Using data‐driven methods to explore the predictability of surface soil moisture with FLUXNET site data

Soil moisture (SM) is a key variable of land surface‐atmosphere interactions. Data‐driven methods have been used to predict SM, but the predictability of SM has not been well evaluated. This study investigated what variables and methods can be used to better predict SM for leading times of 7 days or longer with a global coverage of FLUXNET site data for the first time. Three machine‐learning models, that is, Bayesian linear regression, random forest, and gradient boosting regression tree, are used for the prediction. Variables including atmospheric forcing, surface soil temperature, time variables (year, day of year, and hour), the Fourier transformation of time variables, and lagged SM (7‐ to 14‐day lagged) were sequentially added into models. A framework with five experiments is designed for factorial exploration of SM predictability. A stepwise method was used to build the best models for each site. The performance of regression models became better when adding more explaining variables in most cases. The results showed that from 50 to 95% of variation of the best models can be explained. The important explaining variables are lagged surface SM, followed by day of year, year, soil temperature, and atmospheric forcing. The predictability of SM depends highly on SM memory characteristics and the persistence of seasonality. The effect of SM memory characteristics on SM prediction as an initial condition question has been widely discussed in this paper. Our results also provide an insight that mechanisms of seasonality effects on SM should be also paid more attention to.     

Estimation of land surface evapotranspiration over complex terrain based on multi‐spectral remote sensing data

Land surface evapotranspiration (ET) plays an important role in energy and water balances. ET can significantly affect the runoff yield of a basin and the available water resources in mountainous areas. The existing models to estimate ET are typically applicable to plains, and excessive data are required to calculate the surface fluxes accurately. This study established a simple and practical model capable of depicting the surface fluxes, while using relatively less parameters. Considering the complex terrain, solar radiation was corrected by importing a series of topographic factors. The water deficit index, a measure of land surface wetness, was calculated by applying the f_c (vegetation fractional cover)‐T_rad (land surface temperature) framework in the two‐source trapezoid model for evapotranspiration model to mountainous areas after corrections of temperature based on altitude variations. The model was successfully applied to the Kaidu River Basin, a basin with few gauges located in the east Tien Shan Mountains of China. Based on the time scale extensions, ET was analyzed at different time scales from 2000 to 2013. The results demonstrated that the corrected solar radiation and water deficit index were reasonably distributed in space and that this model is applicable to ungauged catchments, such as the Kaidu River Basin.     

Soil water content in southern England derived from a cosmic‐ray soil moisture observing system – COSMOS‐UK

Cosmic‐ray soil moisture sensors have the advantage of a large measurement footprint (approximately 700 m in diameter) and are able to operate continuously to provide area‐averaged near‐surface (top 10–20 cm) volumetric soil moisture content at the field scale. This paper presents the application of this technique at four sites in southern England over almost 3 years. Results show the soil moisture response to contrasting climatic conditions during 2011–2014 and are the first such field‐scale measurements made in the UK. These four sites are prototype stations for a UK COsmic‐ray Soil Moisture Observing System, and particular consideration is given to sensor operating conditions in the UK. Comparison of these soil water content observations with the Joint UK Land Environment Simulator 10‐cm soil moisture layer shows that these data can be used to test and diagnose model performance and indicate the potential for assimilation of these data into hydro‐meteorological models. The application of these large‐area soil water content measurements to evaluate remotely sensed soil moisture products is also demonstrated. Numerous applications and the future development of a national COsmic‐ray Soil Moisture Observing System network are discussed. Copyright © 2016 John Wiley & Sons, Ltd.     

