Trends and shifts in groundwater levels across metropolitan France

Abstract

Groundwater is the principal buffer of France’s water security, yet its long-term response to intensifying climate variability and rising demand remains unresolved. We analyse 2000–2023/2024 in situ water-level records from 2536 piezometers across 717 aquifers, using Theil–Sen trend estimation with Mann–Kendall significance testing, and derive trend estimates at aquifer and city scales. A clear geographic divide emerges: aquifers in the south, southwest and Mediterranean fringe show widespread and locally steep declines, whereas northern and northwestern systems, including the Paris Basin, exhibit statistically significant but smaller-magnitude recoveries. Partitioning the record at 2015 reveals intensified deterioration in many southern and Mediterranean aquifers, while several large systems along the Atlantic margin and in northern France show incipient stabilisation. To our knowledge, this is the first high-resolution, in situ national assessment in metropolitan France to evaluate long-term groundwater trends and their evolution since 2015. These hotspot maps and multi-scale diagnostics can guide extraction limits, prioritise recharge efforts, and strengthen drought planning, setting a national benchmark for groundwater vulnerability under climate stress.

Groundwater, a cornerstone of global water security, sustains ecosystems, agriculture, and human populations worldwide¹. Its naturally low suspended solids, minimal microbial contamination, and stable mineral composition make it a reliable source of drinking water across diverse climatic and geological settings². Globally, depletion of this resource has become an escalating concern, driven by unsustainable withdrawals and changing climate patterns^1,3–6. Falling water tables reduce freshwater availability and disrupt baseflow to rivers, wetlands, and lakes, processes that are critical for maintaining aquatic habitats and buffering drought impacts⁷. Compounding these challenges, overexploitation also poses secondary risks, including irreversible compaction of aquifer systems, land subsidence^8–12, saltwater intrusion in coastal aquifers^13–16, and even contributes to global sea-level rise as depleted groundwater ultimately discharges into the oceans^17,18.

In France, groundwater plays an indispensable socioeconomic role, fulfilling 66% of domestic water demand, 31% of industrial requirements, and 37% of agricultural irrigation needs¹⁹. However, this reliance is threatened by climate change and rising demand, which exacerbate groundwater stress and undermine resilience to climatic fluctuations^20–23. During the 2022 European drought, over 100 French municipalities lost access to drinking water and had to be supplied via truck, while 66 departments reached the highest “crisis” level of drought alert²⁴. Such crises underscore groundwater’s critical role as a strategic reserve during droughts^25,26, yet managing this resource sustainably is complicated by France’s diverse hydrogeological landscape. Sedimentary basins like the Paris Basin exhibit slow recharge rates, karstic systems in Mediterranean regions are prone to rapid contamination, and alluvial plains such as the Rhône Valley experience pronounced seasonal variability. Each system demands regionally tailored strategies to balance extraction with recharge while preventing long-term depletion.

Despite these challenges, a comprehensive nationwide assessment of groundwater trends is absent from the scientific literature, with most studies focused on specific regions or short-term fluctuations. Recent national-scale work, such as the IGRAC well-level trend dataset²⁷, provides valuable baseline information but does not aggregate results into hydrological units or explore temporal shifts within the record. Satellite missions such as GRACE (Gravity Recovery and Climate Experiment)²⁸ allow coarse-scale tracking of terrestrial water storage, but their limited spatial resolution and vertical sensitivity hinder precise characterisation of individual aquifers²⁹. These constraints hinder full compliance with the European Water Framework Directive (WFD)³⁰, which requires quantitative status assessments and extraction limits to ensure sustainable use. This study addresses these gaps by analysing high-resolution, in situ piezometric records across metropolitan France’s hydrological regions and, to our knowledge, providing the first national assessment that aggregates long-term trend estimates at piezometer, aquifer, and city scales and quantifies a pre-/post-2015 shift in trends to identify where conditions are worsening, stabilising, or improving. Two key questions guide these analyses:

1. What are the most recent long-term trends in groundwater levels across metropolitan France over the period 2000–2024?

2. How have groundwater trends shifted over the past two decades, comparing conditions before 2015 with those since 2015?

Groundwater trend studies commonly quantify monotonic change using either parametric regression (e.g., ordinary least squares for linear trends), non-parametric rank-based methods (e.g., Mann–Kendall and Theil-Sen for robust trend detection for piezometric series), or change-point tests (e.g., Pettitt test³¹ for abrupt shifts in groundwater levels), depending on the question of interest and data characteristics. Here, we derive trend magnitude using the non-parametric Theil–Sen slope estimator and assess monotonic trend significance using Kendall’s tau-based testing, a combination widely applied in groundwater and hydrological trend assessments^{20,29,32–39}. These methods are preferred for long piezometric records because they are less sensitive to frequent outliers and departures from normality than parametric regression and remain applicable under irregular observations typical of groundwater monitoring networks^40–42.

The Bureau de Recherches Géologiques et Minières (www.brgm.fr) (BRGM; French Geological Survey) is France’s national authority for geological research and monitoring of subsurface resources, including groundwater. This study uses BRGM’s nationwide piezometric network to analyse groundwater levels. To assess trends at the aquifer scale, piezometers were spatially linked to hydrogeological units defined in the Base de Données des Limites des Systèmes Aquifères (bdlisa.eaufrance.fr.) (BDLISA; Database of Aquifer System Boundaries), the official reference for aquifer delineation in France, using the dataset’s pre-defined attribution indicating which piezometers are included in each unit. Using spatial and hydrogeological variability, these data provide a framework for evaluating groundwater status and informing national water policy.

Data

BRGM piezometric network

The original dataset included groundwater level records from 8093 piezometers distributed across metropolitan France. These piezometers are part of the national piezometric network managed by BRGM, and provide in situ measurements of groundwater levels. After removing piezometers with inadequate monitoring duration (see Methods), we retained 2,536 piezometers with sufficient temporal coverage, each with observations extending to 2023 or 2024, for the trend analysis. The spatial distribution of the original piezometer network and the final analysis subset is shown in Fig. 1a,b, respectively.

Fig. 1.

Spatial distribution of piezometers in metropolitan France from the BRGM network dataset. (a) All 8,093 piezometers in metropolitan France from the original BRGM data (black); (b) Subset of 2536 piezometers in metropolitan France meeting the filtering criteria (blue), retaining only those with a time frame extending into 2023 or 2024 and at least five years of data with observations across all four seasons.

Groundwater behaviour varies markedly across France due to diverse hydrogeological settings. For example, Fig. 2 presents time series from three randomly selected piezometers, illustrating heterogeneity in trends, periodicity, and amplitude.

Fig. 2.

Groundwater level time series from three randomly selected piezometers (BRGM piezometer codes: 00873X0036/F2, 05373X0546/5410, and 09368X0111/OULE), showing monthly mean water levels (in meters) over time.

