Next Article in Journal
Carbonyl Compounds Observed at a Suburban Site during an Unusual Wintertime Ozone Pollution Event in Guangzhou
Previous Article in Journal
Seasonal–Diurnal Distribution of Lightning over Bulgaria and the Black Sea and Its Relationship with Sea Surface Temperature
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Atmosphere
Volume 15
Issue 10
10.3390/atmos15101234
atmosphere-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Accuracy Assessment of NOAA IMS 4 km Products on the Tibetan Plateau with Landsat-8 OLI Images

by
Duo Chu
1,2
1
Tibet Institute of Plateau Atmospheric and Environmental Sciences, Tibet Meteorological Bureau, Lhasa 850000, China
2
Tibet Key Laboratory of Plateau Atmosphere and Environment Research, Science and Technology Department of Tibet Autonomous Region, Lhasa 850000, China
Atmosphere 2024, 15(10), 1234; https://doi.org/10.3390/atmos15101234
Submission received: 9 September 2024 / Revised: 26 September 2024 / Accepted: 9 October 2024 / Published: 15 October 2024
(This article belongs to the Section Atmospheric Techniques, Instruments, and Modeling)
Browse Figures
Versions Notes

Abstract

:
The NOAA IMS (Interactive Multisensor Snow and Ice Mapping System) is a blended snow and ice product based on active and passive satellite sensors, ground observation, and other auxiliary information, providing the daily cloud-free snow cover extent in the Northern Hemisphere (NH) and having great application potential in snow cover monitoring and research in the Tibetan Plateau (TP). However, accuracy assessment of products is crucial for various aspects of applications. In this study, Landsat-8 OLI images were used to evaluate and validate the accuracy of IMS products in snow cover monitoring on the TP. The results show that (1) average overall accuracy of IMS 4 km products is 76.0% and average mapping accuracy is 88.3%, indicating that IMS 4 km products are appropriate for large-scale snow cover monitoring on the TP. (2) IMS 4 km products tend to overestimate actual snow cover on the TP, with an average commission rate of 45.4% and omission rate of 11.7%, and generally present that the higher the proportion of snow-covered area, the lower the probability of omission rate and the higher the probability of commission rate. (3) Mapping accuracy of IMS 4 km snow cover on the TP generally is higher at the high altitudes, and commission and omission errors increase with the decrease of elevation. (4) Compared with less regional representativeness of ground observations, the spatial characteristics of snow cover based on high-resolution remote sensing data are much more detailed, and more reliable verification results can be obtained. (5) In addition to commission and omission error metrics, the overall accuracy and mapping accuracy based on the reference image instead of classified image can better reveal the general monitoring accuracy of IMS 4 km products on the TP area.

1. Introduction

Snow cover is a critical component of both the radiation balance and the hydrological cycle of the Earth system [1,2,3]. The high albedo of snow cover contributes to the suppression of surface air temperature by as much as 4 °C–8 °C compared with snow-free conditions [4,5,6]. By exerting a large-scale cooling effect, snow cover can have a significant impact on the global climate [7,8]. In addition, snow cover serves as a reservoir, which as it melts supplies a considerable amount of water for ecosystems, irrigation, and human consumption [9,10,11].
The Tibetan Plateau (TP), located in the mid-low latitudes of the Northern Hemisphere (NH), is a massive, elevated land on the Eurasia continent and central region of the High Mountain Asia. Snow cover is one of the major surface cover types on the TP with the largest seasonal variations [12,13]. Snow cover constitutes an important alpine environmental and ecological component of the plateau, and snowmelt plays a crucial role in the water supply in the plateau and downstream areas. Furthermore, snow cover affects regional and global climate change by its hydrological effect and the interaction between land surface and atmosphere, while snow cover anomalies on the TP have a strong link with the development of Asian monsoon systems [8,14]. However, excessive snowfall and long-term snow accumulation and duration often cause snow-related hazards for alpine grassland animal husbandry and road access, which has become one of major meteorologically induced hazards in the TP [13,15].
Fast, timely, and accurate snow cover information is essential to understand snow cover variations, water resources management, and snow-related hazard prevention and mitigation [16,17,18]. The in situ and remote sensing observations are two main ways to obtain information on snow cover on the ground. The in situ observations mainly depend on the meteorological station network. It provides high accurate data at point to small scales but has less spatial representativeness [19]. Satellite remote sensing is characterized by large spatial coverage, various spatial and temporal resolutions, low cost, and more consistency and has become the main approach for large-scale snow cover observations. In particular, in the mountain areas with complex terrain and difficult access, remote sensing technology has become the most effective approach to monitoring snow cover variations [17,18]. At present, there are three large-scale snow cover products available for snow cover monitoring from regional to global scales [20,21,22]. The first is optical snow cover products, e.g., MODIS (Moderate Resolution Imaging Spectroradiometer) series products. The second is passive microwave snow products, e.g., SMMR, SSM/I, AMSR-E, and other derivative products. The third is the blended snow cover product derived from various active and passive satellite data. The representative is the NOAA IMS (Interactive Multisensor Snow and Ice Mapping System) products.
IMS data are widely used in snow cover and snow phenology studies due to their high reliability and fine temporal resolution [23,24] and are prone to be more suitable to investigate the spatial distribution of snow cover. The agreement between IMS and ground-based measurements at the daily scale mostly ranges between 80 and 90% during the winter seasons [25]. The mapping accuracy of IMS products under all weather conditions was higher than the accuracy of Terra and Aqua MODIS daily composite products. The accuracy of the IMS is 40% higher than that of MODIS under various weather conditions [25]. IMS snow cover data are more accurate than passive microwave and MODIS snow cover products in China, and clouds make the optical remote sensing, such as MODIS, unable to exert its unique advantages in daily snow cover mapping [26]. The overall accuracy of the IMS is higher than 79% against Landsat TM images. The IMS snow cover product is suitable for larger-scale analysis and applications, with the advantage over MODIS of being able to detect snow over under cloudy conditions [27].
Snow cover on the TP is an important component of the NH cryosphere and can exist at higher-altitude regions throughout the seasons, becoming a unique feature in global snow cover maps [11,12,13]. The main challenge with operational snow cover monitoring from remote sensing in the TP is how to more effectively detect snow cover under cloudy weather. The spatial resolution of the MODIS daily snow product is much higher than that of IMS products, even though over half of daily MODIS images are seriously contaminated due to being covered by clouds. IMS products provide daily cloud-free snow and ice data in the NH, which are not only crucial for operational large-scale snow and ice detection but are also vital parameters for hazard monitoring, numerical weather prediction (NWP), and hydrological models and climate system research. IMS products have great potential for snow cover monitoring in the TP area. However, a systematic evaluation and validation of products is essential before use. Due to less spatial representativeness of ground-point observations, the high-resolution remote sensing data provide more detailed information regarding the spatial distribution of snow cover, and accuracy verification is more accurate and reliable through image-to-image comparisons.
Therefore, by using cloud-free images of Landsat-8 in the typical regions across the TP, the systematic evaluation and validation of IMS 4 km products is conducted in the study, and the spatial distribution and dependence of accuracy error and main causes are analyzed for the first time. The advantages and limitations of IMS 4 km products on the TP are further discussed. The study aims to better apply IMS products in snow cover monitoring on the TP for timely understanding of the spatial and temporal variations of snow cover and the products’ potential in cryospheric hazard prevention and mitigation.

