Monitoring of Snow and Ice Cover in Dalad Banner, Inner Mongolia Based on Hyperspectral Mixture Spectral Decomposition

Snow cover is an active systemic variable in Earth's climate change, not only an important indicator of global climate change but also closely related to Earth's energy balance, atmospheric cycle, water cycle, and carbon and nitrogen cycle. Its high reflectivity, high cold storage capacity, enormous latent heat of phase transition, and role as a source and sink of greenhouse gases make it play a crucial role in the Earth's surface energy balance. Quantitative remote sensing technology is the primary means of acquiring snow cover information to monitor snow cover changes, assess its response to climate change, and conduct snow cover hydrological research. Satellite remote sensing is widely used in the field of snow cover research.

Snow has high reflectivity in the visible light band, but low reflectivity in the shortwave infrared band, while most clouds have high reflectivity in both bands. Therefore, calculating the difference between the visible and shortwave infrared bands using the Normalized Difference in Snow Index (NDSI) can effectively identify snow information in optical remote sensing data. However, because snow and clouds have similar spectral characteristics in the visible light band, snow products developed based on optical remote sensing technology are often affected by insufficient solar radiation and cloud cover, resulting in a large number of missing values ​​and posing certain difficulties in application. Therefore, we use a hybrid spectral decomposition method to extract snow area, which can achieve high accuracy and low false positive rate.

CAS Xiguang Aerospace - Snow and Ice Cover Mapping

Data source: XIGUANG003 hyperspectral imager

Region: Dalad Banner, Inner Mongolia

Date of shooting: January 5, 2024

Snow and ice cover mapping includes the following steps:

1. Radiometric calibration, atmospheric correction, and orthorectification of raw L2A data;
2. Spectral dimensionality reduction is performed using Minimum Noise Transform (MNF).
3. Use spatial consistency measurements to determine the dimensions of the data;
4. Calculate the Pure Pixel Index (PPI) and further screen the variables using the Pure Pixel Index (PPI);
5. The results are visualized in N dimensions, automatically clustered, and endmembers are identified.
6. Spectral mapping was performed using a spectral angle mapper (SAM) and a hybrid tuned matched filter (MTMF) to classify and compute endmember abundance simultaneously.
7. Maps are created based on the mapping results and then fused with high-resolution panchromatic imagery.

△ Figure 1 Reflectance image after atmospheric correction and orthorectification

The study area is located in Dalad Banner, Inner Mongolia, specifically in the southern part of Baotou City, on the south bank of the Yellow River. It covers Dalad Town and the Dalad Photovoltaic Power Station. The main landforms in this area are towns, rivers, mines, and deserts, with significant snow cover observed during satellite transit.

△Figure 2 Comparison of reflectivity of various ground features

The presence of a reflection peak at 760 nm in snow indicates its strong reflectivity towards light at that wavelength. This is because snow is composed of ice crystals, which have a strong scattering effect on light. When light shines on snow, some light is absorbed by the ice crystals, while the rest is scattered in various directions. Among the scattered light, the scattering intensity is strongest at 760 nm, hence the reflection peak at this wavelength. According to Figure 2, the difference in reflectivity at 760 nm can be used to further classify snow cover; that is, by monitoring changes in the reflectivity of light at 760 nm, the melting of snow can be monitored. When snow melts, the ice crystals turn into water, and water scatters light less effectively than ice crystals. Therefore, as snow melts, the reflectivity of light at 760 nm decreases.

After pure pixel calculation and endmember extraction, the final number of retained endmembers was 3, two of which represented snow cover. Therefore, the snow cover was subsequently divided into new snow and old snow (ice-snow mixture), and the snow melting situation was analyzed.