A practical method for estimating dynamic soil stiffness on surface of multi‐layered soil

It is important to estimate the influence of layered soil in soil–structure interaction analyses. Although a great number of investigations have been carried out on this subject, there are very few practical methods that do not require complex calculations. In this paper, a simple and practical method for estimating the horizontal dynamic stiffness of a rigid foundation on the surface of multi‐layered soil is proposed. In this method, waves propagating in the soil are traced using the conception of the cone model, and the impulse response function can be calculated directly and easily in the time domain with a good degree of accuracy. The characteristics of the impedance, that is the transformed value to the frequency domain of the obtained impulse response, are studied using two‐ to four‐layered soil models. The cause of the fluctuation of impedance is expressed clearly from its relation to reflected waves from the lower layer boundary in the model. Copyright © 2005 John Wiley & Sons, Ltd.     

Year‐round estimation of soil moisture content using temporally variable soil hydraulic parameters

Soil moisture plays a key role in the hydrological cycle as it controls the flux of water between soil, vegetation, and atmosphere. This study is focused on a year‐round estimation of soil moisture in a forested mountain area using the bucket model approach. For this purpose, three different soil moisture models are utilised. The procedure is based on splitting the whole year into two complement periods (dormant and vegetation). Model parameters are allowed to vary between the two periods and also from year to year in the calibration procedure. Consequently, two sets of average model parameters corresponding to dormant and vegetation seasons are proposed. The process of splitting is strongly supported by the experimental data, and it enables us to variate saturated hydraulic conductivity and pore‐size characterisation. The use of the two different parameter sets significantly enhances the simulation of two (Teuling and Troch model and soil water balance model‐green–ampt [SWBM‐GA]) out of three models in the 6‐year period from 2009 to 2014. For these two models, the overall Nash‐Sutcliffe coefficient increased from 0.64 to 0.79 and from 0.55 to 0.80. The third model (the Laio approach) proved to be insensitive to parameter changes due to its insufficient drainage prediction. The variability of the warm and cold parameter sets between particular years is more pronounced in the warm periods. The cold periods exhibited approximately similar character during all 6 years.     

A change detection algorithm for terrain surface moisture mapping based on multi‐year passive microwave remote sensing (Examples of SSM/I and TMI Channels)

As an indication of the surface polarized emission, a polarization index (PI) of microwave radiance from the terrain surface (half‐space of canopy‐soil land) is derived from the radiative transfer model. This PI separates the radiance effects of the canopy‐soil moisture and interference from surface roughness and atmosphere, and is suitable to describe the change of terrain surface moisture, especially for extreme drought or flood conditions. As an example, the statistics of the monthly average < PI > from 6 years' data of the Defense Meteorological Satellite Program (DMSP) SSM/I observations at the lowest frequency 19·35 GHz channel as available are applied for the demonstration of the surface moisture status over a large and heterogeneous territory such as China. The deviation of the PI data at the same month from the average < PI > , i.e. Δ_nPI(≡(PI? < PI>)/ < PI>), gives prominence to focusing moisture variation of terrain surface, and its anomaly shows possible drought or flood occurrence in extreme conditions. The Δ_nPI mapping is validated by the typical examples of the drought in China's Shanxi area in May 2001 and the flood around China's Yangtze River in August 1998, respectively. Our approach is recommended for lower frequency channels to minimize the influence from vegetation canopy for future applications (such as the channels of the Advanced Microwave Scanning Radiometer [AMSR‐E] launched in May 2002 and microwave imaging radiometer of China's Fengyun satellite series). When the monthly < PI > and the ground truth of average volumetric moisture < m_v > of the region are correctly evaluated, it is tractable to retrieve the soil land surface moisture by using the PI data at the same month and the same region without much knowledge of surface roughness, vegetation canopy and other factors. As an example, the retrieval of m_v is favourably tested by using the Tropical Rainfall Measuring Mission (TRMM) Tropical Microwave Imager (TMI) data. Copyright © 2007 John Wiley & Sons, Ltd.     