2. Data and Methods

2.1. Study Area

As shown in Figure 1, the study area is the Tibetan Plateau (TP). As a major part of the global mountain regions, the TP is a vast high mountain region with an average altitude of over 4000 m a.s.l., located at the intersection of Central, South, and East Asia. The TP is surrounded by the Himalayas in the south and southwest, Karakorum Mountain in the northwest, western Kunlun and the Altun mountains in the north, Qilian Mountain in the northeast, and the Hengduan parallel mountains in the southeast. The interior of the TP is relatively flat, and the high mountain ranges include the Gangdise, Nyainqentanglha, eastern Kunlun, Tanggula, Bayan Har, and Anyemaqen [13]. Many large river systems in Asia, such as the Yangtze, Yellow, Yarlung Zangbo (upper Brahmaputra), Salween, Mekong, Ganges, and Indus, originated in the TP, which is known as “water tower of Asia” [28]. The TP is dominated by westerlies in the winter and by South and East Asian monsoons in the summer. Spring and autumn are transition seasons between winter and summer atmospheric circulations [13,29].

2.2. IMS Products

IMS snow and ice products are the longest time series of snow cover products developed by NOAA NESDIS (National Environmental Satellite Data Information Service Center) through the fusion of optical and microwave, active and passive sensor data from most available polar orbit and geostationary satellites and ground snow observation data [30]. IMS data have been available since the 1960s and provide daily the cloud-free snow and ice extent of the NH and have been widely used in snow and ice mapping, climate analysis, and operational weather forecasting models [31,32,33]. IMS real-time data can be downloaded for free from the U.S. National Ice Center (NIC) website, and historical data can be downloaded from the U.S. National Snow and Ice Data Center (NSIDC).
The polar azimuth projection centered on the North Pole was used in IMS products, and the data format includes ASCII, GIF, GeoTIFF, and NetCDF. The GeoTIFF format available from 2004 is more common in GIS and image-processing areas so that more users can access the products to conduct applications and research. The pixel values in the products represent different surface types: (1) represents ocean, (2) represents land, (3) represents sea and lake ice, and (4) represents snow cover. The NSIDC has archived IMS products with 24 km resolution since 1997 and 4 km resolution since February 2004. The resolution of IMS products has been further increased to 1 km since 2 December 2014. The IMS 4 km snow cover extent in the NH on 25 February 2018 is shown in Figure 2. The image processing procedure includes downloading IMS products with the GeoTIFF format from the NSIDC website and converting them to the ArcGIS GRID format and transforming the polar azimuth projection into the Albers projection and clip using the TP boundary.

2.3. Landsat-8 OLI Data

Landsat 8 is the eighth satellite of the Landsat series and was launched on 11 February 2013, including all bands of ETM+ sensors. Compared to the 8-bit data of TM and ETM+ sensors, the OLI data are 12 bits and can obtain a higher radiation accuracy and signal-to-noise ratio. Landsat 9, a partnership project of NASA and USGS and almost identical satellite of Landsat 8, was launched on 27 September 2021, and continues to record important earth observation data from space through the constellation of Landsat 8 and 9 [34].
All Landsat-8 data used in the study were downloaded from www.gscloud.cn and a total of 25 cloud-free Landsat-8 OLI images were used for the accuracy assessment of IMS products. The spatial distribution of 25 Landsat-8 images is shown in Figure 1.

2.4. Mapping Methods

The NDSI (Normalized Difference Snow Index) is the main method for mapping snow cover from remotely sensed data and has been used in various earth observation satellites, including the Landsat series [22,35,36]. NDSI values depend on the presence of snow in a pixel. Snow has very high visible (VIS) reflectance and very low reflectance in the shortwave infrared (SWIR), and these spectral signatures are used to detect snow on the ground and distinguish snow from cloud.
An NDSI-based NASA snow cover mapping algorithm (SNOMAP) is applied to generate MODIS global standard snow products [36,37]. Band 4 and band 6 are used to compute the MODIS NDSI, while corresponding bands for Landsat-8 OLI are band 3 and band 6, and NDSI can be calculated using following equation.
NDSI = [Band3 − Band6]/[Band3 + Band6]
Hall et al. [36] indicated that when the NDSI ≥ 0.40, snow cover on the surface can be identified well in satellite images, and most clouds on the images can also be screened, so the threshold for the NDSI is set to 0.40. On this basis, a reflectance of band 5 greater than 0.11 is used to remove the influence of water bodies on snow cover detection using the spectral signature of the low reflectance of water in the near-infrared band and the high reflectance of snow in this band since snow cover and water bodies often have similar NDSI values. The thresholds of the NDSI ≥ 0.40 and band 5 > 0.11 are used in this study for mapping snow cover from Landsat-8 images.
To retrieve the surface reflectance of satellite data, atmospheric correction is performed to remove the atmospheric effects from the satellite-measured signal. In this study, the Landsat-8 OLI image processing procedure is as follows. First, the raw data of OLI 1–7 images are radiometrically calibrated in ENVI 5.6.1 image-processing software, and the digital number (DN) of images is converted into the top-of-atmosphere (TOA) radiance. Second, the TOA radiance of each band is converted into the surface reflectance by using a FLAASH atmospheric correction module with a mid-latitude winter atmospheric model, a rural aerosol model, 4 km average elevation, and 40 km visibility. After obtaining the surface reflectance of bands 1 to 7, the NDSI can be calculated according to Equation (1) by using the reflectance values of bands 3 and 6. If the NDSI ≥ 0.40 and the reflectance of near-infrared band 5 > 0.11, this pixel is identified as snow; otherwise, it is identified as non-snow pixels.
Based on the image processing above, a Landsat-8 NDSI image with spatial resolution of 30 m is aggregated and is resampled to 1 km, and an IMS image on the same day is converted to UTM projection consistent with the NDSI images and is resampled to 1 km spatial resolution. Finally, the accuracy is evaluated according to the pixel values of two images, and the spatial distribution characteristics of accuracy error with elevation are analyzed by combining DEM (Digital Elevation Model) data with the same spatial resolution and projection coordinates.

