Deep tree roots at risk of accelerating groundwater pollution beneath clay-rich aquitards Authors: Peter R. Jørgensen, Klaus Mosthaf, Paul Henning Krogh, Anders J. Hansen, Ida B. Nielsen, Frederik Seersholm, Natascha D. Wagner, Jesper Olsen, Jiayi Qin, Luca Modesti, Jens Aamand, Una Pétursdóttir, Rasmus Thalund-Hansen, Carsten T. Petersen, Bjarne W. Strobel, Marine Lacoste, Massimo Rolle Published: 2026 ABSTRACT Glacial aquitards from the last Pleistocene glaciation cover vast areas of the Northern Hemisphere. Due to their clay-rich composition, these aquitards are considered to protect underlying aquifers from surface contaminations, provided they are not penetrated by widespread deep fractures. In this study we combine ancient DNA (aDNA) with detailed hydrological techniques to investigate contaminant transport in the aquitards – an interdisciplinary approach that was not possible when the current understanding of the dominant role of fractures was established. We show that deep fractures (> 2 m below surface) in the glacial aquitards across Denmark are closed or marginal as pathways of dissolved contaminant transport unless they contain flow channels from deep modern and ancient tree roots. Our study reveals that deep tree roots can compromise natural groundwater protection and accelerate groundwater pollution by opening fractures and generating deep macropores. The results highlight that thorough attention needs to be paid to land-use history and widespread legacy and future soil pollution prior to reforestation and afforestation of aquitards above drinking water resources. We furthermore demonstrate that the protection of aquitard areas with less thickness than historical rooting depths and high degrees of lithological heterogeneity (related to aquitard glacial till types) is crucial to preserve drinking water resources and to restore groundwater quality beneath glacial aquitards. Introduction Glacial deposits of clayey tills (i.e., glacial aquitards) from Pleistocene ice ages stretch over millions of square kilometers on the land surface of temperate Northern and Central Europe, Russia and North America (Fig. 1A, B). The glacial tills are heterogeneous, low-permeability deposits that share similar lithological and hydrogeological characteristics across large distances and continents1,2. Many of these areas have been occupied by indigenous forests and parkland which spanned much of the Northern hemisphere in past Holocene time.Fig. 1Ice cover in the last ice age in the Northern Hemisphere, investigated aquitard locations, and investigation techniques in this study.A Extension of ice cover in the last glacial maximum in the Northern Hemisphere, approximately 26,000–16,000 BP (adapted from2). B Location of study sites, hydraulic tests, and contaminant transport experiments (see Table 1 in Methods section for description). C Laboratory LUC16 and in-situ field17 tests applied for hydraulic and contaminant transport studies (Table 1). a: LUC flexible wall permeameter, b: Large undisturbed column (LUC) samples inside permeameters exhibiting flexible wall membranes to transfer simulated confining pressure in hydraulic and transport tests, c: Dismantling of fractured LUC sample after testing (Holbæk 3.1–3.6 mbgs). d: Preparation of in-situ field hydraulic and transport tests (Salløv 5.2 mbgs), e: Dye tracer infiltration (Brilliant blue FCF) in infiltration basins (Salløv 5.2 − 6 mbgs), f: Mapping of dye tracer transport patterns in excavated geological profile after in-situ field tests (Flakkebjerg2 ground level to 6 mbgs). Ice cover in the last ice age in the Northern Hemisphere, investigated aquitard locations, and investigation techniques in this study. A Extension of ice cover in the last glacial maximum in the Northern Hemisphere, approximately 26,000–16,000 BP (adapted from2). B Location of study sites, hydraulic tests, and contaminant transport experiments (see Table 1 in Methods section for description). C Laboratory LUC16 and in-situ field17 tests applied for hydraulic and contaminant transport studies (Table 1). a: LUC flexible wall permeameter, b: Large undisturbed column (LUC) samples inside permeameters exhibiting flexible wall membranes to transfer simulated confining pressure in hydraulic and transport tests, c: Dismantling of fractured LUC sample after testing (Holbæk 3.1–3.6 mbgs). d: Preparation of in-situ field hydraulic and transport tests (Salløv 5.2 mbgs), e: Dye tracer infiltration (Brilliant blue FCF) in infiltration basins (Salløv 5.2 − 6 mbgs), f: Mapping of dye tracer transport patterns in excavated geological profile after in-situ field tests (Flakkebjerg2 ground level to 6 mbgs). In many regions, glacial aquitards are relied upon as natural protection of underlying aquifers from surface pollution, due to their generally low permeability. The degree of protection depends on the thickness and hydrogeological properties of the aquitards1,3. The thickness of the aquitards range from a few meters to more than 100 meters, however, in large areas it is less than 20 m4,5. In many such areas glacial aquitards have provided particular low protection, and aquifers and water supply wells have become extensively contaminated by chemicals from widespread agricultural, industrial, and urban sources6,7 which are potential hazards to the environment and human health. Numerous studies conducted in North America and Scandinavia over the last four decades have shown that glacial aquitards exhibit a surficial zone approximately 5–20 m in thickness which contains abandoned root macropores and deep vertical fractures1,2. This zone is characterized by vertical saturated hydraulic conductivities (Ks) elevated by one to three orders of magnitude compared to the unfractured aquitard till material (Ksm) i.e., the clayey matrix without fractures. These elevated hydraulic conductivities are widely attributed to fractures, which are considered the most probable deep pathways of enhanced flow into aquifers where permeable lithological heterogeneities, such as sand lenses or erosional windows, are absent in the aquitards1,2,5,8–10. Investigators generally assume that deep root macropores, which are also widespread in the aquitards, have marginal importance for flow in the deposits and they have largely been neglected. The understanding that fractures are major controls of groundwater flow in glacial aquitards is built on field studies carried out with indirect hydrogeological approaches such as slug tests, pumping tests, environmental tracers, infiltration of tracers, and water balance modeling1,2,5,9,11–14. Whilst these studies have provided evidence for the widespread occurrence of deep pathways for rapid water movement, the indirect nature of these approaches has not enabled investigators to directly quantify and distinguish flow in fractures from flow in coexisting deep root macropores (primarily located within the fractures). This lack of direct measurements leaves uncertainty about the fundamental mechanisms, biological versus geological, and hence, cause-effect relationships that control groundwater vulnerability to contamination beneath glacial aquitards which is concerning, especially for protection of drinking water resources. To address this research gap, Jørgensen et al.15,16 designed well-defined flexible-wall large undisturbed column (LUC) experiments, and Ouf et al.17 advanced in-situ dye tracer field experiments (Fig. 1C). These techniques enable to distinguish and quantify flow and contaminant behavior in deep fractures and macropore channels inside fractures (see detailed descriptions in Methods section). These approaches are employed in the present study to further clarify the major pathways for groundwater flow and contaminant transport. There are evident reasons to address the role of deep root macropores as flow and contaminant pathways in aquitards: (i) abandoned deep root macropores are widespread observed features along fractures in glacial aquitards1,5,9,18–20 and similar unlithified sedimentary aquitards21–23, (ii) roots from trees and shrubs are well known to increase infiltration in soils24, and (iii) reforesting and afforesting (hereafter termed reforesting) is recognized as global strategy25 to absorb excessive man-made atmospheric carbon dioxide for mitigating global warming. Recent studies show that trees and shrubs have the potential to develop much deeper root systems than previously believed, and deeper than most agricultural crops and other plants26. Descending roots have a preference to follow old root channels27 and fractures in glacial aquitards and similar fine grained deposits18. Tree roots exhibit exceptional plasticity within and between tree species in response to the sub-soil environment28,29. Development of deep root systems is in particular regulated (pull and push effect) by depth of water tables and water table fluctuations28,29. Water tables in glacial aquitards are primarily controlled by seasonal precipitation surplus30, topography, and leakage into underlying aquifers, of which the latter can be amplified by pumping in aquifers. This in turn may lower water tables further and increase rooting depths by pull-down feedback28. Root systems from trees can grow several meters per year20,27 and rapidly adapt to these conditions. Through biomechanical processes, deep tree roots can create new, reopen old, and expand existing deep root channels and fractures27 in response to changing patterns and composition of tree vegetation, lowering of groundwater tables28,29, and increased atmospheric carbon dioxide and temperatures31. In this study we present new data obtained using the direct methods advanced by Jørgensen et al.16 and Ouf et al.17 and we provide a regional synthesis of the current and previous studies15,19,32–53 conducted with these methods (Fig. 1C and Table 1 in Methods). We combine these studies with investigations of ancient DNA (aDNA), radiometric dating, X-ray imaging, geochemistry, and glacial geology to assess the origin and hydrological role of deep relic root macropores in glacial tills deposited under the last glacial maximum (approximately 26,000–16,000 years BP54) across Denmark (Fig. 1B and Table 1). We use the insights gained from these investigations and observations of current deep rooting of trees, as a blueprint for evaluating the possible effects of trees and reforestation on future groundwater quality beneath glacial tills and similar fine-grained aquitards. This is the first time such a wide interdisciplinary evidence approach has been integrated, which was not possible when the current understanding of flow and contaminant migration in glacial aquitards was established. The specific objectives of the study are to (i) investigate the geographical extent, origins, and hydraulic activity of deep root macropores, (ii) quantify flow and transport behavior in fractures by relic root macropores located inside the fractures versus flow and transport in fractures without root macropores, (iii) evaluate the control of flow and the impact on groundwater vulnerability beneath glacial aquitards by fractures, root macropores, and aquitard lithological heterogeneity, respectively, with focus on transport of persistent mobile pesticides, and (iv) assess the implications of these findings for groundwater protection strategies. Results Fractures, tree roots, and deep root macropores Figure 2 shows tree roots (2A) and relic type root macropores (2B-D) observed in vertical profiles at the study sites (Supplementary Figs. 1–5 and Table 1). The conceptual geological model in Fig. 2B shows the relationship between relic root macropores (diameter >0.5 mm), fractures, weathering