BDLISA hydrogeological reference

The BDLISA is the national hydrogeological reference framework for France, designed to delineate and characterize aquifer systems across varying spatial scales. Version 3 of the BDLISA database provides an updated spatial and attribute-based representation of France’s hydrogeological units. It is structured into three hierarchical levels of utilization—local (niveau 3), regional (niveau 2), and national (niveau 1)—with increasing levels of aggregation. Each unit is characterized by lithology, aquifer nature (e.g., permeable or impermeable), aquifer state (e.g., confined or unconfined), and dominant hydrogeological environment (e.g., porous, fractured, karstic). The dataset also records the vertical stacking order of aquifers using the ordreabseh attribute (ordre = order; abseh = aquifer basin system ensemble horizon), which indexes units from top to bottom to resolve multi-layered systems in terms of their relative stacking order (lower values correspond to units higher in the stack and higher values to units lower in the stack), and does not represent physical depth. Delivered in standard GIS formats (Shapefile, GeoPackage, File Geodatabase), the database enabled spatial linkage of piezometers to aquifers. For this analysis, BRGM piezometers were assigned to BDLISA units at the local scale to support aquifer-based trend interpretation (see Methods for details). Figure 3 shows the spatial coverage of BDLISA aquifers that were included in the analysis, i.e., those to which piezometers were assigned.

Fig. 3.

Aquifer Coverage of Selected Piezometers. The 2536 selected piezometers are linked to mapped aquifer units at the local BDLISA scale. Blue shading indicates the 717 aquifers associated with these piezometers that were used for long-term analysis.

Piezometer-level long-term trend analysis

To address the first research question—“What are the most recent long-term trends in groundwater levels across metropolitan France over the period 2000–2024?”, we applied the Theil–Sen slope estimator (hereafter Sen’s slope), expressed in meters per year (m yr⁻¹), together with the Mann–Kendall test to 2536 selected piezometers. Each piezometer is shown as a point, with slope values mapped using a continuous color scale. Figure 4 presents both positive and negative long-term trends, each subdivided into statistically significant and non-significant categories. Statistical significance is defined as p ≤ 0.05, while p > 0.05 indicates non-significant results.

Fig. 4.

Spatial distribution of piezometers exhibiting statistically significant and non-significant groundwater trends based on Sen’s slope (m yr⁻¹) and Mann–Kendall tests. (a) Statistically significant negative trends (n = 1,052; Sen’s slope ≤ 0, p ≤ 0.05), with darker red indicating stronger declines. (b) Non-significant negative trends (n = 517; Sen’s slope ≤ 0, p > 0.05), with symbol size scaled by p. (c) Statistically significant positive trends (n = 509; Sen’s slope ≥ 0, p ≤ 0.05), with darker green indicating stronger increases. (d) Non-significant positive trends (n = 458; Sen’s slope ≥ 0, p > 0.05), where greener shades represent larger slope magnitudes; symbol size reflects p. The full dataset of piezometer trends is provided in Supplementary Table S1 (S1_piezometers_long_term.csv).

Figure 4a shows statistically significant negative trends in groundwater levels concentrated in southwestern France, along the Mediterranean coast, and in smaller clusters in central France. The most pronounced declines occur in the southwest and Mediterranean coastal areas. In contrast, Fig. 4c shows that positive trends are rare in the south and appear mainly as scattered clusters in the west and southwest. Their magnitudes are generally lower than those observed for significant negative trends. The overall pattern is one of marked spatial contrast, with contiguous areas dominated by declining slopes and only isolated pockets of increase. Figures 5 and 6 present these results separately by region, illustrating statistically significant negative and positive trends, respectively. Additionally, Table 1 summarises these results for each region, reporting the mean and median Sen’s slope values along with the proportion of piezometers exhibiting significant negative, significant positive, and non-significant trends.

Fig. 5.

Number of piezometers exhibiting statistically significant negative groundwater trends (Sen’s slope < 0 m yr⁻¹, p ≤ 0.05) by region and slope class. Each row corresponds to a metropolitan region of France, and each column represents a Sen’s slope interval.

Fig. 6.

Number of piezometers exhibiting statistically significant positive groundwater trends (Sen’s slope ≥ 0 m yr⁻¹, p ≤ 0.05) by region and slope class. Each row corresponds to a metropolitan region of France, and each column represents a Sen’s slope interval.

Table 1.

Summary statistics of long-term groundwater level trends from piezometers in each region of metropolitan France.

Region	No. of piezometers	Mean of Sen’s slopes	Median of Sen’s slopes	Median of negative sen’s slopes	Median of positive Sen’s slopes	% Negative significant	% Positive significant	% Non−significant
Auvergne-Rhône-Alpes	185	−0.0204	−0.0139	−0.0308	0.0122	55.14	14.59	30.27
Bourgogne-Franche-Comté	107	0.0031	−0.0055	−0.0193	0.0199	31.78	23.36	44.86
Bretagne	53	0.0069	−0.0008	−0.0055	0.0149	5.66	15.09	79.25
Centre-Val de Loire	203	−0.0676	−0.0528	−0.0616	0.0184	78.33	6.9	14.78
Corse	25	−0.0117	−0.0062	−0.0261	0.0263	40	32	28
Grand Est	289	−0.0096	−0.0056	−0.0172	0.0125	43.94	17.65	38.41
Hauts-de-France	190	−0.0342	−0.0205	−0.0369	0.0208	47.89	11.58	40.53
Normandie	130	−0.0499	−0.0152	−0.0437	0.0157	46.92	14.62	38.46
Nouvelle-Aquitaine	665	−0.0162	0.0028	−0.0312	0.021	27.37	27.82	44.81
Occitanie	302	−0.0785	−0.011	−0.0312	0.0144	46.69	18.21	35.1
Pays de la Loire	205	0.0161	−0.0004	−0.0281	0.0303	32.68	26.34	40.98
Provence-Alpes-Côte d’Azur	124	−0.0571	−0.0103	−0.0227	0.0175	45.97	16.13	37.9
Île-de-France	55	0.0374	0.0063	−0.0216	0.0544	30.91	36.36	32.73

Metrics include the number of piezometers, mean and median Sen’s slope (m yr⁻¹) values, medians for positive and negative Sen’s slopes, and the percentage of piezometers showing statistically significant negative trends, statistically significant positive trends, and non-significant trends.

Viewing Figs. 4, 5, and 6 together, significant increases are concentrated and generally modest. Nouvelle-Aquitaine dominates the positive counts (largest totals in 0.0–0.1, tapering through 0.1–0.4 and extending sparsely up to 0.8–0.9), with smaller clusters in Grand Est, Pays de la Loire, Occitanie, and Auvergne-Rhône-Alpes; most other regions register only a few positives and rarely above + 0.4, and Corse shows none. By contrast, significant declines are numerous and widespread. Centre-Val de Loire and Grand Est record the highest totals in − 0.1–0.0 and substantial counts in − 0.2 to − 0.1, with large regional burdens also evident in Nouvelle-Aquitaine, Occitanie, Bourgogne–Franche-Comté, Hauts-de-France, and Provence-Alpes-Côte d’Azur. Occitanie is notable for declines spanning almost the full negative range, including entries beyond − 0.6. Several regions, including Occitanie, exhibit declines extending into stronger bins (≤ − 0.5), with isolated cases occurring at ≤  − 1.0. Even regions with few positives, such as Bretagne and Corse, display negative trends, indicating that declines are not only more prevalent but also more widely distributed and more likely to reach higher magnitudes than positive changes.