2.5. Evaluation Methods

The error matrix is used for accuracy evaluation to compare IMS products and Landsat-8 snow cover images. In the binary snow cover images, the number of pixels with snow on both Landsat 8 and the IMS is set to A, the number of pixels with snow on Landsat 8 but no snow on the IMS is set to B, the number of pixels with no snow on Landsat 8 but snow on the IMS is set to C, and the number of pixels with no snow on both images is set to D. The final error matrix established is shown in Table 1.
Based on the error matrix, many evaluation metrics can be derived. Among these, overall accuracy (OA) is the ratio of the number of pixels with snow or no snow in both images to the total number of pixels. Producer’s accuracy (PA) or mapping accuracy refers to the ratio of the number of pixels with snow in both images to the total number of snow pixels in Landsat 8 images as the reference, also known as the recall rate. User’s accuracy (UA) refers to the ratio of the number of pixels with snow in both images to the total number of IMS snow pixels, also known as snow classification accuracy. Non-snow detection rate (NSDR) is the ratio of the number of snow-free pixels in Landsat 8 and IMS products to all snow-free pixels in the classification results.
In the error matrix, if there is snow or no snow in both images, it indicates that snow cover is correctly detected by the IMS. If both are inconsistent, it shows that the IMS has an error in snow cover detection. There are two different errors: omission error, missing snow when snow is present, and commission error, detecting non-snow features as snow, also referred to as the false alarm rate. The omission error (OE) is the ratio of the number of pixels with snow on Landsat 8 but no snow on the IMS to the total number of Landsat-8 snow pixels, while the commission error rate (CE) refers to the ratio of the number of pixels with no snow on Landsat 8 but snow on the IMS to the total number of no-snow pixels in Landsat 8 as reference images.
OA = (A + D)/(A + B + C + D) × 100
PA = A/(A + B) × 100
UA = A/(A + C) × 100
OE = B/(A + B) × 100
CE = C/(C + D) × 100
NSDR = D/(B + D) × 100
In addition to the metrics above, the kappa coefficient (Κ) is also used to evaluate the consistency between two images as a comprehensive metric. A kappa coefficient > 0.80 can be considered as high consistency between the map to be evaluated and the reference map, a kappa coefficient of 0.60–0.80 can be considered as excellent consistency between two maps, a kappa coefficient of 0.40–0.60 indicates medium consistency, a kappa coefficient of 0.20–0.40 indicates poor consistency, and a kappa coefficient < 0.20 indicates very poor consistency [38].

3. Results and Discussions

3.1. Accuracy Evaluation

In the study, the 25 cloud-free Landsat-8 OLI images across the TP during snow seasons from 2013 to 2019 were selected as reference data. The detailed information and spatial distribution of all images are given in Table 2 and Figure 1. In terms of imaging time, most of images were acquired in winter seasons with a total of 19 images, of which 13 images were acquired in January. In other snow seasons, the images obtained were 1–3 per month. In the TP, the highest number of snow cover days occurs during the winter season, and there are more clear-sky days over the plateau during the winter, which is favorable for snow cover monitoring from satellite observations [39]. The procedure of Landsat image processing is shown in Section 2.4. Landsat-8 NDSI images finally were compared with an IMS snow cover image on the same day with same projection and spatial resolution. The results of the first 10 images are shown in Figure 3, and the final accuracy evaluation results are shown in Table 2 and Figure 4 and Figure 5.
As can be seen from Table 2 and Figure 4, the overall accuracy of IMS 4 km products is high, with a minimum of 54.3% and a maximum of 93.9%. The average overall accuracy is 76.0%, and for most of the images, it is greater than 70%. Yang et al. used Landsat-5 TM to evaluate the accuracy of IMS 4 km snow products on the TP, and the results showed that overall accuracy reached 79.2% and its accuracy decreased with the decrease in snow depth; the commission errors were most common in forest areas; and IMS products are more suitable for large-scale snow cover detection and analysis and are not contaminated by clouds in comparison with MODIS daily products [27]. Chen et al. assessed the applicability of the IMS 4 km snow cover product in southern China based on the Landsat ETM+ images and found that there was a good agreement between snow cover images from IMS and Landsat ETM+ in plain terrain, such as agricultural land, with an average agreement of over 85%, whereas in forest and mountain areas, its accuracy is below 75%, and the IMS overestimates snow cover by over 50% [26]. In patchy snow-covered areas, the accuracy of the IMS 4 km product is even worse, and the overestimation of snow-covered areas is more obvious. Moreover, in the areas with complex terrain, the accuracy of snow cover detection is more problematic due to the coarse spatial resolution of the IMS and mixed pixels [26]. Table 2 also shows that mapping accuracy is higher than overall accuracy, with a minimum of 53.5%, a maximum of 99.5%, and an average of 88.3%. The user’s accuracy ranges from 29.7% to 96.0%, with an average of 62.3%, lower than the mapping accuracy and overall accuracy. The snow-free classification accuracy represented by the non-snow detection rate is very high, ranging from 31.8% to 98.4%, with an average of 84.9%.
In addition, the detection errors of IMS products on the TP are further investigated and are shown in Figure 5 and Table 2. The omission or missing errors indicate that the IMS did not correctly detect snow cover on the ground as represented by corresponding Landsat-8 images and the IMS missed snow cover on the ground. The lowest missing rate of IMS 4 km is 0.5%, which occurred on 12 January 2017, followed by 20 February 2014 and 19 January 2017, with a missing rate of 1.2% and 1.6%, respectively. The missing rate is less than 2% in three reference images, which means that snow cover in these reference images was almost effectively detected by the IMS. The maximum missing rate is 46.5%, which occurred on 8 January 2018, showing that nearly half of the snow-covered areas were missed by IMS detection, followed by 31.1% and 27.8%; that is, around one-third and one-quarter of snow cover on the reference images were missed and were not detected by the IMS. Average omission error rate of IMS 4 km was 11.7%. The commission errors mean that the IMS detects non-snow features as snow. The lowest commission error rate was on 10 April 2015, with 7.7%, followed by 9.7% on 11 November 2013, and the rest of commission errors were higher than 10%. The maximum commission error rate was 82.9%, which appeared on an IMS snow cover image on 19 January 2017, followed by 12 January 2017 with 82.1%, and the remaining was less than 80%. The average commission error rate is 45.4%, which is 33.7% higher than the average omission error rate. The study also found that commission errors were obviously higher than omission errors for IMS snow cover products [32].
In addition to the indicators above, the kappa coefficient is one of the commonly used indicators to express classification accuracy. As presented in Table 2, the minimum kappa coefficient between images of IMS 4 km and Landsat 8 is 0.1440, and the remaining are all greater than 0.2. There are 13 pairs of images with a kappa coefficient less than 0.4, and the kappa coefficient of the remaining 12 pairs of images is greater than 0.4, accounting for around half of the pairs of images to be evaluated. The maximum kappa coefficient is 0.8763, followed by 0.6863, which are two pairs of images with a kappa coefficient greater than 0.6. The average kappa coefficient of 25 image pairs is 0.3942, indicating that agreement between the IMS 4 km snow cover and Landsat-8 reference images is reasonable.
The IMS 4 km detection errors show that the average commission error rate is obviously higher than the omission error rate. The former is 45.4%, and the latter is 11.7%, with a difference of 33.7%, which means that the IMS product overestimates actual snow cover on the surface. The overall accuracy and mapping accuracy of IMS 4 km products are high, but the kappa coefficient is low, which is mainly attributed to the high commission error rate of IMS 4 km products. By comparing IMS 4 km and MODIS products, it was found that the IMS relatively overestimates the actual snow-covered area. The main reason is that due to low resolution of the IMS, the ability to detect patchy snow cover on the ground is limited, and the commission error rate is higher than the omission error rate [40]. The coarse remote sensing data do not represent well the detailed information of patchy snow cover on the TP and, instead, just consider such areas as completely snow covered, resulting in overestimating the surface snow cover [41,42,43].
The studies conducted in North America show that the IMS overestimates the actual snow cover and the commission error rate is higher than omission error rate [25,44], which is consistent with the conclusions from this study. Furthermore, even though multi-source data and a manual mapping method are used, the accuracy of IMS snow products still remains uncertain under thick cloud conditions. When clouds cover the ground for several successive days, the IMS produces a current snow cover map in the NH through updating snow cover information of the previous day. As a result, the IMS often overestimates the snow cover on the surface in some regions [31].

