A comparison of the use of local legacy soil data and global datasets for hydrological modelling a small-scale watersheds: Implications for nitrate loading estimation
Introduction
Hydrological models support decision-making regarding various issues, such as water resource planning, flood prevention, contamination mitigation, etc. (Beven, 2012). To reduce the uncertainty in model outputs, realistic input data are needed (Robinson et al., 2016). Soils play a crucial role in rainfall-runoff processes and constituent loading. Soil properties that relate to the rate of infiltration, or ability to store water, significantly affect the water balance in watersheds (Geroy et al., 2011). Additionally, the impact of soils on hydrological processes and ion sorption also affect nutrient loss (Gaines and Gaines, 1994, Kurunc et al., 2011). Moreover, soil properties vary spatially (Biggar and Nielsen, 1976, Iqbal et al., 2005) and with depth in the soil profile (Franzluebbers, 2002). Some hydropedological properties, e.g. hydraulic conductivity, may even vary with season (Šípek et al., 2019). Uncertainty in soil information can outweigh the uncertainty of climate change impact, as shown by Folberth et al. (2016) in the case of crop yield modelling by global, gridded crop models.
One of the most popular tools for modelling hydrological processes in watersheds is SWAT (Gassman et al., 2007). It is commonly used for the estimation of water balance and nonpoint source water pollution, especially by nutrients, and the estimation of the effectiveness of best management practices (e.g. Strauch et al., 2013) or the possible impact of climate change (e.g. Bhatta et al., 2019). The importance of the input soil data resolution on the output of the SWAT model has been shown by many authors (Bouslihim et al., 2019, Bhandari et al., 2018, Bossa et al., 2012, Mednick, 2010, Moriasi and Starks, 2010, Romanowicz et al., 2005). Although it is possible to obtain comparable results for the watershed outlet, spatial model performance is expected to decrease significantly with less reliable soil data, as shown by Tavares Wahren et al. (2016). In addition, a few authors have investigated the impact of soil input data of different resolution on nitrate loading (Chaplot, 2005, Cotter et al., 2003, Geza and McCray, 2008) and have observed considerable effects of these different soil input data.
There are several ways to address soil property inputs for models. Global datasets, such as the FAO soil map (Batjes, 1997, Nachtergaele et al., 2010) or SoilGrids (Hengl et al., 2017), provide soil data for the whole world and are readily available. These data are suitable for large-scale studies (Abbaspour et al., 2015). In most cases, traditional soil maps with soil classes are used. The mean soil properties are then estimated for particular soil classes from databases (Čerkasova et al., 2018, Mbungu and Kashaigili, 2017) or in combination with field surveys (Cordeiro et al., 2018, Kmoch et al., 2019, Lima et al., 2013). This approach has the uncertainty relating to spatial delimitation of patches representing soil classes. Another way is to spatially predict soil properties by digital soil mapping (DSM) (Ma et al., 2019, Piikki and Söderström, 2019, McBratney et al., 2003). Soil properties are predicted using known information from particular pedons represented by spatial points. Approaches to these predictions include interpolation methods, regression models or machine learning methods. Hydropedological properties can be predicted directly from existing measurements, but, more often, only information about particle size distribution (PSD), including the proportion of clay, silt and sand fractions and organic carbon content (OC), is available. Hydropedological properties must be derived by pedotransfer functions (PTFs). These two approaches can provide comparable results (Tóth et al., 2018). For the SWAT model, the DSM approach was used in a study by Santra et al. (2011), in which the basic soil properties (PSD and organic carbon) were interpolated by regression kriging, the soil hydraulic properties were derived by PTFs and the resulting pixels were aggregated by a fuzzy approach. In comparison to the map of dominant soil classes, the fuzzy approach performed better for stream flow prediction. Ziadat et al. (2015) developed a tool called SLEEP (soil-landscape estimation evaluation program) which is capable of spatially estimating soil properties from point surveys with a digital terrain model and the remote sensing data required to derive covariates for fitting a linear regression model. Testing the performance of the SWAT model with soil data from SLEEP, kriging and FAO maps showed that the model with SLEEP-derived data performed similarly to the FAO map-based model and better than the kriging-based model. Tavares Wahren et al. (2016) tried to solve the problem of soil data scarcity and improve the spatial distribution of soil depths by using the SoLIM tool. Although stream flow prediction was comparable to that from the model with the base soil data, these authors demonstrated reduced uncertainty in model parameters after calibration.
In the case of large-scale watersheds, more detailed soil data may not have a significant effect on hydrological model performance, but in the case of smaller studies, more detailed soil information may have a more pronounced effect (Mukundan et al., 2010). Conducting soil surveys for particular modelling studies can require time and money. The aim of this study is to determine whether SoilGrids, the most detailed global soil data set currently available, is sufficient for smaller watershed modelling studies of nitrate loading by comparison with a dataset derived from local soil legacy data (SLD). The differences in streamflow output, possible parameter range and implications for water balance are assessed. The implications of soil datasets and the resulting difference in the effects of water balance on nitrate loading are investigated.