zones (oxidized and unoxidized), and the occurrence of coarse-grained geological layers (i.e., lithological heterogeneity) in the profiles, which are representative of glacial till aquitards described widely elsewhere1.Fig. 2Tree roots and relic root macropores systems.A Extension and architecture of tree roots along fractures controlled by tree species and water table (WT) depth; deep root systems from 10–14 years old Common Oaks (Quercus robur) and shallow roots from beech trees (Fagus sylvatica) on fractures surfaces (two most common tree species 6000–3000 BP and after 3000 years BP, respectively). Notice high densities and dimorphism106 of fine roots (<2 mm in diameter) displayed by Oaks to increase water uptake near water table (full size images and further examples in Supplementary Figs. 3–5). B Conceptual model showing relationship between geology, fractures, residual roots, and water table fluctuation in the glacial tills of the study. Greyish fractures (hereafter termed grey fractures) and reddish-brown fractures (hereafter termed red fractures) in the diagram are representing weathered fractures exhibiting depletion and accumulation of secondary Fe/Mn-oxides, respectively, by translocation9,23,43,59. “Ghost-fractures” are (invisible) unweathered fractures identified by root macropores or roots occurring along lines in excavations (Fig. 2E and Supplementary Fig. 5) and breaking of till along slickensides9. C Ancient, deteriorated tree root macropores (flattened) along grey fracture surface at Salløv site, 2.8 mbgs in INF2. D Recent root macropores (i.e., containing root remains) along surface of red fracture, 4 mbgs at Gedser site. E Ancient root macropores in cross section of vertical fracture in unweathered glacial till, 5.4 mbgs at Salløv site INF2 (See also Supplementary Fig. 5 photographs 4–6). Tree roots and relic root macropores systems. A Extension and architecture of tree roots along fractures controlled by tree species and water table (WT) depth; deep root systems from 10–14 years old Common Oaks (Quercus robur) and shallow roots from beech trees (Fagus sylvatica) on fractures surfaces (two most common tree species 6000–3000 BP and after 3000 years BP, respectively). Notice high densities and dimorphism106 of fine roots (<2 mm in diameter) displayed by Oaks to increase water uptake near water table (full size images and further examples in Supplementary Figs. 3–5). B Conceptual model showing relationship between geology, fractures, residual roots, and water table fluctuation in the glacial tills of the study. Greyish fractures (hereafter termed grey fractures) and reddish-brown fractures (hereafter termed red fractures) in the diagram are representing weathered fractures exhibiting depletion and accumulation of secondary Fe/Mn-oxides, respectively, by translocation9,23,43,59. “Ghost-fractures” are (invisible) unweathered fractures identified by root macropores or roots occurring along lines in excavations (Fig. 2E and Supplementary Fig. 5) and breaking of till along slickensides9. C Ancient, deteriorated tree root macropores (flattened) along grey fracture surface at Salløv site, 2.8 mbgs in INF2. D Recent root macropores (i.e., containing root remains) along surface of red fracture, 4 mbgs at Gedser site. E Ancient root macropores in cross section of vertical fracture in unweathered glacial till, 5.4 mbgs at Salløv site INF2 (See also Supplementary Fig. 5 photographs 4–6). Figure 2A and Supplementary Figs. 3–5 show penetration of tree roots in outcrops at the Gedser, Helnæs, and Videbæk study sites (Fig. 1B). The outcrops reveal dense systems of tree roots along fractures to depths from a few meters to over 10 meters below ground surface (mbgs) depending on tree species (Beech and Stem Oak, respectively) and depth to groundwater tables28,29. These observed rooting depths align with observations in North America, where tree roots in glacial tills have been reported at 9 to about 15–20 mbgs5,18. Systematic collection of data on the maximum depth of tree roots in relation to tree species and sub-surface conditions represents a research gap that has yet to be addressed27,28. Nonetheless, the available evidence shows that dense systems of fine tree roots can potentially extend, or have extended, to the base of the previously described hydraulically active fractured zones in glacial aquitards, provided suitable hydrological conditions are - or have been – present. Such conditions encompass deep natural water table fluctuations28,29, lowering of water tables (WT) by groundwater pumping55, and past Holocene altithermal periods with lower precipitation56. The relationship between hill-slope, WT fluctuation, and depths of root macropores and relic roots at the Salløv and Bulbro sites are shown in Fig. 3A and Supplementary Figs. 6 and 7. Note in particular the selective “pull-down”28,29,57 of deep root channels at the Salløv site by deep WT fluctuations (INF 2) driven by aquitard leakage along dislocated and folded sand layers in the aquitard (displayed in Supplementary Fig. 6).Fig. 3Fossil roots and paleo-root macropore flow paths.A Salløv – Bulbro hill-slope profile (750 m long; 7x vertical exaggeration) showing findings of fossil roots from ancient Alnus glutinosa trees (common alder) below ground water table by the foot of the hill (25–28 m above sea level) and aDNA from Salix subgenus Vetrix clade (willow shrubs and trees) along paleo-root macropores clogged by Fe/Mn-oxide translocation (hydraulic barrier) at high-ground (42−45 m above sea level); see Supplementary Fig. 6 for further hydrogeological details and “pull-down” of paleo-root macropores by water table28,29. B Specimens of fossil roots from Common alder trees (Alnus glutinosa) dating from 2800 to 7,250 BP in Bulbro site (Bulbro INF1-3) and Fe/Mn-oxide filled paleo-root macropores (hydraulic barrier in Salløv site INF2) in which the aDNA from Salix subgenus Vetrix species were found. C Brilliant blue dye tracer flow paths observed in infiltration tests, a: Dyed root macropore flow paths on exposed surfaces of grey fractures and obstruction of flow by red fractures clogged by translocated Fe/Mn-oxides in Salløv site, b: Dye tracer flow along root macropores and associated fracture desiccation channel segments formed along the root macropores (further explanation in text). c: CT-scan of fracture desiccation channel (grey shaded area) along root macropores in fracture (scan resolution = 369 µm) (full size image in Supplementary Fig. 18C). Fossil roots and paleo-root macropore flow paths. A Salløv – Bulbro hill-slope profile (750 m long; 7x vertical exaggeration) showing findings of fossil roots from ancient Alnus glutinosa trees (common alder) below ground water table by the foot of the hill (25–28 m above sea level) and aDNA from Salix subgenus Vetrix clade (willow shrubs and trees) along paleo-root macropores clogged by Fe/Mn-oxide translocation (hydraulic barrier) at high-ground (42−45 m above sea level); see Supplementary Fig. 6 for further hydrogeological details and “pull-down” of paleo-root macropores by water table28,29. B Specimens of fossil roots from Common alder trees (Alnus glutinosa) dating from 2800 to 7,250 BP in Bulbro site (Bulbro INF1-3) and Fe/Mn-oxide filled paleo-root macropores (hydraulic barrier in Salløv site INF2) in which the aDNA from Salix subgenus Vetrix species were found. C Brilliant blue dye tracer flow paths observed in infiltration tests, a: Dyed root macropore flow paths on exposed surfaces of grey fractures and obstruction of flow by red fractures clogged by translocated Fe/Mn-oxides in Salløv site, b: Dye tracer flow along root macropores and associated fracture desiccation channel segments formed along the root macropores (further explanation in text). c: CT-scan of fracture desiccation channel (grey shaded area) along root macropores in fracture (scan resolution = 369 µm) (full size image in Supplementary Fig. 18C). Fine roots from trees undergo constant replacement through root turnover31, leaving behind abandoned root macropores and fracture channels (Fig. 2B–D and Supplementary Figs. 5 and 7). Only a few examples of residual roots were found below crop roots (extending to 2–3 mbgs, Supplementary Figs. 6 and 16) in the agricultural sites of this study, which is due to the limited preservation of organic root residuals under aerobic conditions, as well as the absence of recent forest vegetation at most of these sites. However, we found fossil roots and residual DNA fragments from trees (Fig. 3A, B and Supplementary Figs. 6 and 7) in the bottom-end of the extensive systems of relic root macropores across the Bulbro and Salløv study sites (Supplementary Figs. 8 and 9–15) - both of which have more than a thousand years of agricultural history with absence of forest vegetation58, (see Site description in methods). The fossil roots were preserved by water logged anaerobic soil conditions beneath stable high water tables exhibited at the Bulbro site (i.e., from tree species insensitive to water-logging as shown in the following), whereas degraded DNA (Supplementary Fig. 12) was preserved at the Salløv site along Fe/Mn-oxide filled root macropores that extended into unoxidized clayey till beyond the bottom of the INF2 excavation at 6 mbgs (Fig. 3A, B and Supplementary Figs. 6 and 8). Sequencing of aDNA and radiocarbon dating applied to the fossil roots from the Bulbro site (Supplementary Figs. 8G and 9–15, and Supplementary Table 1 and 2) revealed that the roots originated from 2800 to 7250 year old common alder (Alnus glutinosa) trees and shrubs. Similarly ancient DNA from willow trees and shrubs (Salix subgenus Vetrix clade) was found along the abandoned macropores at the Salløv site (description of sampling in Methods section). At the Salløv site, reliable radiocarbon dating of the abandoned root macropores was not possible because macropore-linings and infillings represent mixed systems open to exchange of radiocarbon, as opposed to the closed system of the fossil roots. However, an age at the order of several thousand years was estimated55 for the root macropores that contained the willow trees and shrubs aDNA based on the calculated length of time required for the observed secondary Fe-oxides to accumulate in the root macropores by translocation43,59 (Figs. 2B and 3B Salløv INF2). Hence, this age-estimate combined with observed distinct ancient DNA damage patterns for the Salix DNA samples (damage patterns shown in Supplementary Figs. 9 and 12), suggests a similar ancient origin of the willow DNA to that of the common alder fossil roots. Note that willow and common alder were among the most common trees and shrubs in the dense indigenous forest which covered Denmark before 6.000 BP in the Danish hunter-gather stone age58. The fossil roots and aDNA thus backdates the origin of the dense root macropore systems at the Bulbro and Salløv sites (Fig. 3 and Supplementary Figs. 6 and 7) to the indigenous forests that spanned Central Eurasia during early and mid-Holocene55. This provides evidence for very long-term persistence of deep relic root macropores from trees in the glacial aquitards (and likely also in similar fine-grained aquitard sediments such as widespread clays21,22 and saprolites23). Furthermore, it suggests that a likely origin of many deep relic root macropores reported across continents1,5,9 is from ancient indigenous forests (including wide areas that have not been forested since the early Holocene60). The ancient macropores coexist with younger root macropore systems that represent the succession of tree species after the last ice age up to the present58. As mentioned, such younger root