3.2. Impact of Snow-Covered Area on Accuracy Evaluation

The fifth column of Table 2 shows the percentage of snow-covered area (snow pixels) in the total image area (all pixels) in Landsat-8 NDSI images. The snow-covered areas of 25 images account for 14.6–95.1%. In order to find out the impact of snow-covered areas of reference data on IMS accuracy, a simple linear correlation analysis was carried out between the proportion of snow-covered areas (PSCA) and accuracy evaluation metrics, and results are shown in Table 3. There is a very significant positive relationship between the PSCA and user’s accuracy and commission error rate, a very significant negative relationship between the PSCA and missing rate, and a significant positive correlation between the PSCA and mapping accuracy. It shows that the larger the snow cover extent in Landsat-8 images, the higher the snow detection ability of IMS products, the greater the probability of false detection, and the lower the probability of a missing rate. The larger snow cover extent in Landsat-8 images means that snow cover on the surface is more continuous, which is more suitable for the IMS to effectively monitor snow cover on the ground. At the same time, the probability of a commission error is higher, while the missing rate is less accordingly. There is a very significant negative and positive linear correlation between the PSCA and the missing and commission error rate, indicating that the larger the snow cover area in Landsat-8 images, the less the missing rate, and the higher the commission rate. That is to say, the higher the proportion of snow cover in the reference images, the lower the probability of missing snow detection by the IMS, and the higher the probability of false snow detection by the IMS. A very significant negative relationship exists between the PSCA and the snow-free detection rate, which indicates that the monitoring accuracy of IMS products for non-snow cover features decreases with the increase in snow-covered areas in the reference images.
In comparison with significant linear relationships for non-snow cover features, there is no significant linear correlation between the PSCA and overall accuracy, which means that there is no obvious impact of the PSCA on overall accuracy for IMS snow cover monitoring. Therefore, given the impact of snow-covered areas in the reference images on evaluation accuracy, when validating coarse-resolution snow cover products using high-resolution satellite data, it is important to select the representative cloud-free high-resolution images with the relatively average distribution of snow and snow-free areas on the surface as ground truth data.

3.3. Relationships between Metrics

The intercorrelations between accuracy evaluation metrics are shown in Table 3. It is clear that there is a very significant positive and negative linear relationship between the kappa coefficient and the overall accuracy and commission error rate, respectively, indicating that a higher kappa coefficient means a higher overall accuracy and a lower error of false detection by IMS 4 km products for snow cover monitoring. In seven metrics, as shown in Table 3, there is a very significant positive and obvious negative relationship between the overall accuracy and user’s accuracy and commission error, which means that the higher the overall accuracy, the higher the user’s accuracy and the lower the commission error rate of IMS snow products. Significant positive and negative relationships exist between the mapping accuracy and the commission error rate and omission error rate, respectively, and a significant positive relationship with user’s accuracy occurs. It indicates that the higher the mapping accuracy, the lower the missing rate and the higher the false detection rate of IMS snow products. Similar to the producer’s accuracy above, in addition to significant positive relationships between the user’s accuracy and overall accuracy/mapping accuracy, there are significant positive and negative relationships between the user’s accuracy and the commission error and missing error rate, respectively, which shows that the higher the user’s accuracy, the lower the missing rate and the higher the false detection rate of IMS snow products. The negative relationship between the non-snow detection rate and the false detection rate is statistically significant, showing that high snow-free classification accuracy means a lower probability of commission errors for IMS products due to much fewer snow-covered areas. Additionally, when comparing snow cover on the surface detected by the IMS with Landsat-8 images as references, a very significant negative correlation exists between the omission and commission error rates, which means that the higher the omission error rate, the lower the commission error rate in general and vice versa. How to reduce these two errors as much as possible and balance the two errors more effectively are fundamental to improve the accuracy of snow cover monitoring from satellite observations.
Selecting appropriate evaluation metrics is important to evaluate snow cover products and their practical applications, which is of great importance to improve the accuracy of snow cover products. From the analysis above, it is concluded that in addition to the omission and commission errors, the overall accuracy and mapping accuracy (PA or recall rate) based on the high-resolution reference images instead of classified images can better reflect the general accuracy of snow cover products for evaluating the accuracy of low–medium resolution snow cover products using high-resolution satellite data.