Monitoring snow melting can be used for the following applications:

1. Flood forecasting: Snowmelt is one of the main sources of flooding. By monitoring snowmelt, the likelihood and scale of floods can be predicted.
2. Avalanche forecasting: Avalanches are another natural disaster caused by melting snow. By monitoring snow melting, the likelihood and scale of avalanches can be predicted.
3. Water resource management: Snowmelt is a significant source of river runoff. By monitoring snowmelt, changes in river runoff can be predicted, thus enabling water resource management.

△ Figure 3 Abundance distribution map of new snow element.

△ Figure 4 Abundance distribution map of old snow endogenous genus.

Mixture-Tuned Matched Filtering (MTMF) uses a partial unmixing method to find the abundance of endmember spectra. This technique utilizes a "matched filter" (MF) to maximize the response of known endmembers and suppress responses compounded with unknown backgrounds, thus "matching" a known signal. It provides a method for rapidly detecting specific materials based on matching the spectra of endmembers in a library or image, without requiring knowledge of all endmembers in the image scene. Linear spectral unmixing, on the other hand, finds the abundance of materials in image pixels by assuming that the pixel reflectance in each band equals a linear combination of the reflectances of the endmember materials present within the pixel.

Figures 3 and 4 show the results of MTMF, which can visualize the endmember abundance of each pixel and intuitively display the mixed spectral composition of the pixel. The darker the color, the higher the abundance of that endmember. The fresh snow was found to cover the western plains and southern mountains of Dalat Town. The old snow was mainly distributed in the western desert, Dalat Town, and the mining area.The results of subsequent spectral angle mapper (SAM) also confirm this.

△ Figure 5 Classification results of spectral angle mapper

The Spectral Angle Mapper (SAM) matches an image spectrum with a reference spectrum in N-dimensional space. SAM compares the angle between the endmember spectrum (viewed as an n-dimensional vector, where N is the number of bands) and the vector of each pixel in N-dimensional space. A smaller angle indicates a closer match to the reference spectrum. When used on calibrated data, this technique is relatively insensitive to illumination and albedo effects. SAM generates a classification image based on a maximum angle threshold.

Figure 5 represents the final classification results. It can be seen that compared with the MTMF results, the classification effect for the western region is better, effectively reducing the false positive rate. At the same time, it more clearly shows the low snow and ice cover and snow melting phenomenon in Dalat Town in the northeast and the mining area in the southeast.

Statistics on snow and ice coverage area

The statistics on snow and ice coverage are as follows:

Total area covered by satellite imagery: 6,400 square kilometers

Snow cover: 11% (704 square kilometers)

Old snow cover: 6% (384 square kilometers)

By consulting local historical meteorological data, it was found that the highest temperature was around -1℃. Therefore, the main reasons for the melting of snow and ice in this area are likely the following factors:

1. Human activity impact: Densely populated areas and coal mining areas usually have a lot of human activities, such as mining, transportation, and mechanical operations. These activities may change the nature and structure of the surface, reduce the surface roughness, and make it difficult for snow to stay or accumulate on the surface. At the same time, artificial snow removal is carried out in towns and roads. In contrast, the northern areas are mainly villages and farmland with less human interference, so the snow melts more slowly.
2. Cover and facilities: Densely populated areas and coal mining areas typically have a large number of man-made structures such as buildings, equipment and roads. These covers absorb solar radiation, which raises the surface temperature and reduces the formation and maintenance of snow.
3. Heat emissions: Production activities in densely populated areas and coal mines are usually accompanied by a large amount of heat emissions, such as the heat generated by burning coal and the working heat of mechanical equipment. This heat will raise the temperature of the surrounding surface, making it difficult for snow to accumulate in such an environment.

The old snow/ice surface classification results appeared in the Yellow River channel. Combined with meteorological conditions and the date of the Yellow River ice flood season, it was determined to be ice floes composed of ice-water mixture or a partially frozen river channel. Combined with ground data, further analysis of ice floe composition or river freezing can be carried out. The entire process of ice floes in the river channel can be monitored from a larger spatial scale, which also reflects the application value of hyperspectral remote sensing technology.