We collected infiltration measurements from different countries or regions by contacting the data owners or by extracting infiltration data from published literature (Fig. 1). To do this, a data request was sent to potential data owners through different forums and email exchanges. The flyer asked data owners to cooperate in the development of the Soil Water Infiltration Global (SWIG) database by providing infiltration data as well as metadata about experimental conditions (e.g., initial soil moisture content at the start of the experiment and method used), soil properties, land use, topography, geographical coordinates of the sites, and any other relevant information to interpret the data and to increase the value of the database. Infiltration data reported in the literature were digitized and included in the database together with additional information provided in these papers. The digitization approach is discussed in Sect. 2.2. In total, 5023 single infiltration curves were collected, of which 510 infiltration curves were digitized from 74 published papers (Table 1) and 4513 were provided by 68 different research teams (Table 2), being published or unpublished data. The references and correspondences for data supplied by direct communications with researchers are also reported in Table 2. Therefore, users may refer to these references for detailed information about the applied methods or procedures.

In order to digitize infiltration curves reported in the literature, screenshots of the relevant plots were taken, and figures were imported into the plot digitizer 2.6.8 (Huwaldt and Steinhorst, 2015). First, the origin of the axes and the highest x and y values were defined and the diagram plane was spanned. Then, all point values were picked out and an output table with the x–y pairs (time vs. infiltration rate or cumulative infiltration) was generated and stored.

The SWIG database is prepared in *.xlsx with a backup file in *.csv formats containing several datasets. Supplementary data are available at https://doi.org/10.1594/PANGAEA.885492 (Rahmati et al., 2018) . The first dataset, named “I_cm”, contains cumulative infiltration data in centimeter units and is referred to as “Ixxxx”, whereby “xxxx” is the identifier of the individual infiltration test. The corresponding time intervals in hours for the infiltration data are labeled “T_Hour” and named “Txxxx”. The constant or varying pressure or tension heads (if any) during infiltration measurements are also reported in another dataset named “Tension_cm”. The database also contains additional variables and information relevant to the infiltration data provided by data owners or digitized from articles, as listed in Table 3, and which is labeled “Metadata”. Additional soil properties were determined by different standards; therefore, data harmonization might be needed for some of those, especially in the case of water content at field capacity, pH, or wet-aggregate stability. Further information on measurement methods is available from references of the data. Since the geometric mean diameter (dg) and standard deviation (Sg) of soil particle sizes are rarely measured, both parameters were computed using the following equations (Shirazi and Boersma, 1984): dg=exp⁡(a),a=0.01∑i=1nfiln⁡Di,Sg=exp⁡(b),b2=0.01∑i=1nfiln⁡2Di-a2, where fi is the percent of total soil mass having diameters equal to or less than the arithmetic mean of interval limits (Di) that define three main fractions (i) of clay, silt, and sand with mean values of 0.001, 0.026, and 1.025 mm, respectively. For the infiltration data, where the soil texture is unknown, dg and Sg could not be calculated and the data field in the database was left empty. The database also contains the locations of the experimental sites in another dataset named “Locations” that provides the approximate latitude and longitude in decimal degree (dd.dd) format. Table 2 is also provided in the SWIG database in two other worksheets named “Ref. for digitized data” and “Ref. for data provided by owner”.

