Spatial Representativeness Analysis for Snow Depth Measurements of Meteorological Stations in Northeast China

Spatial Representativeness Analysis for Snow Depth Measurements of Meteorological Stations in Northeast China

<p></p><ul style="display: flex;"><li style="flex:1">APRIL 2020 </li><li style="flex:1">W A N G&nbsp;A N D&nbsp;Z H E N G </li></ul><p></p><p>791 </p><p>Spatial Representativeness Analysis for Snow Depth Measurements of <br>Meteorological Stations in Northeast China </p><p>YUANYUAN WANG AND ZHAOJUN ZHENG </p><p>National Satellite Meteorological Center, China Meteorological Administration, Beijing, China </p><p>(Manuscript received 14 June 2019, in final form 20 February 2020) <br>ABSTRACT <br>Triple collocation (TC) is a popular technique for determining the data quality of three products that estimate the same geophysical variable using mutually independent methods. When TC is applied to a triplet of one point-scale in situ and two coarse-scale datasets that have the similar spatial resolution, the TC-derived performance metric for the point-scale dataset can be used to assess its spatial representativeness. In this study, the spatial representativeness of in situ snow depth measurements from the meteorological stations in northeast China was assessed using an unbiased correlation metric r<sup style="top: -0.2409em;">2</sup><sub style="top: 0.1891em;">t,X </sub>estimated with TC. Stations are </p><p>1</p><p>considered representative if r<sup style="top: -0.2409em;">2</sup><sub style="top: 0.1843em;">t,X </sub>$ 0:5; that is, in situ measurements explain no less than 50% of the variations </p><p>1</p><p>in the ‘‘ground truth’’ of the snow depth averaged at the coarse scale (0.258). The results confirmed that TC can be used to reliably exploit existing sparse snow depth networks. The main findings are as follows. 1) Among all the 98 stations in the study region, 86 stations have valid r<sup style="top: -0.2409em;">2</sup><sub style="top: 0.1843em;">t,X </sub>values, of which 57 stations are </p><p>1</p><p>representative for the entire snow season (October–December, January–April). 2) Seasonal variations in r<sub style="top: 0.1843em;">t</sub><sup style="top: -0.2409em;">2</sup><sub style="top: 0.1843em;">,X </sub></p><p>1</p><p>are large: 63 stations are representative during the snow accumulation period (December–February), whereas only 25 stations are representative during the snow ablation period (October–November, March–April). 3) The r<sup style="top: -0.2409em;">2</sup><sub style="top: 0.1843em;">t,X </sub>is positively correlated with mean snow depth, which largely determines the global decreasing </p><p>1</p><p>trend in r<sup style="top: -0.2409em;">2</sup><sub style="top: 0.1843em;">t,X </sub>from north to south. After removing this trend, residuals in r<sup style="top: -0.2409em;">2</sup><sub style="top: 0.1842em;">t,X </sub>can be explained by heterogeneity features concerning elevation and conditional probability of snow presence near the stations. </p><p></p><ul style="display: flex;"><li style="flex:1">1</li><li style="flex:1">1</li></ul><p></p><p>1. Introduction </p><p>Validation of microwave snow depth products with ground truth data is key to improving inversion algorithms. However, owing to the high spatial variability of snow depth, the validation process can be quite challenging. An in situ snow depth measurement can only be representative over a very small spatial </p><p>scale (Clark et al. 2011; Trujillo et al. 2007), whereas </p><p>satellite-derived snow depth represents the mean value