Piezometers were assigned to the cities in which they are located, with Table 2 listing those with negative median Sen’s slopes and Table 3 showing those with positive values; in both cases, only piezometers with statistically significant trends (p ≤ 0.05) were considered. Cities with the steepest negative piezometer medians are spread across multiple French regions but are especially concentrated in the south. The strongest decline is in Saignon (Vaucluse, Provence-Alpes-Côte d’Azur, − 3.69), followed by Fontès (Hérault, Occitanie, − 2.42), Avrechy (Oise, Hauts-de-France, − 1.63), Cabrières (Hérault, Occitanie, − 1.29), and Valliguières (Gard, Occitanie, − 1.21). Occitanie accounts for many of these top declines, including Coustaussa (Aude, − 1.12), Rivesaltes (Pyrénées-Orientales, − 1.07), Vailhan (Hérault, − 0.95), and Faugères (Hérault, − 0.80), as well as Sainte-Colombe-de-la-Commanderie (Pyrénées-Orientales, − 0.68).

Table 2.

Cities with the steepest negative long-term Sen’s slope (m yr⁻¹) trends (2000–2023/2024).

Order	City_or_commune	Region	Department	Median Sen’s slope	No. piezometers
1	Saignon	Provence-Alpes-Côte d’Azur	Vaucluse	−3.69	1
2	Fontès	Occitanie	Hérault	−2.42	1
3	Avrechy	Hauts-de-France	Oise	−1.63	1
4	Cabrières	Occitanie	Hérault	−1.29	1
5	Valliguières	Occitanie	Gard	−1.21	1
6	Ombrée d’Anjou	Pays de la Loire	Maine-et-Loire	−1.12	1
7	Coustaussa	Occitanie	Aude	−1.12	1
8	Rivesaltes	Occitanie	Pyrénées-orientales	−1.07	1
9	Savignac-sur-Leyze	Nouvelle-Aquitaine	Lot-et-Garonne	−0.96	1
10	Vailhan	Occitanie	Hérault	−0.95	1
11	Pécorade	Nouvelle-Aquitaine	Landes	−0.89	1
12	Castelnau-Tursan	Nouvelle-Aquitaine	Landes	−0.87	1
13	Mussidan	Nouvelle-Aquitaine	Dordogne	−0.8	1
14	Faugères	Occitanie	Hérault	−0.8	1
15	Valréas	Provence-Alpes-Côte d’Azur	Vaucluse	−0.79	1
16	Burosse-Mendousse	Nouvelle-Aquitaine	Pyrénées-atlantiques	−0.78	1
17	Lannecaube	Nouvelle-Aquitaine	Pyrénées-atlantiques	−0.73	2
18	Sainte-Colombe-de-la-Commanderie	Occitanie	Pyrénées-orientales	−0.68	1
19	Montpon-Ménestérol	Nouvelle-Aquitaine	Dordogne	−0.67	1
20	Chennebrun	Normandie	Eure	−0.67	1
21	Gabian	Occitanie	Hérault	−0.66	1
22	Plaissan	Occitanie	Hérault	−0.66	1
23	Geaune	Nouvelle-Aquitaine	Landes	−0.63	2
24	Dému	Occitanie	Gers	−0.61	1
25	Idrac-Respaillès	Occitanie	Gers	−0.6	1
26	Mourèze	Occitanie	Hérault	−0.58	1
27	Cournonsec	Occitanie	Hérault	−0.53	1
28	Pignan	Occitanie	Hérault	−0.52	1
29	Polastron	Occitanie	Gers	−0.51	1
30	Feugarolles	Nouvelle-Aquitaine	Lot-et-Garonne	−0.51	1
31	Orée d’Anjou	Pays de la Loire	Maine-et-Loire	−0.5	1
32	Broût-Vernet	Auvergne-Rhône-Alpes	Allier	−0.5	1
33	Eugénie-les-Bains	Nouvelle-Aquitaine	Landes	−0.5	2
34	Eysines	Nouvelle-Aquitaine	Gironde	−0.5	1
35	Passage	Nouvelle-Aquitaine	Lot-et-Garonne	−0.47	1
36	Agen	Nouvelle-Aquitaine	Lot-et-Garonne	−0.46	1
37	Réaup-Lisse	Nouvelle-Aquitaine	Lot-et-Garonne	−0.46	1
38	Saint-Rémy-sur-Avre	Centre-Val de Loire	Eure-et-Loir	−0.43	1
39	Lagleygeolle	Nouvelle-Aquitaine	Corrèze	−0.42	1
40	Lagruère	Nouvelle-Aquitaine	Lot-et-Garonne	−0.42	1

Shown are the city, its department, and region, along with the median Sen’s slope of all piezometers within the city boundary that exhibit statistically significant trends (p ≤ 0.05). The complete list of 928 cities is provided in Supplementary Table T2 (T2_cities_negative_significant_long_term.csv).

Table 3.

Cities with the steepest positive long-term Sen’s slope (m yr⁻¹) trends (2000–2023/2024).

Order	City_or_commune	Region	Department	Median Sen’s slope	No. piezometers
1	Saint-Philbert-du-Peuple	Pays de la Loire	Maine-et-Loire	0.99	1
2	Nieul-lès-Saintes	Nouvelle-Aquitaine	Charente-Maritime	0.89	1
3	Saint-Julien-d'Armagnac	Nouvelle-Aquitaine	Landes	0.74	1
4	Paris	Île-de-France	Paris	0.67	1
5	Vaux-sur-Mer	Nouvelle-Aquitaine	Charente-Maritime	0.63	1
6	Avy	Nouvelle-Aquitaine	Charente-Maritime	0.59	1
7	Bassens	Nouvelle-Aquitaine	Gironde	0.58	1
8	Tourtoirac	Nouvelle-Aquitaine	Dordogne	0.58	1
9	Arces	Nouvelle-Aquitaine	Charente-Maritime	0.57	1
10	Isle-Adam	Île-de-France	Val-d'Oise	0.51	1
11	Lille	Hauts-de-France	Nord	0.5	1
12	Campbon	Pays de la Loire	Loire-Atlantique	0.49	1
13	Clisse	Nouvelle-Aquitaine	Charente-Maritime	0.4	1
14	Chapelle-Bâton	Nouvelle-Aquitaine	Vienne	0.4	1
15	Saint-Georges-des-Coteaux	Nouvelle-Aquitaine	Charente-Maritime	0.38	2
16	Scientrier	Auvergne-Rhône-Alpes	Haute-Savoie	0.36	1
17	Cessey-sur-Tille	Bourgogne-Franche-Comté	Côte-d'Or	0.35	1
18	Bordes	Nouvelle-Aquitaine	Pyrénées-Atlantiques	0.34	3
19	Saint-Bris-des-Bois	Nouvelle-Aquitaine	Charente-Maritime	0.32	1
20	Trets	Provence-Alpes-Côte d’Azur	Bouches-du-Rhône	0.32	2
21	Alligny-Cosne	Bourgogne-Franche-Comté	Nièvre	0.31	1
22	Allons	Nouvelle-Aquitaine	Lot-et-Garonne	0.31	1
23	Beaufort-en-Anjou	Pays de la Loire	Maine-et-Loire	0.31	1
24	Fontcouverte	Nouvelle-Aquitaine	Charente-Maritime	0.3	1
25	Teich	Nouvelle-Aquitaine	Gironde	0.3	1
26	Baugé-en-Anjou	Pays de la Loire	Maine-et-Loire	0.28	2
27	Orée d’Anjou	Pays de la Loire	Maine-et-Loire	0.28	2
28	Montsoreau	Pays de la Loire	Maine-et-Loire	0.27	1
29	Arthaz-Pont-Notre-Dame	Auvergne-Rhône-Alpes	Haute-Savoie	0.26	1
30	Mauges-sur-Loire	Pays de la Loire	Maine-et-Loire	0.26	2
31	Valencisse	Centre-Val de Loire	Loir-et-Cher	0.25	1
32	Losse	Nouvelle-Aquitaine	Landes	0.25	1
33	Bergerac	Nouvelle-Aquitaine	Dordogne	0.24	1
34	Noyant-Villages	Pays de la Loire	Maine-et-Loire	0.24	2
35	Puttelange-aux-Lacs	Grand Est	Moselle	0.24	1
36	Mazé-Milon	Pays de la Loire	Maine-et-Loire	0.23	2
37	Labruguière	Occitanie	Tarn	0.23	1
38	Cholet	Pays de la Loire	Maine-et-Loire	0.23	3
39	Tour-d'Auvergne	Auvergne-Rhône-Alpes	Puy-de-Dôme	0.23	1
40	Lempzours	Nouvelle-Aquitaine	Dordogne	0.22	1