3.4. Spatial Dependence of Evaluation Accuracy

The accuracy evaluation of IMS 4 km products on the TP based on Landsat-8 OLI images shows that the average commission error rate is 45.4% and the average omission error rate is 11.7%. IMS 4 km products tend to overestimate snow cover on the ground. The accuracy of IMS 4 km products varies greatly in different regions in the TP based on the 25 Landsat-8 images, with the omission error rate ranging from 0.5% to 46.5% and the commission error rate ranging from 7.7% to 82.9%. In order to analyze spatially the topographic dependence of the evaluation accuracy of IMS 4 km products in the TP, two images with the highest commission errors and omission errors were selected from the IMS products from Table 2 and were compared with corresponding Landsat 8 images, and the spatial distribution characteristics and causes of detecting errors were examined in combination with DEM data. The SRTM (Shuttle Radar Topography Mission) DEM released by the USGS EROS (Earth Resources Observation and Science Center) was used. The original spatial resolution of the SRTM DEM is 90 m, which is resampled to 1 km spatial resolution consistent with the IMS 4 km and Landsat-8 snow cover map.
The highest omission error rate of IMS 4 km snow cover images occurred on 8 January 2018, with a 46.5% missing rate and a 23.1% commission error rate, while the overall accuracy was 70.6%. The mapping accuracy was the lowest among all images at 53.5%. The spatial distribution of accuracy errors of the IMS 4 km image is shown in Figure 6a. A total of 5357 snow pixels were correctly detected by the IMS, 4652 snow pixels were missed by the IMS, while a total of 6242 snow-free pixels were mistakenly detected as snow pixels by the IMS, and non-snow pixels both in IMS and Landsat-8 images is 20,815 in total. According to Figure 6a and DEM data, most of the snow-covered pixels both in Landsat and the IMS are located in the northwest and southeastern corners with higher elevation, whereas snow pixels in Landsat and non-snow pixels in the IMS are mainly in the surrounding area of snow-covered pixels. Snow pixels in the IMS and snow-free pixels in Landsat as commission errors are mainly distributed in the north and south with lower elevations. The snow-free pixels both in Landsat and the IMS are mainly distributed in the broad central region of the image where the elevation is relatively lower. The hypsographic curve in Figure 6b shows the accuracy error with a 100 m interval of elevation. The lowest elevation of snow-covered pixels both in Landsat and the IMS is 3638 m, the highest elevation is 5728 m, and the average elevation is 4625 m. Most of the snow pixels are located in the area above 4500 m, accounting for 50.1% of all snow-covered pixels both in Landsat and the IMS. The lowest elevation of pixels with snow in Landsat but no snow in the IMS is 3236 m, with a highest point of 5480 m and an average elevation of 4520 m. The total area above 4500 m accounts for 37.0% of all pixels missed by IMS detection. The lowest elevation of pixels with no snow in Landsat but snow in the IMS is 2981 m, with a highest point of 5425 m and an average elevation of 4332 m. The total area with above 4500 m of elevation is 16.1% of all pixels falsely detected as snow by the IMS. The lowest point of snow-free pixels both in Landsat and the IMS is 2786 m, with a highest elevation of 5333 m and an average elevation of 4144 m. The total area above 4500 m accounts for 5.0% of all snow-free pixels both in Landsat and the IMS.
As analyzed above, it is clear that the average elevation of pixels with snow both in Landsat and the IMS is the highest, followed by pixels with snow in Landsat but non-snow in the IMS, while the average elevation of snow-free pixels both in Landsat and the IMS is the lowest. The main reason is that in the TP the higher altitudes are not only more favorable for snowfall due to the orographic effect, but the higher altitudes also make snow cover on the ground more stable and less patchy due to lower temperatures, which results in a higher mapping accuracy metric for satellite-based remote sensing products such as the IMS. The massive high mountain topography of the TP is the main condition for the presence and persistence of snow cover on the plateau at the mid–low latitudes of the NH [12,13,18]. Snow cover on the TP shows strong elevation dependence: the higher the altitude, the higher the snow cover frequencies, the longer the snow cover duration, and the more stable the intra-annual variation [13]. At higher elevations, the increase in precipitation appears to compensate for the increase in air temperature such that the snow-free period has decreased [45]. However, snow cover at lower altitudes can persist for a shorter duration and makes snow easier to be fragmented and melt due to higher temperatures, causing snow cover on the ground to be more likely to be missed or falsely identified by IMS products, eventually increasing snow detection errors. Snow cover on the TP shows altitudinal dependence and the length of the snow-covered season appears to be decreasing at lower elevations because of the increase in air temperatures [12,13], while snow cover distribution is generally temperature-driven at low altitudes [45].
The highest commission error rate occurred on January 19, 2017, with up to 82.9%, while its overall accuracy and kappa coefficient are the lowest in all images, with 54.3% and 0.1440, respectively. Based on DEM data, the spatial distribution of accuracy errors and main causes were analyzed. As given in Figure 7a, the snow-covered pixels both in Landsat and the IMS is 1671, most of which are located in the central image. IMS products failed to detect 265 snow-covered pixels in the Landsat image, which covers a small area in the north of the image. There is a large area where there was not snow in the Landsat but snow in the IMS image, which is mainly located in the north and south of image, with a total number of pixels of 16,714 and a commission error rate that is the highest in the 25 images. Total snow-free pixels both in Landsat 8 and IMS images is 3437, which are mainly distributed in the north of image.
The accuracy error of the IMS 4 km product with a 100 m interval of elevation is shown in Figure 7b. The lowest elevation of snow-covered pixels both in Landsat and the IMS is 4014 m, with a highest elevation of 5399 m and an average elevation of 4610 m. Most of these pixels are located in areas above 4500 m, accounting for 63.3% of all snow pixels both in Landsat and the IMS. The lowest elevation of pixels with snow in Landsat but no snow in the IMS is 4244 m, with a highest point of 5150 m and an average altitude of 4600 m. The area above 4500 m accounts for 61.9% of all pixels missed by the IMS. The lowest elevation of pixels with no snow in Landsat but snow in the IMS is 3892 m, with a highest altitude of 5389 m and an average altitude of 4560 m. The area with above 4500 m accounts for 67.3% of all pixels. The lowest altitude of snow-free pixels both in Landsat and the IMS is 4167 m, with a highest point of 5176 m and an average altitude of 4482 m. The area above 4500 m is 40.9% of all snow-free pixels.
It is evident that the average elevation of snow-covered pixels both in Landsat and the IMS, as snow cover pixels correctly detected by IMS products, is the highest, with the highest maximum altitude among evaluation metrics, followed by the average altitude of all pixels that were missed and falsely detected by the IMS, respectively, whereas the average elevation of snow-free pixels both in Landsat and the IMS, as snow-free areas correctly detected by IMS products, is the lowest among different metrics. Snow cover on the TP is strongly associated with terrain features with longer duration in the high mountains and shorter duration in the vast interior and river valleys of the TP. The higher snow cover frequency corresponds well with the huge mountains [13,21]. The elevation is one of the major factors affecting the accuracy of IMS snow cover monitoring in the TP. In terms of mapping accuracy and non-snow detection, it generally presents that the higher the elevation, the higher the mapping accuracy for satellite remote sensing, while the lower the elevation, the higher the accuracy of snow-free detection by satellite remote sensing. However, there is no obvious relationship between overall accuracy and elevations since the accuracy of snow-free detection is involved in the overall accuracy.
Snow cover distribution depends on the terrain height on the TP, and the duration of snow persistence varies in different elevation ranges and generally becomes longer with increases in the terrain elevation [3,45,46]. The higher altitude is more favorable to persistence, spatial integrity, and stability of snow cover on the ground due to lower temperatures, and mapping accuracy tends to increase. On the contrary, at lower altitudes, the higher temperatures are more likely to cause snow melting and snow being more fragmented spatially, leading to an increasing rate of missing and false detection of snow cover on the surface by satellite remote sensing, while in areas with much lower altitudes, a much higher temperature may lead to rapid snow melting or snow-free areas being dominant, which mainly presents snow-free features on the ground and can be represented by a snow-free detection metric.
In a word, the mapping accuracy of IMS 4 km products on the TP area is generally higher at higher altitudes. With a decrease in altitudes, the spatial distribution of snow cover on the ground tends to be patchy and fragmented, which is more likely to cause an increase in missing and false detection rates for snow cover monitoring. Since snow-free detection is involved in the overall accuracy metric and the high probability of a snow-free area generally is in the river valley and basins at the lower altitudes, there are no elevation dependences for the overall accuracy metric in the TP area.