structures preferably extend along preexisting root macropores27 (Supplementary Fig. 16C). This may reopen, connect, and expand older relic root macropore structures in response to reforestation with deep-rooted tree species (e.g., Stem Oak; the most common tree species in Danish reforestation projects61). Identification of flow and transport patterns Flow paths in the geological profiles were investigated in the LUC and in-situ hydraulic tests, and with combinations of these (Figs. 1C and 3C) by infiltration of dye tracer (Brilliant Blue62 (BB)). In the A, B, and C soil profile horizons occupying the upper 1–1.5 m of geological profiles (Fig. 2B) the dye tracer flow paths were dominated by earthworm burrows, root holes, and vertical desiccation fractures (i.e., fractures formed through atmospheric desiccation and freeze-thaw processes9) connecting with the soil surface (Fig. 2B and Supplementary Fig. 16B, D). The desiccation fractures exhibited up to 0.5–2 cm open apertures in the uppermost 1–2 m of the soil profiles during summers (Supplementary Fig. 17A) under which conditions dye tracer flow was evenly distributed along entire fracture surfaces (Supplementary Fig. 17B1). This observed flow pattern corroborates well-established knowledge of preferential flow in clay-rich soils, e.g., Jarvis63. About 1–1.5 mbgs, at the transition from the soil profiles to the underlying parent material (i.e., glacial tills) dyed earthworm burrows terminated and most root macropores entered fractures in which they interconnected with dense systems of dyed relic root casts which followed the deep grey, red, and non-visible “ghost-fractures” (Figs. 2B, 3C, and Supplementary Figs. 17–19). Below 1 to 2 mbgs dye tracer flow in the fractures was consistently absent in all study sites except for the dye tracer flow observed in the root macropores and in fracture desiccation channels developed along these (Fig. 3C, Supplementary Figs. 17B2, 17–19, and Supplementary Video 1 showing excavation of fractures after dye tracer infiltration). The fracture desiccation channels along the root macropores constituted 0.01 to 0.1 m wide, open fracture segments (Fig. 3C and Supplementary Figs. 18C, D and 20B). Wider open fracture sections formed by overlapping root desiccation channels associated with closely spaced root macropores (Fig. 2A and Supplementary Fig. 3) were further identified in dye tracer experiments at the Havdrup site19. In the Salløv, Bulbro, and Holbæk sites, the general 2 m maximum depth of open desiccation fractures was measured during a five-month drought in 2018 where the highest soil moisture deficit for more than a century was recorded in Denmark64. This suggests that purely atmospherically driven desiccation of fractures is generally confined to the upper 1–2 meters in the aquitards under the current climate conditions, whereas the vertical extension of narrow desiccation channels along root macropores observed in the deep fractures (Fig. 3Cb,c) is driven by the original roots drawing water from the clay matrix surrounding them. The root macropores and associated desiccation channels formed interconnected network facilitating very rapid dye tracer macropore flow along the fractures throughout the vadose zone in the in-situ field tests at the Bulbro, Flakkebjerg2 and Havdrup sites (Supplementary Figs. 19E and 20). At the Salløv site, the rapid dye tracer flow was retarded by the clogging with secondary Fe/Mn-oxides of the deep root macropores containing the Salix aDNA described in the previous section (hydraulic barrier in Fig. 3A, B and Supplementary Fig. 6). Furthermore, the dye tracer tests revealed that in 2 out of the 12 study sites, dyed relic root macropore systems extended vertically along weathered red fractures and “ghost- fractures” beyond the maximum depth of the excavations (e.g., beyond 6 mbgs at Salløv INF2 location, Supplementary Fig. 6). At three locations the relic root macropores from common alder, a species insensitive to water logging65, penetrated 1–2 m below current and historical water table depths following invisible “ghost-fractures” into grey unweathered clays (Supplementary Figs. 8A–F and 19E). At five study sites, the dyed root macropores interconnected (by root hydrotrophism57 and dimorphism) with shallow aquifers and water-bearing sand/gravel intrabeds (Supplementary Fig. 20C-I), some of which provided aquitard leakage into underlying ground water (Fig. 3 and Supplementary Fig 6 Salløv INF2). These root macropores included the more than 7000 years old root macropore systems at the Bulbro site, still dominating extended vertical flow to the present day (Supplementary Fig. 20I). No dye tracer transport was observed in fractures more than 3–4 m below aquitard water tables (including “ghost-fractures” in Fig. 2B). Thus, in summary, relic root macropores and root desiccation fracture channels provided rapid connected major flow paths, in some cases throughout the vadose zone and about 1–2 m into the permanently water-saturated zone of the glacial aquitards. The root macropore flow paths connected with the soil surface through interconnection with earthworm borrows and soil root channels which were extensively utilized by current crop roots to 2–3 mbgs. Dyed root macropores interconnected with water-bearing coarse-grained lithological heterogeneities, some of which provided aquitard leakage into underlying aquifers. Fractures per se (i.e., without root macropores inside) were consistently occluded or formed very narrow flow paths from approximately 1–2 mbgs at the study sites. Hydraulic conductivity in aquitard profiles Water-saturated flow and integrated contaminant transport behavior along fractures at the study sites (Fig. 1B) were quantified in LUC and in-situ hydraulic tests (Table 1, and Supplementary Figs. 1 and 2) shown in Fig. 4A–C.Fig. 4Compilation of LUC and in-situ field tests from the study sites (Fig. 1B) showing the distribution of flow between fractures with and without root macropores, and flow in the clay matrix.A Water-saturated hydraulic conductivity of LUC samples with fractures/macropores shown together with sub-samples of intact LUC clay matrix (i.e., without fractures and macropores). B Calculated effective fracture apertures in the LUC and field tests. Note that water fluxes along fractures and fracture channels scale with the aperture cubed, i.e., increasing aperture (2beff) from e.g., 10 µm to 100 µm enables 1000 times higher water flux. C Effective fracture apertures related to the maximum depth of seasonal water table (WT) fluctuations (i.e., vadose zone) in the glacial till aquitards of the study. Numerical parameter values for A-C are given in Supplementary Table 5 and 6. D Average fracture spacings observed in the excavations at the study sites and represented in the LUC and in-situ tests from the excavations; note that the fractures spacings are generally smaller in the tests than the average in the excavations and therefore represent maximum values of fracture and root macropore frequencies over depth. Compilation of LUC and in-situ field tests from the study sites (Fig. 1B) showing the distribution of flow between fractures with and without root macropores, and flow in the clay matrix. A Water-saturated hydraulic conductivity of LUC samples with fractures/macropores shown together with sub-samples of intact LUC clay matrix (i.e., without fractures and macropores). B Calculated effective fracture apertures in the LUC and field tests. Note that water fluxes along fractures and fracture channels scale with the aperture cubed, i.e., increasing aperture (2beff) from e.g., 10 µm to 100 µm enables 1000 times higher water flux. C Effective fracture apertures related to the maximum depth of seasonal water table (WT) fluctuations (i.e., vadose zone) in the glacial till aquitards of the study. Numerical parameter values for A-C are given in Supplementary Table 5 and 6. D Average fracture spacings observed in the excavations at the study sites and represented in the LUC and in-situ tests from the excavations; note that the fractures spacings are generally smaller in the tests than the average in the excavations and therefore represent maximum values of fracture and root macropore frequencies over depth. The saturated hydraulic conductivity (Ks) of the LUC samples (Fig. 4A) reveals that the samples which contained relic root macropores along fractures (and earthworm burrows in LUCs from the upper 1 m) have Ks values in the range 10−8 to 10−4 m/s, which is up to 4 orders of magnitude higher than the LUC samples without root macropores inside fractures. The latter exhibit Ks values in the range 10−9 to 10−8 m/s which is at maximum only 4 times higher than flow in the unfractured matrix (Ksm) in the same samples (also shown in Fig. 4A). Hence, the values in hydraulic conductivity between the LUC samples without root macropores along fractures were to a high degree controlled by the hydraulic conductivities of the unfractured matrix and differences between these (i.e., lithological heterogeneity). The overall vertical decrease of Ks values in the upper 1–3 mbgs observed for the LUC samples containing root macropores is due to increasing spacings between fractures (Fig. 4D) that contain root macropores and increasing macropore collapse42 and clogging of macropores with fines (Supplementary Fig. 18A) and translocated Fe/Mn-oxides with depths23,55 (Fig. 3B Salløv site INF2). Effective aperture and flow in fractures and root-macropores The effective fracture apertures (2beff)17 shown in Fig. 4B describe the water-saturated flow in the fractures observed in the hydraulic tests. The results show that the fractures and fracture sections containing relic root macropores in the sub-soil have 2beff values between 14 and 174 µm. According to the cubic law66, flow through fractures scales with the cube of the aperture and therefore the variation of flow is very strong within the exhibited range of apertures. Note furthermore, that these aperture values are represented at a regional scale, which agrees with the origin of the root macropores dating back to dense indigenous forests that covered Denmark58 (and large areas of the Northern Hemisphere) through the early and mid-Holocene. This in turn is consistent with the fact that most 2beff aperture values are 1 to 2 orders of magnitude smaller than the observed diameters of roots and root macropores from current trees shown in Fig. 2A and Supplementary Fig. 5, which corroborate that many of the former (e.g., Fig. 2C) are older deteriorated structures (long agricultural history at most of the study sites). The fractures without root macropores or with extensively clogged and occluded root macropores inside (e.g., Fig. 2C and Salløv site BB5 and BB6 from INF2 in Fig. 3) exhibit a remarkably consistent and narrow range of small 2beff values from 0 to 12 µm with 75% < 6 µm and often represent channels related to the roughness of fracture surfaces40 along otherwise occluded fractures (see Supplementary Fig. 30). These small aperture values are independent of the type of till, current vegetation, land-use in the study sites (Table 1), and the method of determination (i.e., LUC versus in-situ technique). According to the cubic law that relates effective fracture aperture and flow, the deep vertical fractures that contain large open root channels can entirely dominate water flow and contaminant migration over other flow paths below 1–2 mbgs, and consequently, the role of fractures without open root macropores is marginal or negligible. The range of 2beff values determined by the LUC and in-situ tests in this study corroborate the range of fracture hydraulic