Section snippets
The study area
The Olešná reservoir is located in the eastern part of the Czech Republic (49° 38′ N, 18° 18′ E), and its contributing watershed area covers approximately 33 km2 (Fig. 1). The reservoir serves for protection against flooding of a nearby residential area, as a water source for a company producing wood pulp, and for the recreational activities of swimming and fishing. The elevation of the watershed ranges from 300 m to 860 m (mean 401 m). The long-term mean annual precipitation is 995 mm, and the
Legacy data-based soil property prediction
Mean average error (MAE) and root mean square error (RMSE) for predicting the percentage proportion of particles <0.01 mm are summarized in Table 2. The prediction by random forest resulted in relatively low error according to comparison with the used training dataset, but performance decreased after comparison with the independent control dataset. The prediction error was lower for the topsoil layer than for the subsoil layer. Performance improved slightly after residual interpolation.
Mean
Prediction of soil properties
Prediction accuracy is in accordance with that of other studies (e.g. Nussbaum et al., 2018), including those describing SoilGrids development (Hengl et al., 2017). Generally, the pattern of soil texture and OC is consistent in both datasets, but SoilGrids shows systematically finer soil texture and higher OC content, which was also observed in France (Vaysse and Lagacherie, 2015). SoilGrids gives a good representation of the coarsest signal in global soil property variation (Hengl et al., 2017
Conclusions
The lack of detailed spatially distributed soil property data still limits the use of hydrological models. DSM methods seem to promise approaches for estimating soil properties and increase the relevance of hydrological model output. Globally, the most detailed product, SoilGrids, provides soil property data at high resolution, which may allow applications in smaller catchment areas. In this study, we have shown a comparison with our own DSM data prepared from local legacy data in the case of
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgement
The research was funded by the University of Ostrava from internal project SGS02/PřF/2019-2020.
The authors kindly thank reviewers for their constructive and inspiring.
The English language was reviewed by James P. Leckie.
References (76)
- et al.
A continental-scale hydrology and water quality model for Europe: calibration and uncertainty of a high-resolution large-scale SWAT model
J. Hydrol.
(2015) - et al.
Beyond model metrics: the perils of calibrating hydrologic models
J. Hydrol.
(2019) A manifesto for the equifinality thesis
J. Hydrol.
(2006)- et al.
Evaluation and application of a SWAT model to assess the climate change impact on the hydrology of the Himalayan River Basin
Catena
(2019) - et al.
Analyzing the effects of different soil databases on modeling of hydrological processes and sediment yield in Benin (West Africa)
Geoderma
(2012) - et al.
Understanding the effects of soil data quality on SWAT model performance and hydrological processes in Tamedroust watershed (Morocco)
J. Afr. Earth Sci.
(2019) - et al.
Estimation and potential improvement of the quality of legacy soil samples for digital soil mapping
Geoderma
(2007) - et al.
Development of a hydrology and water quality model for a large transboundary river watershed to investigate the impacts of climate change – a SWAT application
Ecol. Eng.
(2018) Impact of DEM mesh size and soil map scale on SWAT runoff, sediment, and NO3-N loads predictions
J. Hydrol.
(2005)- et al.
Spatial prediction of major soil properties using Random Forest techniques – a case study in semi-arid tropics of South India
Geoderma Reg.
(2017)
Introducing a new open source GIS user interface for the SWAT model
Environ. Model. Software
Water infiltration and soil structure related to organic matter and its stratification with depth
Soil Tillage Res.
Effects of soil data resolution on SWAT model stream flow and water quality predictions
J. Environ. Manage.
Identification of nitrate leaching hot spots in a large area with contrasting soil texture and management
Agric. Water Manage.
Uncertainty in the spatial prediction of soil texture. Comparison of regression tree and Random Forest models
Geoderma
Mapping continuous depth functions of soil carbon storage and available water capacity
Geoderma
A QGIS plugin to tailor SWAT watershed delineations to lake and reservoir waterbodies
Environ. Model. Software
Estimation of shallow groundwater discharge and nutrient load into a river
Ecol. Eng.
Digital soil mapping of arable land in Sweden – validation of performance at multiple scales
Geoderma
The effect of sampling frequency and strategy on water quality modelling driven by high-frequency monitoring data in a boreal catchment
J. Hydrol.
Soil data for biophysical models in Victoria, Australia: current needs and future challenges
Geoderma Reg.
Sensitivity of the SWAT model to the soil and land use data parametrisation: a case study in the Thyle catchment
Belgium. Ecol. Modell.
