Overall, this gives us worldwide coverage of our live updates and future projections. Each incoming satellite picture renews the data accessible in our APIs.

Models and input data are combined and regionally modified to work consistently across a broad spectrum of geographical situations. This approach is reasonable and produces reliable and practically consistent outcomes. Because older techniques are often less detailed, they cannot achieve the precision of newer models.

Our satellite-based irradiance simulations can determine solar radiation intensity without the requirement for ground sensors at the desired site. Irradiance calculations based on satellite data range from physically accurate to simply empirical.

Four Key Steps in Generation Solar Data Input and Algorithms

  • Satellite Imagery

Each raw satellite picture is processed in a couple of seconds via geo-coding and quality check algorithms before being sent onto our albedo simulations.

  • Albedo Imagery

The most recent satellite picture is then combined with a collection of digital images for the same place from the previous month. We construct a visualization of the ‘background,’ generally referred to as the ‘albedo,’ using predictive methods, which helps us distinguish cloud cover from snow, bright beach, ocean glare, or other characteristics of the images misinterpreted as the cloud.

  • Cloud Opacity

The existing satellite imagery is then decomposed into cloudy and cloud-free zones. We next use a unique 3D cloud modeling methodology to vertically split cloud cover into levels and characterize the thickness of each level in places identified to have cloud cover. The overall aggregate density is referred to as the ‘cloud opacity.’

Solar Radiation Estimates

In the fourth phase, we employ a slightly different version of the REST2 clear-sky exposure simulation to provide exact estimates of the solar irradiance accessible to cloud-free regions using the most recent global aerosol and water vapor concentration. We utilize cloud opacity to determine the expected quantity of solar radiation reaching the Earth’s surface in cloudy places.

Throughout this stage, we employ a mixture of peer-reviewed methodologies, industry-standard models, and customized algorithms. Furthermore, our extraction model was widely acknowledged as the best performing technique for this purpose worldwide.

Forecasting Cloud Cover

After determining and allocating cloud opacity sectors to the proper vertical layers, we use a mixture of NWP wind fields and image processing algorithms to anticipate where cloud opacity will travel ahead, out to a 4-hour timeframe.

The cloud opacity sector is then projected forward in time using an 18-member array of NWP wind forecasts and image processing vector patterns obtained from the prior hour of satellite data. In our API prediction data, all 18 ensemble elements are integrated to provide statistical forecast data, which includes a P50 prediction as well as accompanying P10 and P90 forecasts.

Our cloud opacity projections are then utilized to provide solar radiation projections for the GHI, DNI, and DHI. An array of five statistical weather models is used to enhance projections out to 7 days. We add machine learning-based bias adjustments on the 24-hour horizon, which are implemented based on previous cloud opacities from satellite images.

Leave a comment

Your email address will not be published. Required fields are marked *