aperture values (1–159 µm) estimated in previous studies of glacial aquitards conducted between 2 to approximately 20 mbgs in Northern Europe and North America12,14,52,67,68. Nonetheless, these studies do not distinguish and quantify the possible contribution of flow by relic root macropores inside the fractures (observed in many cases). However, the same possible effect of root macropores inside fractures as determined in this study was reported (although not distinguished and quantified) in field experiments in a North American glacial aquitard40,68. A subsequent LUC study conducted in the same aquitard40 revealed the same effective fracture aperture range (2beff = 8–11 µm) for fractures without root macropores (occurring immediately below depths of observed root casts in fractures) as determined in the current study (i.e., 2beff < 5–12 µm in Fig. 4C). Similar low hydraulic aperture values (2b = 0–5 µm) were suggested from hydraulic tests with small intact fractured columns with no roots69. Depths of flow in fractures and root macropores Open root macropores along fractures penetrated 1–1.4 m below maximum depth of water table fluctuations at the study sites (i.e., into the permanently water-saturated zone, Fig. 3A) whereas fractures without root macropores occluded (2beff < 5 µm) 2–4 m below the maximum depth of the water tables (Fig. 4C), which agrees with the dye tracer infiltration experiments described above. The depth to permanently water-saturated conditions in the aquitards varied between 2–3 to >8 mbgs (Table 1), the latter occurring primarily at locations with elevated topography and enhanced vertical drainage provided by permeable aquitard lithological heterogeneity and groundwater pumping in underlying aquifers55 (Figs. 2B and 3, and Supplementary Fig. 6). The occlusion with increasing depth of the fractures without root macropores inside was additionally observed in hydraulic tests in which the LUC permeameter confining stress was increased above the in-situ stress to simulate increasing depositional overload (Supplementary Fig. 21). By contrast, fractures with root macropores remained hydraulically active under elevated effective soil stress exceeding 15–20 mbgs, which aligns with maximum depths reported for tree root systems18 and the widespread hydraulically active zones in glacial aquitards traditionally attributed to open fractures1,2. Discussion The consistency between the direct flow measurements in Fig. 4, aDNA analyses, and the field observations of tree roots, provides evidence that deep open root macropores from tree vegetation are widespread dynamic controls of flow and transport in surficial aquitards at a regional scale, extending to the maximum historical rooting depths of trees and shrubs (i.e., beyond 10 mbgs). There is similarly consistent evidence that flow and transport within the fractures themselves below 1–2 mbgs are comparably marginal or negligible regardless of the till type, current vegetation, or land-use (Table 1). The small contribution to flow by the fractures without open root macropores (Fig. 4A, B) is greatly insufficient (Supplementary Fig. 22) to account for 1–3 orders of magnitude difference between aquitard vertical Ks-values and clay matrix Ksm-values observed widely in surficial glacial aquitards, which has conventionally been attributed to flow in fractures1,2,5. Instead, the elevated Ks values in aquitards can be accounted for when including the measured flow along relic root macropores and associated fracture channels within the fractures. This provides a strong indication that deep relic root macropores represent major contributors to flow in the hydraulic active surficial zone observed widely in glacial aquitards; this is particularly evident when including the origin of the root macropores dating back to indigenous Holocene forests and woodland that spanned vast regions in Northern and Central Europe, Russia and North America. The indicated absence of open fractures (2beff < 5 µm in Fig. 4B, C) below rooting depths and water tables implies that elevated aquitard Ks values (Ks > 1 ∙ 10−9 m/s) in those depths are primarily controlled by lithological heterogeneity (i.e. coarse-grained layers, lenses, liners, and bodies in the tills (Fig. 2B) and textural heterogeneity of the till matrix (Fig. 4A and Supplementary Fig. 6)) forming random interconnected pathways of leakage through the aquitards. Glacial tills and such embedded coarse-grained layers often exhibit increasing glaciotectonic deformation (dislocation and folding of layers) and heterogeneity in the following order of till types70,71: glaciolacustrine till, lodgement till, hummocky till/flow tills, and terminal tills/push moraines (glaciotectonites70). Active vertical flow along such randomly interconnected pathways corroborates the observations in the current study (Fig. 2B and Supplementary Figs. 6 and 20) and has previously been demonstrated to dominate aquifer recharge and contaminant pathways into underlying aquifers10,47,48,72. To demonstrate the implications of our findings for groundwater quality beneath glacial aquitards, we performed numerical simulations (Figs. 5 and 6) of the migration of persistent mobile organic contaminants (PMOCs) through 5 m of glacial till into an underlying aquifer described in Fig. 5A (boundary conditions and parameters are described in Methods and Supplementary Table 6). The featured PMOC compound is the fungicide metabolite N,N-dimethylsulfamide (DMS) which is not adsorbed and exhibits no or very slow degradation in the sub-soil environment73 (Supplementary Table 6). This aligns with DMS is the most frequently detected pesticide metabolite in Danish drinking water wells and was found in 24.3 % of drinking water supply well-screens in 2014–2023 together with other PMOC pesticides (4-bis-amido-3,5,6 trichlorobenzenesulfonate; Desphenylchloridazone; 2,6-Dichlorbenzamide, and DEET) and PFAS compounds, primarily underneath glacial aquitards74. The transport simulations are conducted with no adsorption and degradation75 over the determined range of effective fracture apertures in Fig. 4B describing both flow and contaminant transport16,17,42,43,49 (See Supplementary Extended method description 1.1). The simulations cover three scenarios with different hydraulic conditions in the matrix domain (matrix hydraulic conductivities Ksm of 10−10, 10−9, and 10−8 m/s and a conservative description of transport by total-porosity piston flow) to mirror the characteristic range of lithological heterogeneity and glaciotectonic deformation in the till types mentioned above.Fig. 5Mathematical modeling of the impact of persistent mobile pesticides and similar persistent mobile organic compounds (represented by the fungicide metabolite, DMS) on groundwater (aquifer) underneath clayey till based on LUC, in-situ hydraulic tests (Fig. 4A-C) and contaminant transport studies integrated in the tests (see Table 1).A Schematic illustration of the model domain showing water and contaminant fluxes in model simulations including solute exchange between fractures and matrix by flow and diffusion (purple arrows). The simulations show B The arrival time for 10% of the infiltrated pesticide concentration (C/C0 = 0.1 of constant infiltration with 1 µg/L) at the bottom of the aquitard and C The pesticide flux into the aquifer by the time of 10% breakthrough (calculated at C/C0 = 0.1). Note that water fluxes vary with the cube of the fracture aperture, i.e., increasing aperture (2beff) from 10 µm to 60 µm (exhibited in-situ in field experiments, Fig. 4B) decreases the arrival time of DMS by up to 1000 years and increases DMS fluxes into the aquifer by up to 200 times which occur particularly under summer rain storm events and buildup of water tables in autumn30,45,50 (see Supplementary Fig. 6). D Acceleration of aquifer contamination with DMS from 20 years old residual agricultural soil pollution as result of introducing one root macropore channel (1 cm horizontal length) along a fracture (root macropore blueprint shown in CT-scan Fig. 3Cb,c and Supplementary Fig. 20B). The fracture initially contains no root macropores or only clogged root macropores (example from Salløv site, Fig. 4B and Supplementary Fig. 5). a: shows the 25% largest apertures of fractures with and without root macropore channels applied in the simulations (adapted from Fig. 4B together with Ksm = 10−9 m/s), b and c: shows the simulated breakthrough of DMS in the aquifer with and without, respectively, introducing the open root macropore along the fracture. Initial residual DMS concentration in the leaching from the contaminated soil is 0.23 µg/L (Albers et al.73) in both simulated scenarios (b and c). See further details in Methods. Mathematical modeling of the impact of persistent mobile pesticides and similar persistent mobile organic compounds (represented by the fungicide metabolite, DMS) on groundwater (aquifer) underneath clayey till based on LUC, in-situ hydraulic tests (Fig. 4A-C) and contaminant transport studies integrated in the tests (see Table 1). A Schematic illustration of the model domain showing water and contaminant fluxes in model simulations including solute exchange between fractures and matrix by flow and diffusion (purple arrows). The simulations show B The arrival time for 10% of the infiltrated pesticide concentration (C/C0 = 0.1 of constant infiltration with 1 µg/L) at the bottom of the aquitard and C The pesticide flux into the aquifer by the time of 10% breakthrough (calculated at C/C0 = 0.1). Note that water fluxes vary with the cube of the fracture aperture, i.e., increasing aperture (2beff) from 10 µm to 60 µm (exhibited in-situ in field experiments, Fig. 4B) decreases the arrival time of DMS by up to 1000 years and increases DMS fluxes into the aquifer by up to 200 times which occur particularly under summer rain storm events and buildup of water tables in autumn30,45,50 (see Supplementary Fig. 6). D Acceleration of aquifer contamination with DMS from 20 years old residual agricultural soil pollution as result of introducing one root macropore channel (1 cm horizontal length) along a fracture (root macropore blueprint shown in CT-scan Fig. 3Cb,c and Supplementary Fig. 20B). The fracture initially contains no root macropores or only clogged root macropores (example from Salløv site, Fig. 4B and Supplementary Fig. 5). a: shows the 25% largest apertures of fractures with and without root macropore channels applied in the simulations (adapted from Fig. 4B together with Ksm = 10−9 m/s), b and c: shows the simulated breakthrough of DMS in the aquifer with and without, respectively, introducing the open root macropore along the fracture. Initial residual DMS concentration in the leaching from the contaminated soil is 0.23 µg/L (Albers et al.73) in both simulated scenarios (b and c). See further details in Methods. The simulations in Fig. 5B, C confirm that fractures which contain root macropores (2beff = 17–123 µm in Fig. 4B–C) have the potential to entirely dominate pesticide breakthrough and fluxes into aquifers in locations and areas with aquitard thickness smaller than past and modern rooting depths, whereas the impact of fractures without root macropores (2beff = 5–12 µm in Fig. 4B–C) on the pesticide fluxes is two to five orders of magnitude smaller and therefore negligible compared to the impact of the root macropores. For effective fracture apertures 2beff < 30 µm, the simulations show that the pesticide impact on the aquifer is increasingly controlled by the hydraulic conductivity (Ksm) of the matrix domain (i.e., the degree of till lithological heterogeneity and clay matrix texture (Fig. 2B, Table 