1. Introduction
Soil moisture is a key variable within earth system dynamics from regional to pedon scales (Famiglietti et al., 1998; Korres et al., 2015). Identification and prediction of soil moisture patterns at different scales are important in a wide range of agronomic, hydrological, pedological, and environmental studies (e.g., Grayson et al., 2002; Lin et al., 2006). However, obtaining accurate information on soil moisture at an appropriate temporal and spatial scales is still challenging (e.g. Crow et al., 2012; Vereecken et al., 2008). This shortage of information has impeded the modeling, predication, and management of water resources (Owe et al., 1982; Grayson et al., 1997; Zhu and Lin, 2010; Shi et al., 2015). Linking soil moisture mapping with monitoring can provides a more integrated approach to understanding soils and water resources (Dunn and Lilly, 2001; Joshi and Mohanty, 2010; Ma et al., 2017), and provides a way of combatting the general decline of field hydrology relative to modelling (Beven et al., 2020).
Previous studies have indicated that the spatial pattern of soil moisture is highly dependent on various controlling factors such as parent material, soil, landuse/vegetation, topography and climate (Grayson et al., 1997; Famiglietti et al., 1998; Western et al., 1999; Baldwin et al., 2017). However, separating the relative importance of individual factors is usually extremely difficult as some of the influencing factors are interdependent due to the scale-dependence, dry-wet cycles, and interface exchange from top to deep soil layers (Zeleke and Si, 2006; Famiglietti et al., 1998; Western et al., 1999). During wet conditions, subsurface lateral flow may be important, whereas during dry conditions, the surface vertical flow at the onset of precipitation has more influence (Grayson et al., 1997; Gómez-Plaza et al., 2001). The correlation length of soil moisture ranges from near zero to a thousand meters. It has been found to be highly dependent on the scale of the study and on the density of the measurements (Western and Blöschl, 1999), with a clear direct proportionality of increasing correlation lengths with increasing scales having been observed. For the on-site soil moisture monitoring system or networks, the correlation measurements typically include the several sites, whereas for the measurements characterizing soil moisture variation within the field, the area can cover several ha areas, or even much large areas for satellite-based investigations. It is evident that the potential exists for more accurate estimations of soil moisture by using synergistic approaches among a variety of earth observation methodologies.
The procedures for associating the relatively stable spatial patterns of soil moisture and specific soil hydrologic processes to precipitation events are sorely lacking. The temporal variations associated with field measurements of soil moisture have been displayed within numerous studies that have shown the static, terrain-derived indices rarely explained more than 50% of the soil moisture variability (Famiglietti et al. 1998; Western et al. 1999; Baldwin et al., 2017). Although climate seasonality and localized vs. non-local hydrological fluxes exert a strong influence on surface or near-surface soil moisture distribution (Grayson et al., 1997; Western et al., 1999; D’Odorico and Porporato, 2010), less is understood regarding how the subsurface soil moisture spatial patterns change with climate seasonality and seasonal changes in hydrological fluxes. By integrating spatially-distributed soil physical properties, topographical variables and meteorological data, Schwärzel et al. (2009) developed a novel approach in model-based mapping of soil moisture within forested sites. This approach provided a more objective description of soil moisture variability than the traditional mapping by an integrated ecological approach. Other research using a modification of this approach showed that both the spatial distribution and temporal evolution of soil moisture may be investigated at multiple scales (e.g., Deiana et al., 2008). Another potential improvement was the integration of geophysical scanning with real-time soil moisture monitoring and dye staining (Guo et al., 2019). Real-time soil moisture monitoring may provide information concerning actual hydrologic dynamics and the timing of preferential flow occurrence, while dye staining patterns may validate the spatial distribution of preferential flow pathways as revealed by geophysical scans. The synthesis of these methods may enhance the data accuracy and associated processes, thus increasing our understanding of hydrological processes in various soils and landscapes (Ma et al., 2019).
Spatial dependency is commonly characterized and quantified by geostatistical methods, such as autocorrelation and variogram analysis (Zhao et al., 2011). An additional way to analyze the spatial patterns of soil moisture and their connection to regional characteristics is through an empirical orthogonal function (EOF) analysis (Jawson and Niemann, 2007; Zhao et al., 2012). The EOF analysis decomposes a dataset into a series of orthogonal spatial patterns. These patterns may be correlated with regional characteristics to identify whether the characteristics have an influence on the most important tendencies of the soil moisture. Utilizing correlation analyses, these underlying (stable) patterns of soil moisture variations can be connected to parameters derived from topography, soil, vegetation, land management and meteorology. Yoo and Kim (2004) investigated the spatial and temporal variability of field-scale soil moisture and concluded that there is no simple and unique mechanism that can be applied to explain the evolution of the soil moisture field. Wagenet (1998) summarized the main factors that influence soil moisture from the pore scale to the global scale, and recognized soil and topography as important local controls of soil moisture variations. During dry periods, soil moisture distribution may be conceptualized as being controlled predominantly by soil properties, whereas during wet periods the topography is the controlling factor within a landscape. However, the pedon/plot scale results have not been easily transferable to hillslope and catchment scales, and both theoretical and empirical approaches have been used to quantify hydrological dynamics based on such ‘point-scale’ data that tends to over- or underestimate parameters and fluxes (Sidle et al., 2017). While the temporal frequency for the manual measurement of soil moisture has been found to normally be about weekly-or monthly-based, the automated monitoring data may be hourly-based or daily-based (Korres et al., 2015).
In this study, part of our soil moisture dataset contains manually measured data that offers better spatial coverage with many sites, but with limited temporal frequency as weekly measurements were made. Another part of our soil moisture dataset contains automatically monitored data that offers better temporal frequency with repetitive ten-min to one-hour measurements, but limited spatial coverage with a limited number of selected sites. It is expected that the product of this combination captures the temporal variation of soil moisture from the automatically monitored data and improved spatial resolution using the information from manual measurements. The objectives of this study were to: (1) derive the dominant soil moisture spatial-temporal patterns based on multi-year datasets; (2) determine how these patterns are controlled by terrain, soil, and vegetation as a function of scale, wetness, and depth; and (3) provide a possible way to integrate spatial-extensive datasets with temporal-extensive datasets.