Shown are the city, its department, and region, along with the median Sen’s slope of all piezometers within the city boundary that exhibit statistically significant trends (p ≤ 0.05). The complete list of 465 cities is provided in Supplementary Table T3 (T3_cities_positive_significant_long_term.csv).

Nouvelle-Aquitaine also accounts for many steep declines, including Savignac-sur-Leyze (Lot-et-Garonne, Nouvelle-Aquitaine, − 0.96), Pécorade (Landes, Nouvelle-Aquitaine, − 0.89), Castelnau-Tursan (Landes, Nouvelle-Aquitaine, − 0.87), Burosse-Mendousse (Pyrénées-Atlantiques, Nouvelle-Aquitaine, − 0.78), Lannecaube (Pyrénées-Atlantiques, Nouvelle-Aquitaine, − 0.73), Mussidan (Dordogne, Nouvelle-Aquitaine, − 0.80), and Montpon-Ménestérol (Dordogne, Nouvelle-Aquitaine, − 0.67). These communes lie within the Adour–Garonne and Aquitaine sedimentary basins, where groundwater is heavily exploited for irrigation and climate-driven declines in recharge exacerbate stress on alluvial aquifers^43,44. A few scattered steep entries occur outside this basin, including Ombrée d’Anjou (Maine-et-Loire, Pays de la Loire, − 1.12) in the west, Valréas (Vaucluse, Provence-Alpes–Côte d’Azur, − 0.79) in the southeast near the Rhône Valley, and Chennebrun (Eure, Normandie, − 0.67) in the northwest. In addition, along the border of Normandie and Centre-Val de Loire, and more broadly across much of Centre-Val de Loire, many cities also register mild negative trends. However, these declines are much less steep than those observed in southern France.

Negative median Sen’s slopes are scattered throughout France, but the steepest long-term declines concentrate in the south and southwest. They form a broad belt running from the Mediterranean coast of Occitanie through the inland valleys and coastal plains of Nouvelle-Aquitaine and into the hills and lowlands of Provence-Alpes-Côte d’Azur, spanning the Languedoc coast, the Aquitaine Basin, and the Pyrenean foothills. Farther north and northwest, declines also appear in Avrechy (Oise, Hauts-de-France), Chennebrun (Eure, Normandie), and Ombrée d’Anjou (Maine-et-Loire, Pays de la Loire), though they are fewer and less intense, leaving the southern half of the country as the area most affected by steep reductions. This pattern reflects the Mediterranean climate of Provence-Alpes-Côte d’Azur and southern Occitanie, where hot, dry summers and concentrated winter rainfall restrict recharge and heighten drought stress, while the north and Atlantic west benefit from a temperate oceanic climate with cooler summers and more evenly distributed precipitation. Population pressures exacerbate this divide: dense urban and peri-urban centers, such as Marseille, Montpellier, and Toulouse, intensify groundwater demand in the south, whereas northern regions remain less densely populated outside major cities like Paris and Lille.

In comparison with Table 2, cities with positive long-term Sen’s slopes in Table 3 are smaller in absolute value than the negative ones, so recoveries remain modest relative to the steep southern declines. The largest increases occur in Saint-Philibert-du-Peuple (Maine-et-Loire, Pays de la Loire, + 0.99), Nieul-lès-Saintes (Charente-Maritime, Nouvelle-Aquitaine, + 0.89), Saint-Julien-d’Armagnac (Landes, Nouvelle-Aquitaine, + 0.74), and Paris (Île-de-France, + 0.67). In western France, improvements appear as localized pockets rather than a continuous swath: several communes in Charente-Maritime—Nieul-lès-Saintes, Vaux-sur-Mer (+ 0.63), Avy (+ 0.59), and Arces (+ 0.57)—and nearby Tourtoirac (Dordogne, + 0.57) and Bassens (Gironde, + 0.58). Similar modest long-term gains are also visible in most cities of Île-de-France. Overall, positives are modest and dispersed, but they signal areas where groundwater levels show encouraging signs of stability, with most of the higher positive slopes falling below about + 1.0 and commonly in the + 0.2 to + 0.6 range, whereas negative slopes extend much further, often between − 0.5 and − 2.0 and reaching below − 3.0 in the steepest cases.

Aquifer-level long-term trend analysis

The preceding section examined piezometer-level trends to identify regions potentially affected by water scarcity. Here, the analysis is extended to the aquifer scale to capture broader spatial patterns of potential risk. Details of the aquifer-level trend calculations are provided in the Methods, and the resulting Sen’s slope heatmap for aquifers is shown in Fig. 7. Because multiple aquifers overlap spatially, some units lower in the stack are not visually distinguishable in this representation; therefore, we focus on the 20 aquifers with the strongest negative and the 20 aquifers with the strongest positive trends in Figs. 8 and 9, which allows clear identification of the most strongly trending aquifers.

Fig. 7.

Aquifer-Level Heatmap of Sen’s Slope (m yr⁻¹) Trends Across Metropolitan France. (a) Sen’s slope values for 717 aquifers. (b) Cube root–normalized Sen’s slope values to enhance visual contrast. Aquifers are arranged vertically by ordreabseh, with shallower units (low ordreabseh) at the top and deeper units (high ordreabseh) below. Colors indicate positive long-term trends in green, stable conditions in yellow, and negative long-term trends in red. The complete dataset underlying this figure is provided in Supplementary Table S2 (S2_BDLISA_all_aquifers_long_term.csv).

Fig. 8.

Twenty aquifers with the most negative Sen’s slope (m yr⁻¹) values, representing the strongest declines in groundwater levels. (a) Sen’s slope values with corresponding p-values and (b) the spatial distribution of these twenty aquifers. Darker red shades denote more negative slopes, indicating greater rates of decline.

Fig. 9.

Twenty aquifers with the most positive Sen’s slope (m yr⁻¹) values, representing the strongest increases in groundwater levels. (a) Sen’s slope values with corresponding p-values and (b) the spatial distribution of these twenty aquifers. Darker green shades denote higher slopes, indicating more pronounced recovery.

Figure 7 reveals most negative aquifer trends (indicated by red shading), particularly in northern, central, and southern coastal regions near Montpellier and Marseille, where significant declines in groundwater levels have been observed. These reductions are likely driven by excessive abstraction, especially in densely populated urban centers and agriculturally intensive zones, compounded by limited recharge due to climate variability^45–47. For instance, Montpellier and Marseille, both highly urbanized areas within the Provence-Alpes-Côte d’Azur region, experience substantial groundwater extraction, further straining the region’s resources. Prolonged drought conditions in these regions have also diminished natural recharge, exacerbating the decline.