of a microwave footprint with a size of 25 km 3 25 km or larger (Vander Jagt et al. 2013). If satellitederived snow depth is directly compared with point measurements, the obtained errors are likely dominated by representativeness errors due to the variability of the snow depth field on subgrid scales as opposed to snow depth inversion model errors </p><p>(Brasnett 1999; Tustison et al. 2001; Chang et al. 2005; Liston 1999, 2004). </p><p>Snow cover is a key component in the global water cycle and directly impacts the Earth’s energy balance and climate dynamics (Cohen 1994). Remote sensing is the most efficient way to regularly measure snow cover and depth on global and regional scales (Armstrong </p><p>and Brodzik 2002; Foster et al. 2011). The Scanning </p><p>Multichannel Microwave Radiometer (SMMR), Special Sensor Microwave Imager (SSM/I), and Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E) have been routinely used to retrieve snow depth and snow water equivalent (SWE) since the 1970s (Che et al. 2016). Satellite snow products are increasingly used for modeling and monitoring in various fields such as hydrology (Berezowski et al. 2015), climate research (Bormann et al. 2012), glaciology (Stroeve et al. 2005), and numerical weather prediction </p><p>(Brasnett 1999). </p><p>To evaluate the spatial representativeness of the point-scale snow depth, most studies attempted to obtain the difference between the point measurement and the area average, and argued that a point measurement is representative if its value deviates less than </p><p>Corresponding author: Yuanyuan Wang, <a href="mailto:[email protected]" target="_blank">wangyuany@ </a><a href="mailto:[email protected]" target="_blank">cma.gov.cn </a></p><p>DOI: 10.1175/JHM-D-19-0134.1 </p><p>Ó 2020 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the <a href="/goto?url=http://www.ametsoc.org/PUBSReuseLicenses" target="_blank">AMS Copyright </a></p><p><a href="/goto?url=http://www.ametsoc.org/PUBSReuseLicenses" target="_blank">Policy </a>(<a href="/goto?url=http://www.ametsoc.org/PUBSReuseLicenses" target="_blank">www.ametsoc.org/PUBSReuseLicenses</a>). </p><p>Unauthenticated | Downloaded 10/05/21 12:39 PM UTC </p><p>792 </p><p>J O U R N A L&nbsp;O F&nbsp;H Y D R O M E T E O R O L O G Y </p><p>VOLUME 21 </p><p>10% from the area average (Neumann et al. 2006; questions,&nbsp;which have not been fully explored in previ- </p><p>Molotch and Bales 2005, 2006; Rice and Bales 2010; ous&nbsp;TC studies. Meromy et al. 2013; Grünewald and Lehning 2013; </p><p>1) How&nbsp;representativeness varies with season? <br>Grünewald et al. 2013). This method requires a dense <br>Representativeness of a station is not a constant sampling network, based on which upscaling to the </p><p>(Bohnenstengel et al. 2011); it can change con- </p><p>coarse scale can be achieved by using spatial modeling siderably from the snow accumulation period to methods. Although this method has been successthe snow ablation period owing to the variafully applied at the watershed scale, it is of limited tions in the spatial heterogeneity of snow depth use for estimating the spatial representativeness of </p><p>(Molotch and Bales 2005; Winstral and Marks </p><p>sparse meteorological stations that provide only one <br>2014). Some researchers argued that the obserin situ observation for a satellite footprint. Since it vations need to be selected with the specific is logistically prohibitive to carry out extensive snow objective of representing either the accumulation surveys or set up dense networks over hundreds of opor the ablation season process (Molotch and erational meteorological stations, the limitations of <br>Bales 2005). Understanding the seasonal variapoint measurements at these stations in adequately tions in representativeness can help us choose the representing snow depth for the surrounding area have most representative stations according to the been questioned but not explored in detail (Blöschl time of the snow depth product and hence make </p><p>1999; Neumann et al. 2006; Derksen et al. 2003; Chang </p><p>full use of the existing networks. </p><p>et al. 2005; Grünewald and Lehning 2013; Meromy </p><p>2) What factors in the vicinity of stations play a </p><p>et al. 2013). </p><p>dominant role in determining representativeness? <br>A promising way to evaluate the representativeness <br>Understanding the dominant factors has two advanof a point-scale dataset is the triple collocation (TC) tages. First, it provides an indirect approach to validate technique, which estimates the data quality of three representativeness assessments. Strong heterogeneity mutually independent datasets without treating any usually results in low representativeness; thus, the dataset as perfectly observed ‘‘truth’’ (Stoffelen 1998). representativeness assessments are generally reason- <br>TC has now become a standard procedure in compreable if they are strongly correlated with heterohensive satellite validation processes, especially in soil geneity features. Second, dominant factors can be </p><p>moisture research (Scipal et al. 2008; Dorigo et al. 2010, </p><p>used to predict representativeness, which is poten- </p><p>2015; Chen et al. 2017; Gruber et al. 2016a,b, 2017). </p><p>tially useful in choosing the representative locations <br>When TC is applied to a triplet containing one pointfor new sites. scale and two coarse-scale datasets that have the similar </p><p></p><ul style="display: flex;"><li style="flex:1">spatial resolution, performance metrics associated with </li><li style="flex:1">The remainder of this paper is organized as follows. </li></ul><p>the point-scale dataset indicate its spatial representa-&nbsp;Section 2 introduces the TC technique and how TC is tiveness, assuming that the instrumental random error&nbsp;used to evaluate station representativeness. Section 3 can be neglected (Gruber et al. 2013, 2016a; Chen et al.&nbsp;describes the study region, datasets, TC implementation 2017). The most prominent feature of using TC to assess&nbsp;process, and the method of extracting heterogeneity feathe spatial representativeness is that it is data-driven and&nbsp;tures. Results and discussion are presented in sections 4 does not need field surveys or dense sampling net-&nbsp;and 5, respectively. works. The credibility of using the TC-derived correlation metric or random error variances in representing the closeness of the point-scale data to the coarse-scale </p><p>2. Introduction&nbsp;of the TC technique </p><p>a. TC&nbsp;approaches </p><p>ground truth has been confirmed at densely instrumented validation sites by Miralles et al. (2010) </p><p>and Chen et al. (2017). </p><p>The most commonly used error model for TC analysis is the following model (Gruber et al. 2016a): <br>Validations of microwave snow depth and soil mois- </p><p>ture share a high degree of similarity. The success of TC applications in soil moisture studies has prompted </p><p>X<sub style="top: 0.274em;">i </sub>5 a<sub style="top: 0.274em;">i </sub>1 b<sub style="top: 0.274em;">i</sub>t 1 «<sub style="top: 0.274em;">i </sub>, </p><p>(1) us to adopt this technique for snow depth studies.&nbsp;where X<sub style="top: 0.1322em;">i </sub>(i 2 {1, 2, 3}) are three collocated and indeTo the best of our knowledge, this study is the first at-&nbsp;pendent datasets of the same geophysical variable linetempt to apply TC to evaluate the spatial representa-&nbsp;arly related to the true underlying value t with additive tiveness of point-scale snow depth measurements from&nbsp;zero-mean random errors «<sub style="top: 0.1323em;">i</sub>. The terms X<sub style="top: 0.1323em;">i</sub>, t, «<sub style="top: 0.1323em;">i </sub>are meteorological stations. Besides assessing the spatial&nbsp;all random variables; a<sub style="top: 0.1323em;">i </sub>and b<sub style="top: 0.1323em;">i </sub>are the intercepts and representativeness, we investigated the answers to two&nbsp;slopes, respectively, representing systematic additive </p><p>Unauthenticated | Downloaded 10/05/21 12:39 PM UTC </p><p></p><ul style="display: flex;"><li style="flex:1">APRIL 2020 </li><li style="flex:1">W A N G&nbsp;A N D&nbsp;Z H E N G </li></ul><p></p><p>793 </p><p>(6) </p><p>8</p><p>Q<sub style="top: 0.274em;">12</sub>Q<sub style="top: 0.274em;">13 </sub>Q<sub style="top: 0.274em;">23 </sub></p><p>and multiplicative biases of dataset X<sub style="top: 0.1323em;">i </sub>with respect to the true signal t. </p><p>2</p><p>«</p><p>&gt;&gt;&gt;</p><p>s</p><p>5 Q<sub style="top: 0.2787em;">11 </sub></p><p>222</p><p>1</p><p>&gt;&gt;&gt;&gt;</p><p>There are four main underlying assumptions for the </p><p>error model of TC (Zwieback et al. 2012; Gruber </p><p>et al. 2016a,b): (i) linearity between the true signal and the observations; (ii) signal and error stationarity; (iii) error orthogonality: independence between the errors and the true signal, that is, Cov(t, «<sub style="top: 0.1275em;">i</sub>) 5 0; and (iv) zero error cross correlation: independence between the errors of X<sub style="top: 0.1323em;">i </sub>and X<sub style="top: 0.1323em;">j</sub>, that is, Cov(«<sub style="top: 0.1323em;">i</sub>, «<sub style="top: 0.1323em;">j</sub>) 5 0, for i ¼ j. </p><p>&gt;&gt;&lt;</p><p>Q<sub style="top: 0.274em;">12</sub>Q<sub style="top: 0.274em;">23 </sub>Q<sub style="top: 0.274em;">13 </sub>s<sup style="top: -0.326em;">2</sup><sub style="top: 0.2221em;">« </sub>5 Q<sub style="top: 0.2787em;">22 </sub></p><p>.</p><p>2</p><p>&gt;&gt;&gt;&gt;&gt;&gt;&gt;&gt;&gt;:</p><p>Q<sub style="top: 0.274em;">13</sub>Q<sub style="top: 0.274em;">23 </sub>Q<sub style="top: 0.274em;">12 </sub></p><p>2</p><p>s<sub style="top: 0.2221em;">« </sub>5 Q<sub style="top: 0.2787em;">33 </sub></p><p>3</p><p>Since s<sup style="top: -0.2835em;">2</sup><sub style="top: 0.222em;">« </sub>is the absolute random error variance af- </p><p>i</p><p>fected by the dynamic range of the data, Draper et al. (2013) proposed relative error variance (fMSE<sub style="top: 0.1323em;">i</sub>), which is calculated by normalizing the error variances with the corresponding dataset variances: <br>Following McColl et al. (2014), the covariances between the different datasets are calculated as follows: </p><p>Cov(X<sub style="top: 0.2787em;">i</sub>, X<sub style="top: 0.2787em;">j</sub>) 5 E(X<sub style="top: 0.2787em;">i</sub>X<sub style="top: 0.2787em;">j</sub>) 2 E(X<sub style="top: 0.2787em;">i</sub>)E(X<sub style="top: 0.2787em;">j</sub>) </p><p>s<sub style="top: 0.222em;">«</sub><sup style="top: -0.2835em;">2 </sup></p><p>i</p><p>5 b<sub style="top: 0.2787em;">i</sub>b<sub style="top: 0.2787em;">j</sub>s<sup style="top: -0.326em;">2</sup><sub style="top: 0.2173em;">t </sub>1 b<sub style="top: 0.2787em;">i</sub>Cov(t, «<sub style="top: 0.2787em;">j</sub>) 1 b<sub style="top: 0.2787em;">j</sub>Cov(t, «<sub style="top: 0.2787em;">i</sub>) </p><ul style="display: flex;"><li style="flex:1">fMSE<sub style="top: 0.274em;">i </sub>5 </li><li style="flex:1">.</li><li style="flex:1">(7) </li></ul><p></p><p>Q<sub style="top: 0.2787em;">ii </sub></p><p>1 Cov(«<sub style="top: 0.274em;">i</sub>, «<sub style="top: 0.274em;">j</sub>), <br>(2) <br>Combining (7), (4), and (3), fMSE<sub style="top: 0.1323em;">i </sub>can be written as </p><p>follows: where s<sup style="top: -0.2882em;">2</sup><sub style="top: 0.2173em;">t </sub>5 var(t). Using the assumptions of error or- </p><p>thogonality and zero error cross correlation, the equation is reduced to (3): </p><p></p><ul style="display: flex;"><li style="flex:1">s<sub style="top: 0.2173em;">«</sub><sup style="top: -0.2882em;">2 </sup></li><li style="flex:1">s<sub style="top: 0.2173em;">«</sub><sup style="top: -0.2882em;">2 </sup></li></ul><p></p><p></p><ul style="display: flex;"><li style="flex:1">i</li><li style="flex:1">i</li></ul><p></p><p>fMSE<sub style="top: 0.2787em;">i </sub>5 </p><p>5</p><p></p><ul style="display: flex;"><li style="flex:1">,</li><li style="flex:1">(8) </li></ul><p></p><p>u<sup style="top: -0.326em;">2</sup><sub style="top: 0.2173em;">i </sub>1 s<sup style="top: -0.2315em;">2</sup><sub style="top: 0.2173em;">« </sub>b<sup style="top: -0.326em;">2</sup><sub style="top: 0.2173em;">i </sub>s<sub style="top: 0.2173em;">t</sub><sup style="top: -0.2315em;">2 </sup>1 s<sup style="top: -0.2315em;">2</sup><sub style="top: 0.2173em;">« </sub></p><p></p><ul style="display: flex;"><li style="flex:1">i</li><li style="flex:1">i</li></ul><p></p><p>(</p><p>b<sub style="top: 0.274em;">i</sub>b<sub style="top: 0.274em;">j</sub>s<sup style="top: -0.326em;">2</sup><sub style="top: 0.2173em;">t </sub>, </p><p>b<sub style="top: 0.2787em;">i</sub>b<sub style="top: 0.2787em;">j</sub>s<sup style="top: -0.326em;">2</sup><sub style="top: 0.2221em;">t </sub>1 s<sup style="top: -0.326em;">2</sup><sub style="top: 0.2221em;">« </sub>, for i 5 j for i ¼ j </p><ul style="display: flex;"><li style="flex:1">where b<sup style="top: -0.3213em;">2</sup><sub style="top: 0.2173em;">i </sub>s<sub style="top: 0.2173em;">t</sub><sup style="top: -0.2882em;">2 </sup>represents the signal and s<sub style="top: 0.2173em;">«</sub><sup style="top: -0.2882em;">2 </sup>represents the </li><li style="flex:1">Q<sub style="top: 0.274em;">ij </sub>[ Cov(X<sub style="top: 0.274em;">i</sub>, X<sub style="top: 0.274em;">j</sub>) 5 </li><li style="flex:1">,</li><li style="flex:1">(3) </li></ul><p></p><p>i</p><p>noise (Gruber et al. 2016a; McColl et al. 2014); thus, </p><p>fMSE<sub style="top: 0.1323em;">i </sub>is not only a measure of relative error, but also a measure of signal-to-noise ratio (SNR). Furthermore, fMSE<sub style="top: 0.1323em;">i </sub>is related to the linear correlation coefficient of </p><p>i</p><p>where s<sup style="top: -0.2881em;">2</sup><sub style="top: 0.2174em;">« </sub>5 var(«<sub style="top: 0.1182em;">i</sub>), representing the variance of </p><p>i</p><p>random error in dataset X<sub style="top: 0.1276em;">i</sub>. Since there are six equations (Q<sub style="top: 0.1275em;">11</sub>, Q<sub style="top: 0.1276em;">12</sub>, Q<sub style="top: 0.1276em;">13</sub>, Q<sub style="top: 0.1276em;">22</sub>, Q<sub style="top: 0.1276em;">23</sub>, Q<sub style="top: 0.1276em;">33</sub>) but seven unknowns (b<sub style="top: 0.1842em;">1</sub>, b<sub style="top: 0.1842em;">2</sub>, b<sub style="top: 0.1842em;">3</sub>, s<sub style="top: 0.1181em;">« </sub>, s<sub style="top: 0.1181em;">« </sub>, s<sub style="top: 0.1181em;">« </sub>, s<sub style="top: 0.1181em;">t</sub>), the system is <br>X<sub style="top: 0.1323em;">i </sub>with the underlying true signal t (denoted by r<sub style="top: 0.1937em;">t,X </sub>). According to McColl et al. (2014), the relationship be- </p><p>i</p><p></p><ul style="display: flex;"><li style="flex:1">1</li><li style="flex:1">2</li><li style="flex:1">3</li></ul><p></p><p>underdetermined. It can be solved by defining a new variable u<sub style="top: 0.1323em;">i </sub>5 b<sub style="top: 0.1323em;">i</sub>s<sub style="top: 0.1323em;">t</sub>. Then, the equations can be rewritten as in (4): tween r<sub style="top: 0.189em;">t,X </sub>and the ordinary least squares (OLS) slope b<sub style="top: 0.1323em;">i </sub>can be written as in (9): </p><p>i</p><p>(</p><p>b<sub style="top: 0.274em;">i</sub>s<sub style="top: 0.274em;">t </sub></p><p>pffiffiffiffiffiffi </p><p>u<sub style="top: 0.2741em;">i</sub>u<sub style="top: 0.2741em;">j</sub>, </p><p>for i ¼ j </p><p>r<sub style="top: 0.274em;">t,X </sub></p><p>5</p><p></p><ul style="display: flex;"><li style="flex:1">.</li><li style="flex:1">(9) </li></ul><p></p><p>i</p><p>Q<sub style="top: 0.2788em;">ii </sub><br>Q<sub style="top: 0.2788em;">ij </sub>5 </p><p></p><ul style="display: flex;"><li style="flex:1">.</li><li style="flex:1">(4) </li></ul><p>u<sup style="top: -0.326em;">2</sup><sub style="top: 0.2221em;">i </sub>1 s<sup style="top: -0.326em;">2</sup><sub style="top: 0.2221em;">« </sub>, for&nbsp;i 5 j </p><p>i</p><p>Combining (7), (8), and (9), we obtain (10): <br>Now there are six equations and six unknowns, and the system can be solved. Variable u<sup style="top: -0.3212em;">2</sup><sub style="top: 0.2174em;">i </sub>, which provides estimates of the sensitivity of datasets X<sub style="top: 0.1276em;">i </sub>to ground truth changes (Gruber et al. 2016a), can be written as follows: </p><p>Q<sub style="top: 0.274em;">ii </sub>2 s<sup style="top: -0.2882em;">2</sup><sub style="top: 0.2173em;">« </sub></p><p>s<sub style="top: 0.2173em;">«</sub><sup style="top: -0.2882em;">2 </sup></p><p>b<sup style="top: -0.3213em;">2</sup><sub style="top: 0.2173em;">i </sub>s<sup style="top: -0.2882em;">2</sup><sub style="top: 0.2173em;">t </sub>r<sup style="top: -0.326em;">2</sup><sub style="top: 0.2221em;">t,X </sub></p><p></p><ul style="display: flex;"><li style="flex:1">5</li><li style="flex:1">5</li></ul><p></p><p><sup style="top: -0.3071em;">i </sup>5 1 2 <sup style="top: -0.3071em;">i </sup>5 1 2 fMSE<sub style="top: 0.2787em;">i </sub>. (10) </p><p>i</p><p></p><ul style="display: flex;"><li style="flex:1">Q<sub style="top: 0.274em;">ii </sub></li><li style="flex:1">Q<sub style="top: 0.274em;">ii </sub></li><li style="flex:1">Q<sub style="top: 0.274em;">ii </sub></li></ul><p></p><p>Equation (10) indicates that r<sup style="top: -0.2835em;">2</sup><sub style="top: 0.222em;">t,X </sub>and fMSE<sub style="top: 0.1323em;">i </sub>are </p><p>8</p><p>Q<sub style="top: 0.2787em;">12</sub>Q<sub style="top: 0.2787em;">13 </sub></p><p>i</p><p>21<br>21<br>2</p><p>t</p><p>&gt;&gt;&gt;</p><p>u 5 b s&nbsp;5 </p><p>complementary. When fMSE<sub style="top: 0.1323em;">i </sub>is 0.5, the coefficient of determination r<sup style="top: -0.2882em;">2</sup><sub style="top: 0.2173em;">t,X </sub>for the linear error model is 0.5, and </p><p>Q<sub style="top: 0.2787em;">23 </sub></p><p>&gt;&gt;&gt;&gt;</p><p>i</p><p></p><ul style="display: flex;"><li style="flex:1">pffiffiffiffiffiffi </li><li style="flex:1">&gt;</li></ul><p>&gt;&lt;</p><p>the correlation coefficient of X<sub style="top: 0.1323em;">i </sub>with t is 0:5 (’0.71). </p><p>Q<sub style="top: 0.2787em;">12</sub>Q<sub style="top: 0.2787em;">23 </sub>Q<sub style="top: 0.2787em;">13 </sub>u<sup style="top: -0.326em;">2</sup><sub style="top: 0.2173em;">2 </sub>5 b<sup style="top: -0.326em;">2</sup><sub style="top: 0.2173em;">2</sub>s<sup style="top: -0.326em;">2</sup><sub style="top: 0.2173em;">t </sub>5 </p><p></p><ul style="display: flex;"><li style="flex:1">.</li><li style="flex:1">(5) </li></ul><p></p><p>&gt;&gt;</p><p>b. Representativeness&nbsp;analysis of point-scale data with TC </p><p>&gt;&gt;&gt;&gt;&gt;&gt;&gt;:</p><p>Q<sub style="top: 0.2787em;">13</sub>Q<sub style="top: 0.2787em;">23 </sub>Q<sub style="top: 0.2787em;">12 </sub></p><p></p><ul style="display: flex;"><li style="flex:1">2</li><li style="flex:1">2</li><li style="flex:1">2</li></ul><p></p><p>u<sub style="top: 0.2173em;">3 </sub>5 b<sub style="top: 0.2173em;">3</sub>s<sub style="top: 0.2173em;">t </sub>5 </p><p>While TC is a powerful tool for estimating random errors and removing systematic differences between the <br>The estimation equation for error variances can be&nbsp;signal variance component of observations, it is affected </p><ul style="display: flex;"><li style="flex:1">written as follows: </li><li style="flex:1">by representativeness errors (Yilmaz and Crow 2014). </li></ul><p></p><p>Unauthenticated | Downloaded 10/05/21 12:39 PM UTC </p><p>794 </p><p>J O U R N A L&nbsp;O F&nbsp;H Y D R O M E T E O R O L O G Y </p><p>VOLUME 21 </p><p>TC assumes that the three datasets represent the same&nbsp;regions in China (Li et al. 2008) and is characterized by signal, which is very unlikely given that the three datasets&nbsp;taiga snow (Sturm et al. 1995). The region with a total can have very different spatial measurement support&nbsp;area of 1.26 3 10<sup style="top: -0.3496em;">6 </sup>km<sup style="top: -0.3496em;">2 </sup>encompasses the provinces of (McColl et al. 2014; Gruber et al. 2016a). When a triplet&nbsp;Heilongjiang, Jilin, Liaoning, and the eastern part of consists of one point-scale in situ dataset and two coarse-&nbsp;Inner Mongolia. The regional climate includes warm scale datasets that have the similar spatial resolution, the&nbsp;temperate, medium temperate, and subarctic zones. high-resolution signal in the point-scale dataset cannot be&nbsp;Annual precipitation is approximately 430–680&nbsp;mm, of detectable for coarse-scale datasets and therefore be re-&nbsp;which 5%–10% is snowfall (He et al. 2013; Zhang et al. garded as error (Gruber et al. 2016a). In other words, TC&nbsp;2016). There are three mountain ranges (Daxinganling, will penalize the point-scale dataset for its limited rep-&nbsp;Xiaoxinganling, and Changbaishan Mountains) and two resentativeness at the coarse scale, whereas no repre-&nbsp;large plains (Songnen and Sanjiang) in the region. sentativeness error is assigned to the error estimates of&nbsp;Primary land cover types are forest (40%), farmland the coarse-scale datasets (Gruber et al. 2016a; Yilmaz and&nbsp;(30%), and grassland (20%). Figure 1 shows the spatial Crow 2014). This characteristic of TC opens an oppor-&nbsp;pattern of tree cover (%) and elevation (m) in the tunity for evaluating the spatial representativeness of&nbsp;study region. point-scale data efficiently, which has been proved feasi- </p>

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    15 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us