First approximations of soil moisture retention curves using the filter‐paper method

Relationships between gravimetric soil moisture content (w) and matric potential (ϕ), and between volumetric soil moisture content (θ_v) and pressure head (h) were approximated for the unsaturated zone on Long Island, New York. Soil samples were collected from two sites using a hand auger. The soil moisture content was determined using the filter‐paper (w_f) and gravimetric (w) methods, respectively. The w_f was then used in an empirical equation to estimate ϕ_m. Each set of ϕ_m and w was combined with a straight‐line empirical model to obtain a w(ϕ_m) relationship. Soil ϕ_m was converted to h, and w to the volumetric moisture content θ_v, in order to produce a θ_v(h) curve. Graphical and statistical comparison showed that the resulting θ_v(h) curves fell within one order of magnitude of similar curves generated by a more sophisticated non‐linear model developed previously. The simplicity and low cost of the filter‐paper approach described in this study recommends it for preliminary studies of hydraulic properties in the unsaturated zone. Copyright © 2000 John Wiley & Sons, Ltd.     

Estimation of soil water content in watershed using artificial neural networks

Soil water content (SWC) is an important factor in transfer processes between soil and air, contributing to water and energy balances, and quantifying it remains a challenge. This study uses artificial neural networks (ANNs) to analyse spatial and temporal variation of SWC in a Brazilian watershed, based on climate information, soil physical properties and topographic variables. Thirty eight input variables were tested in 200 models. The outputs were compared with 650 gravimetric moisture measurements collected at 26 points (25 field studies). The results show that it is possible to estimate SWC efficiently (Nash-Sutcliffe statistic, NS = 0.77) using topographic data, soil physical properties and rainfall. If only climate information is considered, modelling is less efficient (NS = 0.28). Using many variables does not necessarily improve performance. Alternatively, SWC can be estimated by simplified models using rainfall and topographic maps information, although the performance is less good (NS = 0.65).     

Improving the representation of soil moisture by using a semi‐analytical infiltration model

Soil moisture is widely recognized as a fundamental variable governing the mass and energy fluxes between the land surface and the atmosphere. In this study, the soil moisture modelling at sub‐daily timescale is addressed by using an accurate representation of the infiltration component. For that, the semi‐analytical infiltration model proposed by Corradini et al. (1997) has been incorporated into a soil water balance model to simulate the evolution in time of surface and profile soil moisture. The performances of this new soil moisture model [soil water balance module‐semi‐analytical (SWBM‐SA)] are compared with those of a precedent version [SWBM‐Green–Ampt (GA)] where the GA approach was employed. Their capability to reproduce in situ soil moisture observations at three sites in Italy, Spain and France is analysed. Hourly observations of quality‐checked rainfall, temperature and soil moisture data for a 2‐year period are used for testing the modelling approaches. Specifically, different configurations for the calibration and validation of the models are adopted by varying a single parameter, that is, the saturated hydraulic conductivity. Results indicate that both SWBMs are able to reproduce satisfactorily the hourly soil moisture temporal pattern for the three sites with root mean square errors lower than 0.024 m³/m³ both in the calibration and validation periods. For all sites, the SWBM‐SA model outperforms the SWBM‐GA with an average reduction of the root mean square error of ~20%. Specifically, the higher improvement is observed for the French site for which in situ observations are measured at 30 cm depth, and this is attributed to the capability of the SA infiltration model to simulate the time evolution of the whole soil moisture profile. The reasonable models performance coupled with the need to calibrate only a single parameter makes them useful tools for soil moisture simulation in different regions worldwide, also in scarcely gauged areas. Copyright © 2013 John Wiley & Sons, Ltd.     