4. Conclusions

In this study, the accuracy of IMS 4 km products for snow cover monitoring on the TP is evaluated using Landsat-8 OLI snow cover images, and the spatial distribution and dependence of accuracy error and main factors are analyzed for the first time. The main conclusions drawn are as follows:
(1)
The overall accuracy of IMS 4 km products is relatively high, ranging from 54.3% to 93.9%, with an average accuracy of 76.0%. Average mapping accuracy and snow-free detection accuracy are 88.3% and 84.9%, respectively. IMS 4 km products are applicable for large-scale snow cover detection on the TP.
(2)
The average omission error is 11.7%, and the commission error is 45.4%. IMS 4 km products overestimate snow cover on the TP and generally present that with an increasing proportion of snow-covered areas, the probability of omission errors tends to decrease while the probability of commission errors tends to increase. Therefore, when evaluating the accuracy of low- to medium-resolution snow cover products using high-resolution satellite images, it is important to select typical cloud-free images in which snow and snow-free areas are more averagely distributed as reference data.
(3)
Elevation is one of the important factors that affect the accuracy of IMS products on the TP, and the mapping accuracy of IMS 4 km products on the TP is elevation-dependent in general, which presents that mapping accuracy is higher at higher-altitude areas, while the missing and false detection rates tend to increase with a decrease in elevation.
(4)
At present, the spatial resolution of IMS snow and ice products has been improved to 1 km. However, there is still a challenge to reduce detection errors and uncertainties of IMS products on the TP due to patchiness and the rapid change of snow cover on the TP. As compared to IMS products as the binary data, the fractional snow cover is a more useful measure of snow cover and can more accurately represent the actual snow cover on the surface. Therefore, it is very important to use fractional snow cover products with higher spatial and temporal resolution to monitor and study snow cover variations in the TP so as to better reveal the spatial distribution and temporal evolution characteristics of snow cover on the TP.

Funding

This research work was financially supported by the Second Tibetan Plateau Scientific Expedition and Research (STEP) Program (2019QZKK010312;2019QZKK0603), the Key Science and Technology Project of Tibet Autonomous Region (XZ202201ZD0005G01), and the National Natural Science Foundation of China (41561017).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy.

Acknowledgments

The author would like to acknowledge the NSIDC for providing the IMS products and the USGS EROS for providing the Landsat-8 images and the SRTM DEM data.

Conflicts of Interest

The author declares no conflict of interest.