Figure 8 shows the 20 aquifers with the most severe negative trends, with Sen’s slope values ranging from − 3.69 to − 0.22. These aquifers are concentrated in regions of high water demand, such as the south, southwest, and central areas, including regions like Provence-Alpes-Côte d’Azur, Nouvelle-Aquitaine, and Occitanie. However, Île-de-France is not included among the most severely impacted aquifers, indicating that its groundwater trends are more stable. In contrast, the central, southern, and southwestern aquifers face the most alarming imbalances, where high abstraction rates persist despite limited recharge, worsened by climate-induced factors like reduced rainfall and higher evaporation rates, as reported in earlier work⁴⁵. Table 4 details the BDLISA properties and attribution of these 20 aquifers.

Table 4.

Twenty aquifers with the strongest negative long-term Sen’s slope (m yr⁻¹) values.

Order	Codeeh	Ordreabseh	Niveaueh	Incluseh	Etateh	Natureeh	Milieueh	Themeeh	Origineeh	Sen’s Slope	p	Lower CI	Upper CI	Area_km2
1	561AB00	99	3	561AB	2	6	1	2	3	−3.69	0.02025891	−7.21	−0.6	361.606978
2	681AB01	200	3	681AB	2	7	1	4	1	−0.97	8.69E−08	−1.37	−0.61	201.963643
3	334AE01	15,030,100	3	334AE	X	7	1	2	1	−0.9	2.42E−28	−1.05	−0.76	5637.64472
4	657AB00	200	3	657AB	3	6	1	2	2	−0.54	2.72E−25	−0.62	−0.46	661.723237
5	113AL07	12,130	3	113AL	3	6	4	2	1	−0.5	4.79E−29	−0.56	−0.43	2068.15294
6	657AC00	200	3	657AC	3	6	1	2	2	−0.48	1.99E−09	−0.62	−0.32	148.580005
7	681AB02	200	3	681AB	2	5	3	4	1	−0.4	7.38E−16	−0.5	−0.31	120.863158
8	334AG13	15,040,210	3	334AG	1	7	1	2	1	−0.37	9.63E−101	−0.38	−0.36	9155.56319
9	334AC05	15,020,100	3	334AC	1	5	1	2	1	−0.37	2.22E−38	−0.41	−0.32	7383.84914
10	358AE03	27,040,100	3	358AE	3	5	8	2	1	−0.37	2.65E−59	−0.41	−0.33	27,293.7054
11	534AH00	150	3	534AH	2	5	3	2	2	−0.34	0.0038433	−0.57	−0.11	352.720043
12	356AB01	26,020,100	3	356AB	3	5	5	2	1	−0.33	2.71E−24	−0.39	−0.28	13,550.6091
13	119AA01	16,750	3	119AA	3	5	1	2	1	−0.32	1.28E−74	−0.34	−0.31	4321.85333
14	334AK01	200	3	334AK	3	5	3	2	1	−0.27	1.81E−34	−0.3	−0.24	15.5070872
15	643AB00	100	3	643AB	X	6	X	2	3	−0.27	0.00010862	−0.4	−0.12	102.433723
16	565AJ00	5000	3	565AJ	1	5	3	2	3	−0.26	0.00207438	−0.43	−0.09	672.335753
17	681AE00	100	3	681AE	3	6	9	4	3	−0.23	4.72E−06	−0.35	−0.14	148.630007
18	681AF00	100	3	681AF	3	6	4	4	3	−0.23	0.00360002	−0.44	−0.07	138.647729
19	573AA01	10	3	573AA	2	5	3	2	1	−0.23	0.00036596	−0.35	−0.1	34.7495642
20	561AB01	10	3	561AB	1	5	1	2	1	−0.22	9.59E−09	−0.3	−0.15	164.213337

Aquifers are identified using the BDLISA dataset. Reported properties include negative Sen’s slope values (2000–2023/2024) and their significance (p-values). The complete dataset of negative aquifers is available in Supplementary Table T4 (T4_BDLISA_negative_aquifers_long_term.csv).

Conversely, Fig. 9 shows the 20 aquifers with the highest positive Sen’s slope values, indicating areas of groundwater recovery. These positive trends are concentrated primarily in the north and central regions of France, with the largest values occurring in Île-de-France and Hauts-de-France, where Sen’s slope values range from + 0.42 to + 0.12. Smaller pockets of positive trends are also observed in the southwest along the Atlantic coast and scattered regions in the south. Few positive trends appear in the Mediterranean or southeastern regions. Most of the ranked aquifers exhibit statistically significant positive trends (many with p ≈ 0.00), though a few of the highest values show p > 0.05, suggesting varying levels of statistical certainty due to higher variability and/or shorter records. These aquifers benefit from relatively lower abstraction pressures, more consistent rainfall, and favourable recharge conditions. Table 5 details the BDLISA properties and attribution of these 20 aquifers.

Table 5.

Twenty aquifers with the strongest positive long-term Sen’s slope (m yr⁻¹) values.

Order	Codeeh	Ordreabseh	Niveaueh	Incluseh	Etateh	Natureeh	Milieueh	Themeeh	Origineeh	Sen’s Slope	p	Lower CI	Upper CI	Area_km2
1	121AB30	19,001	3	121AB	3	5	9	2	1	0.42	0.167129	−0.07	8.21	5430.76
2	344AA03	2E + 07	3	344AA	1	6	4	2	1	0.3	9.86E−09	0.21	0.38	837.383
3	362AE15	3E + 07	3	362AE	1	5	4	2	1	0.29	4.44E−15	0.22	0.35	240.746
4	127AA01	26,910	3	127AA	3	5	1	2	1	0.25	9.89E−103	0.22	0.27	20,974.2
5	516AG00	50	3	516AG	1	6	1	2	2	0.18	4.79E−45	0.17	0.2	7.87526
6	691AC02	1000	3	691AC	2	6	4	3	1	0.18	3.20E−08	0.12	0.24	8.74546
7	167AC02	50,000	3	167AC	2	5	4	3	1	0.18	0.012105	0.04	0.32	50.8972
8	149AC01	49,100	3	149AC	1	5	9	3	1	0.17	2.89E−16	0.13	0.2	2514.39
9	600AA01	10	3	600AA	2	6	4	2	1	0.16	2.48E−17	0.13	0.18	160.972
10	127AA05	26,930	3	127AA	3	5	1	2	1	0.16	3.14E−99	0.15	0.17	21,366.5
11	141AG01	41,000	3	141AG	3	6	1	2	1	0.14	1.48E−24	0.12	0.17	402.771
12	710AE01	2	3	710AE	2	5	1	1	1	0.14	2.24E−48	0.13	0.16	131.445
13	332AA05	1E + 07	3	332AA	3	6	1	2	1	0.14	2.19E−50	0.12	0.15	1913.8
14	170AA02	50,000	3	170AA	2	5	4	3	1	0.12	0.115332	−0.02	0.27	130.044
15	113AI01	9005	3	113AI	X	7	1	2	1	0.12	7.33E−44	0.11	0.13	8655.38
16	113AK01	10,000	3	113AK	3	6	1	2	1	0.12	7.33E−44	0.11	0.13	6690.88
17	651AA00	600	3	651AA	3	5	3	2	2	0.12	6.97E−21	0.1	0.14	79.5187
18	344AA04	2E + 07	3	344AA	1	6	4	2	1	0.12	6.66E−21	0.1	0.13	4375.89
19	127AA99	26,900	3	127AA	3	5	1	2	1	0.12	1.91E−48	0.1	0.13	69,862.2
20	370AC01	5E + 07	3	370AC	2	6	2	3	1	0.12	1.23E−16	0.09	0.14	519.468

Aquifers are identified using the BDLISA dataset. Reported properties include positive Sen’s slope values (2000–2023/2024) and their significance (p). The complete dataset of positive aquifers is available in Supplementary Table T5 (T5_BDLISA_positive_aquifers_long_term.csv).