The SWIG database (Rahmati et al., 2018) consists of 5023 soil water infiltration measurements spread over nearly all continents (Fig. 2). Data were derived from 54 countries (Table 4). The largest number of data sources were provided by scientists in Iran (n=38), China (n=23), and the USA (n=15), whereby one data source might contain several water infiltration measurements. The SWIG database covers measurements from 1976 to 2017. A sparse coverage was obtained for the higher latitudes of the Northern Hemisphere (above 60∘) including Norway, Finland, Sweden, Iceland, Greenland, and Russia. The lack of reports with infiltration data from most countries of the former Soviet Union as well as the Sahelian and Saharan countries is also notable, as well as the small number of infiltration data from Australia. Figure 3 shows the number of samples by climatic zone (Rubel et al., 2017; Kottek et al., 2006). The majority of the data are from warm temperate, fully humid climate (49 %); arid steppe climate and warm temperate climate with dry summer are the second and third most represented climate zones with 22 and 12 %, respectively. Figures 4 and 5 show the frequency of experimental sites, respectively, by the World Reference Base (WRB) (IUSS, 2006) and USDA soil taxonomy systems (USDA, 2014) based on the SoilGrids dataset (Hengl et al., 2017). Regarding the WRB classification system (Fig. 4), in total, 35 WRB reference soil subgroups are included among experimental sites, where 55 % of the experimental sites comprised four subgroup classes of Haplic Acrisols (8 %), Haplic Luvisols (11 %), Haplic Calcisols (15 %), and Haplic Cambisols (21 %). A total of 29 soil suborders classes of USDA soil taxonomy are included in this study (Fig. 5) with Udalfs (9 %), Orthents (9 %), and Ustolls (9 %). Thus, the wide spatial and temporal distribution of infiltration data from this database provides a comprehensive view of the infiltration characteristics of many soils in the world which can be used in future studies.

Textural information (clay, silt, and sand content) is available for 3842 out of 5023 collected infiltration curves (∼ 76 %). The infiltration measurements cover nearly all soil textural classes according to the USDA classification, except for the sandy clay and silt textural class (Fig. 6), which makes the SWIG database a valuable data source for comprehensive studies. To complete the large dataset, the open-access SWIG database might be amended with information regarding those soils poorly or altogether unrepresented by the existing database, including those not usually considered by infiltration studies, such as soils with extremely high stone content (Poesen, 2018). Loam, sandy loam, silty loam, and clay loam contributed with 19, 18, 14, and 13 % (Table 5) to the infiltration measurements, respectively. Table 5 shows that infiltration measurements are almost equally distributed among textures when these are categorized in three major classes: course- (1092), medium- (1238), and fine- to moderately fine-textured soils (1447). Table 6 reports on the soil properties that are available in the SWIG database and it gives some simple statistics such as mean, minimum, maximum, median, and coefficient of variation. Bulk density (available for 66 % of infiltration measurements) and organic carbon content (available for 62 % of infiltration measurements) are two other soil properties besides texture that have the highest frequency of availability. Saturated hydraulic conductivity, initial soil water content, saturated soil water content, calcium carbonate equivalent, electrical conductivity, and pH are available in 22 to 38 % of infiltration data. The other soil properties have a frequency lower than 10 %.

Different instruments were used to measure soil water infiltration (Table 8). About 32 % (1595 out of 5023) of the measurements were carried out using different types of ring infiltrometers. The most frequently used methods are the disc infiltrometer methods (disc, mini-disc, and micro-disc, hood, and tension infiltrometers), which have been used in about 51 % of the experiments. About 5 % of the data were submitted to the database without specifying the measurement method (251 infiltration tests) and around 12 % of the measurements were carried out with other methods not listed above (Table 7).

Land use is known to potentially impact soil structure and then water infiltration into soils (e.g., Ilstedt et al., 2007; Waterloo et al., 2007). Consequently, we collected information on the type of land use at all experimental sites where available. In general, the type of land use was reported in 3818 out of 5023 infiltration curves (∼ 76 %) and this information is reported in the Metadata dataset. For simplicity, we grouped all reported land use types into 22 major groups (Table 8). A frequency analysis showed that agricultural land use, i.e., cropped land, irrigated land, dryland, and fallow land, is the most frequently reported land use in the database with about 53 % (2019 out of 3818) of all land uses. With 22 %, grasslands are the second most frequently represented land use type. Pasture with 6 % and forest with 5 % are ranked as the third- and fourth-largest reported land use types. The 18 remaining land use types all together cover only 545 experimental sites (less than 15 %).