References

  1. Bormann, K.J.; Brown, R.D.; Derksen, C.; Painter, T.H. Estimating snow-cover trends from space. Nat. Clim. Change 2018, 8, 924–928. [Google Scholar] [CrossRef]
  2. Musselman, K.N.; Clark, M.P.; Liu, C.; Ikeda, K.; Rasmussen, R. Slower snowmelt in a warmer world. Nat. Clim. Change 2017, 7, 214–219. [Google Scholar] [CrossRef]
  3. You, Q.; Wu, T.; Shen, L.; Pepin, N.; Zhang, L.; Jiang, Z.; Wu, Z.; Kang, S.; AghaKouchak, A. Review of snow cover variation over the Tibetan Plateau and its influence on the broad climate system. Earth Sci. Rev. 2020, 201, 103043. [Google Scholar] [CrossRef]
  4. Robinson, D.A.; Dewey, K.F.; Heim, R.R. Global snow cover monitoring: An update. Bull. Am. Meteorol. Soc. 1993, 74, 1689–1696. [Google Scholar] [CrossRef]
  5. Déry, S.J.; Brown, R.D. Recent Northern Hemisphere snow cover extent trends and implications for the snow-albedo feedback. Geophys. Res. Lett. 2007, 34, 60–64. [Google Scholar] [CrossRef]
  6. Choi, G.; Robinson, D.A.; Kang, S. Changing Northern Hemisphere snow seasons. J. Clim. 2010, 23, 5305–5310. [Google Scholar] [CrossRef]
  7. Lievens, H.; Demuzere, M.; Marshall, H.-P.; Reichle, R.H.; Brucker, L.; Brangers, I.; de Rosnay, P.; Dumont, M.; Girotto, M.; Immerzeel, W.W.; et al. Snow depth variability in the Northern Hemisphere mountains observed from space. Nat. Commun. 2019, 10, 4629. [Google Scholar] [CrossRef]
  8. Li, W.; Guo, W.; Qiu, B.; Xue, Y.; Hsu, P.-C.; Wei, J. Influence of Tibetan Plateau snow cover on East Asian atmospheric circulation at medium-range time scales. Nat. Commun. 2018, 9, 4243. [Google Scholar] [CrossRef] [PubMed]
  9. Hale, K.E.; Jennings, K.S.; Musselman, K.N.; Livneh, B.; Molotch, N.P. Recent decreases in snow water storage in western North America. Commun. Earth Environ. 2023, 4, 170. [Google Scholar] [CrossRef]
  10. Siirila-Woodburn, E.R.; Rhoades, A.M.; Hatchett, B.J.; Huning, L.S.; Szinai, J.; Tague, C.; Nico, P.S.; Feldman, D.R.; Jones, A.D.; Collins, W.D.; et al. A low-to-no snow future and its impacts on water resources in the western United States. Nat. Rev. Earth Environ. 2021, 2, 800–819. [Google Scholar] [CrossRef]
  11. Barnett, T.P.; Adam, J.C.; Lettenmaier, D.P. Potential impacts of a warming climate on water availability in snow-dominated regions. Nature 2005, 438, 303–309. [Google Scholar] [CrossRef] [PubMed]
  12. Qin, D.; Liu, S.; Li, P. Snow cover distribution, variability, and response to climate change in western China. J. Clim. 2006, 19, 1820–1833. [Google Scholar]
  13. Chu, D.; Liu, L.; Wang, Z. Snow cover on the Tibetan Plateau and topographic controls. Remote Sens. 2023, 15, 4044. [Google Scholar] [CrossRef]
  14. Wu, G.; Liu, Y.; He, B.; Bao, Q.; Duan, A.; Jin, F.-F. Thermal controls on the Asian summer monsoon. Sci. Rep. 2012, 2, 404. [Google Scholar] [CrossRef] [PubMed]
  15. Cui, P.; Ge, Y.; Li, S.; Li, Z.; Xu, X.; Zhou, G.G.; Chen, H.; Wang, H.; Lei, Y.; Zhou, L.; et al. Scientific challenges in disaster risk reduction for the Sichuan–Tibet Railway. Eng. Geol. 2022, 309, 106837. [Google Scholar] [CrossRef]
  16. Eckerstorfer, M.; Vickers, H.; Malnes, E.; Grahn, J. Near-real time automatic snow avalanche activity monitoring system using Sentinel-1 SAR data in Norway. Remote Sens. 2019, 11, 2863. [Google Scholar] [CrossRef]
  17. Liang, T.; Zhang, X.; Xie, H.; Wu, C.; Feng, Q.; Huang, X.; Chen, Q. Toward improved daily snow cover mapping with advanced combination of MODIS and AMSR-E measurements. Remote Sens. Environ. 2008, 112, 3750–3761. [Google Scholar] [CrossRef]
  18. Gao, J.; Williams, M.W.; Fu, X.; Wang, G.; Gong, T. Spatiotemporal distribution of snow in eastern Tibet and the response to climate change. Remote Sens. Environ. 2012, 121, 1–9. [Google Scholar] [CrossRef]
  19. Immerzeel, W.; Droogers, P.; de Jong, S.; Bierkens, M. Large-scale monitoring of snow cover and runoff simulation in Himalayan river basins using remote sensing. Remote Sens. Environ. 2009, 113, 40–49. [Google Scholar] [CrossRef]
  20. Che, T.; Li, X.; Jin, R.; Huang, C. Assimilating passive microwave remote sensing data into a land surface model to improve the estimation of snow depth. Remote Sens. Environ. 2014, 143, 54–63. [Google Scholar] [CrossRef]
  21. Pu, Z.; Xu, L.; Salomonson, V.V. MODIS/Terra observed seasonal variations of snow cover over the Tibetan Plateau. Geophys. Res. Lett. 2007, 34, 137–161. [Google Scholar] [CrossRef]
  22. Zhang, H.; Zhang, F.; Zhang, G.; Che, T.; Yan, W.; Ye, M.; Ma, N. Ground-based evaluation of MODIS snow cover product V6 across China: Implications for the selection of NDSI threshold. Sci. Total Environ. 2019, 651, 2712–2726. [Google Scholar] [CrossRef]
  23. Brown, R.D.; Robinson, D.A. Northern Hemisphere spring snow cover variability and change over 1922–2010 including an assessment of uncertainty. Cryosphere 2011, 5, 219–229. [Google Scholar] [CrossRef]
  24. Chen, X.; Long, D.; Liang, S.; He, L.; Zeng, C.; Hao, X.; Hong, Y. Developing a composite daily snow cover extent record over the Tibetan Plateau from 1981 to 2016 using multisource data. Remote Sens. Environ. 2018, 215, 284–299. [Google Scholar] [CrossRef]
  25. Chen, C.; Lakhankar, T.; Romanov, P.; Helfrich, S.; Powell, A.; Khanbilvardi, R. Validation of NOAA-Interactive Multisensor Snow and Ice Mapping System (IMS) by comparison with ground-based measurements over continental United States. Remote Sens. 2012, 4, 1134–1145. [Google Scholar] [CrossRef]
  26. Chen, X.; Jiang, L.; Yang, J.; Pan, J. Validation of ice mapping system snow cover over southern China based on Landsat Enhanced Thematic Mapper Plus imagery. J. Appl. Remote Sens. 2014, 8, 084680. [Google Scholar] [CrossRef]
  27. Yang, J.; Jiang, L.; Ménard, C.B.; Luojus, K.; Lemmetyinen, J.; Pulliainen, J. Evaluation of snow products over the Tibetan Plateau. Hydrol. Process 2015, 29, 3247–3260. [Google Scholar] [CrossRef]
  28. Immerzeel, W.W.; Van Beek, L.P.H.; Bierkens, M.F.P. Climate change will affect the Asian water towers. Science 2010, 328, 1382–1385. [Google Scholar] [CrossRef]
  29. Chu, D. Remote Sensing of Land Use and Land Cover in Mountain Region; Springer Nature: Singapore, 2020. [Google Scholar]
  30. Ramsay, B.H. The interactive multisensor snow and ice mapping system. Hydrol. Process 1998, 12, 1537–1546. [Google Scholar] [CrossRef]
  31. Helfrich, S.R.; McNamara, D.; Ramsay, B.H.; Baldwin, T.; Kasheta, T. Enhancements to, and forthcoming developments in the Interactive Multisensor Snow and Ice Mapping System (IMS). Hydrol. Process 2007, 21, 1576–1586. [Google Scholar] [CrossRef]
  32. Frei, A.; Lee, S.Y. A comparison of optical-band based snow extent products during spring over North America. Remote Sens. Environ. 2010, 114, 1940–1948. [Google Scholar] [CrossRef]
  33. Chang, A.T.C.; Foster, J.L.; Hall, D.K. Effects of forest on the snow parameters derived from microwave measurements during the BOREAS winter field campaign. Hydrol. Process 1996, 10, 1565–1574. [Google Scholar] [CrossRef]
  34. Masek, J.G.; Wulder, M.A.; Markham, B. Landsat 9: Empowering open science and applications through continuity. Remote Sens. Environ. 2020, 248, 111968. [Google Scholar] [CrossRef]
  35. Hall, D.K.; Riggs, G.A.; Salomonson, V.V. Theoretical Basic Document (ATBD) for the MODIS Snow and Sea Ice-Mapping Algorithms; NASA/Goddard Space Flight Center: Greenbelt, MD, USA, 2001; p. 20771. [Google Scholar]
  36. Hall, D.K.; Riggs, G.A.; Salomonson, V.V. Development of methods for mapping global snow cover using moderate resolution imaging spectroradiometer data. Remote Sens. Environ. 1995, 54, 127–140. [Google Scholar] [CrossRef]