Piezometer-level trend shift over time

To address the second research question—" How have groundwater trends shifted over the past two decades, comparing conditions before 2015 with those since 2015?", We compared Sen’s slope (m yr⁻¹) trends between 2004–2014 and 2015–2023/2024 to assess whether groundwater declines worsened or improved or was stable in the recent decade relative to the earlier one. Of the 2536 piezometers in our research, 1897 had continuous data for both decades (see Methods for selection of these data) and were used for this experiment. Figure 10 visualizes the difference in Sen’s slope values between the two periods for each of these 1897 piezometers, with red indicating worsening trends (lower Sen’s slope in 2015–2023/2024 compared to before 2015), green indicating improving trends (higher Sen’s slope in 2015–2023/2024 compared to before 2015).

Fig. 10.

Spatial distribution of changes in Sen’s slope (m yr⁻¹) before and after 2015 for 1897 piezometers across France. Maps show the difference in groundwater trends (Sen’s slope after 2015 minus before 2015). Maps show the difference in groundwater trends (Sen’s slope after 2015 minus before 2015). Panels (a) and (b) depict negative changes (declining trends) for 925 piezometers, while panels (c) and (d) depict positive changes (increasing trends) for 972 piezometers. Panel (a) shows 344 piezometers where both periods have statistically significant trends (p₁ ≤ 0.05 and p₂ ≤ 0.05). Panel (b) includes piezometers where at least one period is not statistically significant: 235 with only the post-2015 period significant (upward triangles), 173 with only the pre-2015 period significant (downward triangles), and 173 with both periods non-significant (squares). Panel (c) shows 276 locations where both periods are significant (p₁ ≤ 0.05 and p₂ ≤ 0.05). Panel (d) includes locations where at least one period is not significant: 384 with only the post-2015 period significant (upward triangles), 218 with only the pre-2015 period significant (downward triangles), and 96 with both periods non-significant (squares). Color intensity represents the magnitude of slope change, with darker reds indicating stronger negative changes and darker greens indicating stronger positive changes. The full dataset of piezometer trend shifts is provided in Supplementary Table S3 (S3_piezometers_trend_shift.csv).

As shown in Fig. 10, most piezometers maintained consistent behavior after 2015 compared to before, either continuing their declines, recoveries, or relatively stable levels. However, in some areas, clear directional shifts are also evident. Panels (a) and (b) show widespread negative changes, with clusters of steep post-2015 declines in southern France, particularly along the Mediterranean coast of Occitanie and Provence-Alpes-Côte d’Azur, as well as in the Garonne basin of Nouvelle Aquitaine. Many of these sites record slope changes below − 0.5, and some exceed − 1.0. In contrast, panels (c) and (d) show positive shifts that are concentrated mainly in the Atlantic and northern corridor. These improvements are generally modest, within the range of 0.0 to 0.3, but rare high-magnitude recoveries above + 0.7 appear in isolated pockets of central and northern France. Many of the shifts are statistically significant in both periods, while some are significant only before 2015 and others only after 2015.

Figures 10, 11, and 12 indicate that since 2015, negative shifts have been widespread, with the largest counts of declines (Sen’s slope change ≤  − 0.1) in Nouvelle-Aquitaine, Grand Est, Occitanie, Centre-Val de Loire, Auvergne-Rhône-Alpes and Provence–Alpes–Côte d’Azur. In Nouvelle-Aquitaine, the sharpest declines are concentrated in the south-central area around the northern foothills of the Pyrenees and parts of the Garonne basin, with several sites between − 0.5 and ≤  − 1.0. In the southern part of France, particularly in Occitanie and Provence–Alpes–Côte d’Azur, the steepest post-2015 declines occur along the Mediterranean coast in Gard, Hérault, Aude, and Bouches-du-Rhône, as well as inland in Vaucluse and parts of the southern Massif Central and lower Rhône Valley, with many sites below − 0.5 and some at ≤  − 1.0. Positive shifts are fewer in most regions and mainly small (0.0–0.2), but the Atlantic and north corridor shows improvement: Nouvelle Aquitaine records the highest number of positive shifts, in some cases exceeding its negatives, and includes rare high-magnitude gains (≈0.6–0.7 to ≥ 1.0), while Hauts-de-France, Pays de la Loire, Bretagne and Normandie also have more positives than negatives, with scattered bins near 0.9–1.0. Nouvelle-Aquitaine, Centre-Val de Loire, and Grand Est have more piezometers than other regions, which increases the number of sites counted in both positive and negative shift classes. These regions are hydrogeologically diverse, so different piezometers can show opposite changes between 2004–2014 and 2015–2023/2024.

Fig. 11.

Number of piezometers with negative shifts in groundwater trends (Sen’s slope change < 0 m yr⁻¹, p₁ ≤ 0.05 or p₂ ≤ 0.05), where p₁ and p₂ refer to trends in 2004–2014 and 2015–2023/2024, respectively, shown by region and Sen’s slope change range.

Fig. 12.

Number of piezometers with positive shifts in groundwater trends (Sen’s slope change ≥ 0 m yr⁻¹, p₁ ≤ 0.05 or p₂ ≤ 0.05), where p₁ and p₂ refer to trends in 2004–2014 and 2015–2023/2024, respectively, shown by region and Sen’s slope change range.

Tables 6 and 7 describe city-level Sen’s slope changes from the trend-shift analysis. The steepest post-2015 declines occur mainly in the Mediterranean south, led by Valliguières (Occitanie, Gard; − 4.61), Montauroux (Provence-Alpes-Côte d’Azur, Var; − 2.56), Bosc (Occitanie, Hérault; − 2.43), Veyras (Auvergne–Rhône-Alpes, Ardèche; − 2.14), and Motte (Provence-Alpes-Côte d’Azur, Var; − 1.60). Additional sharp drops are observed in Gignac (Occitanie, Hérault; − 1.40) and Rochefort-du-Gard (Occitanie, Gard; − 1.34), along with Matelles (Occitanie, Hérault; − 1.23), Saint-Gély-du-Fesc (Occitanie, Hérault; − 1.14), and Pignan (Occitanie, Hérault; − 1.11). Geographically, the worst deterioration is concentrated in southern Occitanie and coastal Provence-Alpes-Côte d’Azur, where a dense cluster of cities registers post-2015 declines stronger than − 1.0, and the weakening extends into some parts of Nouvelle-Aquitaine, though with more moderate declines. In the north and the rest of central France, very steep deteriorations after 2015 are rare; instead, most negative cases appear as localized cities, such as those in central and southern Centre-Val de Loire, with declines that are noticeably less sharp than those observed in the main southern hotspots.