Modelling soil moisture in a high‐latitude landscape using LiDAR and soil data

Soil moisture has a pronounced effect on earth surface processes. Global soil moisture is strongly driven by climate, whereas at finer scales, the role of non‐climatic drivers becomes more important. We provide insights into the significance of soil and land surface properties in landscape‐scale soil moisture variation by utilizing high‐resolution light detection and ranging (LiDAR) data and extensive field investigations. The data consist of 1200 study plots located in a high‐latitude landscape of mountain tundra in north‐western Finland. We measured the plots three times during growing season 2016 with a hand‐held time‐domain reflectometry sensor. To model soil moisture and its temporal variation, we used four statistical modelling methods: generalized linear models, generalized additive models, boosted regression trees, and random forests. The model fit of the soil moisture models were R² = 0.60 and root mean square error (RMSE) 8.04 VWC% on average, while the temporal variation models showed a lower fit of R² = 0.25 and RMSE 13.11 CV%. The predictive performances for the former were R² = 0.47 and RMSE 9.34 VWC%, and for the latter R² = 0.01 and RMSE 15.29 CV%. Results were similar across the modelling methods, demonstrating a consistent pattern. Soil moisture and its temporal variation showed strong heterogeneity over short distances; therefore, soil moisture modelling benefits from high‐resolution predictors, such as LiDAR based variables. In the soil moisture models, the strongest predictor was SAGA (System for Automated Geoscientific Analyses) wetness index (SWI), based on a 1 m² digital terrain model derived from LiDAR data, which outperformed soil predictors. Thus, our study supports the use of LiDAR based SWI in explaining fine‐scale soil moisture variation. In the temporal variation models, the strongest predictor was the field‐quantified organic layer depth variable. Our results show that spatial soil moisture predictions can be based on soil and land surface properties, yet the temporal models require further investigation. Copyright © 2017 John Wiley & Sons, Ltd.     

Spatio‐temporal distribution of near‐surface and root zone soil moisture at the catchment scale

Soil moisture is highly variable both spatially and temporally. It is widely recognized that improving the knowledge and understanding of soil moisture and the processes underpinning its spatial and temporal distribution is critical. This paper addresses the relationship between near‐surface and root zone soil moisture, the way in which they vary spatially and temporally, and the effect of sampling design for determining catchment scale soil moisture dynamics. In this study, catchment scale near‐surface (0–50 mm) and root zone (0–300 mm) soil moisture were monitored over a four‐week period. Measurements of near‐surface soil moisture were recorded at various resolutions, and near‐surface and root zone soil moisture data were also monitored continuously within a network of recording sensors. Catchment average near‐surface soil moisture derived from detailed spatial measurements and continuous observations at fixed points were found to be significantly correlated (r² = 0·96; P = 0·0063; n = 4). Root zone soil moisture was also found to be highly correlated with catchment average near‐surface, continuously monitored (r² = 0·81; P < 0·0001; n = 26) and with detailed spatial measurements of near‐surface soil moisture (r² = 0·84). The weaker relationship observed between near‐surface and root zone soil moisture is considered to be caused by the different responses to rainfall and the different factors controlling soil moisture for the soil depths of 0–50 mm and 0–300 mm. Aspect is considered to be the main factor influencing the spatial and temporal distribution of near‐surface soil moisture, while topography and soil type are considered important for root zone soil moisture. The ability of a limited number of monitoring stations to provide accurate estimates of catchment scale average soil moisture for both near‐surface and root zone is thus demonstrated, as opposed to high resolution spatial measurements. Similarly, the use of near‐surface soil moisture measurements to obtain a reliable estimate of deeper soil moisture levels at the small catchment scale was demonstrated. Copyright © 2007 John Wiley & Sons, Ltd.     

Estimation of actual evaporation using precipitation and soil moisture records in arid climates

There is little knowledge available about infiltration and evaporation processes in wadi channels in arid regions. This work was conducted to determine the actual evaporation from bare soils in wadi channels in the south-western region of the Kingdom of Saudi Arabia. The estimation of soil evaporation is highly dependent on the availability of moisture in the upper layers of alluvial wadis, in which the areal rainfall, flood hydrograph and soil properties play a significant part. The study was conducted by estimating the actual evaporation using soil moisture data, precipitation and runoff depths in a representative basin. The results are compared with potential rates. The actual rates were 1.5 mm/day immediately after a rainy day and then decreased to 0.42 mm/day. The minimum rate was about 0.1–0.2 mm/day during the dry season. The potential rates were about 9.5 mm/day in June and July, decreasing to 3.5 mm/day in December and January.     

