① Non-uniform terrain
In coastal areas and some other arid areas, the ground albedo may change drastically within a relatively small range. Although today’s satellite navigation performance is far superior to earlier satellite platforms, satellite navigation is not always accurate. Therefore, the dynamic range of a given hypothetical location may include two nearby observation points with completely different albedo, and the multi-time gap historical dynamic range cannot be effective in these situations.
In the SUNY model, there is a sorting method that is applicable to all data observation points, and this method can effectively solve complex terrain problems and effectively correct the remaining solar geometric structure effects that are not completely caused by multiple dynamic ranges. This step assumes that for any particular time interval, within a certain period (such as 1 month), the nth highest clear sky index (GHI/GHIclear) must be at least equal to the given clear sky percentage x. The rank order n and the percentage x depend on the dominant cloud amount in the observation location/time period. The dominant cloud amount can be estimated based on existing low-resolution databases, such as NASA Surface Solar Energy (SSE) or NREL CSR (NREL
2012). For example, in the 30-day period in Arizona, n=8 and x=100%; and in November in Seattle, n=1 and x=100%. In other words, in the specified time interval and month, assuming that at least 8 days in June in Arizona are sunny, and assuming that only one day in November in Seattle is sunny, then less than 8 sunny days indicates that the lower limit is lowered. Before using the SolarGIS model to process the satellite image, the geometric structure of the satellite image needs to be corrected to significantly reduce the mixed pixel impact caused by satellite navigation fluctuations.
The dynamic range and sorting methods have good performance in most locations, but when there is snow on the ground, this method may not work. There are two main reasons:
· The ground will become very bright, especially in dry grassland and wasteland. This brightness greatly reduces the dynamic range of the model (see Figure 1 for details), thereby reducing the accuracy of the model. And in some cases, the ground brightness may even exceed the upper limit of the dynamic range.
·The lower limit of the drag window cannot capture the rapid changes in albedo due to heavy snow.
The challenge of satellite inversion model operation is to firstly detect whether there is snow on the ground, and secondly to avoid problems caused by the reduction of dynamic range and rapid changes.
In the SUNY/SolarAnywhere model, the ground area snow information can be obtained from external data sources. External data sources include: Interactive Multi-Sensor Ice and Snow Mapping System (IMS2012) (available worldwide) and National Hydrological Remote Sensing Center (NOHRSC2012) (available only in the United States). These two data sources update the ground area snow data and the ground resolution data within a few kilometers on a daily basis.
In the SUNY original model, the lower limit drag window of the model is reduced, and the dynamic range is immediately reduced in advance when snow is found, so as to process the dynamic range logic. However, this method will still cause large deviations. On the one hand, it is impossible to effectively distinguish cloudy weather conditions in snow-covered areas in a short period of time; on the other hand, because the dynamic range can be reduced to close to zero, especially in barren areas. Snow-covered area.
The current version of the SUNY model uses the IR channel of the satellite to directly infer the cloud index CI of the snow-covered area.
The IR channel method is a purely empirical method that uses the brightness temperature and ground temperature of each satellite IR channel (see Table 1 for details) (for example, through reanalysis and summary or obtained from a climate summary) and represents a variety of climate environments The GHI measurement values of several North American observation sites are fitted by multiple fittings. Under normal operating conditions, this empirical model is not as accurate as the physical model or the semi-empirical model, but its performance has been greatly improved under snow conditions. Another advantage of the IR channel that measures the brightness temperature is the same as the visible sensor. It can continuously monitor the calibration of the channel using the recorded temperature of the satellite, and it can display the relationship between attenuation and satellite changes without operating adjustments.
|Satellite imager channel||Wavelength range (μm)||Ground resolution at the sub-satellite point||Main detection object|
|1.Visible||0.55~0.75||1km||Clouds, reflectivity, smoke|
|2.Shortwave infrared spectroscopy||3.80~4.00||4 km||Clouds and smoke|
|3.Moisture infrared spectroscopy||6.30~6.70||8 km||Clouds and water vapor|
|4.Surface temperature infrared spectrum||10.20~11.20||4 km||Clouds, water vapor and surface temperature|
|5.Long wave infrared spectroscopy||12.80~13.80||4 km||Clouds and water vapor|
In terms of operation, whenever NOHSRC detects a snowy environment, the SolarAnywhere/SUNY version 3 model can switch from the semi-empirical visible model to the IR mode.
In the SolarGIS model, snow detection is done internally by multi-spectral channels (one visible channel, up to three infrared (R) channels, and auxiliary meteorological parameters). This method is designed on the basis of Durr and Zelenka’s research and development work. The auxiliary snow depth and air temperature data come from NOAA’s Global Forecast System (GFS) database. First, convert the calibrated pixel value into three indexes: ① Snow Cover Index (NDSI); ② Infrared (IR) Cloud Index; ③ Time Variation Index. The reflectance values of the visible channel and the three infrared spectrum channels, as well as the spectral index, variation index, solar geometric structure, and auxiliary data from the meteorological model are all used in a decision tree classifier, and different pixels are assigned to different categories ( The number and choice of spectral indices depend on the mission of the satellite-Meteosat, MTSAT, GMS and GOES). Therefore, for each data point, a category ID (snow, snow-free land, snow-free water, clouds, and unclassified parts) is obtained for each time interval (see Figure 2 for details). Then use the post-classification filtering method to enhance the classification results, aiming to clear the geographical isolation category of the day, and check the consistency in the subsequent time. The classification results are used to determine special cases of high surface albedo (snow covered areas, salt beds, white sand areas). The IR cloud index derived from the infrared channel replaces the visible channel cloud index.