In order to predict infiltration parameters from infiltration measurements, we classified the SWIG database infiltration curves in two groups: (i) infiltration curves that were obtained under the assumption of 1-D infiltration and (ii) infiltration curves that were obtained under 3-D flow conditions. We fitted the three-parameter infiltration equation of Philip (Kutílek and Krejča, 1987), Eq. (6), to the 1-D experimental data and the simplified form of Haverkamp et al. (1994), Eq. (7), to the 3-D experimental data: I1-D=St12+A1t+A2t32,I3-D=St+2-β3Ksat+γS2RDθs-θit. We reduced the number of parameters in Eq. (6) by defining A1=0.33×Ksat (Philip, 1957) and A2=A where A was assumed to be a constant. In Eq. (7), we put β=0.6 (Angulo-Jaramillo et al., 2000) and the second term between brackets on the right-hand side was assumed to be a constant. Therefore, we simplified the equations as follows: I1-D=St12+0.33Ksatt+At32,I3-D=St+0.47Ksatt+At. In our analysis, we assumed that double-ring infiltrometer measurements result in 1-D infiltration conditions, while the different types of disc infiltration and single-ring infiltrometer measurements lead to 3-D flow conditions that can be captured by Eq. (9). As 1-D or 3-D infiltration conditions are not guaranteed for measurements made with rainfall simulators, Guelph permeameters, Aardvark permeameters, linear and point source methods, and hood infiltrometer measurements, these infiltration curves were not considered in our first analysis. By excluding these methods, 596 infiltration curves were excluded from the fitting to Eqs. (8) and (9). In addition, 251 infiltration curves were also excluded from the fitting to Eqs. (8) and (9) as no indication was available on the measurement method used. In total, 4178 infiltration curves were included in our analysis, of which 828 infiltration curves reflected 1-D and 3350 were considered as the results of 3-D infiltration. As no sufficient information was available on the properties of the sand contact layer, we did not correct 3-D infiltration measurements. Finally, the selected infiltration curves were fitted to Eq. (8) or (9) using the lsqnonlin command in Matlab^™.

The fitting results of Eq. (8) to the single infiltrometer data are shown in Table 9. R2 values were higher than 0.9 in 97 % of the cases and higher than 0.99 in 77 % of the cases. Fitting Eq. (9) to the 3-D infiltration curves data, R2 values higher than 0.9 and 0.99 were obtained in 94 and 68 % of the cases, respectively. The statistics for the fitting process as well as the fitted parameters of two mentioned models are reported in the SWIG database in an additional dataset labeled “Statistics”. For infiltration curves excluded from the analysis, an empty cell is reported.

The average values of estimated Ksat and sorptivity (S), using Eq. (8) or (9) as well as measured Ksat for different soil texture classes extracted from the current database, are reported in Table 10. The measured values of Ksat were obtained by other means by the contributors and tabulated in the SWIG database. More detailed information of how Ksat was calculated in individual cases can be found in the references linked to those data points. Comparison between estimated (Ksat-es) and measured (Ksat-m) values of Ksat (Table 10) reveals that there is reasonably good agreement between measurements and estimation, except for loamy sand (with mean Ksat-es=62 cm h-1 vs. Ksat-m=25 cm h-1), sandy loam (with mean Ksat-es=32 cm h-1 vs. Ksat-m=41 cm h-1), silt loam (with mean Ksat-es=27 cm h-1 vs. Ksat-m=3 cm h-1), and silty clay (with mean Ksat-es=26 cm h-1 vs. Ksat-m=45 cm h-1) textural classes. However, the only significant difference between measured and estimated Ksat values was found for the silt loam textural class (Table 10) applying an independent t test.

We also compared our estimated Ksat values from the infiltration measurements from the SWIG database with Ksat values from databases that have been published in the literature (Table 11). The validity of our estimated Ksat values is confirmed by comparing the order of magnitude of the difference between these values, and those tabulated in previous studies, for the various different soil classes. Some of these databases like that of Clapp and Hornberger (1978) and Cosby et al. (1984) have been used to parameterize land surface models. Most of the Ksat values in the listed databases have been obtained from laboratory-scale measurements often performed on disturbed soil samples. In most of the reported databases Ksat is controlled by texture, with the highest mean values obtained for the coarse-textured soils and the lowest mean values for the fine-textured soils. This is not the case for the Ksat values obtained from the SWIG database. Clayey soils have a mean value that is similar to the coarser textured soils. This may be partly explained by the fact that the measurements collected in the SWIG database are obtained from field measurements on undisturbed soils. It was observed that the standard deviation of Ksat in the SWIG database is typically larger than the standard deviations obtained from the databases in the literature. This indicates that texture is apparently not the most important control on Ksat values. However, one would also pose that much of the lack of correlation between soil texture and predicted Ksat from the SWIG database is related to the lack of soil structural information, such as macro porosity quantification or other possible soil attributes. Indeed, many of the datasets presented in our paper on saturated and near-saturated flow can be used to infer the state of the soil's structure, namely its macroporosity, by using the slope of the near-saturated conductivity curve, via Philip's “flow-weighted mean pore-size” analysis. White and Sully (1987) have discussed this in a great detail. Zhang et al. (2015) is another example of where tension infiltrometers can be used to describe the temporal dynamics of the macroporosity which characterizes soil structure. This could inspire researchers to collect such information when conducting additional soil infiltration measurements and include this in the database in the future. This finding indicates that present parameterization in current land surface models, which are mainly based on texture, may severely underestimate the variability of Ksat. In addition, it shows that also mean values are not dominantly controlled by textural properties. Other land surface properties such as land use and crusting may, in fact, be much more important.

In order to demonstrate the potential of the SWIG database for analyzing infiltration data and for developing pedotransfer functions, principal component analysis (PCA) was performed and biplots were generated to show both the observations and the original variables in the principal component space (Gabriel, 1971).

In a biplot, positively correlated variables are closely aligned with each other and the larger the arrows the stronger the correlation. Arrows that are aligned in opposite directions are negatively correlated with each other and the magnitude of the arrows is again a measure of the strength of the correlation. Arrows that are aligned 90∘ to each other show typically no correlation. Figures 7 and 8 show the results of two PCAs. The first PCA (Fig. 7) shows the relationship between soil textural properties, S and Ksat, based on 3267 infiltration measurements. The first two principal components explain 74.5 % of the variability in the data. Figure 7 shows a positive correlation between Ksat and S (0.527) and the largest values for both variables are found in clay soils. Clay content appears to only be weakly correlated with Ksat and S as is also shown by the correlation coefficients of 0.112 and 0.025, respectively. Figure 8 shows the biplot of soil textural properties with Ksat, S, organic carbon content, and bulk density in the principal component space – based on 1910 infiltration measurements. The first two principal components still explain 55 % of the variability. Neither S nor Ksat showed appreciable correlations with available soil properties. Only Ksat and S are correlated (arrows are aligned but small) with a value of 0.29. Organic carbon and bulk density show a negative correlation with a calculated value equal to -0.51. It also shows that, for example, the sandy clay loam textural class (yellow dots) shows a wide spread in organic matter content and bulk densities. These analyses show that the examined basic soil properties do not contain enough information to properly estimate Ksat and S. However, the SWIG database provides additional information such as land use, initial water content, and slope that might prove to be good predictors. A further analysis in this respect is however beyond the scope of this paper. More importantly, the present analysis in combination with the results provided in Table 11 shows that a texture-dominated derivation of Ksat values, as implemented in most land surface models, does not provide adequate means to estimate Ksat.

Similar to any other databases, the data presented in the SWIG database may be subject to different error sources and uncertainties. These include (1) transcription errors that occurred when implementing the measurement data into the EXCEL spreadsheets, (2) inaccuracy and uncertainties in determining related soil properties such as textural properties, (3) violation of the underlying assumption when performing the experiments, and (4) uncertainty (variability) in estimated soil hydraulic properties due to the different measurement methods. Unfortunately, none of these errors or uncertainty sources are under the control of the SWIG database authors, and quantification of these sources is often difficult, since the required information is often lacking. The uncertainty and variability related to the applied measurement techniques for estimated soil hydraulic properties may be assessed as information on the applied techniques is available; however, some of these methods may only have been used in few cases, making a statistical analysis difficult.