Use of multi‐platform,multi‐temporal remote‐sensing data for calibration of a distributed hydrological model: an application in the Arno basin,Italy

Images from satellite platforms are a valid aid in order to obtain distributed information about hydrological surface states and parameters needed in calibration and validation of the water balance and flood forecasting. Remotely sensed data are easily available on large areas and with a frequency compatible with land cover changes. In this paper, remotely sensed images from different types of sensor have been utilized as a support to the calibration of the distributed hydrological model MOBIDIC, currently used in the experimental system of flood forecasting of the Arno River Basin Authority. Six radar images from ERS‐2 synthetic aperture radar (SAR) sensors (three for summer 2002 and three for spring–summer 2003) have been utilized and a relationship between soil saturation indexes and backscatter coefficient from SAR images has been investigated. Analysis has been performed only on pixels with meagre or no vegetation cover, in order to legitimize the assumption that water content of the soil is the main variable that influences the backscatter coefficient. Such pixels have been obtained by considering vegetation indexes (NDVI) and land cover maps produced by optical sensors (Landsat‐ETM). In order to calibrate the soil moisture model based on information provided by SAR images, an optimization algorithm has been utilized to minimize the regression error between saturation indexes from model and SAR data and error between measured and modelled discharge flows. Utilizing this procedure, model parameters that rule soil moisture fluxes have been calibrated, obtaining not only a good match with remotely sensed data, but also an enhancement of model performance in flow prediction with respect to a previous calibration with river discharge data only. Copyright © 2006 John Wiley & Sons, Ltd.     

Active microwave remote sensing for soil moisture measurement: a field evaluation using ERS‐2

Active microwave remote sensing observations of backscattering, such as C‐band vertically polarized synthetic aperture radar (SAR) observations from the second European remote sensing (ERS‐2) satellite, have the potential to measure moisture content in a near‐surface layer of soil. However, SAR backscattering observations are highly dependent on topography, soil texture, surface roughness and soil moisture, meaning that soil moisture inversion from single frequency and polarization SAR observations is difficult. In this paper, the potential for measuring near‐surface soil moisture with the ERS‐2 satellite is explored by comparing model estimates of backscattering with ERS‐2 SAR observations. This comparison was made for two ERS‐2 overpasses coincident with near‐surface soil moisture measurements in a 6 ha catchment using 15‐cm time domain reflectometry probes on a 20 m grid. In addition, 1‐cm soil moisture data were obtained from a calibrated soil moisture model. Using state‐of‐the‐art theoretical, semi‐empirical and empirical backscattering models, it was found that using measured soil moisture and roughness data there were root mean square (RMS) errors from 3·5 to 8·5 dB and r² values from 0·00 to 0·25, depending on the backscattering model and degree of filtering. Using model soil moisture in place of measured soil moisture reduced RMS errors slightly (0·5 to 2 dB) but did not improve r² values. Likewise, using the first day of ERS‐2 backscattering and soil moisture data to solve for RMS surface roughness reduced RMS errors in backscattering for the second day to between 0·9 and 2·8 dB, but did not improve r² values. Moreover, RMS differences were as large as 3·7 dB and r² values as low as 0·53 between the various backscattering models, even when using the same data as input. These results suggest that more research is required to improve the agreement between backscattering models, and that ERS‐2 SAR data may be useful for estimating fields‐scale average soil moisture but not variations at the hillslope scale. Copyright © 2004 John Wiley & Sons, Ltd.     