1, and Supplementary Fig. 22) becomes increasingly important). Thus, for apertures 2beff < 5 µm, representing observed fractures without root macropores below groundwater tables (Fig. 4C), the impact is negligible whereas the lithological heterogeneity and clay matrix texture accounts for one to two orders of magnitude difference in pesticide fluxes and a corresponding decrease of arrival times of contaminants in the aquifer. In the absence of open relic root macropores only the fractures observed above and immediately below water tables (Fig. 4C) exhibiting 2beff values of about 10 µm have a notable impact on pesticide transport when embedded in a uniform low permeability matrix17,76. The strong impact of the root macropores on water flow and solute transport implies that the risk of groundwater contamination (groundwater vulnerability) can shift in time and space with changing patterns and composition of tree vegetation (e.g., afforestation and transition from dominance of shallow-rooted beech to deep-rooted oak trees, and similar, by reforestation, Fig. 2A). These vegetation changes may lead to expansion and deepening of open root macropores and become reinforced by lowering of groundwater tables28,31 due to groundwater abstraction77 which can make watershed locations with high groundwater protection priority particularly vulnerable. Such conditions are suggested to exist underneath the former apple tree orchard at the Havdrup study site (rooting depths to potentially 10 mbgs20) located on ancient arable land (Table 1). This site is now exhibiting a high density of open large-aperture root macropores (2beff = 23–123 µm in Fig. 4B) and observed rapid pesticide transport9,19 into a local aquifer at 5–6 mbgs after the water table has been lowered by groundwater abstraction in the area55,78 (see site description in Methods section). In comparison, at the Salløv site adjacent to the orchard and located in the same aquitard55 with no record of recent forest vegetation, similar deep paleo-root macropores originating from ancient Salix trees and shrub vegetation described above were clogged (i.e., 2beff = 0–12 µm, BB5 and BB6 in Fig. 4B) due to long-term accumulation of Fe/Mn-oxide55 (Figs. 2B and 3B (Salløv INF2)). In Fig. 5D, the fracture root channel effective apertures exhibited underneath the Havdrup orchard site (modern tree vegetation) and Salløv site (pre-historic forest vegetation), respectively (Fig. 4B), are used as a blueprint to simulate the impact of afforestation on groundwater quality by DMS contamination represented by a nearby 20 years old residual soil pollution observed in the same aquitard73. The simulations show the impact on the aquifer in Fig. 5A by introducing one new open root macropore flow channel along a fracture (representative of the Havdrup orchard and Hinnerup forest study sites) versus the same fracture without open root macropores (representing the clogged pre-historic root macropores in the hydraulic barrier at the Salløv site in Fig. 3A and Supplementary Fig. 6). Note that these two cases are spanning the 25% largest deep fracture flow paths with and without root macropores, respectively, determined in the LUC and in-situ tests in this study (Fig. 5Da). The simulations reveal that without introducing the large aperture fracture root channel (Fig. 5Db), the DMS infiltration into the aquifer is significantly halted (>40 years) and concentrations are suppressed below drinking-water standard (DWS). This is due to low infiltration through the aquitard layer and strong diffusive exchange of the DMS between the small aperture fracture and aquitard clay matrix (i.e., matrix-diffusion1,49) that smoothens out breakthrough concentrations. In contrast, the infiltration rate is increased and the matrix diffusion process is marginal under the rapid transport in the introduced large-aperture root macropore channel49, resulting in up to an 81% breakthrough of the infiltrating DMS concentrations (13 times increase of DMS concentration and 86 times increase in DMS mass-flux entering the aquifer) which exceeds the DWS in the aquifer recharge less than 1 month after infiltration along the root macropore has been introduced (Fig. 5Dc). Note that possible slow degradation of DMS in the subsoil aquitard layer (DT50 653 to 2990 days73) will only marginally influence the rapid aquifer impact of DMS in the case with one root macropore whereas it will greatly reduce any impact in the case with no root macropores due to the long residence time in the aquitard layer44 for this case. The uncertainty of the simulations shown in Fig. 5D is marginal with respect to the calculated cause-effect relationship between root channel apertures and aquifer contaminant impact which has been replicated in multiple experimental studies16,17,42,43,49 (see also modeling in Methods section). However, there are site-specific uncertainties about the degree of impact of new tree roots on aquifer vulnerability due to variability in composition of tree species, hydrological and geological variation, the degree of expansion of preexisting root channels, contaminant source species and concentration, as well as forest managing. This variability is a separate important issue which is beyond the scope of this study. Our aim with the modeling is to demonstrate the exceptional impact on groundwater vulnerability and contamination of introducing just one representative open root channel macropore in a system with no or significantly clogged macropores represented by the Salløv agricultural site (Fig. 3A, B and Supplementary Fig. 6) influenced by a relatively minor legacy PMOC contaminant source. Thus, the simulations demonstrate that broad invasions of new deep roots and open root macropore channels by reforestation can be at risk of accelerating aquifer contamination with multiple legacy and future sources of surface pollutions such as widespread residual pesticides and PFAS in soils and atmospheric deposition6,47,73,79,80. This in turn, may lead to drinking water scarcity in areas relying on groundwater which already struggle to comply with drinking water quality standards due to such contamination6,74. To avoid such damage, attention needs to be paid on the land-use history and potential PMOC soil contaminant sources prior to and after reforesting when this is located above drinking water resources. To further illustrate the implications of the findings in Fig. 4 and simulations in Fig. 5B, C on overall aquifer protection strategies, Fig. 6 shows the impact from a PMOC pollution source located in three separate aquifer areas (in a hypothetical aquitard-aquifer system), each with different aquifer recharge controlled by different degrees of aquitard lithological heterogeneity and glaciotectonic deformation underneath rooting depth (i.e., Ksm = 10−10, 10−9, and 10−8 m/s by piston flow adopted from Fig. 5B, C and corresponding to aquifer recharge rates of 2, 20 and 180 mm/y, respectively). The simulations demonstrate that below depths of hydraulically active root macropores, the PMOC impact (i.e., breakthrough time and fluxes) to the aquifer changes strongly as function of aquitard lithological heterogeneity and glacial deformation. Notice that possible slow degradation of DMS73 (and other similar pesticide compounds) and time-variable preferential flow (e.g., stormflow events, see Supplementary Fig. 6) in the aquitard layer have the effect of increasing the contrast of impact between the three scenarios (addressed in detail by Jørgensen et al.44 and Stenemo et al.45). The simulations imply that targeted regulation and remediation of surface pollution in areas with aquitard thickness less than rooting depths (Fig. 2A, 4C) and in areas with identified lithological heterogeneity and aquitard leakage44,45 into aquifers (i.e., Fig. 6Bc) can markedly assist to improve groundwater quality in underlying aquifers and keep contaminant concentrations in water supply wells below drinking water standards. This, in turn, can prevent the closure of drinking-water wells and the associated large social and economic damage.Fig. 6Schematic illustration of aquifer vulnerability to a persistent mobile pesticide contaminant source (PMOC) in glacial aquitard layer underneath the depth of open root macropores (based on conceptual diagram in Fig. 2B and simulations in Fig. 5B, C).A Groundwater catchment with three scenarios with same surface infiltration but different aquifer recharge (2 mm/yr, 20 mm/yr, and 180 mm/yr) the latter representing the different degrees of matrix domain hydraulic conductivity (Ks = 10−10, 10−9, and 10−8 m/s) and fracture effective apertures 2beff ≤ 5 µm in Fig. 4C. B Cross section (X-X’) of aquitard-aquifer system showing evolution of relative PMOC flux over time into the aquifer simulated for the three scenarios of aquifer recharge (a, b, c). Note the large contrast of PMOC arrival times and impact to the aquifer between the three scenarios. Influence of typical differences in vertical hydraulic gradients between sub-areas (not shown) are marginal relative to the impact of the displayed three orders of magnitude difference in aquitard hydraulic conductivity. Schematic illustration of aquifer vulnerability to a persistent mobile pesticide contaminant source (PMOC) in glacial aquitard layer underneath the depth of open root macropores (based on conceptual diagram in Fig. 2B and simulations in Fig. 5B, C). A Groundwater catchment with three scenarios with same surface infiltration but different aquifer recharge (2 mm/yr, 20 mm/yr, and 180 mm/yr) the latter representing the different degrees of matrix domain hydraulic conductivity (Ks = 10−10, 10−9, and 10−8 m/s) and fracture effective apertures 2beff ≤ 5 µm in Fig. 4C. B Cross section (X-X’) of aquitard-aquifer system showing evolution of relative PMOC flux over time into the aquifer simulated for the three scenarios of aquifer recharge (a, b, c). Note the large contrast of PMOC arrival times and impact to the aquifer between the three scenarios. Influence of typical differences in vertical hydraulic gradients between sub-areas (not shown) are marginal relative to the impact of the displayed three orders of magnitude difference in aquitard hydraulic conductivity. In summary, this study provides rich multidisciplinary evidence that deep root macropores from ancient indigenous forests and modern trees and shrubs, together with aquitard lithological heterogeneity (related to different glacial till types), define aquifer vulnerability beneath glacial aquitards across Denmark. Contrary to current understanding, fractures per se (i.e., without containing root macropores) play a marginal or negligible role for aquitard flow and dissolved contaminant transport. Instead, widespread formation of root macropores and root channels inside fractures dominate vertical water flow and contaminant transport to depths from 1–2 mbgs to more than 10 mbgs based on observed tree rooting depths in this study. The deep tree roots can compromise natural groundwater protection by opening fractures, re-open deteriorated historical macropores, and expanding new macropore structures. These selectively connect (searching root hydrotropism and dimorphism) with deep water tables developed by aquitard leakage into underlying aquifers, which can be reinforced by groundwater pumping. The formation of such new deep root channel systems by reforestation efforts can significantly accelerate groundwater contamination at sites with legacy soil contamination from previous land use and future pollutions such as multiple source pesticides and PFAS chemicals. Due to absence of deep open fractures per se, groundwater flow below tree rooting