Table 6.

Cities with the steepest negative post-2015 Sen’s slope (m yr⁻¹) changes.

Order	City_or_commune	Region	Department	Median Sen’s Slope Diff	No. piezometers
1	Valliguières	Occitanie	Gard	−4.61	1
2	Montauroux	Provence-Alpes-Côte d’Azur	Var	−2.56	1
3	Bosc	Occitanie	Hérault	−2.43	1
4	Veyras	Auvergne-Rhône-Alpes	Ardèche	−2.14	1
5	Motte	Provence-Alpes-Côte d’Azur	Var	−1.6	1
6	Gignac	Occitanie	Hérault	−1.35	1
7	Rochefort-du-Gard	Occitanie	Gard	−1.34	1
8	Tricot	Hauts-de-France	Oise	−1.28	1
9	Bondues	Hauts-de-France	Nord	−1.23	1
10	Matelles	Occitanie	Hérault	−1.23	1
11	Saint-Gély-du-Fesc	Occitanie	Hérault	−1.14	1
12	Pignan	Occitanie	Hérault	−1.11	1
13	Roquebrussanne	Provence-Alpes-Côte d’Azur	Var	−1.08	1
14	Lille	Hauts-de-France	Nord	−1.01	1
15	Saint-Hilaire-de-Beauvoir	Occitanie	Hérault	−0.99	1
16	Baixas	Occitanie	Pyrénées-Orientales	−0.95	1
17	Pia	Occitanie	Pyrénées-Orientales	−0.9	1
18	Estagel	Occitanie	Pyrénées-Orientales	−0.89	1
19	Draguignan	Provence-Alpes-Côte d’Azur	Var	−0.88	1
20	Trets	Provence-Alpes-Côte d’Azur	Bouches-du-Rhône	−0.88	1
21	Montlaur	Occitanie	Aude	−0.78	1
22	Plou	Centre-Val de Loire	Cher	−0.69	1
23	Aubagne	Provence-Alpes-Côte d’Azur	Bouches-du-Rhône	−0.64	1
24	Bourdic	Occitanie	Gard	−0.64	1
25	Paris	Île-de-France	Paris	−0.59	1
26	Rougeries	Hauts-de-France	Aisne	−0.58	1
27	Laval	Pays de la Loire	Mayenne	−0.58	1
28	Saint-Christophe-sur-Avre	Normandie	Eure	−0.57	1
29	Bannost-Villegagnon	Île-de-France	Seine-et-Marne	−0.53	1
30	Roquetaillade	Occitanie	Aude	−0.51	1
31	Tramoyes	Auvergne-Rhône-Alpes	Ain	−0.46	1
32	Roquette-sur-Var	Provence-Alpes-Côte d’Azur	Alpes-Maritimes	−0.46	1
33	Moussac	Occitanie	Gard	−0.44	1
34	Vailly	Grand Est	Aube	−0.42	1
35	Maillet	Centre-Val de Loire	Indre	−0.39	1
36	Créon-d'Armagnac	Nouvelle-Aquitaine	Landes	−0.39	3
37	Égriselles-le-Bocage	Bourgogne-Franche-Comté	Yonne	−0.38	1
38	Saint-Michel	Occitanie	Tarn-et-Garonne	−0.38	1
39	Lieuvillers	Hauts-de-France	Oise	−0.37	1
40	Ségry	Centre-Val de Loire	Indre	−0.37	1

Shown are the city, its department, and the median Sen’s slope difference (slope₂ − slope₁) of all contained piezometers within the city boundary, where slope₁ (2004–2014) or slope₂ (2015–2023/2024) is significant (p₁ or p₂). The complete list of 628 cities is available in Supplementary Table T6 (T6_cities_negative_p1_or_p2_significant_trend_shift).

Table 7.

Cities with the steepest positive post-2015 Sen’s slope (m yr⁻¹) changes.

Order	City_or_commune	Region	Department	Median Sen’s Slope Diff	No. piezometers
1	Lagleygeolle	Nouvelle-Aquitaine	Corrèze	3.45	1
2	Gex	Auvergne-Rhône-Alpes	Ain	1.73	1
3	Mazé-Milon	Pays de la Loire	Maine-et-Loire	1.62	3
4	Faugères	Occitanie	Hérault	1.45	1
5	Pécorade	Nouvelle-Aquitaine	Landes	1.4	1
6	Castelnau-Tursan	Nouvelle-Aquitaine	Landes	1.25	1
7	Chaignes	Normandie	Eure	1.25	1
8	Laires	Hauts-de-France	Pas-de-Calais	1.16	1
9	Blacqueville	Normandie	Seine-Maritime	1.12	1
10	Loubejac	Nouvelle-Aquitaine	Dordogne	1.05	1
11	Geaune	Nouvelle-Aquitaine	Landes	1.02	2
12	Saint-Georges-de-Montclard	Nouvelle-Aquitaine	Dordogne	0.99	1
13	Ermenouville	Normandie	Seine-Maritime	0.92	1
14	Broût-Vernet	Auvergne-Rhône-Alpes	Allier	0.9	1
15	Eugénie-les-Bains	Nouvelle-Aquitaine	Landes	0.89	2
16	Burosse-Mendousse	Nouvelle-Aquitaine	Pyrénées-Atlantiques	0.88	1
17	Caudrot	Nouvelle-Aquitaine	Gironde	0.88	1
18	Congerville-Thionville	Île-de-France	Essonne	0.85	1
19	Saint-Avold	Grand Est	Moselle	0.82	1
20	Cierrey	Normandie	Eure	0.8	1
21	Rocquemont	Normandie	Seine-Maritime	0.74	1
22	Veauville-lès-Quelles	Normandie	Seine-Maritime	0.73	1
23	Saint-André-de-l'Eure	Normandie	Eure	0.71	1
24	Villers-Saint-Frambourg	Hauts-de-France	Oise	0.7	1
25	Scientrier	Auvergne-Rhône-Alpes	Haute-Savoie	0.7	1
26	Lannecaube	Nouvelle-Aquitaine	Pyrénées-Atlantiques	0.67	2
27	Pizou	Nouvelle-Aquitaine	Dordogne	0.65	1
28	Celle-Lévescault	Nouvelle-Aquitaine	Vienne	0.65	1
29	Val de Louyre et Caudeau	Nouvelle-Aquitaine	Dordogne	0.65	1
30	Montpon-Ménestérol	Nouvelle-Aquitaine	Dordogne	0.64	1
31	Morvillers-Saint-Saturnin	Hauts-de-France	Somme	0.64	1
32	Noyant-Villages	Pays de la Loire	Maine-et-Loire	0.59	2
33	Beaufort-en-Anjou	Pays de la Loire	Maine-et-Loire	0.58	3
34	Saint-Romain	Nouvelle-Aquitaine	Vienne	0.57	1
35	Saint-Yzan-de-Soudiac	Nouvelle-Aquitaine	Gironde	0.56	1
36	Chuelles	Centre-Val de Loire	Loiret	0.54	1
37	Uchacq-et-Parentis	Nouvelle-Aquitaine	Landes	0.53	1
38	Saint-Césaire	Nouvelle-Aquitaine	Charente-Maritime	0.51	1
39	Simacourbe	Nouvelle-Aquitaine	Pyrénées-Atlantiques	0.5	1
40	Maisnil	Hauts-de-France	Pas-de-Calais	0.49	1

Shown are the city, its department, and region, and the median Sen’s slope difference (slope₂ − slope₁) of all contained piezometers within the city boundary, where slope₁ (2004–2014) or slope₂ (2015–2023/2024) is significant (p₁ or p₂). The complete list of 554 cities is provided in Supplementary Table T7 (T7_cities__positive_p1_or_p2_significant_trend_shift).

