Update of the definition of vegetative filter strip scenarios for Europe w/ appendices

knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 1 Update of the definition of vegetative filter strip scenarios for Europe Report number 120333-1 Project number 120333 Author Dr. Stefan Reichenberger Dr. Thorsten Pohlert Jorge Olivares Rivas Date March 19, 2025 Sponsor Crop Life Europe 9 rue Guimard 1040 Brussels Belgium knoell Germany GmbH Page 2 of 83 Project No.: 120333 Report No.: 120333-1 2 GENERAL INFORMATION Sponsor: Crop Life Europe 9 rue Guimard 1040 Brussels Belgium Monitoring Scientist: Dr. Robin Sur Service provider: knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim Germany Project number: Report number: 120333 120333-1 Substance: n.a. Author: Dr. Stefan Reichenberger Dr. Thorsten Pohlert Jorge Olivares Rivas Report completion date: March 19, 2025 knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 3 EXECUTIVE SUMMARY Environmental risk assessment procedures for pesticides in the context of approval and authorization at EU and national level must keep up with scientific developments and therefore have to be updated regularly. The standardised FOCUS surface water approach for the estimation of predicted environmental concentrations of pesticides in surface water and sediment consists of a total of four tiers (Steps 1 to 4). Mitigation measures for spray drift and surface runoff can be included in Step 4. While spray drift inputs are typically mitigated with drift-reducing nozzles and no-spray zones, the most widely implemented mitigation measure to reduce the input of pesticides to surface waters via surface runoff and erosion are vegetative filter strips (VFS). The SWAN tool for higher-tier surface water exposure assessment implements fixed runoff and erosion reduction efficiencies according to the report of the FOCUS Landscape and Mitigation Working Group. Since SWAN version 3 the tool also includes the option to apply mitigation using the model VFSMOD, a numerical, mechanistic, dynamic and event-based model for predicting runoff reduction, sediment deposition, and pesticide trapping across a vegetative filter strip (VFS). VFSMOD was applied by Brown et al. (2012) to develop realistic worst case VFS scenarios for calculating the reduction of pesticide input to surface water for the four FOCUS runoff scenarios. The resulting scenarios were subsequently included in SWAN. Since then, with the publication of substantial scientific developments, VFSMOD has been improved in its underlying assumptions and process descriptions. This requires also updating the original VFS scenarios of Brown et al. (2012) using the new VFSMOD version. Moreover, new EU soil profile and land cover data have become available, and the United Kingdom has left the EU. Hence, the population of the spatial cumulative distribution functions (CDFs) of the relative reduction of incoming pesticide load (∆P) by VFS needed to be updated. The objective of this study was therefore to update and improve the SWAN-VFSMOD scenarios for EU pesticide risk assessment. In this study, 64 new spatial CDFs of ∆P have been created. Out of these, the four CDFs corresponding to the most relevant combination of input factors in the context of FOCUS were selected for the new SWAN-VFSMOD scenarios. From those, the soil profiles corresponding to the 10th percentile of ∆P (thus, the 90th percentile worst case in space) were selected for parameterization of VFSMOD in SWAN 5.2. A comparison of the new spatial CDFs with predicted trapping efficiencies ∆P obtained with the previous SWAN-VFSMOD scenarios and the old VFSMOD version yielded that the new SWANVFSMOD scenarios are overall more conservative than the old ones. Analysis of the new CDFs with regard to soil texture confirmed that the approach of selecting overall 90th percentile worst case soil profiles for the VFS scenarios is reasonable. As expected, some texture classes have higher vulnerability than others. The general ranking of texture classes with respect to ∆P (Coarse > Medium > Fine > Medium Fine > Very Fine) reflects the soil hydraulic properties; notably the predicted saturated conductivity is on average lowest for soils with texture classes Medium Fine (very high silt content) and Very Fine (very high clay content). In conclusion, this study has derived a set of updated VFSMOD scenarios for FOCUS Step 4 simulations. The new 90th percentile worst case profiles for SWAN-VFSMOD correspond to the state of the art of VFS scientific knowledge and the most recent available soil profile databases and land cover data. The updated SWAN-VFSMOD scenarios are both more realistic and more protective than the old ones. While the derived SWAN-VFSMOD scenarios represent the best estimate of a 90th percentile worst case given the available data, this report may also offer a blueprint for revising the scenarios as new data become available. Potential future updates of the SWAN-VFSMOD scenarios should include first a revision of the FOCUS scenario maps because of climate change. knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 4 GLOSSARY Term Definition ∆E relative reduction of incoming eroded sediment load by the Vegetative Filter Strip (VFS) ∆P relative reduction of incoming pesticide load by the VFS ∆Q relative reduction of total inflow (runoff + rainfall) by the VFS ∆R relative reduction of incoming surface runoff by the VFS CDF Cumulative Distribution Function CLC CORINE Land Cover CORINE Coordination of Information on the Environment CRU Climatic Research Unit DegT50 Degradation time: Time taken for 50 % of the substance to disappear from a compartment by degradation processes DegT90 Degradation time: Time taken for 90 % of the substance to disappear from a compartment by degradation processes FC Field capacity water content (VFSMOD parameter) FOCUS FOrum for the Coordination of pesticide fate models and their USe HYPRES HYdraulic PRoperties of European Soils Kd linear adsorption coefficient; ratio of concentration in the sorbed and the liquid phase Kfoc Adsorption/desorption coefficient normalised for organic carbon content in soil following a Freundlich adsorption isotherm Koc Linear adsorption coefficient normalised to soil organic carbon content Kom Linear adsorption coefficient normalised to soil organic matter content Kow n-Octanol/ water partitioning coefficient MARS Monitoring Agricultural ResourceS MUSLE Modified Universal Soil Loss Equation MUSS Modified version of MUSLE for small catchments; meaning of acronym unknown NUTS Nomenclature des Unités Territorials Statistiques (Nomenclature of Territorial Units for Statistics) OS saturated water content (VFSMOD parameter) ptf pedotransfer function PECsed Predicted environmental concentration in sediment PECsw Predicted environmental concentration in surface water SAV Green-Ampt average suction at wetting front (VFSMOD parameter) SGDBE Soil Geographical Database of Europe SMU Soil Mapping Unit SPADE Soil Profile Analytical Database for Europe STU Soil Typological Unit SWAN Surface Water Assessment eNabler UH Unit Hydrograph or Hydrologic Utility VFS Vegetative (or Vegetated) Filter Strip VFSMOD Vegetative Filter Strips Modeling System VKS vertical saturated conductivity (VFSMOD parameter) VL VFS length in flow direction (VFSMOD parameter), often referred to as “buffer strip width” knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 5 TABLE OF CONTENTS GENERAL INFORMATION ....................................................................................................... 2 EXECUTIVE SUMMARY ........................................................................................................... 3 GLOSSARY ................................................................................................................................... 4 TABLE OF CONTENTS .............................................................................................................. 5 LIST OF TABLES ......................................................................................................................... 7 LIST OF FIGURES ....................................................................................................................... 8 1 INTRODUCTION ............................................................................................................. 10 2 MATERIALS AND METHODS ...................................................................................... 11 2.1 Update of spatial basis and determination of soil profiles to be simulated .................. 11 2.2 Parameterization and running of VFSMOD .................................................................. 14 2.2.1 Treatment of soil data ................................................................................................ 14 2.2.2 Derivation of VFSMOD parameters.......................................................................... 16 2.2.3 New versions of UH and VFSMOD .......................................................................... 17 2.2.4 Setup and running of VFSMOD simulations ............................................................ 20 2.3 Spatial aggregation and determination of 90th percentile worst case profiles ............. 21 2.3.1 Calculation of area fractions for each soil profile ..................................................... 21 2.3.2 Spatial aggregation to CDFs ...................................................................................... 22 2.3.3 Determination of 90th percentile worst case profiles and associated parameters ...... 22 2.4 Comparison of new CDFs with previous SWAN-VFSMOD scenarios / VFSMOD version 22 2.5 Analysis with regard to soil texture ................................................................................. 23 3 RESULTS AND DISCUSSION ........................................................................................ 25 3.1 Area covered by CDFs ...................................................................................................... 25 3.2 Spatial CDFs ...................................................................................................................... 26 3.2.1 CORINE vs. noCORINE ........................................................................................... 26 3.2.2 Effect of Rainfall duration ......................................................................................... 28 3.2.3 Effect of Koc ............................................................................................................. 29 3.2.4 Variant A vs. Variant B ............................................................................................. 30 3.3 Final choice of scenario ..................................................................................................... 35 3.4 Comparison of new CDFs with previous SWAN-VFSMOD scenarios / old VFSMOD version 36 3.5 Analysis with regard to soil texture ................................................................................. 39 4 CONCLUSIONS AND OUTLOOK ................................................................................ 44 5 REFERENCES .................................................................................................................. 45 6 ANNEX ............................................................................................................................... 48 6.1 Resources ............................................................................................................................ 48 6.2 Spatial extent of the FOCUS R scenarios ........................................................................ 49 6.3 Description of tables from SGDBE .................................................................................. 52 6.4 Description of SPADE2 tables .......................................................................................... 53 knoell Germany GmbH Page 6 of 83 Project No.: 120333 Report No.: 120333-1 6 6.5 Description of SPADE14 tables ........................................................................................ 56 6.6 Detailed method to determine number of profile / scenario combinations to be modelled.............................................................................................................................. 63 6.7 Description of spatial aggregation procedure ................................................................. 65 6.7.1 Spatial input tables..................................................................................................... 65 6.7.2 Recoding of land use for SPADE14 .......................................................................... 67 6.7.3 Calculation of area fractions for the different profiles .............................................. 69 6.7.4 Calculation of spatial CDFs ....................................................................................... 72 6.7.5 Determine 90th percentile worst case profile from CDF............................................ 75 6.8 Complete VFSMOD parameterization ............................................................................ 77 6.9 Manning’s roughness coefficient for VFS (RNA) ........................................................... 77 6.10 Confusion matrix for predicting missing profile texture using the dominant topsoil texture class of the STU .................................................................................................... 79 6.11 List of delivered files (digital annex) ................................................................................ 80 knoell Germany GmbH Page 7 of 83 Project No.: 120333 Report No.: 120333-1 7 LIST OF TABLES Table 2 List of countries after intersecting the FOCUS scenario shapefiles with the map of administrative boundaries (NUTS0) ............................................................................... 13 Table 3 List of profile-specific VFSMOD parameters ................................................................ 17 Table 4 90th percentile worst-case profiles and profile-specific VFSMOD parameters for the proposed combination and the four R scenarios ............................................................. 35 Table 5 Pesticide reduction efficiencies (∆P) predicted with old SWAN-VFSMOD scenarios and old vfsm executable, and 10th percentile profiles of the new CDFs 1) created using the new vfsm executable ................................................................................................. 37 Table 6 Positions of old SWAN/VFSMOD scenarios (with old VFSMOD executable) on new CDFs 1) ............................................................................................................................ 38 Table 7 Cumulative area fractions for each texture class corresponding to the 10th percentile pesticide reduction efficiency (∆P) of the whole CDF (all texture classes) for the combination selected for the SWAN-VFSMOD scenarios (Koc 100 L kg-1, duration 8 h, CORINE, variant B) ................................................................................................... 41 Table 8 Cumulative area fractions for each texture class corresponding to the 10th percentile pesticide reduction efficiency (∆P) of the whole pooled CDF (all texture classes and scenarios combined) for the combination selected for the SWAN-VFSMOD scenarios (Koc 100 L kg-1, duration 8 h, CORINE, variant B) ...................................................... 43 Table A 1 Variables contained in stuorg.dbf .................................................................................. 52 Table A 2 Variables contained in stu_sgdbe.dbf (only the relevant ones are listed)...................... 53 Table A 3 Variables contained in spadev211.dbf (Hannam et al., 2009) ....................................... 53 Table A 4 Variables contained in cleaned-up profile/horizon table gethorizons_SPADE2_all_20240304.xlsx ..................................................................... 55 Table A 5 Variables contained in spade14_stu.xlsx ....................................................................... 57 Table A 6 Variables contained in prof_all (Daroussin, 1999) ........................................................ 57 Table A 7 Variables contained in hor_all (Daroussin, 1999) ......................................................... 59 Table A 8 Variables contained in cleaned-up profile/horizon table gethorizons_SPADE14_all_20240304.xlsx ................................................................... 61 Table A 9 Columns of table stuorg-stu_sgdbe-spade14-spade2-R1234.xlsx ................................. 64 Table A 10 Variables contained in R1234_grouped_20240411.xlsx and R1234_CORINE_grouped_20240411.xlsx .................................................................... 65 Table A 11 Variables contained in R1234_NUTS_agg_leftjoin_stuorg.xlsx ................................... 66 Table A 12 Variables contained in R1234_NUTS_CLC_agg_leftjoin_stuorg.xlsx ......................... 66 knoell Germany GmbH Page 8 of 83 Project No.: 120333 Report No.: 120333-1 8 Table A 13 Land use codes used in SPADE2 (Hannam et al., 2009)................................................ 67 Table A 14 Classification rules applied to SPADE14 land use descriptions .................................... 68 Table A 15 Variables contained in of CLC_link_table.csv or noCLC_link_table.csv ..................... 71 Table A 16 Variables contained in VFSMOD result tables R1234_30mm_noWT_out_ 20240526.xlsx (Variant A) and R1234_30mm_noWT_out_20240610.xlsx (Variant B). 73 Table A 17 Variables contained in intermediate data frame totalAreaFracPerSMU ........................ 74 Table A 18 Variables contained in intermediate data frame smuAreasNew ..................................... 74 Table A 19 Variables contained in output CDF files (.csv) .............................................................. 74 Table A 20 Extracted 90th percentile worst case profiles for all CDFs ............................................. 75 LIST OF FIGURES Figure 1 CORINE Land Cover 2018 map of Europe. Source: https://sdi.eea.europa.eu/............. 13 Figure 2 Updates VFSMOD approach for pesticide residues. Source: Muñoz-Carpena et al. (2022). ............................................................................................................................. 19 Figure 3 The SGDBE texture triangle. Source: https://www.researchgate.net/figure/Textureclasses-after-CEC-1985-Expressing-lateral-variability-BULLET-A-STU-canhave_fig2_237642627 .................................................................................................... 24 Figure 4 Area fractions of FOCUS scenario shapefiles (for the extent of the EU27) covered by the new spatial CDFs. ..................................................................................................... 25 Figure 5 Example CDF (R3, Koc = 100 L kg-1, duration = 8 h, and variant B) with (top) and without (bottom) intersection with CORINE Land Cover (CLC) 2018 class 2. ............ 27 Figure 6 Example CDF (FOCUS R1, intersect with CLC 2018 class 2, Koc = 100 L kg-1, variant B, duration = 1 h (top) and 8 h (bottom). ....................................................................... 28 Figure 7 Example CDF (FOCUS R4, intersection with CLC 2018 class 2, duration = 1 h, variant B, Koc = 100 L kg-1 (top) and 10000 L kg-1 (bottom). .................................................. 30 Figure 8 Example CDF (FOCUS R4, intersection with CLC 2018 class 2, duration = 1 h, Koc = 100 L kg-1, variant A (top) and variant B (bottom). ....................................................... 31 Figure 9 Example CDF (FOCUS R4, intersection with CLC 2018 class 2, duration = 1 h, Koc = 10000 L kg-1, variant A (top) and variant B (bottom)). .................................................. 32 Figure 10 Example CDF (FOCUS R4, intersection with CLC 2018 class 2, duration = 8 h, Koc = 100 L kg-1, variant A (top) and variant B (bottom). ....................................................... 33 Figure 11 Example CDF (FOCUS R4, intersection with CLC 2018 class 2, duration = 8 h, Koc = 10000 L kg-1, variant A (top) and variant B (bottom). ................................................... 34 Figure 12 Spatially weighted boxplots of predicted pesticide trapping efficiency (∆P) according to knoell Germany GmbH Page 9 of 83 Project No.: 120333 Report No.: 120333-1 9 texture class for all simulated profiles for each scenario (Koc 100 L kg-1, duration 8 h, CORINE, variant B). Boxes: 25th and 75th percentiles, whiskers: min and max. Spatial weighting with with area of the scenario/profile combination was applied. Horizontal line: 10th percentile ∆P (90th percentile worst case) of whole CDF (all texture classes) for option CORINE. Texture classes are C = Coarse, M = medium, MF = medium fine, F = fine, VF = very fine, O = organic. The sample size is given below each box. ........ 40 Figure 13 Spatially weighted boxplots of predicted pesticide trapping efficiency (∆P) according to texture class for all simulated profiles for all scenarios combined (Koc 100 L kg-1, duration 8 h, CORINE, variant B). Boxes: 25th and 75th percentiles, whiskers: min and max. Spatial weighting with area of the scenario/profile combination was applied. Horizontal line: 10th percentile of ∆P of the combined CDF over all scenarios for option CORINE. Texture classes are C = Coarse, M = medium, MF = medium fine, F = fine, VF = very fine, O = organic. The sample size is given below each box. ....................... 42 Figure A 1 Extent of the FOCUS R1 scenario shapefile (FOCUS, 2001) after intersection with CORINE Land Cover 2018 class 2 (agricultural land) ................................................... 49 Figure A 2 Extent of the FOCUS R2 scenario shapefile (FOCUS, 2001) after intersection with CORINE Land Cover 2018 class 2 (agricultural land) ................................................... 50 Figure A 3 Extent of the FOCUS R3 scenario shapefile (FOCUS, 2001) after intersection with CORINE Land Cover 2018 class 2 (agricultural land) ................................................... 51 Figure A 4 Extent of the FOCUS R4 scenario shapefile (FOCUS, 2001) after intersection with CORINE Land Cover 2018 class 2 (agricultural land) ................................................... 52 Figure A 5 Tabulated values of Manning’s n for different vegetation types and bare surfaces. Source: Muñoz-Carpena and Parsons (2022) ................................................................. 78 Figure A 6 Confusion matrix for all combinations of profiles with texture information and STU (n = 1233) ............................................................................................................................ 79 knoell Germany GmbH Page 16 of 83 Project No.: 120333 Report No.: 120333-1 16 2.2.2 Derivation of VFSMOD parameters Profile-specific VFSMOD parameters (Table 2) were calculated from the effective single-horizon profiles using the following stepwise procedure: 1) Calculate Mualem-Van Genuchten parameters, using HYPRES continuous pedotransfer functions (Wösten et al., 1998) for soils within the validity limit of the HYPRES continuous ptfs (0.5 % < SILT < 99 % AND 0.5 % < CLAY < 90 % AND 0.1 % < OM < 30 % AND 0.5 kg/dm3 < DB). For soils outside these validity limits, the HYPRES class pedotransfer functions (Wösten et al., 1998) were used 2) Calculate VFSMOD parameters • Vertical saturated conductivity VKS: use Mualem-Van Genuchten parameter K0 from HYPRES ptfs • Average suction at wetting front SAV: calculated by Rafael Muñoz-Carpena in Mathematica according to Neuman (1976) • Saturated water content OS: (use theta_s from HYPRES, corrected for stone content, assuming a stone porosity of 10 %); the reduction of OS will lead to a deeper wetting front, but it does not affect the value of SAV • Field capacity water content FC: read off from Van Genuchten water retention curve at pF2 • FC is also used as initial water content OI • Manning’s roughness coefficient RNA: refers to the surface characteristics of the VFS; depends on the vegetation and state of the VFS; RNA does not depend on the VFS soil profile nor on the field soil → same value (0.40 s m-1/3) chosen for all profiles (identically to Brown et al., 2012); see also tabulated values from the VFSMOD manual (Muñoz-Carpena and Parsons, 2022) for comparison (Figure A 5) • Other parameters: see VFSMOD parameterization guidance (Ritter et al., 2023) In Brown et al. (2012) the average suction at the wetting front SAV was defined as a function of the FOCUS scenario, i.e. as a property of the soil of the arable field. However, since SAV is used for simulating infiltration in the VFS, it has to be a property of the VFS soil. Hence, in our study SAV is specific to the VFS soil profile. Of the 1897 scenario / profile combinations, there were 239 with an R horizon in less than 1 m depth. The presence of shallow R horizons implies that the assumptions of the Green-Ampt approach (homogeneous soil, no obstacles to the wetting front) are not met, unless the rocks are karstic or very fractured. An impermeable rock layer poses the same limitation to percolation as a permanent shallow water table. Hence, assuming that the R layer is impermeable, the profile can be modelled with the shallow water table option (cf. Bach et al., 2017). Note, however, that the assumption of impermeability is a worst case because many types of hard rock have fissures or cracks (e.g. granite) or a certain permeability (e.g. sandstone, limestone). To test the impact of using the shallow water option for shallow soils on the final CDFs (90th percentile worst case profiles and predicted pesticide reduction efficiencies (∆P)), two variants were set up in VFSMOD: • Variant A: Profiles with hard rock in < 100 cm depth were simulated as infiltration on a soil with a limiting horizon, using the shallow water table option in VFSMOD (Muñoz-Carpena et al., 2018); no changes for the other profiles with limiting horizons >100 cm were made, as the infiltration will not typically run that deep in normal infiltration events • Variant B: All profiles were simulated with classical Green-Ampt infiltration (no limiting horizons in the soil) knoell Germany GmbH Page 17 of 83 Project No.: 120333 Report No.: 120333-1 17 Table 2 List of profile-specific VFSMOD parameters Parameter unit description remarks VKS m s-1 vertical saturated conductivity SAV m suction at wetting front Used only for Green-Ampt approach FC m3 m-3 field capacity water content of VFS soil Used only for pesticide degradation OS m3 m-3 saturated water content of VFS soil Stone content considered (assuming a stone porosity of 10 %) OI m3 m-3 initial water content of VFS soil Set equal to FC VGN (-) Van Genuchten n Used only for shallow water table option VGALPHA m-1 Van Genuchten alpha Used only for shallow water table option VGM (-) Van Genuchten m Used only for shallow water table option; set to 1 -1/VGM WTD m water table depth Used only for shallow water table option; Set equal to depth to hard rock 2.2.3 New versions of UH and VFSMOD UH (Muñoz-Carpena and Parsons, 2022; Muñoz-Carpena and Parsons, 2004) is a utility written in Fortran to create inputs for VFSMOD: rainfall hydrographs, surface runoff hydrographs and eroded sediment loads. The name UH stands for Unit Hydrograph or Hydrologic Utility. It was already used by Brown et al. (2012) to create rainfall (.irn) and runoff (.iro) hydrographs as input for VFSMOD. Unfortunately, the rainfall and runoff hydrographs produced by Brown et al (2012) using an older version of UH could not be retrieved anymore. Hence, they had to be recreated. The new internal method in VFSMOD for determining the median particle size d50 (Reichenberger et al., 2023) uses peak runoff rate as an input. However, the peak runoff rate depends on the temporal resolution of the runoff hydrograph. Hence, it was necessary to make the temporal resolution of the UH output hydrographs configurable. Moreover, it was decided to use the MUSS equation (Williams, 1995) for calculating eroded sediment yield, analogously to PRZM for the FOCUS scenarios (FOCUS, 2001). These modifications required the development of a new version of UH by the VFSMOD developer. After iterative beta-testing of the new UH version a final version was obtained UH v. 3.07, dated 27/02/2023. The new UH v 3.07 is able to • produce rainfall and runoff hydrographs with user-defined temporal resolution • produce erosion estimated with MUSS equation (used in FOCUS PRZM) Since the VFSMOD version used by Brown et al. (2012) (v.4.2.4) several improved process decriptions were implemented in VFSMOD. First, new pesticide trapping equations were implemented in VFSMOD to address the concerns formulated in the MAgPIE report (Alix et al., 2017): “The regulatory status of VFSMOD in the EU regulatory process is currently uncertain. The model is recommended for use here given its general validation status in the scientific literature and because it is able to reflect changes in buffer efficacy based on e.g. changes in antecedent moisture conditions. Additional work is recommended outside of the MAgPIE process to reach a conclusion on the regulatory acceptability of the model in the EU. A particular issue is evaluation of coupling of the basic VFSMOD code with the knoell Germany GmbH Page 18 of 83 Project No.: 120333 Report No.: 120333-1 18 regression equation for pesticide transfer across vegetated filter strips reported by Sabbagh et al. (2009).” To corroborate and improve the predictive capability of the Sabbagh equation, Reichenberger et al. (2019) compiled additional experimental VFS data from the available literature. The enlarged dataset (n = 244 values of relative pesticide load reduction ∆P) was used to recalibrate the Sabbagh equation, the recently proposed Chen equation (Chen et al., 2016) and a set of “reduced” Sabbagh equations with fewer independent variables. Moreover, an alternative, regression-free mass balance approach was developed and tested, building on the initial mass balance proposed in Muñoz-Carpena et al. (2015). The recalibrated Sabbagh equation fitted the observed data well (coefficient of determination R2 = 0.819), and its coefficients were corroborated by both a rigorous k-fold cross-validation analysis and a maximum-likelihood-based calibration and uncertainty analysis conducted with DREAM_ZS (Vrugt, 2016). However, the regression-free mass-balance approach also fitted the observed data quite well (R2 = 0.741) and much better than the old Sabbagh equation with the original coefficients (R2 = 0.528). Because of its advantage of being mechanistic and its overall good predictive performance while being conservative, the mass balance approach is now recommended as the default option for regulatory modelling of pesticide trapping in VFS with VFSMOD (Ritter et al., 2023). The mass-balance trapping equation of Reichenberger et al. (2019) is given as: ∆P 100% =min⁡[(𝑉𝑖+𝐾𝑑𝐸𝑖);( ∆𝑄 100%𝑉𝑖+∆𝐸 100%𝐾𝑑𝐸𝑖)]⁡ (𝑉𝑖+𝐾𝑑𝐸𝑖) (eq. 1) with ∆P relative reduction of total incoming pesticide load (%) Vi incoming run-on volume (L) (run-on is the overland flow entering the VFS from upslope) Kd linear sorption coefficient (L kg-1); Ei incoming eroded sediment load (kg) ∆Q relative reduction of total inflow (rainfall + run-on) (%) ∆E relative reduction of incoming eroded sediment load (%) The minimum expression for the conservation of mass was later shown to be unnecessary, because ∆Q and ∆E cannot get larger than 100 %. This simplifies the equation to ∆P 100% =(∆𝑄 100%𝑉𝑖+∆𝐸 100%𝐾𝑑𝐸𝑖)]⁡ (𝑉𝑖+𝐾𝑑𝐸𝑖) (eq. 1b) The Kd in eqs. 1 and 1b is calculated in VFSMOD as Kd = Koc * OCP/100, with Koc being the linear adsorption coefficient normalized to organic carbon (user input) and OCP the organic carbon content (%) of the eroded sediment (assumed equal to the organic carbon content of the uppermost horizon of the field soil, i.e. the soil of the R scenario parameterized in PRZM). The second new feature in VFSMOD was the internal calculation of median particle size d50, which is the most important sediment characteristic for sediment trapping. The rationale for the development of this calculation was that the default value of d50 = 20 µm chosen by Brown et al (2012) seemed too high, thus leading to an overestimation of sediment load reduction ∆E. The value of 20 µm had been chosen as a low percentile from an experimental d50 dataset (n = 185) which was however likely not representative of edge-of-field runoff (cf. discussion in Reichenberger et al., 2023). Since errors in knoell Germany GmbH Page 19 of 83 Project No.: 120333 Report No.: 120333-1 19 sediment trapping efficiency ΔE propagate to ΔP, for strongly-sorbing compounds an accurate prediction of ΔE is crucial for a reliable prediction of ΔP. Reichenberger et al. (2023) first compiled an extensive dataset of 93 d50 values and 17 candidate explanatory variables compiled from heterogeneous data sources and methods, and subsequently derived an improved dynamic d50 parameterization equation for use in regulatory VFS scenarios. The dataset was analysed first using machine learning techniques (Random Forest, Gradient Boosting) and Global Sensitivity Analysis (GSA) as a dimension reduction technique and to identify potential interactions between explanatory variables. Using the knowledge gained, a parsimonious multiple regression equation with 6 predictors was developed and thoroughly tested. Since three of the predictors are event-specific (eroded sediment yield, rainfall intensity and peak runoff rate), predicted d50 vary dynamically across event magnitudes and intensities. The internal, dynamic d50 calculation option is now recommended as the standard option for regulatory modelling with VFSMOD (Ritter et al., 2023). The third new feature, the new leaching / remobilization algorithm (Muñoz-Carpena et al., in preparation; Muñoz-Carpena et al., 2022) was developed to improve the accuracy of the calculation of pesticide residues in the mixing layer of the VFS soil at the end of a runoff event as well as the accuracy of the pesticide mass remobilized at the next runoff event. Leaching in the VFS is now calculated fully mechanistically using an analytical solution of the chemical convective-dispersive transport equation (CDE) with sorption under Green-Ampt infiltration (Huang and van Genuchten, 1995). The key features of the new remobilization approach (Figure 2) are that i) only dissolved pesticides in the mixing layer are remobilized, and ii) non-remobilized pesticide residues do not disappear, but are added to the incoming pesticide mass trapped in the mixing layer and are thus carried over to the next event, considering degradation. However, in this study remobilization was not relevant because only one single event was simulated, i.e. no series of events like in FOCUS step4. Figure 2 Updates VFSMOD approach for pesticide residues. Source: Muñoz-Carpena et al. (2022). Finally, a shallow water table option (Muñoz-Carpena et al., 2018; Lauvernet and Muñoz-Carpena, 2018) has already been available for several years. This option allows simulating the impact of a permanent or seasonal shallow water table on VFS efficiency. Moreover, the option can also be used to simulate the impact of a shallow impermeable rock layer. The basic concept of the SWINGO algorithm (Shallow Water table INfiltration algorithm; Muñoz-Carpena et al., 2018) is the following: In the first phase, infiltration is calculated using a time-explicit infiltration solution based on a combination of knoell Germany GmbH Page 20 of 83 Project No.: 120333 Report No.: 120333-1 20 approaches by Salvucci and Entekhabi (1995) and Chu (1997). Once the wetting front reaches the water table (i.e. the VFS profile is completely saturated), the boundary condition changes. The infiltration rate at the soil surface is then controlled by lateral subsurface flow, with the subsurface flow rate following Dupuit-Forchheimer assumptions (Muñoz-Carpena et al., 2018). During preliminary VFSMOD simulations numerical problems with snowmelt events (only relevant for FOCUS R1 scenario) emerged. The reason was that these events do not have rainfall, while a mean rainfall rate is needed for the d50 algorithm. Moreover, there was a problem with the .owq output files: Sometimes they contained numbers having negative exponents with three digits, which were then not recognized as numbers during the postprocessing in Python. Iterative beta-testing of the new VFSMOD version resulted in a final, consolidated version: vfsm.exe v.4.5.2, dated 21/03/2023. The new version also contains a modified algorithm to determine the number of numerical nodes N for the numerical solution of overland flow: N = 90/99 (VL - 1) + 11 with VL being the VFS length in flow direction (m) 2.2.4 Setup and running of VFSMOD simulations VFSMOD v. 4.5.2 was run for all 1897 valid combinations of FOCUS scenario and soil profile for example rainfall/runoff events, closely following the method of Brown et al. (2012). First, eight rainfall/runoff events (two for each R scenario) were created with the tool UH v. 3.07 (Univ. Florida) using the following settings: • Rainfall volume: 30 mm • Event duration: 1 h / 8 h • Slope length: 100 m • Runoff and erosion parameters according to Brown et al. (2012) (see Table 6 in Brown et al., 2012) • Erosion calculation with MUSS equation (like in FOCUS PRZM); this equation was not available in earlier versions of UH • Temporal resolution of created rainfall hydrographs (.irn) and field runoff hydrographs (.iro): 15 min UH input and output files can be found in the digital annex (see section 6.11). Brown et al. (2002) selected a rainfall volume of 30 mm in order to produce sufficient surface runoff so that complete infiltration in the VFS occurred only rarely. The rainfall durations of 1 h and 8 h were chosen by Brown et al (2012) for the following reasons: The high-intensity, short-duration event (30 mm in 1 hour) was conceptualised as a convective rainstorm event as it occasionally occurs in European summers, while the low-intensity, long-duration event (30 mm in 8 hours) was conceptualised as a depression-led, advective rainfall event. VFSMOD was run for two example compounds (Koc = 100 L kg-1 and 10000 L kg-1) and a VFS length in flow direction (“VFS width”) VL of 10 m. Hence, in total 1897 scenario/profile combinations * 2 event durations * 2 Koc values * 2 variants (A and B) = 15176 VFSMOD simulations were performed (1 simulation = 1 event). The two Koc values were chosen by Brown et al. (2012) to reflect contrasting sorption behaviour. While for Koc = 100 L kg-1 the compound will be predominantly transported in the dissolved phase of the surface runoff, for Koc = 10000 L kg-1 the dissolved and particle-bound fractions will be of the same order of magnitude. In contrast to the simulations of Brown et al. (2012) new process descriptions were used that did not exist in the old VFSMOD version (v. 4.2.4) used by Brown et al. (2012) (see section 2.2.3): knoell Germany GmbH Page 21 of 83 Project No.: 120333 Report No.: 120333-1 21 • trapping equation: mass balance approach (Reichenberger et al., 2019; Muñoz-Carpena et al., 2015) • median particle size d50: internal calculation with multiple regression equation (Reichenberger et al., 2023) • improved leaching/remobilization algorithm (Muñoz-Carpena et al., in preparation; MuñozCarpena et al., 2022) Moreover, a minor change was made to the parameter SS (average spacing of grass stems): The value was changed from 1.63 cm (corresponding to ryegrass (perennial); Muñoz-Carpena and Parsons, 2022) to a slightly more conservative value of 2.15 cm (corresponding to a grass mixture; Muñoz-Carpena and Parsons, 2022). For the VFSMOD simulations, a tailing time of 100 % was added to the .irn file in order to make sure that all runoff water has either infiltrated or run off before the end of the simulation. VFSMOD simulations were created, launched and postprocessed using the knoell tool swan2pyVFSMOD. swan2pyVFSMOD is a Python wrapper around VFSMOD which reproduces the functionalities of SWAN-VFSMOD. The target variable was the relative reduction of pesticide load (ΔP) by the VFS. All VFSMOD input and output can be found in the digital annex. 2.3 Spatial aggregation and determination of 90th percentile worst case profiles The workflow to determine 90th percentile worst case profiles was: 1) Calculate area of scenario/profile combinations 2) Calculate spatial CDFs of target variable ΔP (pesticide load reduction efficiency) for each combination of FOCUS scenario and treatment • 4 R scenarios • 2 different Koc values (100 and 10000 L kg-1) • 2 event durations (1 h and 8 h) • 2 VFSMOD variants (A and B) • 2 spatial variants: FOCUS scenario shapes clipped with CORINE Land Cover class 2 or not → 64 CDFs 3) Extract profile that corresponds to the 10th spatial percentile of ΔP (90th percentile worst case) The procedure was implemented and performed in R. All created CDFs can be found in the digital annex. 2.3.1 Calculation of area fractions for each soil profile Calculating area fractions of soil profiles per Soil Mapping unit (SMU) was challenging and timeconsuming, due to • complex and different data structures of SPADE2 and SPADE14 profile databases • occurrence of duplicates (both in SPADE2 and SPADE14 identical profiles with different IDs occur) • land use in SPADE14 only being described in text form → a script for assigning SPADE2 land use code had to be developed First, the textual descriptions of land use (field LU of EST_PROF tables from SPADE14 database) were classified according to the 23 numeric land use codes of SPADE2 (see Hannam et al., 2009, p. 25) using an R script (see sections 6.7.2 for the general workflow and 0 for the code (script recodeLUspade14.R)). The calculation of areas for each soil / scenario combination is partly based on an R script by Nils Kehrein (Bayer). However, several modifications and additions were made to the code. A fully transparent and reproducible procedure is in place now (see sections 6.7.3 for the general workflow and 0 for the code (script import_extended.R)). knoell Germany GmbH Page 22 of 83 Project No.: 120333 Report No.: 120333-1 22 The following rules (similar to those used by Hannam et al. (2009)) were used to assign area fractions covered by a given profile within a STU, considering the dominant and secondary land uses of the STU (USE_DOM and USE_SEC, respectively) and the actual land use of the profile (USE): 1) If USE = USE_DOM and USE_DOM is not 0 (i.e. dominant land use is known) and USE_SEC = 0 (i.e. secondary land use unknown or non-existing), the profile is assigned an area fraction of 0.8 2) Else if USE = USE_DOM and USE_DOM is not 0 and USE_SEC is not 0 (i.e. there is a secondary land use), the profile is assigned an area fraction of 0.6 3) Else if USE = USE_SEC and USE_SEC is not 0, the profile is assigned an area fraction of 0.3 4) Else the profile is assigned an area fraction of 0.1. 5) In the case that there is more than one profile in the same STU with the same land use, the area fractions calculated above are evenly distributed over these profiles. 6) Finally the area fraction of the SMU covered by the soil profile is calculated. The result of the procedure is a table of area fractions (see Table A 15 for the list of variables). Two variants have been produced: • CLC_link_table.csv (with intersection with CORINE Land Cover; see section 2.1) • noCLC_link_table.csv (without intersection with CORINE Land Cover) The two tables are included in the digital annex. 2.3.2 Spatial aggregation to CDFs In the next step, spatial cumulative distribution functions of the target variable ∆P were calculated using the area fractions calculated above. The general workflow is described in section 6.7.4, and the code (createSpatialCDF.R) can be found in the digital annex. As a plausibility check, for each FOCUS scenario the fraction of the scenario polygon area that is covered by the corresponding CDF was also calculated. 2.3.3 Determination of 90th percentile worst case profiles and associated parameters After the CDFs had been created, the next steps were: • Determine profile that corresponds to the 10th spatial percentile of ΔP (90th percentile worst case). For this we used a binary search algorithm in R. • Extract scenario parameters of selected profiles using MS Access. The general workflow is described in section 6.7.5, and the code (profileID_90th_PPerc.R) can be found in the digital annex (see section 6.11). 2.4 Comparison of new CDFs with previous SWAN-VFSMOD scenarios / VFSMOD version A series of test runs (4 FOCUS R scenarios * 2 Koc values * 2 durations) were done with the VFSMOD version (v. 4.2.4) and parameterization from SWAN 5.01. The objective was to check how conservative assessments with SWAN 5.01 are in comparison with the new spatial CDFs, and how ΔP will change if we switch from SWAN 5.01 to the 10th percentile of the new CDFs. The settings of the 16 test runs were as follows: • same vfsm.exe (v. 4.2.4) as in the portable version of SWAN 5.01 • Koc 100 and 10000 L kg-1 • 30 mm rainfall, rainfall durations 1 h and 8 h, VL = 10 m (like in Brown et al., 2012) • 4 FOCUS R scenarios with the same VFSMOD parameterization as in SWAN 5.01 knoell Germany GmbH Page 23 of 83 Project No.: 120333 Report No.: 120333-1 23 • same runoff and rainfall hydrographs (.iro and .irn) as in the simulations used for the new CDFs • same eroded sediment load and pesticide input These results were compared with the spatial CDFs obtained using the current vfsm.exe (v. 4.5.2) and recommended parameterization (variant B = classical Green-Ampt for all profiles). To isolate the effect of the different trapping equations (old Sabbagh equation (Sabbagh et al., 2009) vs. mass balance approach (Reichenberger et al., 2019)), we also calculated ∆P according to the mass balance approach for the 16 test simulations. Subsequently we i) compared ∆P of the 16 test runs with ∆P of the 10th percentile profiles from the new CDFs. ii) determined the position of the 16 test simulations on the corresponding spatial CDFs produced with the new approach (using the script get_CumAreapercent_ of_old_VFSMOD.R; the script is included in the digital annex (see section 6.11)) 2.5 Analysis with regard to soil texture In this report we aim to address the standing statement on the SWAN page at ESDAC (https://esdac.jrc.ec.europa.eu/projects/swan): “As the uncertainty whether the mitigation being modelled with VFSmod is achieved for all the other soil textures that the FOCUS scenarios cover still has to be independently evaluated, VFSmod has only rarely been accepted for use in EU level assessments and decision making.” Prior to the development of this report and to ensure that its scope addresses the comment, we engaged EFSA personnel to identify the best way to address this comment. The discussion revealed that this could be properly addressed by demonstrating “a best available estimate of a 90th spatial percentile of a target trapping effectiveness, for each complete area in each of the maps showing spatial area covered by each FOCUS runoff scenario” (EFSA expert comment, 07/08/2024). This fits very well with the main scope of this study: to derive VFS scenarios for SWANVFSMOD which, for each FOCUS R scenario, reflect an overall 90th percentile worst case (see section 2.3). In addition, to provide additional insights and the demonstrate the robustness of the selected scenarios, we complemented the analysis by segmenting the VFSMOD results also with respect to texture classes. Preliminary results (still without spatial weighting) were presented and discussed at the VFSMOD workshop in Piacenza on 03/09/2024 (https://symposiumpesticide.org/xvii-pesticide-symposiumpiacenza-workshop/). There it was agreed by the participants (including EFSA representatives) and included in the workshop resolutions that the final results of the analysis should be included in this report and an upcoming peer-reviewed publication. For this analysis, the following procedure was followed: • First, a SGDBE texture class (Figure 3; Le Bas et al., 1998; CEC, 1985) was assigned to each soil profile simulated with VFSMOD. Note that these are the same texture classes for which the HYPRES class pedotransfer functions (Wösten et al., 1998) have been defined. Profiles with an organic matter content greater than 30 % were classified as “organic” (O). • Subsequently descriptive statistics of ∆P were calculated for each combination of scenario, texture class, Koc and event duration). Spatial weighting was applied in the calculation of these statistics, i.e. each scenario / profile combination was weighted with its area. • Finally, the descriptive statistics were plotted as boxplots and compared with ∆P obtained for the 90th percentile worst case profiles (cf. section 2.3.3). In addition, “pooled” boxplots (i.e. boxplots combining all 4 R scenarios) were created and compared with the 10th percentile of ∆P of the combined CDF over all scenarios. Also these boxplots were spatially weighted. Apart from the graphical comparison, for each texture class also the cumulative area fraction corresponding to the ∆P of the 90th percentile worst case profile was determined using a binary search algorithm. knoell Germany GmbH Page 24 of 83 Project No.: 120333 Report No.: 120333-1 24 Figure 3 The SGDBE texture triangle. Source: https://www.researchgate.net/figure/Textureclasses-after-CEC-1985-Expressing-lateral-variability-BULLET-A-STU-canhave_fig2_237642627 knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 25 3 RESULTS AND DISCUSSION 3.1 Area covered by CDFs For the EU27, 48-61 % of the FOCUS scenario shapefile area are covered by the new spatial CDFs (Figure 4). These numbers are plausible because i) not all STUs have profiles, and ii) profiles usually cover only a part of the STU. There is no systematic difference in area coverage between the variants CORINE and noCORINE (cf. section 2.1). Figure 4 Area fractions of FOCUS scenario shapefiles (for the extent of the EU27) covered by the new spatial CDFs. There is still an open question whether CORINE land cover should be considered at all in the area calculation, i.e. whether an intersection with CLC class 2 (agricultural land) should be made or not. This issue is not discussed in Brown et al. (2012). The problem is that land use from the SGDBE (the dominant and secondary land use of each STU in the stu_sgdbe.dbf) has already been considered in the derivation of the FOCUS scenario shapefiles (FOCUS, 2001; Hollis, 2007): “Having identified those STU’s with soil characteristics representative of each Scenario, it was then necessary to ensure that each of the selected STU’s in the 1:1,000,000 scale Soil Geographic Database of Europe are also associated with at least some of the crops that characterise each scenario. This was also done through the STU attribute database of the Soil Geographic Database of Europe. In this database, each STU is characterised by two land use classes defining its ‘dominant’ and ‘secondary’ land use. The land use classes included in the Soil Geographic Database of Europe are defined in Table 3.4-1 and those used to identify the associated STU’s for each Scenario are shown in Table 3.4-3, together with the range of crops defined for each scenario (see section 3.3). The distribution of each Scenario within Europe was then mapped using the ArcView GIS software. Initially, the soil types corresponding to each scenario were selected by identifying all map units in the 1:1,000,000 scale Soil Geographic Database of Europe that contained an STU with attributes corresponding to those defined in Tables 3.4-2 and 3.4-3.” knoell Germany GmbH Page 32 of 83 Project No.: 120333 Report No.: 120333-1 32 Figure 9 Example CDF (FOCUS R4, intersection with CLC 2018 class 2, duration = 1 h, Koc = 10000 L kg-1, variant A (top) and variant B (bottom)). knoell Germany GmbH Page 33 of 83 Project No.: 120333 Report No.: 120333-1 33 Figure 10 Example CDF (FOCUS R4, intersection with CLC 2018 class 2, duration = 8 h, Koc = 100 L kg-1, variant A (top) and variant B (bottom). knoell Germany GmbH Page 34 of 83 Project No.: 120333 Report No.: 120333-1 34 Figure 11 Example CDF (FOCUS R4, intersection with CLC 2018 class 2, duration = 8 h, Koc = 10000 L kg-1, variant A (top) and variant B (bottom). An overview of the 90th percentile worst case profiles determined for the 64 CDFs is given in R1234_90th_perc_profiles_with_VFSMOD_params_noWT_20240610.xlsx in the digital annex (see section 6.11). The main results can be summarized as follows: • For all 32 combinations of scenario, CORINE/noCORINE, Koc and duration, the 90th percentile worst case profile is different between variant A and variant B. knoell Germany GmbH Page 35 of 83 Project No.: 120333 Report No.: 120333-1 35 • For variant A, in 20 out of 32 cases the 90th percentile worst-case profile was indeed one with depth to hard rock < 1 m. • For 1 h rainfall duration, deltaP (%) was only marginally different between variants A and B. • For 8 h rainfall duration, deltaP (%) was slightly or moderately different between variants A and B for R2-R4, with the largest difference being observed for R4. Differences were marginal for R1. 3.3 Final choice of scenario 64 CDFs of ∆P have been created. However, only four are needed. Finally, the combination Koc = 100 L kg-1 / duration = 8 h / variant B / CORINE has been selected for the new SWAN-VFSMOD scenarios to be included in SWAN 5.2, for the following reasons: • Koc = 100 L kg-1 is more relevant for most substances on the market than Koc = 10000 L kg-1 • duration = 8 h is closer to rainfall intensities in FOCUSsw than 1 h • variant B: classical Green-Ampt for all profiles - runs much faster than variant A (this becomes important when FOCUS repair becomes official and the evaluation period is increased to 20 years) - is easier to set up in SWAN than variant A - since very shallow soils have been removed, the obstacle to infiltration posed by hard rock horizons is only relevant for large rainfall/runoff events, which are rather infrequent (for instance, the return period of a 30 mm rainfall event in FOCUSsw is 1.3 years for R1 and 1.1 years for R3 and R4; only for R2 it is below 1 year) • Intersection with CORINE Land Cover (CLC) class 2: see discussion in section 3.1 The 90th percentile worst-case profiles and their parameters for the proposed combination are presented in Table 3. There are only four profile-specific VFSMOD parameters • vertical saturated conductivity VKS • saturated water content OS • field capacity water content FC • average suction at wetting front SAV Table 3 90th percentile worst-case profiles and profile-specific VFSMOD parameters for the proposed combination and the four R scenarios FOCUS R scenario profile and position on CDF profile-specific VFSMOD parameters Profile ID ∆P 1) cum. area fraction 1) VKS SAV OS FC % % m s-1 m m3 m-3 m3 m-3 R1 FR27 20.63 9.99 5.94E-07 0.078 0.384 0.341 R2 330549_1 56.32 10.00 1.88E-06 0.048 0.451 0.356 R3 330562_1 29.97 10.38 1.07E-06 0.064 0.428 0.352 R4 ES29 44.83 9.88 2.01E-06 0.015 0.413 0.349 1) for the following settings: use CORINE, Koc = 100 L kg-1, rainfall = 30 mm, event duration = 8 h, VL = 10 m, variant B (classical Green-Ampt) knoell Germany GmbH Page 36 of 83 Project No.: 120333 Report No.: 120333-1 36 3.4 Comparison of new CDFs with previous SWAN-VFSMOD scenarios / old VFSMOD version An overview of predicted trapping efficiencies for the example events (30 mm rainfall, VL = 10 m, duration 1 or 8 h) is given in Table 4. The main findings are: • For Koc = 100 L kg-1, the mass balance trapping equation always yielded substantially lower pesticide reduction efficiency (∆P) than the old Sabbagh equation (Sabbagh et al., 2009). • For Koc = 10000 L kg-1, the mass balance trapping equation yielded on average slightly higher pesticide reduction efficiency (∆P) than the old, now deprecated trapping equation of Sabbagh et al. (2009), largely due to R3. R3 is the R scenario with the highest topsoil clay content, and clay has a strongly negative effect on ∆P in the old Sabbagh equation. • For Koc = 100 L kg-1, the new 10th percentile profiles always yielded substantially lower (more conservative) ΔP than the old scenarios with the old VFSMOD version v.4.2.4 (mean difference of ΔP 39.6 % for CORINE and 36.5 for noCORINE) • For Koc = 10000 L kg-1, ΔP of the new 10th percentile profiles and ΔP of the old scenarios / VFSMOD version were on average the same (mean difference of ΔP 0.78 % for CORINE and 0.07 for noCORINE). In three out of eight cases the new profiles yielded higher ΔP than the old scenarios / VFSMOD version; however, as the ∆P obtained with the mass balance show, this was not due to the hydrology, but due to the old trapping equation of Sabbagh et al. (2009) used in SWAN 5. An overview of the positions (i.e. cumulative probabilities) of the old SWAN/VFSMOD scenario on the CDFs is given in Table 5. The main results can be summarized as follows: • When the trapping efficiencies for the 16 test runs were calculated with the old Sabbagh equation, positions of the test runs on the new CDFs of ∆P were always higher than 10 % (i.e. less conservative than the 90th percentile worst case), except for three cases where the Sabbagh equation yielded a cumulative probability of zero or almost zero, and two cases where the mass balance approach yielded a cumulative probability of 8.5 % (for noCORINE). • Positions on the CDF were considerably less variable with the mass balance trapping equation than with the old Sabbagh equation • CORINE vs. no CORINE had only a slight influence on the position on the CDF. It has to be noted that the results of this exercise cannot be completely extrapolated to the SWANVFSMOD tool, because in this simulation exercise we used realistic rainfall and runoff hydrographs generated with the UH tool (analogously to Brown et al., 2012). Results may be slightly different with the rectangular, low-intensity rainfall hydrographs from SWAN (as long as the daily rainfall volume does not exceed 48 mm, rainfall intensity in SWAN is only 2 mm/h due to the disaggregation method used for converting daily PRZM output to the hourly .p2t file). However, this analysis provided enough evidence to conclude that the new SWAN-VFSMOD scenarios are overall more conservative than the old ones. knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 37 Table 4 Pesticide reduction efficiencies (∆P) predicted with old SWAN-VFSMOD scenarios and old vfsm executable, and 10th percentile profiles of the new CDFs 1) created using the new vfsm executable FOCUS scenario pesticide reduction efficiency predicted with new 10th percentile profiles (variant B 2)) pesticide reduction efficiency predicted with old SWAN-VFSMOD scenarios and old vfsm executable difference (old VFSMOD with Sabbagh) - 10th percentile profile difference (old VFSMOD with mass balance) - 10th percentile profile event duration CORINE noCORINE Sabbagh eq. mass balance CORINE noCORINE CORINE noCORINE Koc ∆P ∆P ∆P ∆P difference (∆P) difference (∆P) difference (∆P) difference (∆P) L kg-1 h % % % % % % % % R1 10000 1 68.78 68.78 70.62 72.89 1.85 1.85 4.11 4.11 8 74.26 74.31 80.93 79.06 6.66 6.61 4.80 4.75 R2 1 79.07 79.55 76.17 83.19 -2.90 -3.38 4.13 3.64 8 90.03 94.82 100.00 93.66 9.97 5.18 3.64 -1.15 R3 1 69.72 69.80 52.91 73.68 -16.81 -16.88 3.96 3.88 8 77.94 78.18 65.52 81.32 -12.43 -12.66 3.38 3.14 R4 1 52.12 52.12 64.69 59.55 12.57 12.57 7.44 7.44 8 70.82 70.82 78.13 72.46 7.30 7.30 1.64 1.64 R1 100 1 9.15 9.15 62.35 15.18 53.20 53.20 6.04 6.04 8 20.63 20.80 72.65 32.71 52.02 51.85 12.09 11.92 R2 1 21.97 23.93 68.68 26.37 46.71 44.75 4.40 2.45 8 56.32 77.68 94.26 71.47 37.95 16.58 15.16 -6.21 R3 1 11.56 11.80 44.69 17.70 33.13 32.89 6.14 5.90 8 29.97 30.78 57.30 38.69 27.33 26.51 8.72 7.91 R4 1 12.75 12.75 55.11 23.67 42.36 42.36 10.92 10.92 8 44.83 44.83 68.54 47.58 23.71 23.71 2.75 2.75 1) For a 30 mm rainfall event and VL = 10 m 2) Variant B: classical Green-Ampt infiltration for all profiles knoell Germany GmbH Page 38 of 83 Project No.: 120333 Report No.: 120333-1 38 Table 5 Positions of old SWAN/VFSMOD scenarios (with old VFSMOD executable) on new CDFs 1) Results of old SWAN/VFSMOD scenarios with old VFSMOD exe Positions of old scenarios / exe on new CDFs (variant B) FOCUS scenario relative reduction of total inflow relative reduction of eroded sed. load relative reduction of incoming runoff relative reduction of pesticide load with Sabbagh et al. (2009) with mass balance approach Sabbagh eq. mass balance CORINE noCORINE CORINE noCORINE Koc duration ∆Q 2) ∆E 2) ∆R 2) ∆P ∆P cum. area fraction cum. area fraction cum. area fraction cum. area fraction L kg-1 h % % % % % % % % % R1 10000 1 13.23 98.30 -25.40 70.62 72.89 43.77 40.37 76.40 73.96 8 31.15 99.51 3.99 80.93 79.06 44.25 40.39 37.46 34.94 R2 1 23.68 99.30 -18.11 76.17 83.19 0.02 0.01 22.85 16.51 8 70.42 99.95 54.17 100.00 93.66 98.80 98.73 10.51 8.50 R3 1 15.77 97.50 -22.90 52.91 73.68 0.00 0.00 56.38 54.13 8 37.22 99.47 12.05 65.52 81.32 0.00 0.01 22.82 21.29 R4 1 22.97 99.07 -7.52 64.69 59.55 68.13 67.75 50.71 47.85 8 47.10 99.85 26.03 78.13 72.46 20.30 20.46 11.91 12.50 R1 100 1 13.23 98.30 -25.40 62.35 15.18 99.93 99.94 44.98 41.35 8 31.15 99.51 3.99 72.65 32.71 98.40 98.13 35.36 32.53 R2 1 23.68 99.30 -18.11 68.68 26.37 71.88 65.72 16.24 12.19 8 70.42 99.95 54.17 94.26 71.47 23.20 16.71 10.51 8.50 R3 1 15.77 97.50 -22.90 44.69 17.70 89.68 89.79 33.23 33.31 8 37.22 99.47 12.05 57.30 38.69 57.26 55.28 20.76 18.52 R4 1 22.97 99.07 -7.52 55.11 23.67 91.30 90.65 38.82 36.18 8 47.10 99.85 26.03 68.54 47.58 32.35 30.55 11.91 12.50 1) For a 30 mm rainfall event and VL = 10 m 2) ∆Q, ∆E and ∆R for the old SWAN/VFSMOD scenarios are only given as additional information to ease the interpretation of predicted ∆P. knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 39 3.5 Analysis with regard to soil texture In the following, only the comparison for the combination of factors is discussed from which the proposed scenarios for SWAN-VFSMOD were selected: Koc 100 L kg-1, duration 8 h, CORINE, variant B (classical Green-Ampt). However, the whole set of boxplots and corresponding tables (.xlsx) can be found in the digital annex. Results for the individual scenarios are shown in Figure 12 and Table 6, while the “pooled” results (all scenarios combined) are shown in Figure 13 and Table 7. From the individual FOCUS scenarios, the comparison of the spatially weighted boxplots of ∆P for the five SGDBE (or HYPRES) texture classes with the 10th percentile of ∆P (i.e. the 90th percentile worst case ∆P) of the whole CDF (all texture classes), yielded that for most combinations of scenario and texture class the ∆P for the 90th percentile worst case profiles are conservative (Table 6). Only for R1 / VF (very fine; > 60 % clay) and R2 / MF (medium fine; silt) the class median of ∆P was lower than the 10th spatial percentile of ∆P for the whole CDF (Figure 12). However, these texture classes occur only rarely (n = 11 and 7, respectively) and cover only a small area fraction of the respective scenario. The number of organic soils (class O) was too small for any interpretation. Note that since both the boxplots and the horizontal lines come from the same statistical population (the spatial CDFs of ∆P for all simulated soils belonging to the R scenario), by definition there must be data points below the red horizonal line which correspond to 10 % of the area of the CDF. The “intersection point” of the red line and the boxplots, i.e. the positions of the 10th percentile ∆P for the whole CDF on the CDFs for the individual texture classes is given in Table 6. For several combinations of scenario and texture class the cumulative area percentage of the intersection point was greater than 10 %. However, this is normal, because some texture classes have higher vulnerability than others. It is virtually impossible that an overall 90th percentile worst case over all soils (the red line in the boxplots) could constitute a 90th percentile worst case also for each texture class. For some texture classes, the cumulative area percentage will be higher than 10 %, and for others lower (or even 0). While it would be possible to shift the red line downwards to obtain a ≥ 90th percentile worst case also for the most vulnerable texture class, this would result in an extreme overall worst case. For instance, for R1 / class VF the 10th percentile ∆P is 15.154 %. For the whole CDF for R1 (all texture classes together) this ∆P value corresponds to a cumulative area percentage of 0.56 %. We would thus have an overall 99.44th percentile worst case. Results for the “pooled” boxplots (boxplots combining all 4 R scenarios) were similar to those for the individual scenarios. Only for the relative rare and extreme texture classes “medium fine” and “very fine” the class median of ∆P was (slightly and moderately, resp.) lower than the 10th spatial percentile of ∆P for the whole CDF (Figure 13). The cumulative area percentages of the intersection point were 54.7 % for “medium fine” and 73.3 % for “very fine” (Table 7). For the other texture classes they were close to or exactly zero. If we shift the red line downwards we observe similar patterns as for the CDFs per scenario: For instance, for class MF (all scenarios together) the 10th percentile ∆P is 19.842 %. For the whole CDF (all scenarios and texture classes together) this ∆P value corresponds to a cumulative area percentage of 2.66 %, i.e. an overall 97.34th percentile worst case. In summary, this analysis confirms that the approach of selecting overall 90th percentile worst case soil profiles for the VFS scenarios is reasonable. Some texture classes have higher vulnerability than others. The general ranking of ∆P (C > M > F > MF > VF) reflects the hydraulic parameter sets obtained with the HYPRES pedotransfer functions; notably predicted saturated conductivity is on average lowest for soils with texture classes MF and VF. knoell Germany GmbH Page 40 of 83 Project No.: 120333 Report No.: 120333-1 40 Figure 12 Spatially weighted boxplots of predicted pesticide trapping efficiency (∆P) according to texture class for all simulated profiles for each scenario (Koc 100 L kg-1, duration 8 h, CORINE, variant B). Boxes: 25th and 75th percentiles, whiskers: min and max. Spatial weighting with with area of the scenario/profile combination was applied. Horizontal line: 10th percentile ∆P (90th percentile worst case) of whole CDF (all texture classes) for option CORINE. Texture classes are C = Coarse, M = medium, MF = medium fine, F = fine, VF = very fine, O = organic. The sample size is given below each box. knoell Germany GmbH Page 41 of 83 Project No.: 120333 Report No.: 120333-1 41 Table 6 Cumulative area fractions for each texture class corresponding to the 10th percentile pesticide reduction efficiency (∆P) of the whole CDF (all texture classes) for the combination selected for the SWAN-VFSMOD scenarios (Koc 100 L kg-1, duration 8 h, CORINE, variant B) FOCUS Scenari o 10th percentile ∆P of whole CDF (all texture classes) HYPRES texture class sample size N closest ∆P in CDF of ∆P for texture class ∆P cumulative area cumulative area fraction of CDF for texture class area fraction of whole CDF 3) (all texture classes) % % ha ha ha-1 ha ha-1 R1 20.625 C 56 53.766 14234.2 0.050 1) 0.0016 M 346 23.809 35036.8 0.009 0.0041 MF 100 20.625 611345.8 0.249 2) 0.0708 F 178 20.796 62812.1 0.038 0.0073 VF 11 18.714 233915.7 0.970 0.0271 R2 56.315 C 83 59.363 3210.5 0.007 0.0042 M 64 56.315 43646.7 0.210 0.0569 MF 7 44.582 27634.5 1.000 0.0360 F 18 42.938 5421.3 0.140 0.0071 R3 29.971 C 75 61.776 60.8 0.000 0.0000 M 225 30.438 31933.9 0.025 0.0143 MF 27 29.971 160401.3 0.550 0.0718 F 84 30.071 29395.1 0.083 0.0132 VF 5 16.629 2627.2 0.130 0.0012 O 3 27.868 17393.3 0.998 0.0078 R4 44.834 C 93 47.632 1241.0 0.001 0.0001 M 286 44.834 660297.4 0.086 0.0550 MF 15 45.493 342763.9 0.471 0.0286 F 97 43.967 207478.1 0.095 0.0173 VF 4 62.165 62111.3 0.988 0.0052 O 3 28.698 96681.6 0.998 0.0081 1) The binary search algorithm always finds a solution: If ∆P from the 10th percentile profile is below the whole distribution of ∆P for a given texture class (cf. Figure 12), the first row of the CDF (the row with the lowest ∆P) is selected. The determined cumulative area fraction then corresponds to the area fraction of this row. However, the true cumulative area fraction for these cases (highlighted in red) is zero. 2) Cumulative area fractions > 10 % are highlighted in bold. 3) Calculated as the ratio of the value in column “cumulative area” and the total area of the whole CDF (all texture classes). The values in this column should add up to 0.1 for each scenario; however, this is only approximately the case (the sums are between 0.104 for R2 and 0.114 for R4). The reason is that the binary search algorithm does not interpolate, but only finds the ∆P value closest to the target value . knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 48 6 ANNEX 6.1 Resources Datasets: Datasets used in this project are available for download at: • SGDBE: https://esdac.jrc.ec.europa.eu/resource-type/european-soil-database-soil-properties • SPADE2: https://esdac.jrc.ec.europa.eu/content/spade-2-soil-profile-analytical-database-europeversion-10 • SPADE14: https://esdac.jrc.ec.europa.eu/content/spade-14 Models: The VFSMOD version used in this project (v.4.5.2) as well as UH v.3.0.7 are available for download at https://abe.ufl.edu/faculty/carpena/vfsmod/updates.shtml knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 49 6.2 Spatial extent of the FOCUS R scenarios Figure A 1 Extent of the FOCUS R1 scenario shapefile (FOCUS, 2001) after intersection with CORINE Land Cover 2018 class 2 (agricultural land) knoell Germany GmbH Page 50 of 83 Project No.: 120333 Report No.: 120333-1 50 Figure A 2 Extent of the FOCUS R2 scenario shapefile (FOCUS, 2001) after intersection with CORINE Land Cover 2018 class 2 (agricultural land) knoell Germany GmbH Page 51 of 83 Project No.: 120333 Report No.: 120333-1 51 Figure A 3 Extent of the FOCUS R3 scenario shapefile (FOCUS, 2001) after intersection with CORINE Land Cover 2018 class 2 (agricultural land) knoell Germany GmbH Page 52 of 83 Project No.: 120333 Report No.: 120333-1 52 Figure A 4 Extent of the FOCUS R4 scenario shapefile (FOCUS, 2001) after intersection with CORINE Land Cover 2018 class 2 (agricultural land) 6.3 Description of tables from SGDBE The SGDBE is the geometry part of the European Soil Database v2.0 (EC, 2004) and can be obtained here: https://esdac.jrc.ec.europa.eu/content/european-soil-database-v20-vector-and-attribute-data Table A 1 Variables contained in stuorg.dbf variable description unit key or attribute remarks SMU Soil Mapping Unit n.a. part of key STU Soil Typological Unit n.a. part of key STUs are not localized within the SMU; SMU and STU have a m:n relationship: a given SMU can have several STUs, and a given STU can occur in several SMUs PCAREA area fraction of the SMU covered by the STU % attribute knoell Germany GmbH Page 53 of 83 Project No.: 120333 Report No.: 120333-1 53 Table A 2 Variables contained in stu_sgdbe.dbf (only the relevant ones are listed) Variable description unit key or attribute remarks STU Soil Typological Unit n.a. key USE_DOM dominant land use in STU n.a. attribute numerical code (Table A 13) USE_SEC secondary land use in STU n.a. attribute numerical code (Table A 13) 6.4 Description of SPADE2 tables SPADE2 consists only of a single table containing profile and horizon data (spade2v11.dbf). Details are given in Hannam et al. (2009). The variables contained in spadev211.dbf are listed in Table A 3. Table A 3 Variables contained in spadev211.dbf (Hannam et al., 2009) variable description unit key or attribute remarks SMU Soil Mapping Unit code n.a. part of key STU Soil Typological Unit code n.a. part of key COUNTRY Country ID n.a. part of key USE Land use class associated with profile data n.a. part of key numerical code (Table A 13) SOIL Soil name FAO Legend 1974 modified CEC 1985 n.a. attribute PCAREA area fraction of the SMU covered by the STU % attribute STUarea/SMUarea * 100 % HORIZON horizon designation according to FAO Nomenclature n.a. part of key DEPTH_UP Upper limit of horizon cm attribute DEPTH_LO Lower limit of horizon cm attribute CLAY Clay <0.002 mm esd, % oven dry weight (at 105°C) % attribute SILT Silt 0.002-0.05 mm esd, % oven dry weight (at 105°C) % attribute SAND_TOT Sand 0.05-2 mm esd, % oven dry weight (at 105°C) % attribute SAND_01 Sand 0.05-0.1 mm esd, % oven dry weight (at 105°C) % attribute SAND_02 Sand 0.1-0.2 mm esd, % oven dry weight (at 105°C) % attribute SAND_05 Sand 0.2-0.5 mm esd, % oven dry weight (at 105°C) % attribute SAND_20 Sand 0.5-2 mm esd, % oven dry weight (at 105°C) % attribute knoell Germany GmbH Page 54 of 83 Project No.: 120333 Report No.: 120333-1 54 variable description unit key or attribute remarks STONES Stone content as volume % % attribute PH_KCL pH in 0.1 M KCl solution n.a. attribute PH_KCLSD standard deviation of pH in KCl n.a. attribute PH_H2O pH in H2O solution (soil: water ratio 1:2.5) n.a. attribute PH_H2OSD standard deviation of pH in H2O n.a. attribute PH_CA pH in 0.01M CaCl2 solution n.a. attribute PH_CASD Standard Deviation of pH in CaCl2 n.a. attribute OC Organic Carbon content % attribute OC_SD Organic Carbon standard deviation % attribute DB Bulk Density measured kg dm-3 attribute DB_CALC Bulk Density calculated kg dm-3 attribute DB_UK_PTF Bulk density derived from UK pedotransfer function kg dm-3 attribute TEXT1 Dominant surface textural class n.a. attribute TEXT2 Secondary surface textural class n.a. attribute WR Dominant annual average water regime class n.a. attribute WM1 Water management in agricultural land n.a. attribute WM2 Purpose of water management system n.a. attribute WM3 Type of water management syste n.a. attribute SOIL_NEW New soil type assigned to STU from data provider n.a. attribute USE_NEW New land use class assigned to STU from data provider n.a. attribute numerical code (Table A 13) knoell Germany GmbH Page 55 of 83 Project No.: 120333 Report No.: 120333-1 55 Table A 4 Variables contained in cleaned-up profile/horizon table gethorizons_SPADE2_all_20240304.xlsx variable description unit key or attribute remarks FOCUS_scen FOCUS scenario n.a. part of key maxdepth_cm maximum depth for averaging 1) over horizons cm attribute set to 100 ProfileID2 SPADE2 profile ID created by knoell n.a. part of key created as concatenation of STU and USE FOCUSscen_PR ofileID2 concatenation of FOCUS_scen and ProfileID2 n.a. attribute SOIL Soil name FAO Legend 1974 modified CEC 1985 n.a. attribute USE Land use class associated with profile data n.a. part of key numerical code (Table A 13) HORIZON horizon designation according to FAO Nomenclature n.a. part of key isRock binary classifier (1 = is hard rock, 0 = otherwise) n.a. attribute auxiliary variable startRock binary classifier ( 1 = is uppermost hard rock layer; 0 = otherwise) attribute auxiliary variable startdepth_rock start depth of hard rock cm attribute auxiliary variable maxdepth_final final depth for averaging over horizons cm attribute calculated as min (startdepth_rock, 100 cm) DEPTH_UP Upper depth of horizon cm attribute DEPTH_LO Lower depth of horizon cm attribute thickness thickness of horizon cm attribute min(depth_lo, maxdepth_final) minimum of DEPTH_LO and maxdepth_final cm attribute auxiliary variable thickness_above _maxdepth_final thickness of horizon above maxdepth_final cm attribute weight for weighted averaging CLAYnorm normalized clay content attribute normalized so that the sum of the three yields 100 % SILTnorm normalized silt content attribute SANDnorm normalized sand content attribute STONES volumetric stone content % attribute OC organic carbon content attribute DB bulk density measured kg dm-3 attribute DBNEW final bulk density kg dm-3 attribute If DB is present and is NOT flagged as an outlier, DBNEW = DB. If DB is not present, then DBNEW = DBPTF 2). If DB is present but it is an outlier, than knoell Germany GmbH Page 56 of 83 Project No.: 120333 Report No.: 120333-1 56 variable description unit key or attribute remarks DBNEW = DBPTF is used, only if DBPTF is not an outlier; else DBNEW = DB. For all layers where HORIZON starts with R DBNEW was set to 2.4 QA data quality level attribute 1 = complete observations and direct application of HYPRES continuous ptfs, textural data sum-up to 100% and textural classes meet the data range restriction and observed DB is not an outlier; 2 = same as 1 but textural data were normalized to sum-up to 100%; 3 = same as 1, but missing DB values, HYPRES continuous ptfs were applied after DB prediction using the ptfs of Hollis et al. (2012); 4 = same as 3, but textural data were normalized to sum up to 100%; 5 = others than {1,…,4, 6, NA}; 6 = rock or organic layer; NA = application of ptfs not possible due to data incompleteness missing_depth_R checking column attribute always FALSE 1) depth-weighted averaging in order to create an effective single-horizon profile for VFSMOD 2) DBPTF = bulk density calculated using pedotransfer functions of Hollis et al. (2012) 6.5 Description of SPADE14 tables A historic overview on the development of a harmonised soil profile analytical database for Europe is given in Kristensen et al. (2019). Further information on Spade-1.4 are given in Breuning-Madsen and Jones (1995) and Breuning-Madsen et al. (2015). The Spade-1.4 database consists of a directory structure that are named with the 2-letter ISO code for 28 European States. In each directory, two Excel files are present, e.g. XX-EST_HOR.xls and XX-EST_PROF.xls. All 28 XX-EST_HOR.xls files were row-wisely appended to form one large HOR_all.csv file. The same was done to form one large PROF_all.csv. After that, HOR_all.csv and PROF_all.csv were joined via the unique field PROF_NUM. SPADE 14 also contains a table linking the profiles to the Soil Typological Units (Links_prof_stu.xlsx). We renamed this table to spade14_stu.xlsx (Table A 5) to avoid confusion with other datasets. knoell Germany GmbH Page 57 of 83 Project No.: 120333 Report No.: 120333-1 57 Table A 5 Variables contained in spade14_stu.xlsx variable description unit key or attribute remarks STU_DOM Soil Typological unit n.a. part of key SPADE14 only contains profiles for dominant STUs (i.e. the STUs covering the largest area fraction in a given SMU) CNTR_STU concatenation of country and STU_DOM n.a. redundant because country code already contained in PROF_NUM PROF_NUM Profile ID n.a. part of key country code + number country country code n.a. redundant because country code already contained in PROF_NUM Table A 6 Variables contained in prof_all (Daroussin, 1999) variable description unit key or attribute remarks PROF_NUM profile id n.a. key COUNTRY country code n.a. attribute redundant because country code already contained in PROF_NUM) SOIL full 1974 (modified cec 1985) FAO legend soil name n.a. attribute SOIL90 soil type according to FAO 1990 n.a. attribute TEXT1 dominant surface textural class n.a. attribute TEXT2 secondary surface textural class n.a. attribute PM dominant parent material n.a. attribute semantic description (free text) SMU soil mapping unit identifier to which the profile refers (as given by author) n.a. attribute STU soil typogical unit identifier to which the profile refers (as given by author) n.a. attribute LU dominant land use n.a. attribute semantic description (free text) GWL_HI class of mean highest level of a permanent or perched groundwater table n.a. attribute GWL_LO class of mean lowest level of a permanent or perched groundwater table n.a. attribute DR_EFF_A mean effective root depth for a??? cm attribute knoell Germany GmbH Page 64 of 83 Project No.: 120333 Report No.: 120333-1 64 Table A 9 Columns of table stuorg-stu_sgdbe-spade14-spade2-R1234.xlsx No. variable description source value 1 SMU Soil Mapping Unit stuorg 2 STU Soil Typological Unit stuorg 3 PCAREA.x area percentage of SMU covered by STU stuorg 4 USE_DOM dominant land use of STU stu_sgdbe 5 USE_SEC secondary land use of STU stu_sgdbe 6 SRC source of columns 1-3 „stuorg” 7 USE land use of soil profile spade2 or spade14 8 COUNTRY country of soil profile 9 PROFILE profile ID 10 SRC.x source of columns 7-9 „spade14“ or „spade2“ 11 SMU.y Soil Mapping Unit R1234_NUTS 12 CNTR_CODE country code R1234_NUTS 13 NUTS_NAME country name R1234_NUTS 14 EU_STAT EU membership status R1234_NUTS “T” or “F” 15 FOCUS_scen FOCUS scenario R1234_NUTS 16 COUNTRY_SPADE14 country code SPADE14 R1234_NUTS 17 COUNTRY_SPADE2 country code SPADE2 R1234_NUTS 18 TotalArea_SMU_CNTR_ FOCUSscen Total area of combination SMU / country / FOCUS scenario R1234_NUTS 19 stuorg_SMU Soil Mapping Unit R1234_NUTS 20 STU.y Soil Typological unit R1234_NUTS 21 PCAREA.y area fraction of the SMU that is covered by the STU (%) R1234_NUTS 22 SRC.y source of columns 11-21 “R1234_NUTS” or blank knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 65 6.7 Description of spatial aggregation procedure 6.7.1 Spatial input tables The following tables were used for the creation of spatial cumulative distribution functions (CDFs) (section 6.7.4): • R1234_grouped_20240411.xlsx (657 records) • R1234_CORINE_grouped_20240411.xlsx (637 records) These tables contain the total area of each unique combination of country, FOCUS scenario and Soil Mapping Unit. Variables are given in Table A 10. For the CDF creation .csv versions of these tables were used. Table A 10 Variables contained in R1234_grouped_20240411.xlsx and R1234_CORINE_grouped_20240411.xlsx field description unit key or attribute remarks CNTR_ID NUTS country code part of key NAME_ENGL NUTS country name attribute (in national language) EU_STAT EU membership status attribute T (TRUE) or F (FALSE) SMU Soil Mapping Unit part of key FOCUS_scen FOCUS scenario part of key R1, R2, R3 or R4 SumΟfShape _Area area of unique combination of country / SMU / FOCUS scenario m2 attribute The following tables were used as inputs for calculating the area fractions covered by each soil profile (section 6.7.3): • R1234_NUTS_agg_leftjoin_stuorg.xlsx (2560 records; variables see Table A 11) • R1234_NUTS_CLC_agg_leftjoin_stuorg.xlsx (13293 records; variables see Table A 12) The creation of these two tables is described in section 2.1. knoell Germany GmbH Page 66 of 83 Project No.: 120333 Report No.: 120333-1 66 Table A 11 Variables contained in R1234_NUTS_agg_leftjoin_stuorg.xlsx field description unit remarks SMU Soil Mapping Unit CNTR_CODE NUTS country code NUTS_NAME NUTS country name (in national language) EU_STAT EU membership status T (TRUE) or F (FALSE) FOCUS_scen FOCUS scenario R1, R2, R3 or R4 COUNTRY_SPADE14 country code used in SPADE14 COUNTRY_SPADE2 country code used in SPADE2 TotalArea_SMU_CNTR_FOC USscen area of unique combination of country / SMU / FOCUS scenario m2 stuorg_SMU SMU from stuorg table from stuorg.dbf STU Soil Typological Unit from stuorg.dbf PCAREA area fraction of the SMU covered by the STU % from stuorg.dbf Table A 12 Variables contained in R1234_NUTS_CLC_agg_leftjoin_stuorg.xlsx field description unit remarks SMU Soil Mapping Unit CNTR_CODE NUTS country code NUTS_NAME NUTS country name (in national language) EU_STAT EU membership status T (TRUE) or F (FALSE) Code_18 CORINE Land Cover class 211, 212, 213 etc. FOCUS_scen FOCUS scenario R1, R2, R3 or R4 COUNTRY_SPADE14 country code used in SPADE14 COUNTRY_SPADE2 country code used in SPADE2 TotalArea_SMU_CNTR_CLC FOCUSscen area of unique combination of country / SMU / CLC class / FOCUS scenario m2 stuorg_SMU SMU from stuorg table from stuorg.dbf STU Soil Typological Unit from stuorg.dbf PCAREA area fraction of the SMU covered by the STU % from stuorg.dbf knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 67 6.7.2 Recoding of land use for SPADE14 The semantic land use descriptions in the SPADE14 table PROF_all were recoded into Land Use codes according to SPADE2 (Hannam et al. 2009; Table A 13) using R. Table A 13 Land use codes used in SPADE2 (Hannam et al., 2009) Land use code land use description 0 No information 1 Pasture, grassland, grazing land 2 Poplars 3 Arable land, cereals 4 Wasteland, shrub 5 Forest, coppice 6 Horticulture 7 Vineyards 8 Garrigue 9 Bush, macchia 10 Moor 11 Halophile grassland 12 Arboriculture, orchard 13 Industrial crops 14 Rice 15 Cotton 16 Vegetables 17 Olive trees 18 Recreation 19 Extensive pasture, grazing, rough pasture 20 Dehesa (extensive pastoral system in forest parks in Spain) 21 Cultivos enarenados (artificial soils for orchards in SE Spain) 22 Wildlife refuge, land above timberline The classification of the verbatim SPADE14 land use descriptions into SPADE2 numeric land use codes was done sequentially. First, the verbatim SPADE14 land use descriptions were set to lower case in order to ease the matching with regular expressions. Subsequently, a series of rules (Table A 14) was applied sequentially to the land use descriptions using regular expressions. The code is available in the digital annex (recodeLUspade14.R). knoell Germany GmbH Page 68 of 83 Project No.: 120333 Report No.: 120333-1 68 Table A 14 Classification rules applied to SPADE14 land use descriptions Step Pattern in SPADE14 land use description Attributed SPADE2 LU Code 1 Contains “olive” 7 2 Contains “rice” 14 3 Contains “cotton” 15 4 Contains “olive” 7 5 Whole and only word “orchard” 12 6 Contains “grass”, “graz”, “pasture”, “meadow” or “heath” AND does NOT contain “forest” or “arable” or starts with “agric” or starts with “bush” 1 6.a Must contain SPADE2 LU Code 1 from step 6 AND contain “exten”, “rough” or “natural” 19 6.b Must contain SPADE2 LU Code 1 from step 6 AND contain “halophile” 11 7 Contains “marsh” 11 8 Starts with “arable” or contains “cereals” or “crop” AND dos not contain “forest” 3 9 Contains “bush”, "wast”, “shrub”, “schrub”, “scrub”, or “layland” or starts with “non” 4 9.a. Must contain SPADE2 LU Code 5 from step 9 AND contain “recreat” 18 10. Contains “recreat” 18 11. Contains ‘forest”, “pinus” or “wood” AND does not contain “arable”, “agric”, “grass”, “cereals”, “wild” 5 12 Contains “vegetab” 16 13 Contains “horti” 6 14 Contains “moor”, “swamp” or “peat” 10 15 Contains “wildlife” 22 16 Contains “garrigue” 8 17 1) Starts with “agric”, “arable”, “cereals” or “field” 3 18 1) Starts with “orchard” 12 19 1) Starts with “forest” 5 20 1) Starts with “grass” 1 21 All remaining after step 20 0 1) Steps 17-20 address mixed land use classes in SPADE14 such as e.g. “grassland / arable land”. It is assumed that the first term is the dominant land use of the corresponding spatial unit. knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 69 6.7.3 Calculation of area fractions for the different profiles This is a documentation of the final version of the script (import_extended.R). 1) set flag isCLC (TRUE or FALSE) 2) Read in input tables from SGDBE: stu_sgdbe.dbf, stuorg.dbf 3) stuorg: scale sum to STU shares to 100 % for each SMU 4) add dominant and secondary land use types per STU to stuorg (left join of stu_sgdbe to stuorg) 5) Replace USE_DOM = 0 and USE_SEC = 0 with NA 6) import SPADE2 database; 7) import SPADE14 table linking STUs to profiles (spade14_stu.csv); create country code (first two characters of profileID) 8) import table with polygon areas as “R1234_NUTS” R1234_NUTS_CLC_agg_leftjoin_stuorg.xlsx (for case “isCLC = TRUE”) R1234_NUTS_agg_leftjoin_stuorg.xlsx (for case “isCLC = FALSE”) 9) import table with recoded land uses for SPADE14: join land use code to SPADE14 via ProfileID 10) SPADE14: generate combined field STU_PROFID 11) add a field SRC (source) to spade14, spade2, stuorg and R1234_NUTS tables for later identification of columns 12) R1234_NUTS: Create columns SPADE1KEY and SPADE2KEY to enable join with spade14 and spade2 tables 13) left join of R1234_NUTS to spade14 → table spade1R1234_STUCNTR 14) left join of spade1R1234_STUCNTR to stuorg; select only unique combinations → table stuorg_spade14_R1234 15) SPADE2: apply updated land use (in additional column USE_NEW) to column USE where available; Replace -9999 values (No data) in columne USE with 0 16) SPADE2: create ProfileID as concatenation of STU and USE 17) SPADE2: Select only unique combinations of "SMU", "STU", "USE", "COUNTRY", "PROFILE", "SRC") 18) SPADE2 create column SPADE2KEY as concatenation of SMU, STU and country 19) left joint of R1234 to spade2; select only unique combinations → table spade2R1234 20) select only complete cases of stuorg_spade14_R1234 and stuorg_spade2_R1234 21) Remove columns not needed anymore from stuorg_spade14_R1234 and stuorg_spade2_R1234 22) check identity of field names of both tables; order columns so that both tables have the same order; append stuorg_spade14_R1234 and stuorg_spade2_R1234 → table “link”; save as tuorgstu_sgdbe-spade14-spade2-R1234.xlsx 23) link: set all remaining NA values in USE_DOM and USE_SEC to 0 24) link: add area percentage covered by dominant land use (USE_DOM_PERC_AREA): IF there is no secondary land use, set percentage to 80 ELSE IF no dominant land use either: set percentage to NA ELSE (secondary land use present): set percentage to 60 25) link: add area percentage covered by secondary land use (USE_SEC_PERC_AREA): IF there is no secondary land use, set percentage to NA ELSE (secondary land use present): set percentage to 30 26) link: add area percentage of actual land use of soil profile (USE_PERC_AREA) IF USE = USE_DOM and USE_DOM > 0, set percentage to USE_DOM_PERC_AREA ELSE IF = USE_SEC and USE_SEC > 0, set percentage to USE_SEC_PERC_AREA ELSE (use corresponds neither to dominant nor to secondary land use), set percentage to 10 27) link: some consistency checks; remove STUs without valid profiles 28) link: add new field as concatenation of SMU and STU knoell Germany GmbH Page 70 of 83 Project No.: 120333 Report No.: 120333-1 70 29) link: calculate profile area fractions (PROFILE_PERC_AREA) for each SMU_STU (a SMU / STU combination can have more than one profile, and they can have different or the same land use); the fact that due to the columns FOCUS_scen, country (for SPADE2) and 30) CODE_18 (for option CORINE) there are multiple records with the same SMU / STU / profile combination is also dealt with here so that the sums of the area fractions are correct 31) link: check whether sum of PROFILE_PERC_AREA for each SMU_STU combination is <= 100 32) link: calculate variable “fractions” as fractions = STUSHARE / 100 * PROFILE_PERC_AREA / 100; export table link.csv for CDF generation (with prefix CLC_ or noCLC_) 33) calculate final area fraction of SMU covered by a given profile; export results as smu_prof_fracs_tpt.csv (with prefix CLC_ or noCLC_) for quality assurance knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 71 Table A 15 Variables contained in of CLC_link_table.csv or noCLC_link_table.csv No. field description unit remarks 1 SMU Soil Mapping Unit 2 STU Soil Typological Unit 3 STUSHARE area percentage of SMU covered by STU % 4 USE_DOM area percentage of SMU covered by STU 5 USE_SEC dominant land use of STU 6 SRC secondary land use of STU 7 USE source of columns 1-3 8 COUNTRY land use of soil profile 9 PROFILE profile ID 10 SRC.x source of columns 7-9 „spade14“ or „spade2“ 11 SMU.y Soil Mapping Unit 12 CNTR_CODE country code 13 NUTS_NAME country name 14 EU_STAT EU membership status 15 Code_18 code of CORINE Land Cover class (e.g. 211) only in CLC_ link_table.csv 16 FOCUS_scen FOCUS scenario 17 COUNTRY_SPADE14 country code SPADE14 18 COUNTRY_SPADE2 country code SPADE2 19 TotalArea_SMU_CNTR_ CLC_FOCUSscen Total area of combination SMU / country / FOCUS scenario m2 20 stuorg_SMU Soil Mapping Unit 21 STU.y Soil Typological unit 22 PCAREA area fraction of SMU covered by STU % 23 SRC.y source of columns 11-22 24 USE_DOM_PERC_AREA area fraction of SMU/STU combination covered by dominant land use % 25 USE_SEC_PERC_AREA area fraction of SMU/STU combination covered by secondary land use % 26 USE_PERC_AREA area fraction of SMU/STU combination covered by the current land use (field USE) % 27 SMUSTU Concatenated string of SMU and STU 28 PROFILE_PERC_AREA area fraction of SMU/STU combination covered by the current soil profile % 29 fractions area fraction of SMU covered by the current soil profile m2 m-2 30 SMUPROFILE Concatenated string of SMU and PROFILE knoell Germany GmbH Page 72 of 83 Project No.: 120333 Report No.: 120333-1 72 6.7.4 Calculation of spatial CDFs Spatial cumulative distribution functions of the target VFSMOD output variable ∆P (relative reduction of incoming pesticide load by the VFS) were calculated in R. The script createSpatialCDF.R can be found in the digital annex. Input tables: • Table of area fractions covered by the different profiles: CLC_link_table.csv or noCLC_link_table.csv (see Table A 15) • VFSMOD results: R1234_30mm_noWT_out_20240526.xlsx (Variant A) or R1234_30mm_noWT_out_20240610.xlsx (Variant B) (see Table A 16) • Table with areas of unique combination of country / SMU / FOCUS scenario: R1234_CORINE_grouped_20240411.csv or R1234_grouped_20240411_ohne_CNT_NAMES.csv (.csv versions of the Excel files documented in Table A 10) Output: • One CSV file per combination of FOCUS scenarios (R1, R2, .., R4), CORINE or non-CORINE, KOC, DUR and timestep (e.g. R1_KOC100_DUR1_TS15.csv, see Table A 19) • One plot per above mentioned combination with x = deltaP (%) and y = cumulative area percentage (%) • One single file *_sourceForCumArea.calc.csv for QA reasons. The workflow in createSpatialCDF.R was as follows: 1) set flag is.CORINE (TRUE or FALSE) 2) import table with area fractions (CLC_link_table.csv or noCLC_link_table.csv) as “link” 3) import VFSMOD results as “vsmodOutput” 4) import table with areas of country / SMU / scenario combinations (R1234_CORINE_grouped_20240411.csv or R1234_grouped_20240411_ohne_CNT_ NAMES.csv) as “smuAreas” 5) From table link, calculate total area fractions of a given profile per SMU (by summing up); export to intermediate data frame totalAreaFracPerSMU (Table A 17) 6) table smuAreas: remove UK and other non-EU member states 7) table smuAreas: sum up areas for each combination of SMU and FOCUS_scen → new table sumAreasNew 8) join total area fractions (table totalAreaFracPERSMU) with VFSMOD output (table vsmodOutput) via field PROFILE 9) join the resulting table with table sumAreasNew via field SMUSCEN → table z 10) table z: remove invalid records 11) create field ID (concatentation of FOCUS scenario, CORINE or non-CORINE, Koc, duration and timestep) 12) Calculate CDFs using a loop over the unique values of ID; write a .csv file and create a plot for each CDF 13) Write table z as *_sourceForCumArea.calc.csv for quality assurance knoell Germany GmbH Page 73 of 83 Project No.: 120333 Report No.: 120333-1 73 Table A 16 Variables contained in VFSMOD result tables R1234_30mm_noWT_out_ 20240526.xlsx (Variant A) and R1234_30mm_noWT_out_20240610.xlsx (Variant B). field description unit remarks Filename Name of output file before merging n.a. Koc_L_kg linear sorption coefficient normalized to organic carbon (Koc) L kg-1 timestep_irn_iro_min time step of runoff and rainfall hydrographs min value always 15 Project concatenation of FOCUS scenario, rainfall volume and event duration e.g. R1_30_1h FOCUS_scenario FOCUS scenario duration_h duration of rainfall event h 1 or 8 Profile_ID soil profile ID deltaQ_perc relative reduction of total inflow (runoff + rainfall) % deltaE_perc relative reduction of eroded sediment load % deltaR_perc relative reduction of incoming runoff volume % deltaP_perc relative reduction of incoming pesticide load % Vi_m3 surface runoff inflow into the VFS m3 Ei_kg incoming eroded sediment load kg Fph phase distribution coefficient pH Pesticide_input_mi_mg pesticide input into the VFS mg absolute masses Pesticide_output_mo_mg pesticide outflow from the VFS mf_sed_mg pesticide trapped with sediment mfml_mg pesticide trapped in mixing layer mresn_mg pesticide residue remobilization at next event Source_Area_m2 source area of incoming runoff, eroded sediment and pesticides m2 mi_mg_m2 pesticide input into the VFS mg m-2 masses normalized to source area mo_mg_m2 pesticide outflow from the VFS mop_mg_m2 particle-bound pesticide outflow mod_mg_m2 dissolved pesticide outflow mf_mg_m2 pesticide trapped in VFS mfsed_mg_m2 pesticide trapped with sediment mfml_mg_m2 pesticide trapped in mixing layer mfml0_mg_m2 pesticide in mixing layer from last event mres1_mg_m2 Total surface residue (mres1 = mfml + mfsed + mfml0) Tot_res_after_deg_mg_m2 Total residue after degradation (here: 1 d) Dissolved_surf_res_ Dissolved surface residue after degradation knoell Germany GmbH Konrad-Zuse-Ring 25 68163 Mannheim, Germany 80 6.11 List of delivered files (digital annex) List of delivered files (120333_digital_annex_20250319.zip; sent via Cryptshare on 19/03/2025): Folder 0_Geodata contains all spatial data: • Subfolder Processed_FOCUS_scenario_shapefiles: o Subfolder FOCUS_NUTS: FOCUS R scenario shapefiles after intersection with NUTS (see section 2.1) o Subfolder FOCUS_NUTS_CLC2018: FOCUS R scenario shapefiles after intersection with NUTS and CORINE Land Cover 2018 class 2 (see section 2.1) • Subfolder SGDBE: o stuorg.dbf (see section 2.1; Table A 1) o stu_sgdbe.dbf ((see section 2.1; Table A 2) • Subfolder Spatial_tables: o Subfolder Area_fractions ▪ CLC_link_table.csv (see sections 2.3.1, 6.7.3; Table A 15) ▪ noCLC_link_table.csv (see sections 2.3.1, 6.7.3; ; Table A 15) ▪ CLC_smu_prof_fracs_tpt.csv (see section 6.7.3) ▪ noCLC_smu_prof_fracs_tpt.csv (see section 6.7.3) o Subfolder Spatial_input_tables ▪ R1234_grouped_20240411.xlsx (657 records) (see sections 2.1, 6.7.1; Table A 10) ▪ R1234_CORINE_grouped_20240411.xlsx (637 records) (see sections 2.1, 6.7.1; Table A 10) ▪ R1234_NUTS_agg_leftjoin_stuorg.xlsx (see sections 2.1, 6.6, 6.7.1, 6.7.3; Table A 11) ▪ R1234_NUTS_CLC_agg_leftjoin_stuorg.xlsx (see sections 2.1, 6.7.1, 6.7.3; Table A 12) o Intermediate_tables ▪ stuorg-stu_sgdbe-spade14-spade2-R1234.xlsx (see sections 2.1, 6.6) ▪ combinations_from_shapefile (3134 records; see section 6.6) ▪ leftjoin_ProfileID2 (3074 records; see section 6.6) Folder 1_Soil_profile_data contains all soil profile/horizon data: • Subfolder SPADE14 o spade14_stu.csv / .xlsx o PROF_all.csv o HOR_all.csv • Subfolder SPADE2 o spadev211.dbf • Subfolder processed_horizon_data o gethorizons_SPADE14_all_20240304.xlsx (see sections 2.2.1, Table A 8) o gethorizons_SPADE2_all_20240304.xlsx (see sections 2.2.1, Table A 4) o spade2_DBNPTF_VGM_parms_2024-01-26_14-17-31.xlsx (see section 6.10; used by confusionMatrix.R) o spade14_DBNPTF_VGM_parms_2024-01-30_13-05-28.xlsx (see section 6.10; used by confusionMatrix.R) • Subfolder effective_single_horizon_profiles: o 2024_effectiveSingleHoriz_20240305b_SAV_20240504b.xlsx: final scenario / profile combinations for modelling with VFSMOD (1897 records; see section 2.2.1) knoell Germany GmbH Page 81 of 83 Project No.: 120333 Report No.: 120333-1 81 o profileHorizonsMissingValues_20240305.xlsx: List of eliminated scenario / profile combinations due to missing data (1104 records; see section 2.2.1) o effectiveSingleHoriz.xlsx: plain table with scenario / profile combinations (1970 records, since very shallow profiles have not been removed yet; used as input by confusionMatrix.R (see section 6.10) Folder 2_UH contains all UH inputs and outputs (see section 2.2.4): • Subfolder 15_min (15 min = selected time step for output hydrographs) o Subfolder inputs: 8 UH input files (.inp) o Subfolder output: output files for 8 simulations ▪ rainfall hydrographs (.irn) ▪ runoff hydrographs (.iro) ▪ output file hydrograph calculation (.out) ▪ output file erosion calculation (.hyt) ▪ sediment properties input file for VFSMOD (.isd); from this file only the variable CI (eroded sediment concentration in runoff) was used further o irn_to_csv_15min.csv: reformatted rainfall hydrographs (1 row = 1 UH run) o iro_to_csv_15min.csv: reformatted runoff hydrographs (1 row = 1 UH run) Folder 3_VFSMOD contains all VFSMOD inputs and outputs (see section 2.2.4): • R1234_30mm_Koc100_15min_20240526.csv: complete VFSMOD inputs for variant A • R1234_30mm_Koc100_15min_20240610.csv: complete VFSMOD inputs for variant B • R1234_30mm_noWT_out_20240526.xlsx: complete VFSMOD outputs (.owq file) for variant A • R1234_30mm_noWT_out_20240610.xlsx: complete VFSMOD outputs (.owq file) for variant B • 120333_raw_VFSMOD_inputs_outputs_20240526.7z: complete VFSMOD input and output files for all simulations, variant A • 120333_raw_VFSMOD_inputs_outputs_20240610.7z: complete VFSMOD input and output files for all simulations, variant B Folder 4_CDFs contains all spatial CDFs of ∆P that have been created (see sections 2.3.2, 3.2, 6.7.4) • Subfolder variant_A (profiles with hard rock in < 100 cm depth were simulated with the shallow water table option) o Subfolder CDFs (66 files) ▪ 32 CDFs as graph (.png) ▪ 32 CDFs as table (.csv) ▪ sourceForCumAreacalc_variantA.csv (for quality assurance; see section 6.7.4) ▪ CORINEsourceForCumAreaCalc_variantA.csv (for quality assurance; see section 6.7.4) o Subfolder 10th_perc_profiles: ▪ R1234_90th_perc_profiles_with_VFSMOD_params_20240527.xlsx o Subfolder pooled_CDFs: ▪ 8 pooled CDFs (over all 4 R scenarios) as table (.xlsx) (see section 3.5) • Subfolder variant_B (all profiles were simulated with classical Green-Ampt infiltration) o Subfolder CDFs (66 files) ▪ 32 CDFs as graph (.png) ▪ 32 CDFs as table (.csv) ▪ sourceForCumAreacalc_variantB.csv (for quality assurance; see section 6.7.4) ▪ CORINEsourceForCumAreaCalc_variantB.csv (for quality assurance; see section 6.7.4) o Subfolder 10th_perc_profiles: ▪ R1234_90th_perc_profiles_with_VFSMOD_params_noWT_20240610.xlsx o Subfolder pooled_CDFs ▪ 8 pooled CDFs (over all 4 R scenarios) as table (.xlsx) (see section 3.5) knoell Germany GmbH Page 82 of 83 Project No.: 120333 Report No.: 120333-1 82 Folder 5_Boxplots contains spatially weighted boxplots comparing the descriptive statistics of ∆P for the different SGDBE texture classes with ∆P of the 90th percentile worst-case profiles … (see sections 2.5, 3.5) • Subfolder wtd_boxplots_2024-12-11-b o Subfolder variant_A: 1 boxplot for each combination of scenario, Koc, duration and CORINE / no CORINE (32 files) ▪ Subfolder varA_xlsx_full: full CDFs for each texture class (1 .csv file per combination of scenario, Koc, duration and CORINE / no CORINE) ▪ Subfolder varA_xlsx_short: 10th percentile profiles for each texture class (1 .csv file per combination of scenario, Koc, duration and CORINE / no CORINE) o Subfolder variant_B: 1 boxplot for each combination of scenario, Koc, duration and CORINE / no CORINE (32 files) ▪ Subfolder varB_xlsx_full: full CDFs for each texture class (1 .csv file per combination of scenario, Koc, duration and CORINE / no CORINE) ▪ Subfolder varB_xlsx_short: 10th percentile profiles for each texture class (1 .csv file per combination of scenario, Koc, duration and CORINE / no CORINE) • Subfolder wtd_boxplots_2024-12-12_pooled o Subfolder variant_A: 1 pooled boxplot (over all scenarios) for each combination of Koc, duration and CORINE / no CORINE (8 files) ▪ Subfolder varA_xlsx_full: full CDFs for each texture class (1 .csv file per combination of Koc, duration and CORINE / no CORINE) ▪ Subfolder varA_xlsx_short: 10th percentile profiles for each texture class (1 .csv file per combination of Koc, duration and CORINE / no CORINE) o Subfolder variant_B: 1 pooled boxplot (over all scenarios) for each combination of Koc, duration and CORINE / no CORINE (8 files) ▪ Subfolder varB_xlsx_full: full CDFs for each texture class (1 .csv file per combination of Koc, duration and CORINE / no CORINE) ▪ Subfolder varB_xlsx_short: 10th percentile profiles for each texture class (1 .csv file per combination of Koc, duration and CORINE / no CORINE) Folder 6_Comparison_with_previous_SWAN-VFSMOD_scenarios (see sections 2.4, 3.4) • Subfolder test_simulations_vfsm_v424: contains the 16 test simulations with the old VFSMOD executable vfsm v.4.2.4 included in the portable version of SWAN 5 o Subfolder Koc100: contains all VFSMOD input and output files for Koc 100 L kg-1 o Subfolder Koc10000: contains all VFSMOD input and output files for Koc 10000 L kg-1 o merged_owq_Koc100.xlsx: summary table of VFSMOD results for Koc 100 L kg-1 o merged_owq_Koc10000.xlsx: summary table of VFSMOD results for Koc 100 L kg-1 • comparison_vfsm424_with_deltaP_variantB_20250108.xlsx: calculation of ∆P with the mass balance approach; comparison of ∆P with 10th percentile ∆P from new CDFs (see Table 4) • oldVFSMOD_CumAreaPercent_variantB.xlsx: Positions of old SWAN/VFSMOD scenarios (with old VFSMOD executable) on new CDFs (see Table 5) knoell Germany GmbH Page 83 of 83 Project No.: 120333 Report No.: 120333-1 83 Folder 7_Scripts • Subfolder Mathematica contains the Mathematica script (Sav_GA_file2024_20250207.pdf) by Rafael Muñoz-Carpena to calculate the VFSMOD parameter SAV (see section 2.2.2) • Subfolder R_scripts contains the main R scripts used in this study, for reasons of transparency and comprehensible documentation. The R_scripts are the property of knoell Germany GmbH and licensed under GNU GENERAL PUBLIC LICENSE version 3. o recodeLUspade14.R (see section 6.7.2; Table A 14) o import_extended.R (this script performs several tasks; see sections 6.7.3 and 6.6) o createSpatialCDF.R (see section 6.7.4) o profileID_90th_PPerc.R (see section 6.7.5) o get_CumAreapercent_of_old_VFSMOD.R (see section 2.4) o confusionMatrix.R (see sections 2.2.1, 6.10) o GPL-3.txt : license text The scripts can be executed as is in R after setting the working directory to the path of folder 7_Scripts.