depths is to a much higher degree than generally assumed controlled by lithological heterogeneity (e.g., coarse-grained aquitard matrix, silt and sand layers and lamina, and glacio-tectonites). We propose that the mechanisms of flow and contaminant transport uncovered in this study are of key-relevance for understanding groundwater vulnerability and quality beneath glacial aquitards in vast regions of the Northern Hemisphere because of similarity between glacial aquitards and extension of past indigenous forests across continents. Furthermore, the study approach and findings have relevance to other unlithified aquitards such as clays, silts, and saprolites, occurring widely beyond the Pleistocene glaciated areas, in which deep relic root macropores have also been reported and where the same fundamental flow and transport processes control contaminant leakage to underlying aquifers. Methods Study sites and overview of tests The glacial till types considered include basal tills (8 sites), terminal moraine (1 site), hummocky till (1 site), and melt-out tills (2 sites) of late Weichselian age (approximately 22–16 thousand years BP54) which are representative of typical glacial tills described widely elsewhere1,2. Table 1 summarizes the set of LUC laboratory tests and in-situ field tests carried out and analyzed in this study, as well as the major features of geology, hydrology, and land-use at the study sites. The glacial tills had 8–24% clay (Table 1) with clay mineral composition including smectite, vermiculite, and mixed-layer clay as the dominant clay minerals together with non-swelling clay types: illite, kaolinite, and chlorite. In surficial weathered tills, chlorite and iron sulfide minerals were typically absent due to weathering9,55. The tills have relatively low plasticity and measured undrained shear strengths from 50 to >365 kN/m42,50,55 depending on till type and the latter typically increasing with depth of measurement. Postglacial isostatic rebound in the study area varies from 0 to approximately 0.5 mm/y.Table 1Key characteristics of study sites and conducted LUC laboratory and in-situ field hydraulic testsStudy site, land-use, and source of dataA = Arable land,F = Forest/Orchard,O = OtherNumber and method of testsTransport of tracers and contaminants investigated in testsType of glacial till (depth of transition from oxidized to unoxidized till, mbgs)Content of clay in tills (%).Seasonal fluctuations of water tables, (depths of tile drains), and [observed depths of relic root macropores] in mbgsBulbro (A) this study and553 field infiltration experimentsBrilliant Blue (BB) dye tracerBasal till above Paleocene clay and limestone aquifer[1.5 - 1.8]> 121–1.8, (0.5–1), [>3.2mbgs]Farre (A) this study and34–36,394 LUC experimentsPFAS (PFBA), Mecoprop (MCPP), orthophosphate, BBBasal till above sandy aquifer [> 4]18–200.5 - > 4, (1)Flakkebjerg1 (A) this study and1043 LUC experimentsNi, Zn, Cd, Cu, Pb, bromide, BBBasal till over sandy aquifer [4.5]18–251– 4.5a,(1), [5.2]Flakkebjerg2 (A) this study and509 Field infiltration experimentsGlyphosate, AMPA, MCPP, MCPA, bentazon, pendimethalin, propyzamide, azoxystrobin, bromide and BB and UranineMelt-out silty till above lodgement clayey till above sand aquifer [3.8]10–130.5–5.2(1), [5.2b]Grundfør (A)42,432 LUC experimentsMCPP, MCPA, Nitrate, bromide and BBMelt-out till above sandy aquifer (only oxidized) [>3]20–220.5 - > 3, (1), [3]Havdrup (A/F, Apple trees) this study and9,15,19,32,41,45,466 LUC and 3 field infiltration experimentsMCPP, Simazine, prochloraz, metsulfuron-methyl, bromide, BB and UranineBasal till above sandy aquifer and limestone aquifer [2.5–3]18–221–6,(1), [5 to 6]Hinnerup (F, primarily Beech trees)17,42,43,1052 LUC experimentsNitrate, bromide, BBBasal till above sandy aquifer [2.8]20–221–3, (no data), [> 3.5]Holbæk (A) this study and17,493 LUC experimentsBentazone, MCPA, tebuconazole, bromide, BBHummocky till above marginal till [>3]18 –211 − 4.9, (no data), [3.3]Ringe (O)33,37,382 LUC experimentsCreosotes (DNAPL and solutes), nitrate bromide, BBBasal till above sandy aquifer [>4]12–131–3, (no data), [> 2]Salløv (A) this study and17,49,533 LUC and 9 field infiltration experimentsBentazone, MCPA, tebuconazole, bromide, BBBasal till above terminal till above Paleocene clay and limestone aquifer [3–8]13–190.5 - > 8, (0.5–1), [2.5 to > 6b]Skælskør1 (F, Cherry trees) this study and524 LUC experimentsDNAPL creosote, MCPP, dichlorprop, MCPA, 2.4D simazine, atrazine, terbuthylazine, bacteriophage, latex microspheres, BB and UranineBasal till above sandy aquifer [4.3]9–121–3, (no data), [3]Skælskør2 (O) this study and47,483 LUC experimentsMCPP, dichlorprop, MCPA, 2.4D simazine, atrazine, terbuthylazine, BB and UraninePostglacial freshwater layers on top of basal till above sandy aquifer[3.5–4]12–130.5–2,(0.75), [>2]aEstimated from depth of transition from oxidized (reddish brown matrix) to unoxidized clayey till (greyish matrix) (Fig. 2B).bObservation limited by depth of excavation. Table 1Key characteristics of study sites and conducted LUC laboratory and in-situ field hydraulic testsStudy site, land-use, and source of dataA = Arable land,F = Forest/Orchard,O = OtherNumber and method of testsTransport of tracers and contaminants investigated in testsType of glacial till (depth of transition from oxidized to unoxidized till, mbgs)Content of clay in tills (%).Seasonal fluctuations of water tables, (depths of tile drains), and [observed depths of relic root macropores] in mbgsBulbro (A) this study and553 field infiltration experimentsBrilliant Blue (BB) dye tracerBasal till above Paleocene clay and limestone aquifer[1.5 - 1.8]> 121–1.8, (0.5–1), [>3.2mbgs]Farre (A) this study and34–36,394 LUC experimentsPFAS (PFBA), Mecoprop (MCPP), orthophosphate, BBBasal till above sandy aquifer [> 4]18–200.5 - > 4, (1)Flakkebjerg1 (A) this study and1043 LUC experimentsNi, Zn, Cd, Cu, Pb, bromide, BBBasal till over sandy aquifer [4.5]18–251– 4.5a,(1), [5.2]Flakkebjerg2 (A) this study and509 Field infiltration experimentsGlyphosate, AMPA, MCPP, MCPA, bentazon, pendimethalin, propyzamide, azoxystrobin, bromide and BB and UranineMelt-out silty till above lodgement clayey till above sand aquifer [3.8]10–130.5–5.2(1), [5.2b]Grundfør (A)42,432 LUC experimentsMCPP, MCPA, Nitrate, bromide and BBMelt-out till above sandy aquifer (only oxidized) [>3]20–220.5 - > 3, (1), [3]Havdrup (A/F, Apple trees) this study and9,15,19,32,41,45,466 LUC and 3 field infiltration experimentsMCPP, Simazine, prochloraz, metsulfuron-methyl, bromide, BB and UranineBasal till above sandy aquifer and limestone aquifer [2.5–3]18–221–6,(1), [5 to 6]Hinnerup (F, primarily Beech trees)17,42,43,1052 LUC experimentsNitrate, bromide, BBBasal till above sandy aquifer [2.8]20–221–3, (no data), [> 3.5]Holbæk (A) this study and17,493 LUC experimentsBentazone, MCPA, tebuconazole, bromide, BBHummocky till above marginal till [>3]18 –211 − 4.9, (no data), [3.3]Ringe (O)33,37,382 LUC experimentsCreosotes (DNAPL and solutes), nitrate bromide, BBBasal till above sandy aquifer [>4]12–131–3, (no data), [> 2]Salløv (A) this study and17,49,533 LUC and 9 field infiltration experimentsBentazone, MCPA, tebuconazole, bromide, BBBasal till above terminal till above Paleocene clay and limestone aquifer [3–8]13–190.5 - > 8, (0.5–1), [2.5 to > 6b]Skælskør1 (F, Cherry trees) this study and524 LUC experimentsDNAPL creosote, MCPP, dichlorprop, MCPA, 2.4D simazine, atrazine, terbuthylazine, bacteriophage, latex microspheres, BB and UranineBasal till above sandy aquifer [4.3]9–121–3, (no data), [3]Skælskør2 (O) this study and47,483 LUC experimentsMCPP, dichlorprop, MCPA, 2.4D simazine, atrazine, terbuthylazine, BB and UraninePostglacial freshwater layers on top of basal till above sandy aquifer[3.5–4]12–130.5–2,(0.75), [>2]aEstimated from depth of transition from oxidized (reddish brown matrix) to unoxidized clayey till (greyish matrix) (Fig. 2B).bObservation limited by depth of excavation. Key characteristics of study sites and conducted LUC laboratory and in-situ field hydraulic tests Basal till above Paleocene clay and limestone aquifer [1.5 - 1.8] Farre (A) this study and 34–36,39 1– 4.5a, (1), [5.2] 0.5–5.2 (1), [5.2b] Grundfør (A) 42,43 1–6, (1), [5 to 6] Postglacial freshwater layers on top of basal till above sandy aquifer [3.5–4] 0.5–2, (0.75), [>2] aEstimated from depth of transition from oxidized (reddish brown matrix) to unoxidized clayey till (greyish matrix) (Fig. 2B). bObservation limited by depth of excavation. Rainstorms generate events of rapid infiltration and vertical migration of pesticides into the aquitards and shallow aquifers45,50 particularly during summer and build-up of water tables in autumn (see Supplementary Fig. 6) whereas major infiltration and groundwater recharge into underlying aquifers occurs under high water tables in winter30 where no or marginal water-uptake by trees and other vegetation is occurring. The agricultural sites have been cultivated for long time, in some cases (Bulbro, Salløv, Havdrup orchard site) since the Danish Bronze and Iron age58 before which they were occupied by dense indigenous forest. In the area of the Bulbro, Salløv, and Havdrup sites, intensive groundwater abstraction was initiated in 1951 which lowered hydraulic heads by 5–7 m in the regional aquifer78 beneath the aquitard and the water table by approximately 2 m in the aquitard at the Havdrup orchard site55 (occupied by apple trees from 1947 to 1987). Design and strategy of the research The study is built upon well-defined laboratory and field flow and integrated solute transport tests (i.e., controlled experiments) carried out at the study sites (Table 1) using the LUC and in-situ methods developed by Jørgensen et al.15,16 and Ouf et al.17. The study combines and integrates previous observations and new data. The latter include LUC and in-situ field tests at the Flakkebjerg 1 and 2, Skælskør 1 and 2, Salløv, Bulbro and Holbæk sites which are combined with the investigations of aDNA, radiocarbon dating, geochemical dating, TC-scanning and the groundwater modeling presented in this article. The new data are compiled with LUC flow and transport LUC data from previously published studies (Table 1) to synthesize the whole research. The number of LUC samples and in-situ experiments (and dimension of in-situ experiments) in each site were adapted to represent the individual geological profiles (Supplementary Figs. 1 and 2) with respect to geological layering and density of fractures and macropores (described further in the LUC and in-situ experiment’s method sections below). Synthesis of flow and transport investigations:The determination of vertical hydraulic conductivities (Ks) and effective fracture apertures (2beff) in LUC and in-situ field investigations in cited and new investigations (Fig. 4A, B, and 5Da) were based on the same procedures of hydraulic testing, dye tracer infiltration, and integrated solute transport across the study sites (listed in Table 1 and described in Supplementary Text, extended methods description). For previous experiments where 2beff values were not originally calculated36,47,49,50 or new parameter data have become available19 the 2beff were calculated according to Ouf et al.17 and Supplementary Text, extended methods description.The consistency between obtained 2beff values and observed conservative solute transport behavior (bromide and chloride) between experiments and field sites was tested and demonstrated by the match between discrete fracture-matrix (DFM) model simulations of flow and transport and observed flow and transport data in tests when using the determined hydraulic data (i.e., Ks, Ksm, and 2beff values) from the lab and field tests as DF-model key input parameters (Supplementary Figs. 28–30) and keeping all other model parameters constant (Supplementary Tables 5 and 6). The modeling studies confirmed that using the 2beff values and observed matrix Ksm values in DFM simulations provided consistent close approximations to observed flow and conservative transport in experiments with both steady-state and variable flow which was adjusted in the model (as the only calibration parameter) by using the hydraulic gradients applied in the experiments to describe variations in flow rates16,42,43,49 - thus demonstrating model consistency of the cause-effect relationship between fracture aperture and PMOC pesticide transport simulated in Figs. 5 and 6 for different solute and water fluxes.In selected experiments it was additionally tested and demonstrated that consistent reactive solute absorption and degradation parameter values could be estimated by DFM simulation when using the calibrated flow and conservative transport model set-up and adjusting sorption17,49 and degradation values43 simulated by the model to match both observed conservative and reactive transport simultaneously in LUC experiments with reactive solute42,43,49.The consistency between 2beff values obtained with the LUC and the in situ field method, respectively, was evaluated (and demonstrated in this study) by visual comparison of: (i) the 2beff range obtained with the two methods and (ii) comparison of 2beff values obtained with the LUC and in situ method in the same field experiment (Salløv INF2, BB5).Finally, the robustness of DFM modeling to upscaling from the LUC and in-situ test to field scale aquitard thickness and time scale (Figs. 5 and 6) was tested and confirmed in the modeling studies19,46. The determination of vertical hydraulic conductivities (Ks) and effective fracture apertures (2beff) in LUC and in-situ field investigations in cited and new investigations (Fig. 4A, B, and 5Da) were based on the same procedures of hydraulic testing, dye tracer infiltration, and integrated solute transport across the study sites (listed in Table 1 and described in Supplementary Text, extended methods description). For previous experiments where 2beff values were not originally calculated36,47,49,50 or new parameter data have become available19 the 2beff were calculated according to Ouf et al.17 and Supplementary Text, extended methods description. The determination of vertical hydraulic conductivities (Ks) and effective fracture apertures (2beff) in LUC and in-situ field investigations in cited and new investigations (Fig. 4A, B, and 5Da) were based on the same procedures of hydraulic testing, dye tracer infiltration, and integrated solute transport across the study sites (listed in Table 1 and described in Supplementary Text, extended methods description). For previous experiments where 2beff values were not originally calculated36,47,49,50 or new parameter data have become available19 the 2beff were calculated according to Ouf et al.17 and Supplementary Text, extended methods description. The consistency between obtained 2beff values and observed conservative solute transport behavior (bromide and chloride) between experiments and field sites was tested and demonstrated by the match between discrete fracture-matrix (DFM) model simulations of flow and transport and observed flow and transport data in tests when using the determined hydraulic data (i.e., Ks, Ksm, and 2beff values) from the lab and field tests as DF-model key input parameters (Supplementary Figs. 28–30) and keeping all other model parameters constant (Supplementary Tables 5 and 6). The modeling studies confirmed that using the 2beff values and observed matrix Ksm values in DFM simulations provided consistent close approximations to observed flow and conservative transport in experiments with both steady-state and variable flow which was adjusted in the model (as the only calibration parameter) by using the hydraulic gradients applied in the experiments to describe variations in flow rates16,42,43,49 - thus demonstrating model consistency of the cause-effect relationship between fracture aperture and PMOC pesticide transport simulated in Figs. 5 and 6 for different solute and water fluxes. The consistency between obtained 2beff values and observed conservative solute transport behavior (bromide and chloride) between experiments and field sites was tested and demonstrated by the match between discrete fracture-matrix (DFM) model simulations of flow and transport and observed flow and transport data in tests when using the determined hydraulic data (i.e., Ks, Ksm, and 2beff values) from the lab and field tests as DF-model key input parameters (Supplementary Figs. 28–30) and keeping all other model parameters constant (Supplementary Tables 5 and 6). The modeling studies confirmed that using the 2beff values and observed matrix Ksm values in DFM simulations provided consistent close approximations to observed flow and conservative transport in experiments with both steady-state and variable flow which was adjusted in the model (as the only calibration parameter) by using the hydraulic gradients applied in the experiments to describe variations in flow rates16,42,43,49 - thus demonstrating model consistency of the cause-effect relationship between fracture aperture and PMOC pesticide transport simulated in Figs. 5 and 6 for different solute and water fluxes. In selected experiments it was additionally tested and demonstrated that consistent reactive solute absorption and degradation parameter values could be estimated by DFM simulation when using the calibrated flow and conservative transport model set-up and adjusting sorption17,49 and degradation values43 simulated by the model to match both observed conservative and reactive transport simultaneously in LUC experiments with reactive solute42,43,49. In selected experiments it was additionally tested and demonstrated that consistent reactive solute absorption and degradation parameter values could be estimated by DFM simulation when using the calibrated flow and conservative transport model set-up and adjusting sorption17,49 and degradation values43 simulated by the model to match both observed conservative and reactive transport simultaneously in LUC experiments with reactive solute42,43,49. The consistency between 2beff values obtained with the LUC and the in situ field method, respectively, was evaluated (and demonstrated in this study) by visual comparison of: (i) the 2beff range obtained with the two methods and (ii) comparison of 2beff values obtained with the LUC and in situ method in the same field experiment (Salløv INF2, BB5). The consistency between 2beff values obtained with the LUC and the in situ field method, respectively, was evaluated (and demonstrated in this study) by visual comparison of: (i) the 2beff range obtained with the two methods and (ii) comparison of 2beff values obtained with the LUC and in situ method in the same field experiment (Salløv INF2, BB5). Finally, the robustness of DFM modeling to upscaling from the LUC and in-situ test to field scale aquitard thickness and time scale (Figs. 5 and 6) was tested and confirmed in the modeling studies19,46. Finally, the robustness of DFM modeling to upscaling from the LUC and in-situ test to field scale aquitard thickness and time scale (Figs. 5 and 6) was tested and confirmed in the modeling studies19,46. Large undisturbed column tests (LUC) The large undisturbed soil column method (LUC) uses a flexible-wall permeameter technique to simulate sub-soil confining pressures16 (Fig. 1Ca–c). The LUC method enables well-defined direct measurement of controlled flow and integrated contaminant transport behavior in fractured porous media while simulating in-situ confining soil stress and preventing fractures and macropores from opening or collapsing during lab procedures and sampling. The LUC samples were 0.5–1 m long columns with 0.5 m diameter. The columns were collected along prominent weathered fractures with several meters vertical extension (Supplementary Fig. 23). Sampling of the columns was carried out using hand tools to prevent disturbing present fractures in the columns. The columns were enclosed in a tight, prefabricated, flexible polyurethane membrane. The membrane was glued together with the column by injecting fluid membrane material between the LUC sample and the membrane in position and allowing the fluid to solidify. Before detaching the LUCs from the geological formation at their base, they were fixed with a split metal transport cylinder and tightened with clamps. The large undisturbed samples were carefully transported to the lab and prepared for installation in a permeameter. Before removing the transport cylinder, the columns were subjected to a mild vacuum (<4 kPa) to avoid fractures from opening when removing the enclosing pressure from the transport cylinder. The LUC samples were installed in triaxial flexible wall permeameters under isotropic soil stress resembling in-situ stress conditions for the testing of the columns (Fig. 1Ca, Supplementary Fig. 24, and Supplementary Table 3). Bulk water-saturated hydraulic conductivity (Ks) was determined for each LUC sample and, in selected experiments, for 70 mm intact cores sub-samples collected from the unfractured LUC matrix or in excavations immediately next to the LUC sampling. The hydraulic conductivity of the samples was calculated from the observed average volumetric flow obtained from application of more than two different hydraulic gradient across the samples (Supplementary Fig. 24) using: Ksat = Qtotal/(A ∙ i). Where Q = volumetric flow, A = sample cross sectional area, and i = hydraulic gradient32. Dye tracers (Brilliant Blue or Uranine, Supplementary Table 3) were injected over a defined time period to identify and characterize types of flow paths by dismantling the columns after the tests (Supplementary Fig. 24). Root macropore flow paths inside fractures were identified by direct observation of dyed flow channel paths along exposed fracture surface (e.g., Fig. 3C and Supplementary Figs. 16–20) or (LUC samples from Havdrup 1.5 and 2.5 mbgs and Skælskør1 2.5 mbgs) from geochemical evidence i.e., prominent greyish bleached fracture rims and fracture surfaces developed by Fe/Mn-oxide reduction from anaerobic degradation of root organic materials along the fractures23,43. In the Havdrup LUC sample from 4–4.5 mbgs root macropores were not originally inspected17 but inferred from observation of dyed relic root macropores inside fractures as the only type of fracture flow paths in the in-situ field tests carried out at the site19. Dyed flow paths along fractures, solitary macropore flow paths and lithological heterogeneities were characterized and mapped on transparent plastic sheds (Supplementary Figs. 25C and 27F) and by photographic documentation. Prior to dye tracer infiltration the LUC studies integrated investigations of dynamic transport behavior of multiple contaminants which were investigated in detail under well-defined controlled flow-through experiments (see Table 1 for contaminant compounds tested). The LUC experimental procedures are described in further detail in Jørgensen et al.16. Effective apertures (2beff) of fractures and fracture segments which contained biogenic macropores (indicated by preferential transport of dye tracer along fracture flow channels) and fractures without biogenic macropores were calculated separately using the cubic law for dyed fracture segments17 while assuming no flow in fractures with absence of dye tracer transport. Further description of the equivalent effective fracture aperture calculations is provided in Supplementary text, Extended methods description 1.1. In-situ field tests In-situ estimation (Fig. 1Cd–f) of 2beff was based on the technique of Ouf et al.17. The method enables 2beff estimation in single fractures and complex fracture systems based on numerical analysis of visually observed dye tracer penetration patterns17. Dye infiltration experiments at the field scale were carried out in excavated basins (Supplementary Figs. 25–27) located from 0.1–6 mbgs with surface areas increasing with depth from 1 m2 to 40 m2 (Supplementary Table 4) to cover typically overall decreasing densities of fractures containing root macropores with depth (Fig. 4A and D and Supplementary Figs. 7 and 27A). The location of the experiments in the geological profiles was selected to represent the most intensively fractured and macroporous till material in the investigated profiles (Fig. 4D). At the Flakkebjerg2 site, the flow paths were investigated in a continuum from the ground surface and throughout 6 m deep profiles (Supplementary Fig. 27C–F). At the other study sites the focus was on investigating connectivity of preferential flow paths across potential geological or geochemical boundaries within the geological profiles, which could obstruct flow (e.g., the transition from soil to sub-soil, from grey to red fractures, across sand layers, and from unsaturated to permanently water-saturated conditions in the profiles). The tills were naturally water-saturated or water-saturated by infiltrating water before or alongside the dye tracer infiltration experiments. Dye tracer experiments were carried out with 0.01–0.2 m ponding of the dye tracer solution in the infiltration basins. The infiltration basins were drained by horizontal wells installed beneath the basins or by natural sand layers beneath the basins. After dye tracers had been infiltrated, the infiltration experiments were immediately excavated. Root macropore flow paths inside fractures were identified by direct observation of dyed flow channel paths along exposed fracture surface (e.g., Fig. 3C, Supplementary Figs. 16–20, and Supplementary Video 1). Dyed flow paths along fractures, solitary macropore flow paths and lithological heterogeneities were mapped on transparent plastic sheds and recorded using photographic documentation (Supplementary Figs. 25 and 27). Quantification of macropore and fracture densities were carried out by manual measurements and counting except in Flakkebjerg2 field experiments (Supplementary Figs. 2B and 19E) in which counting was performed with automatized image analysis described in81. BB dye tracer penetration lengths along fractures in measured time were determined in sliced sections perpendicular to the fractures, inside fractures (Supplementary Fig. 18) or at arrival into underlying draining wells or natural sand layers. Effective fracture apertures in the experiments were estimated by numerical analysis of the observed dye tracer penetration along individual fractures (Supplementary Figs. 28–30) and breakthrough data of conservative tracers (Extended methods description 1.1 and Supplementary Fig. 31E) by following the procedure described in17. Furthermore, in the Havdrup and Flakkebjerg2 sites transport behavior of pesticide contaminants was studied and characterized in detail under natural and forced gradient field conditions (see Table 1 for chemical compounds tested). Numerical simulation of PMOC migration and aquifer vulnerability The migration of PMOCs (persistent and mobile organic compounds) through the clayey till into aquifers was simulated using a discrete fracture-matrix (DFM) numerical model in the modeling environment COMSOL Multiphysics82. The DFM model describes water flow in both discrete fractures and the matrix based on Darcy’s law, with advective and diffusive exchange between fractures and matrix. Advective, diffusive, and dispersive solute transport in fractures and matrix were described by the advection-dispersion equation, including linear sorption (DMS modeled in the current study exhibit no sorption (Supplementary Table 6)). We considered a cross-section of a 5 m thick glacial till aquitard (Fig. 5A) and describe the pesticide migration through the aquitard layer in discrete fractures and till matrix into an underlying aquifer. The simulations use a fixed fracture spacing of 1 m representing typical deep glaciotectonic and desiccation fractures observed in excavations across Denmark (Fig. 2B and Supplementary Fig. 22B) and a similar typical hydraulic gradient (0.5 m/m) observed across the aquitard layer44,47,83. The simulations assume no sorption or biodegradation of the PMOC. The model parameters used in the simulations are described in Supplementary Tables 5 and 6. The simulated concentrations of DMS infiltrating the aquitard in Fig. 5D represent DMS concentrations (0.23 to 0.20 µg/L) measured in tile drains underneath a past strawberry field 20 years after the primary fungicide source had been applied73 in a study site located nearby the Salløv site of this study and in the same glacial aquitard. The projected decline in DMS concentration in the infiltrating water Cinf (t) due to source depletion, degradation in top soil73, and plant uptake was estimated by linear extrapolation of the observed trend of the DMS concentrations in the tile drains73. The function used for the DMS input in the model was Cinf (t) = −0.023 µg/L ∙ t + 0.23 µg/L, where t represents the time in years over a period of 10 years, until Cinf reaches a concentration of 0 µg/L. Prior to simulations in this study the ability and uncertainty of DFM (including COMSOL), dual porosity (DP), and single continuum (EPM) models were studied and rigorously tested16,17,42,43,49 with respect to describing the flow and transport data obtained from the well-defined LUC and in-situ experiments in Table 1. The studies tested model descriptions of dynamic flow and transport, scale effects, solute properties (sorption, degradation, molecular diffusion), and solute exchange between preferential and matrix flow paths (i.e., flow and matrix diffusion). The outcomes demonstrated that DFM model type simulation provide close approximations of observed flow and solute transport in fractured and macroporous aquitards in the LUC and field experiments (see also Strategy of flow and transport investigations section above). Thus, the DFM models are valid for simulation of the cause-effect relationships and predictions of aquifer groundwater resource presented in the current study (Figs. 5 and 6). Notice, that in the comparison of modeling approaches carried out, the DP and EPM models simulated contaminant concentration errors up to three orders of magnitude when flow and degradation rates were modified with respect to the same base case (simulated correctly by all models) unless they were extensively recalibrated according to flow condition variations42,46. These errors were in particular related to the simplified numerical approach used in these models to describe flow and diffusive solute exchange between fractures/root macropore and matrix domains. For the watershed simulations in Fig. 6, uncertainty is particularly related to geological variability and overlapping degree of lithological heterogeneity between the exhibited different types of clayey till84 and the locations of aquitard leakage in watersheds, i.e., location of groundwater vulnerability areas44,45. Studies suggest that the aggregated impact of geological variability and leakage with respect to aquifer vulnerability can be derived from the variability of aquitard thickness combined with aquifer recharge across watersheds which in turn can be located by combination of geophysical surveying techniques and inverse hydrological modeling85,86. Ancient DNA (aDNA) analysis Fossil roots, root macropores, and soil matrix samples were collected under stringent contamination control protocols; tools and workspaces were cleaned with 5% bleach, 70% ethanol, DNA AWAY™ (Thermo Scientific), and UV irradiation. Up to 200 mg of fossil root material or 500 mg of soil-derived material was processed for DNA extraction using a modified protocol based on Wagner et al.87, including phenol/chloroform/isoamyl alcohol (25:24:1) and PB buffer (Qiagen) for inhibitor removal. Negative extraction controls were included for every six samples. A total of 32 samples (root fragments, sediments, and Fe/Mn oxide coatings) were processed, and aDNA was detected in 16 samples (Supplementary Table 1). Double-stranded DNA libraries were prepared using the BEST protocol88 and sequenced on an Illumina HiSeq4000 platform in 80 bp single-read mode. Sequencing was performed as untargeted shotgun sequencing without locus-specific PCR amplification or hybridization capture enrichment. Libraries were amplified using indexed universal adapter primers as described in the BEST protocol. Indexed libraries were purified with SPRI beads, and library quality was assessed with a Qubit 4 Fluorometer and a 2100 BioAnalyzer (Agilent). Negative controls accompanied all stages of library preparation and sequencing. Sequencing reads were processed with AdapterRemoval (v2.3.0)89 to trim adapters, merge reads, and remove sequences shorter than 25 bp, followed by removal of low-complexity sequences with SGA preprocess (v0.10.15) (dust > 1.0)90. Read mapping and taxonomic assignments were performed using Bowtie2 (v2.2.4)91 and the getLCA script92 (Supplementary Figs. 10 and 11). aDNA damage profiles were assessed with mapDamage (v2.2.0)93 (Supplementary Fig. 12). Mapped reads were sorted using SAMtools (v1.21)94, and consensus sequences were generated using ANGSD (v0.937)95 with stringent coverage thresholds. Sequences with >85% base coverage were aligned with MAFFT(v7.520)96, and phylogenetic trees were constructed under a GTR + G substitution model in MrBayes (v3.2.7a)97 and visualized in FigTree (v1.4.2)98 (Supplementary Figs. 13–15). Ribosomal RNA (rRNA) sequences were mapped to Alnus glutinosa subsp. glutinosa (GenBank acc. MF136526) and Malaisia scandens (GenBank acc. MH135781) as an outgroup), and were aligned together with additional Alnus rRNA sequences from Gryta et al.99, enabling reconstruction of evolutionary relationships using both chloroplast and rRNA markers. All mapped sequencing reads have been deposited on NCBI under BioProject accession PRJEB95955. Radiocarbon dating The 14C dating was carried out in parallel samples with the aDNA analysis (Supplementary Fig. 8G and Supplementary Tables 1 and 2). Plant macrofossil samples (roots) were pre-treated with acid-base-acid to remove carbonate and humic acids, rinsed with MilliQ water and freeze dried. All samples were then transferred to evacuated glass vials containing CuO for combustion to CO2, which were subsequently reduced using H2 to graphite. The graphite was then used for 14C analysis at the 1MV Tandetron accelerator at the Department of Physics, Aarhus University100. All 14C dates are reported as uncalibrated14C ages BP normalized to −25‰ according to international convention using online 13C/12C ratios101, where BP is before present and present is defined as 1950CE. The 14C ages are calibrated to calendar years using the calibration curve IntCal20102 and are reported as calibrated ages calBP. The online calibration program OxCal is used for calibrating the 14C ages into calendar years103. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Video 1 Reporting Summary Transparent Peer Review file Supplementary Information Description of Additional Supplementary Files Supplementary Video 1 Reporting Summary Transparent Peer Review file