The impact of Best Management Practices on simulated streamflow and sediment load in a Central Brazilian catchment
J. Environ. Manage.
Combining digital soil mapping and hydrological modeling in a data scarce watershed in north-central Portugal
Geoderma
A global sensitivity analysis tool for the parameters of multi-variable catchment models
J. Hydrol.
Evaluating Digital Soil Mapping approaches for mapping GlobalSoilMap soil properties from legacy data in Languedoc-Roussillon (France)
Geoderma Reg.
Uncertainty assessment in baseflow nonpoint source pollution prediction: the impacts of hydrographic separation methods, data sources and baseflow period assumptions
J. Hydrol.
A world dataset of derived soil properties by FAO-UNESCO soil unit for global modelling
Soil Use Manage.
Effects of soil data resolution on the simulated stream flow and water quality: application of watershed-based SWAT model. World Environ. Water Resour. Congr. 2018 Watershed Manag. Irrig. Drainage, Water Resour. Plan. Manag. – Sel. Pap. from World Environ
Water Resour. Congr.
Spatial variability of the leaching characteristics of a field soil
Water Resour. Res.
Random forests
Mach. Learn.
System for Automated Geoscientific Analyses (SAGA) v. 2.1.4
Geosci. Model Dev.
Deriving a dataset for agriculturally relevant soils from the Soil Landscapes of Canada (SLC) database for use in Soil and Water Assessment Tool (SWAT) simulations
Earth Syst. Sci. Data
Water quality model output uncertainty as affected by spatial resolution of input data
J. Am. Water Resour. Assoc.
Uncertainty in soil data can outweigh climate impact signals in global crop yield simulations
Nat. Commun.
Cited by (11)
Conversion from farmland to orchards has minor effects on nitrogen biological processes in deep loess deposits
2022, Agriculture, Ecosystems and EnvironmentCitation Excerpt :The large NO3--N reservoir is controlled by multiple nitrogen (N) sources and biological processes, including the nitrification of synthetic N fertilizer, atmospheric N deposition, biological fixation, and the mineralization of organic N (Galloway et al., 2008; Vitousek et al., 2013; Levy-Booth et al., 2014). The N cycling processes in terrestrial ecosystems are well known but most previous studies examined the topsoil < 1 m deep (Zhang et al., 2019b; Krpec et al., 2020). The large amounts of NO3--N, preserved in deep vadose zones, may exist for decades to hundreds of years and become a long-term threat to soil and groundwater (Sebilo et al., 2013; Ding et al., 2014; Yang et al., 2020b; Gurevich et al., 2021).
Evaluation of global and continental scale soil maps for southern Africa using selected soil properties
2022, CatenaCitation Excerpt :While numerous researchers rely on soil data when investigating the environmental impacts of climate change, few provide the reasoning behind selecting a particular soil map. Moreover, while many studies have shown that local maps outperform global or regional maps (Chen et al., 2020; Cramer et al., 2019; Dharumarajan et al., 2021; Krpec et al., 2020; McNicol et al., 2019; Rivas-Tabares et al., 2020; Venter et al., 2021), global and continental soil maps are often used without proper consideration of their reliability. However, an evaluation of map accuracy and comparison across the different available datasets would be useful to aid researchers in selecting the right soil data information.
Using hydropedological characteristics to improve modelling accuracy in Afromontane catchments
2022, Journal of Hydrology: Regional StudiesCitation Excerpt :The same principles can be applied to this study with a number of parameters being sensitive to modelling accuracy including the detailed soil information, the corresponding and detailed digital soil maps which highlight the hydropedological behaviour of the catchments as well as the detailed information obtained from Kuenene et al. (2011) on the soil drainage curves and how these translate into the time taken for lateral water movement to contribute to streamflow. Numerous research studies have highlighted the importance of detailed soil information on improving SWAT model accuracy (Adem et al., 2020; Chen et al., 2016, Krpec et al., 2020, Peschel et al., 2006). Soil information should be coupled with ecological information of the catchments such as the effects of fire, vegetation as well as rates of evapotranspiration on hydrological modelling accuracy.
Effect of sensitivity analysis on parameter optimization: Case study based on streamflow simulations using the SWAT model in China
2021, Journal of HydrologyCitation Excerpt :However, most previous parameter optimization and SA studies did not combine these two processes, thereby resulting in the following problems. (1) In some studies, SA was not performed but instead a small number of parameters were selected for optimization based on the researchers’ experience (Confesor and Whittaker, 2007; Khoi and Thom, 2015; Krpec et al., 2020), which could lead to the omission of important parameters. (2) In other studies, SA was performed, but these studies only ranked the SA results of all selected parameters and then these parameters were directly optimized, rather than those parameters that showed relatively high sensitivity in the SA results (Mulungu and Munishi, 2007; Zhai et al., 2014; Odusanya et al., 2019; Pang et al., 2020).
Soil erosion assessment in the Blue Nile Basin driven by a novel RUSLE-GEE framework
2021, Science of the Total Environment