  37. Hall, D.K.; Riggs, G.A.; Salomonson, V.V.; DiGirolamo, N.E.; Bayr, K.J. MODIS snow-cover products. Remote Sens. Environ. 2002, 83, 181–194. [Google Scholar] [CrossRef]
  38. Congalton, R.G.; Green, K. Assessing the Accuracy of Remotely Sensed Data: Principles and Practices; CRC Press: Boca Raton, FL, USA, 2019. [Google Scholar]
  39. Duo, C.; Yong, Y.; Jiancan, L. The variations of snow cover days over the Tibetan Plateau during 1981–2010. J. Glaciol. Geocryol. 2015, 37, 1461–1472. (In Chinese) [Google Scholar]
  40. Liu, X.; Jin, X.; Ke, C. Accuracy evaluation of the IMS snow and ice products in stable snow covers regions in China. J. Glaciol. Geocryol. 2014, 36, 500–507. (In Chinese) [Google Scholar]
  41. Tang, Z.; Deng, G.; Hu, G.; Zhang, H.; Pan, H.; Sang, G. Satellite observed spatiotemporal variability of snow cover and snow phenology over high mountain Asia from 2002 to 2021. J. Hydrol. 2022, 613, 128438. [Google Scholar] [CrossRef]
  42. Ueno, K.; Tanaka, K.; Tsutsui, H.; Li, M. Snow cover conditions in the Tibetan Plateau observed during the winter of 2003/2004. Arct. Antarct. Alp. Res. 2007, 39, 152–164. [Google Scholar] [CrossRef]
  43. Yu, J.; Zhang, G.; Yao, T.; Xie, H.; Zhang, H.; Ke, C.; Yao, R. Developing daily cloud-free snow composite products from MODIS Terra-Aqua and IMS for the Tibetan Plateau. IEEE T. Geosci. Remote 2016, 54, 2171–2180. [Google Scholar] [CrossRef]
  44. Brubaker, K.L.; Pinker, R.T.; Deviatova, E. Evaluation and comparison of MODIS and IMS snow-cover estimates for the continental United States using station data. J. Hydrometeorol. 2005, 6, 1002–1017. [Google Scholar] [CrossRef]
  45. Pu, Z.; Xu, L. MODIS/Terra observed snow cover over the Tibet Plateau: Distribution, variation and possible connection with the East Asian Summer Monsoon (EASM). Theor. Appl. Climatol. 2009, 97, 265–278. [Google Scholar] [CrossRef]
  46. Jain, S.; Goswami, A.; Saraf, A. Role of elevation and aspect in snow distribution in western Himalaya. Water Resour. Manag. 2009, 23, 71–83. [Google Scholar] [CrossRef]
Atmosphere 15 01234 g001
Figure 1. Study area and location of Landsat-8 OLI images selected for validation. The background image is snow cover extent from IMS 4 km products on 25 December 2018.
Figure 1. Study area and location of Landsat-8 OLI images selected for validation. The background image is snow cover extent from IMS 4 km products on 25 December 2018.
Atmosphere 15 01234 g001
Atmosphere 15 01234 g002
Figure 2. IMS 4 km snow cover extent in the NH on 25 February 2018.
Figure 2. IMS 4 km snow cover extent in the NH on 25 February 2018.
Atmosphere 15 01234 g002
Atmosphere 15 01234 g003
Figure 3. First 10 Landsat-8 band 6-3-2 composite images (a1,a2) and corresponding snow cover maps at 1 km spatial resolution of Landsat-8 (b1,b2) and IMS (c1,c2) in row. The date, path, and row of Landsat-8 OLI images are shown at the top of images.
Figure 3. First 10 Landsat-8 band 6-3-2 composite images (a1,a2) and corresponding snow cover maps at 1 km spatial resolution of Landsat-8 (b1,b2) and IMS (c1,c2) in row. The date, path, and row of Landsat-8 OLI images are shown at the top of images.
Atmosphere 15 01234 g003
Atmosphere 15 01234 g004
Figure 4. Overall accuracy of IMS 4 km products on the TP based on Landsat-8 images.
Figure 4. Overall accuracy of IMS 4 km products on the TP based on Landsat-8 images.
Atmosphere 15 01234 g004
Atmosphere 15 01234 g005
Figure 5. Omission and commission errors of IMS 4 km products on the TP based on Landsat-8 images.
Figure 5. Omission and commission errors of IMS 4 km products on the TP based on Landsat-8 images.
Atmosphere 15 01234 g005
Atmosphere 15 01234 g006
Figure 6. (a) Spatial distribution of accuracy errors of IMS 4 km products on 8 January 2018. (b) Accuracy errors of IMS 4 km products along with elevations on 8 January 2018.
Figure 6. (a) Spatial distribution of accuracy errors of IMS 4 km products on 8 January 2018. (b) Accuracy errors of IMS 4 km products along with elevations on 8 January 2018.
Atmosphere 15 01234 g006
Atmosphere 15 01234 g007
Figure 7. (a) Spatial distribution of accuracy errors of IMS 4 km products on 19 January 2017. (b) Accuracy errors of IMS 4 km products along with elevations on 19 January 2017.
Figure 7. (a) Spatial distribution of accuracy errors of IMS 4 km products on 19 January 2017. (b) Accuracy errors of IMS 4 km products along with elevations on 19 January 2017.
Atmosphere 15 01234 g007
Table 1. Error matrix of accuracy evaluation.
Table 1. Error matrix of accuracy evaluation.
IMS
Landsat-8 OLI SnowNon-snowTotalProducer’s Accuracy
SnowABA + BA/(A + B)
Non-snowCDC + D
TotalA + CB + DA + B + C + D
User’s AccuracyA/(A + C)
Table 2. Results of accuracy assessment of IMS 4 km products based on Landsat-8 images.
Table 2. Results of accuracy assessment of IMS 4 km products based on Landsat-8 images.
Image
No.		Imaging
Date		Path RowPSCA
/%		OE
/%		CE
/%		NSDR
/%		UA
/%		PA
/%		OA
/%		K
L120March20141473672.75.956.673.681.694.180.30.4309
L23January20181463524.79.343.194.940.990.765.30.3382
L329March20141463747.115.737.181.866.984.373.00.4656
L416December20131453614.621.613.895.949.378.485.10.5192
L510April20151453945.74.47.796.291.395.693.80.8763
L619January20141433519.45.520.298.452.994.582.70.5721
L728January20141423730.321.842.285.944.678.264.00.2970
L820February20141433974.21.275.088.279.198.879.80.3107
L924January20151414057.27.649.983.171.292.474.30.4470
L1011November20131404017.713.19.797.065.986.989.70.6863
L1113February20191403715.631.130.092.429.768.969.80.2527
L126January20171403568.43.877.673.272.896.272.90.2287
L131January20171373416.010.821.697.543.989.280.10.4769
L141January20171373765.34.173.377.870.895.971.70.2660
L1518December20141383895.12.378.831.896.097.793.90.2239
L168January20171383919.617.120.395.049.982.980.30.5008
L1716March20141354023.427.816.090.858.072.281.30.5179
L183April20181363965.710.471.362.467.689.666.70.2051
L1921January20151363869.45.967.770.675.994.175.20.3112
L204February20171353831.013.534.891.552.786.571.80.4392
L218January20181333827.046.523.181.746.253.570.60.2900
L2212January20171343783.50.582.187.586.099.586.00.2555
L2319January20171353645.71.682.992.850.098.454.30.1440
L245January20171333635.03.232.797.561.496.877.60.5651
L2511February20181313845.56.9 68.3 84.7 53.2 93.1 59.6 0.2338
Average44.411.7 45.4 84.9 62.3 88.3 76.0 0.3942
Table 3. Intercorrelations between evaluation metrics and snow-covered area ratio.
Table 3. Intercorrelations between evaluation metrics and snow-covered area ratio.
OECENSDRUAPAOAKPSCA
OE1.00
CE−0.56 b1.00
NSDR0.18−0.60 b1.00
UA−0.60 b0.42 c−0.55 b1.00
PA−1.00 a0.56 b−0.180.60 b1.00
OA−0.14−0.36−0.090.56 b0.141.00
K0.02−0.80 a0.51 b0.11−0.020.63 a1.00
PSCA−0.59 b0.83 a−0.75 a0.83 a0.59 b0.11−0.45 c1.00
Note: c represents p < 0.05; b represents p < 0.01; a represents p < 0.001.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Chu, D. Accuracy Assessment of NOAA IMS 4 km Products on the Tibetan Plateau with Landsat-8 OLI Images. Atmosphere 2024, 15, 1234. https://doi.org/10.3390/atmos15101234

AMA Style

Chu D. Accuracy Assessment of NOAA IMS 4 km Products on the Tibetan Plateau with Landsat-8 OLI Images. Atmosphere. 2024; 15(10):1234. https://doi.org/10.3390/atmos15101234

Chicago/Turabian Style

Chu, Duo. 2024. "Accuracy Assessment of NOAA IMS 4 km Products on the Tibetan Plateau with Landsat-8 OLI Images" Atmosphere 15, no. 10: 1234. https://doi.org/10.3390/atmos15101234

APA Style

Chu, D. (2024). Accuracy Assessment of NOAA IMS 4 km Products on the Tibetan Plateau with Landsat-8 OLI Images. Atmosphere, 15(10), 1234. https://doi.org/10.3390/atmos15101234

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop