Detecting a Declining Trend of Multidepth Soil Moisture Over the Mongolian Plateau From 1950 to 2020 Using ERA5-Land Reanalysis Datasets

Soil moisture (SM) is a pivotal element in surface hydrological processes, energy transfer, and mass exchange over the Mongolian Plateau (MP). However, spatial and temporal SM variability on the MP has remained unclear over previous decades due to global warming. Therefore, we conducted a spatio-temporal investigation of SM in the MP from 1950 to 2020 using ERA5-Land reanalysis datasets. Our research detected a declining trend of SM (from −0.003 <inline-formula> <tex-math notation="LaTeX">$\text{m}^{3} / \text{m}^{3}$ </tex-math></inline-formula> per decade to −0.005 <inline-formula> <tex-math notation="LaTeX">$\text{m}^{3} / \text{m}^{3}$ </tex-math></inline-formula> per decade) as well as an increasing trend of soil temperature (ST) (from 0.247 °C per decade to 0.267 °C per decade) along with an increasing soil depth on the MP during the past 70 years. Meanwhile, as the depth increased, the fluctuation degrees of SM and ST gradually declined, and the month for the maximum value emergence was delayed. Statistical analyses, including correlation and Granger causality analysis, suggest that precipitation is the dominant driver of SM dynamics in the MP over the warm season (ST<inline-formula> <tex-math notation="LaTeX">$ > 0~^{\circ }\text{C}$ </tex-math></inline-formula>). The proportion that precipitation being the cause of the SM variation was above 80% across different depths. Additionally, evaporation is a leading factor in triggering SM fluctuations. The percentage of evaporation being the cause of the SM variation was maintained above 60% among different soil layers over the warm season (ST<inline-formula> <tex-math notation="LaTeX">$ > 0~^{\circ }\text{C}$ </tex-math></inline-formula>). Meanwhile, multilayer SM, except for the 100–289 cm layer ones, expressed effective feedback to both precipitation (regional proportions varying between 22.64% and 40.28%) and evaporation (regional proportion varying between 36.76% and 64.72%).


Russia, and the north of China's Inner Mongolia Autonomous 85
Region. This plateau falls under the temperate continental cli-86 mate zone with an annual rainfall of approximately 200 mm 87 and large monthly temperature variations. It has long, cold 88 winters (−23 • C on average) and short, cool summers (16 • C 89 on average). The MP is mountainous in the northwest and 90 dominated by vast bare areas in the southeast and large hills 91 in the middle and east. The altitude gradually decreases from 92 west to east. Influenced by the local climate, the vegetation 93 cover successively spans forests, forest grasslands, typical 94 grasslands, desert grasslands, and Gobi Deserts from north 95 to south. As the fifth-generation reanalysis product of the European 98 Center for Medium-Range Weather Forecasts (ECMWF), 99 ERA5-Land has attracted extensive attention since its advent. 100 ERA5-Land is produced using 4D-Var data assimilation and 101 model forecasts by the ECMWF. The global ERA5-Land 102 monthly averaged data is published from 1950 to date, with a 103 regular lat-lon grid of 0.1×0.1 degrees. All data can be easily 104 acquired through the Copernicus Climate Change Service 105 Climate Data Store (https://www.ecmwf.int/). The  Land was developed by combining model data with abundant 107 observations across the globe to formulate a spatial-temporal 108 continuous and consistent climate reanalysis dataset using the 109 laws of physics. The model used in the production of  Land is the tiled ECMWF scheme for surface exchanges over 111 land, incorporating land surface hydrology (H-TESSEL). For 112 more details about the complicated dynamic physical process 113 of ERA5-Land, readers can refer to [19] and [20]. Four layers 114 of SM and ST are currently available at depths of 0-7 cm 115 (Layer 1), 7-28 cm (Layer 2), 28-100 cm (Layer 3), and 100-116 289 cm (Layer 4). In addition to ST, precipitation and evap-117 oration are also adopted to explore the underlying process 118 of the cause and effect of SM variation due to their strong 119 interdependence, as proven by various previous studies [8], 120 [21], [22], [23], [24], [25], [26], [27]. 121 In this study, we used monthly ERA5-Land multi-layer 122 SM products from 1950 to 2020 to investigate the soil water 123 content variation characteristics. As the successor of ERA-124 Interim, ERA5-Land SM products provide various improve-125 ments and have been thoroughly evaluated since inception, 126 achieving favorable results compared to other reanalysis 127 products [28], [29]. Although numerous of remotely sensed 128 SM products are available, the prevalently existed gap regions 129 and the limited penetration depth of microwave significantly 130 restrict the comprehensive investigation. Given the penetra-131 tion limitation of the sensor signal, remotely sensed products 132 can only obtain SM information at 0-5 cm depth. In com-133 parison, the ERA5-Land reanalysis products can retrieve SM 134 information at 0-289 cm depth. Compared to remotely sensed 135 products, ERA5-Land derived SM has notable superiority 136 in terms of spatio-temporal data continuity. Apart from the 137 ERA5-Land SM, the Global Land Data Assimilation System 138 SM is another widely evaluated and used assimilation-based 139  The format of ERA5-Land product was transformed from the 157 original NC to TIFF using the programming language Python 158 3.5 and the Geospatial Data Abstraction Library [32]. More-159 over, all the parameters were then converted to the coordinate 160 system of GCS_WGS_1984 to be convenient for subsequent 161 investigations and exhibitions.

162
The monthly in-situ measurements were achieved by cal-163 culating the arithmetic means of the three recordings within 164 one month. Specifically, the monthly value of 0-0.1 m was 165 used as the theoretical true value to evaluate ERA5-Land 166 SM of layer 1; the average value of monthly 0.1-0.2 m and 167 0.2-0.3 m was applied to verify ERA5-Land SM of layer 168 2; the average value of monthly 0.3-0.4 m, 0.4-0.5 m, 0.5-169 0.6 m, 0.6-0.7 m, 0.7-0.8m, 0.8-0.9 m, and 0.9-1.0 m was 170 employed to validate ERA5-Land SM of layer 3. The Pearson correlation (R), bias, root mean square 173 error (RMSE), and unbiased root mean square deviation 174 (ubRMSD) were selected to jointly validate the accuracy 175 level of ERA5-Land SM. Moreover, Previous studies have 176 demonstrated close interactions between ST, evaporation, 177 precipitation, and SM [26], [28], [33]. Thus, the R was also 178 utilized in this study to analyze the degree of their temporal 179 relevance.
where X i andX represent the ERA5-Land SM values at pixel     where X t and Y t are two time series, n is the sample capacity, 226 a i , b i , c i , and d i are regression coefficients, ε t and η t are white 227 noise. Equations (7) and (8) demonstrate that the Granger 228 causality test not only considers the interrelationship between 229 the two target variables but also considers the time series auto-230 correlation in each variable. An F-test was then used to check 231 whether the Granger causality hypothesis is tenable (usually 232 set F≤0.05 or 0.1). If the estimation of Y t can be significantly 233 improved by taking X t into consideration compared to merely 234 using past Y t values, it means that previous values of X t have 235 a statistically significant effect on the current value of Y t , and 236 vice versa.

237
Additionally, a significance test with the confidence level 238 of 95% was used throughout the paper to assure the reliability 239 and stability of our findings.

242
Boxplots were used in this study to display the range of error 243 parameters. The horizontal solid lines from top to bottom 244 represent the maximum, first quartile, median, third quar-245 tile, and minimum values [43]. The horizontal dotted line 246 stands for the average, and the red dots indicate potential 247 outliers. As can be seen in figure 2 and table 1, the boxplot 248 revealed an increasing accuracy level as the depth gradually 249 decreased. The Layer 1 SM achieved the superior perfor-250 mance with higher R and smaller bias, RMSE, and ubRMSD. 251  To intuitively depict the SM pattern variation, we firstly 281 compared the annual average SM values in 1950 and 2020. 282 As shown in figure 4, the SM of all layers became drier 283 in 2020 compared to 1950. The dry regions, shown in red 284 and orange, expanded remarkably from the southwest to the 285 east. Meanwhile, the SM around Baikal Lake Basin remained 286 stable. In addition, cumulative distribution function (CDF) 287 curves were drawn to clearly and quantitatively illustrate a 288 statistical difference in the SM value distribution. As shown 289 in figures 4 (c), (f), (i), and (l), the 2020 CDF curves were 290 much higher during 0.1-0.25 m 3 /m 3 than those of 1950, 291 suggesting that there may exist varying degrees of soil water 292 drying trends during the past 70 years. Simultaneously, this 293 integral decreasing variation pattern is roughly consistent 294 with the conclusions of [44], [45]. Considering the exten-295 sive variations in SM in the MP, it is important to conduct 296 further analysis to understand specific spatiotemporal evo-297 lution characteristics and explore the potential evolutionary 298 mechanisms. The monthly evolutionary sequences of the various SM layers 302 were investigated using scatter density distribution ( Figure 5). 303 Every 0.1 • ×0.1 • C grid inside the MP was treated as scatter 304 in this study. The arithmetic average was also displayed to 305 reflect the overall situation of all scatters. We observed an 306 apparent and regular annual fluctuation cycle of the maxi-307 mum SM value between the four layers, which could be up 308

329
Apart from the long-term temporal sequence, boxplots 330 were drawn to explicitly show the statistical properties as 331 well as monthly variations of SM, ST, precipitation, and 332 evaporation in figures 9-12. As displayed in figure 9, SM was 333 mainly concentrated in 0.18-0.35 m 3 /m 3 and the maxi-334 mum revealed periodic fluctuation except for the layer 4, 335 which was in accordance with the CDF curves in figure 336 4 and the temporal sequence in figure 5. In terms of ST, 337 there was a pronounced increasing trend with the begin-338 ning of summer, and a decreasing trend with the setting 339 in of winter. Moreover, as the soil depth increased, the 340 degree of fluctuation gradually decreased, and a month 341 with the maximum value was delayed. This observation 342 indicated that the influence of air temperature on geother-343 mal temperature can be significantly decreased and delayed 344 by a thick soil layer. Figures 11 and 12 show a signif-345 icant positive interdependency between precipitation and 346 evaporation. Evaporation could be effectively enhanced by 347 increasing precipitation and temperature. However, there 348 was a gap between precipitation and evaporation, implying 349 that remaining rainfall infiltrated into the soil, leading to a 350 This gap meant that SM and root absorption did not correlate 376 with decreasing precipitation. Overall, it was indicated that 377 SM reduces as depth increases, and this trend might be trig-378 gered by the combined action of increasing temperature and 379 decreasing precipitation. The impact of precipitation on the 380 surface layer SM declination was verified in a previous study 381 using both remotely sensed and reanalysis products [45]. ST, precipitation, and evaporation are important parameters 385 that exhibit bidirectional interactive responses to SM. When 386 temperature increases, water evaporates from the soil to the 387 air, resulting in a decrease in soil water content [27], [46]. 388 VOLUME 10, 2022     with widely acknowledged findings [28], [49]. Similarly, 404 we observed a significant positive correlation between pre-405 cipitation and SM. It was noticed that when SM was around 406 VOLUME 10, 2022    ). Furthermore, the deep SM layer 440 displayed a more severe degree of drought variation than the 441 shallow layers. In contrast, less than 30% of the area revealed 442 a slightly increasing trend, which was mainly distributed 443 along the northeastern as well as northwestern borders of 444 the MP (rendered in purple). Simultaneously, precipitation 445 expressed a corresponding reducing pattern where the SM 446 decreased and an insignificant increasing trend where the 447 SM also turned wet (Figure 21). Accordingly, evaporation 448 demonstrated a coincident evolution pattern to precipitation, 449 increasing at the point where the precipitation increased (Fig-450  ure 21). Nevertheless, it appears that the ST pattern was not 451 closely linked with the overall SM trend, implying that tem-452 perature might not be the dominant element that triggered SM 453 fluctuation. Even so, it is noteworthy that the ST deep-layer 454 presented a more significant warming trend than the surface 455 layers. Only regions around Lake Baikal remained relatively 456 stable, highlighting the efficient capability of the water body 457 in regional heat regulation among various soil depths.

458
The possible causality between the different SM layers 459 and related parameters was explored using Granger analysis. 460 As shown in figure 23(a)-(d), areas without significant casual 461 links were prevalently distributed in the MP, indicating an 462 inconspicuous interaction between SM and ST. Nevertheless, 463 it is worth noting that the percentage of ST was found to be 464 the unidirectional cause of SM increase from 12% to over 465 50%, expanding from west to east, with a gradual increase 466 in the soil depth. In terms of precipitation, significant uni-467 directional causality relationships appeared in over 50% of 468 the MP areas across all the layers (Figure 23(e)-(h)), illus-469 trating a detectable impact of rainfall on the promotion of 470 SM. Furthermore, in terms of the first three layers, SM and 471 precipitation were found to have evident bidirectional causal-472 ity relationships, which accounted for 21.52%, 39.78%, and 473 30.05% of the MP, respectively. While for the fourth layer, 474 the percentage of bidirectional causality sharply dropped to 475 10.56%, implying that it could be difficult for SM at 100-476 289 cm or deeper to have a significant effect on precipitation. 477 For evaporation, significant bidirectional causality relation-478 ships between SM and evaporation gradually declined as soil 479 depth increased (Figure 23(i)-(l)). Furthermore, the evapora-480 tion was stably maintained to have a far-reaching impact on 481 every layer of SM. SM also presented an evident decreasing 482 effect on evaporation with increasing soil depth. Besides, 483 since ERA5-Land considers both the liquid and solid water 484 content, ERA5-Land SM remains nearly constant over the 485 frozen season. Therefore, precipitation and evaporation could 486 hardly impact SM when ST<0 • C, and substantial influences 487 from precipitation and evaporation are mainly concentrated 488 over the warm season.   In addition to the wetting trend in the vertical direction, 519 SM exhibited a stable trend with increasing soil depth. The 520 seasonal fluctuation of SM gradually decreased as the soil 521 depth increased, implying that deep-layer SM could be imper-522 vious to seasonal climate rhythms. During the downward 523 transportation of water through soil pores, soil layers inter-524 cepted a portion of the water, less water is obtained, and 525 a smaller amplitude was exhibited as the depth increases 526 [55]. Meanwhile, the surface SM was highly variable owing 527 to the direct and synthetic influence of atmospheric con-528 ditions. Moreover, the corresponding ST also presented a 529 VOLUME 10, 2022    To further explore the potential correlation between the 561 dynamics of SM and related atmospheric conditions, this 562 study provides a preliminary analysis of the interactions 563 between these parameters using the Granger causality test. 564 The precipitation was identified as the vital Granger cause of 565 SM in over 80% of the MP across different layers, demon-566 strating the predominant influence of precipitation events 567 on SM dynamics. Conversely, SM at 0-100 cm depth was 568    Simultaneously, the SM-evaporation feedback was not signif-580 icant at high latitudes, where the energy available for evapora-581 tion was small [26]. Moreover, the proportion of ST that could 582 be speculated to be the Granger cause of SM increased with 583 increasing depth. According to a previous investigation [59], 584 the increase in SM (in this study, the SM increased as the 585 depth increased) could result in a higher heat storage capacity, 586 which effectively reduces the irregular fluctuation of the ST 587 originating from unpredictable atmospheric conditions. Thus, 588 the water-heat covariations in the soil could be physically 589 consistent.

591
The results derived from our study were mainly based on the 592 ERA5-Land monthly averaged products. Great efforts have 593 been devoted to evaluating the accuracy of ERA5-Land prod-594 ucts, and ERA5-Land has shown favorable performance com-595 pared to its predecessors as well as other reanalysis products 596 [19], [28], [60]. However, the multi-depth SM, ST, precipi-597 tation, and evaporation dataset are derived from the ERA5-598 Land products, which means inborn uniformity among these 599 datasets. In-depth investigation using mutually independent 600 datasets is expected in future to explore the evolution pattern Xin et al. [61] found that further improvement could be 606 essential for ERA5-Land products to provide accurate pre-607 cipitation patterns in areas with high urbanization levels. 608 Wu et al. [30] thoroughly validated the ERA5-Land SM in 609 China and revealed that the performance of the ERA5-Land 610 SM was mostly determined by SM climatology. In particular, 611 the ERA5-Land SM expressed a relatively large bias in humid 612 and semi-humid areas. In addition to ST, precipitation, and 613 evaporation, many other factors (i.e., soil texture, groundwa-614 ter depth, and vegetation) could also affect the distribution 615 patterns of SM. As a classical measurement method, the 616 Granger causality test has been widely accepted and used in 617 Earth system science studies. However, it may not be suffi-618 ciently rigorous to define a temporal-related phenomenon as 619 FIGURE 23. Granger causality between ST ((a)-(d)), precipitation ((e)-(h)), evaporation ((i)-(l)), and corresponding SM. The red areas are identified with bidirectional causality relationships between the SM and the related climate parameters; the yellow indicates that the related climate parameters are the unidirectional cause of the SM, and the green indicates that the SM is the unidirectional cause of the related climate parameters. Areas without significant causal links are shown in blue.
causality. Therefore, this study conducted a causality analy-620 sis to attempt to explain the reason for the increase in SM soil water resource supply [62], [63]. Therefore, it is primar-629 ily assumed that the fluctuation in SM is a comprehensive  Simultaneously, a pre-existing response of the SM to 648 precipitation was observed. Moreover, the evaporation and 649 land surface SM displayed close interactions, and this bidirec-650 tional interactive response turned unidirectional (the impact 651 of deep layer SM on evaporation remarkably decreased) with 652 increasing depth during the warm period (ST>0 • C). The 653 interaction analysis was expected to improve the understand-654 ing of the spatial and temporal SM dynamics in the MP. 655 However, this study merely conducted a preliminary investi-656 gation of the evolutionary characteristics of SM fluctuations 657 under the background of global climate warming. In particu-658 lar, more in-depth analysis devoted to exploring the intrinsic 659 mechanism of SM change is necessary in the future.

661
The authors would like to thank the anonymous reviewers for 662 their insightful comments on this study. 663 tion of soil moisture from the China land data assimilation system,'' IEEE