Controls of surface soil moisture spatial patterns and their temporal stability in a semi‐arid steppe

Temporal stability of soil moisture spatial patterns has important implications for optimal soil and water management and effective field monitoring. The aim of this study was to investigate the temporal stability of soil moisture spatial patterns over four plots of 105 m × 135 m in grid size with different grazing intensities in a semi‐arid steppe in China. We also examined whether a time‐stable location can be identified from causative factors (i.e. soil, vegetation, and topography). At each plot, surface soil moisture (0–6 cm) was measured about biweekly from 2004 to 2006 using 100 points in each grid. Possible controls of soil moisture, including soil texture, organic carbon, bulk density, vegetation coverage, and topographic indices, were determined at the same grid points. The results showed that the spatial patterns of soil moisture were considerably stable over the 3‐y monitoring period. Soil moisture under wet conditions (averaged volumetric moisture contents > 20%) was more stable than that under dry ( ) or moist ( ) conditions. The best representative point for the whole field identified in each plot was accurate in representing the field mean moisture over time (R² ≥ 0·97; p < 0·0001). The degree of temporal persistence varied with grazing intensity, which was partly related to grazing‐induced differences in soil and vegetation properties. The correlation analysis showed that soil properties, and to a lesser extent vegetation and topographic properties, were important in controlling the temporal stability of soil moisture spatial patterns in this relatively flat grassland. Response surface regression analysis was used to quantitatively identify representative monitoring locations a priori from available soil‐plant parameters. This allows appropriate selection of monitoring locations and enhances efficiency in managing soil and water resources in semi‐arid environments. Copyright © 2010 John Wiley & Sons, Ltd.     

Spatial patterns and temporal stability of soil moisture across a range of scales in a semi‐arid environment

The antecedent soil moisture status of a catchment is an important factor in hydrological modelling. Traditional Hortonian infiltration models assume that the initial moisture content is constant across the whole catchment, despite the fact that even in small catchments antecedent soil moisture exhibits tremendous spatial heterogeneity. Spatial patterns of soil water distribution across three transects (two in a burnt area and one in an unburnt area) in a semi‐arid area were studied. At the transect scale, when the factors affecting soil moisture were limited to topographical position or local topography, spatial patterns showed time stability, but when other factors, such as vegetation, were taken into account, the spatial patterns became time unstable. At the point scale, and in the same areas, topographical position was the main factor controlling time stability. Scale dependence of time stability was studied and local topography and vegetation presence were observed to play an important role for the correlation between consecutive measures depending on the scale. Copyright © 2000 John Wiley & Sons, Ltd.     

Spatial prediction of temporal soil moisture dynamics using HYDRUS‐1D

This article investigates the soil moisture dynamics within two catchments (Stanley and Krui) in the Goulburn River in NSW during a 3‐year period (2005–2007) using the HYDRUS‐1D soil water model. Sensitivity analyses indicated that soil type, and leaf area index were the key parameters affecting model performance. The model was satisfactorily calibrated on the Stanley microcatchment sites with a single point rainfall record from this microcatchment for both surface 30 cm and full‐profile soil moisture measurements. Good correlations were obtained between observed and simulated soil water storage when calibrations for one site were applied to the other sites. We extended the predictions of soil moisture to a larger spatial scale using the calibrated soil and vegetation parameters to the sites in the Krui catchment where soil moisture measurement sites were up to 30 km distant from Stanley. Similarly good results show that it is possible to use a calibrated soil moisture model with measurements at a single site to extrapolate the soil moisture to other sites for a catchment with an area of up to 1000 km² given similar soils and vegetation and local rainfall data. Site predictions were effectively improved by our simple data assimilation method using only a few sample data collected from the site. This article demonstrates the potential usefulness of continuous time, point‐scale soil moisture data (typical of that measured by permanently installed TDR probes) and simulations for predicting the soil wetness status over a catchment of significant size (up to 1000 km²). Copyright © 2012 John Wiley & Sons, Ltd.