With respect to the transcription error, a strong effort has been made to double-check data transcription to prevent or at least to minimize any probable error of this nature. Values of soil properties such as textural composition are known to vary strongly between different laboratories and measurement methods. This is especially true for the finer textural classes like clay. Unfortunately, information on the measurement used to determine soil properties is mostly lacking or insufficient to assess the magnitude of errors or biases. Internationally, there are a number of standard methods used to measure soil properties and several methods may have been applied to measure the reported soil properties. In this regard, no conversion has been made and only raw data are reported in the database. However, we have supplied the references for all data (where available) that can be used to ascertain which methodologies were used, if so desired. Although supplying such information for each soil property may facilitate the use of the database, it would have required considerable additional work that could not be performed at this stage of development. Such additions could form the basis of a second version of the database that any readers should feel free to commence.

The uncertainty with respect to the effect of measurement techniques on quantifying the infiltration process itself may be analyzed from the SWIG database as it provides information on the type of measurement technique used. This analysis is again beyond the scope of this paper. Potential error and uncertainty sources with respect to the use of different measurements are discussed in the Supplement. The uncertainty of estimated soil hydraulic properties from infiltration measurements may be strongly controlled by the person performing the experiment but may also be due the different measurement windows of the methods in terms of measurement volume. The SWIG database provides information to quantify uncertainties introduced by difference in measurement volume and this analysis will be closely related to the assessment of the representative elementary volume, REV (see, for example, the work of Pachepsky on the scaling of saturated hydraulic conductivity).

Careful interpretation of the data, with respect to the details of the experimental and soil conditions, is also required when utilizing the SWIG database. For instance, the cases of soils coded 1211–1420 may at first seem odd, as they display very low infiltration rates for soils of a very high (> 95 %) sand content; however, these unusual findings are explained by the soils being recorded as displaying water repellant characteristics. Another example is estimated values of Ksat from clayey soils showing high values of Ksat (e.g., soils coded 3746 to 3833 in the SWIG database). The Ksat values for these soils were obtained using the single-ring infiltrometer method (Gonzalez-Sosa et al., 2010; Braud, 2015; Braud and Vandervaere, 2015) and were conducted in the field under ponded conditions, with vegetation cut but roots left in place. Macropores could have been activated, leading to an infiltration rate much higher than expected for clayey soils. There were also instances of very high values being obtained for forested land uses, and sometimes for grassland, which is probably explained by the visible cracks in the soil surface present in those cases

We envision that the SWIG database offers a unique opportunity and information source to (1) evaluate infiltration methods and to assess their value in deriving soil hydraulic properties, (2) test different models and concepts for point-scale and grid-scale infiltration processes, (3) develop pedotransfer functions to estimate soil hydraulic properties such as the Mualem–van Genuchten parameters, (4) identify controls on infiltration processes, (5) validate global predictions of infiltration from land surface models, (6) study more complex processes like preferential flow in soils, and (7) highlight the state-of-the-art understanding of the relationships between infiltration and several soil surface characteristics; for example, the SWIG database has already contributed to the scope of Morbidelli et al. (2018) to advance the knowledge of infiltration over sloping surfaces.

We are confident that the SWIG database is just a first step in collecting and archiving infiltration data and we expect that increasing amounts of data will become available in the near future. These data will be archived in the SWIG database and thus made available to the worldwide research community. In this regard, we are interested in receiving existing or newly measured infiltration curves and for this purpose the corresponding author will serve as point of contact or data can be made available through the International Soil Modeling Consortium, ISMC (https://soil-modeling.org/, last access: 1 July 2018), for further archiving in the SWIG database.