Based on Table 7, the largest gains are led by Lagleygeolle (Nouvelle-Aquitaine–Corrèze; + 3.45), Gex (Auvergne-Rhône-Alpes–Ain; + 1.73), Mazé-Milon (Pays de la Loire–Maine-et-Loire; + 1.62), Faugères (Occitanie–Hérault; + 1.45), Pécorade (Nouvelle-Aquitaine–Landes; + 1.24), Chaignes (Normandie–Eure; + 1.25), Castelnau-Tursan (Nouvelle-Aquitaine–Landes; + 1.21), Lières (Hauts-de-France–Pas-de-Calais; + 1.16), Blacqueville (Normandie–Seine-Maritime; + 1.12), Loubéjac (Nouvelle-Aquitaine–Dordogne; + 1.05), Geaune (Nouvelle-Aquitaine–Landes; + 1.02), Saint-Georges-de-Montclard (Nouvelle-Aquitaine–Dordogne; + 0.99), Ermenouville (Normandie–Seine-Maritime; + 0.92), and Congerville-Thionville (Île-de-France–Essonne; + 0.85).

Geographically, recoveries are widespread, with dense clusters in Nouvelle-Aquitaine and a broad continuation across central and northern France: they align along the Loire Valley in Pays de la Loire and are common across Normandie and Hauts-de-France, with additional pockets in Île-de-France and the northern Centre-Val de Loire. By contrast, positive post-2015 signals are only occasional in Occitanie and Provence-Alpes-Côte d’Azur, where Table 6 shows that worsening trends dominate. Overall, while these post-2015 improvements are geographically broad, their absolute magnitudes are generally smaller than the stronger post-2015 deteriorations in Table 6.

Aquifer-level trend shift over time

In this section, we assess aquifer-level trend shifts by estimating Sen’s slope (m yr⁻¹) for each aquifer in two periods, 2004–2014 and 2015–2023/2024, and calculating their difference as post-2015 minus pre-2015. Figure 13 maps these changes, showing widespread negative trends in eastern and southern France, contrasted with mixed signals in central regions, and localized stable and positive changes in the north, west, and southwest. Figure 13 follows the same stacking and visualization approach as Fig. 7, so some units lower in the stack may not be distinguishable in this representation. The 20 aquifers with the most negative post-2015 changes are illustrated in Fig. 14, while the 20 aquifers with stable or positive changes (≥ 0) are shown in Fig. 15.

Fig. 13.

Comparison of aquifer-level groundwater trends between 2004–2014 and 2015–2023/2024. (a) Difference in Sen’s slope values between 2015–2023/2024 and 2000–2014. (b) Same differences cube root–normalized to enhance visual contrast. Aquifers are arranged vertically by ordreabseh, with shallower units (low ordreabseh) at the top and deeper units (high ordreabseh) below. Colors indicate changes in Sen’s slope, with red denoting worsening trends, green improvements, and yellow stable conditions. The complete dataset underlying this figure is provided in Supplementary Table S4 (S4_BDLISA_all_aquifers_trend_shift.csv).

Fig. 14.

Twenty aquifers with the most negative changes in Sen’s slope (m yr⁻¹) values between 2004–2014 and 2015–2023/2024, representing the strongest deteriorations in groundwater trends after 2015. (a) Sen’s slope values with corresponding p-values and (b) the spatial distribution of these twenty aquifers. Darker red shades denote greater negative changes, indicating more substantial declines.

Fig. 15.

Twenty aquifers with the most positive changes in Sen’s slope (m yr⁻¹) values between 2004–2014 and 2015–2023/2024, representing the strongest improvements in groundwater trends after 2015. (a) Sen’s slope values with corresponding p-values and (b) the spatial distribution of these twenty aquifers. Darker green shades denote greater positive changes, indicating more substantial recoveries.

Based on Fig. 14, the 20 aquifers with the strongest post-2015 deteriorations show Sen’s slope differences (2015–2023/2024 minus 2000–2014) ranging from about − 0.32 to − 2.56. These negative shifts are concentrated in southern France, particularly in parts of Occitanie and Provence-Alpes-Côte d’Azur, and along sections of the Mediterranean coast. A few aquifers in Nouvelle-Aquitaine show slight negative changes after 2015, and additional scattered declines appear in northern and central inland basins, including parts of Centre-Val de Loire and Hauts-de-France. Many of these aquifers have a highly significant p-value in one or both decades (often p < 0.01), confirming that the observed worsenings are statistically meaningful. Table 8 details the BDLISA properties and attribution of these 20 aquifers.

Table 8.

Twenty aquifers with the strongest post-2015 declines (m yr⁻¹) in groundwater levels.

Order	Codeeh	Ordreabseh	Niveaueh	Incluseh	Etateh	Natureeh	Milieueh	Themeeh	Origineeh	Sen’s Slope 1	p1	Sen’s Slope 2	p2	Sen’s slope diff.	Area_km2
1	571AE04	10	3	571AE	2	5	3	2	1	1.07	2.15E−17	−1.49	1.30E−07	−2.56	94.7742
2	651AB00	200	3	651AB	3	6	1	2	2	0.13	0.47082	−2.3	3.14E−22	−2.43	184.777
3	534AH00	150	3	534AH	2	5	3	2	2	0.72	0.00744	−1.56	7.08E−06	−2.29	352.72
4	533AK00	860	3	533AK	3	5	5	2	2	1.76	5.00E−08	−0.38	0.00625	−2.14	255.233
5	681AQ02	100	3	681AQ	2	5	5	2	1	0.45	0.10831	−0.9	0.13342	−1.36	26.0556
6	571AE01	10	3	571AE	2	5	3	2	1	0.3	4.18E−06	−1.06	2.19E−10	−1.36	29.8772
7	573AA01	10	3	573AA	2	5	3	2	1	0.35	0.01143	−0.72	0.00239	−1.08	34.7496
8	643AB00	100	3	643AB	X	6	X	2	3	0.06	0.72926	−0.93	1.47E−05	−0.99	102.434
9	631AK01	100	3	631AK	2	5	3	2	1	0.06	0.71887	−0.77	2.61E−12	−0.83	69.2619
10	149AC01	49,100	3	149AC	1	5	9	3	1	0.48	4.69E−37	−0.27	1.06E−15	−0.74	2514.39
11	681AM00	200	3	681AM	2	5	3	4	2	0.03	0.62946	−0.7	1.38E−05	−0.73	403.365
12	643AC01	100	3	643AC	3	6	5	2	1	0.12	3.43E−05	−0.51	5.76E−16	−0.64	243.856
13	681AQ04	100	3	681AQ	2	5	5	2	1	0.13	0.43296	−0.42	0.03992	−0.55	54.3013
14	334AK01	200	3	334AK	3	5	3	2	1	−0.17	7.06E−09	−0.69	1.74E−07	−0.51	15.5071
15	344AA03	20,010,100	3	344AA	1	6	4	2	1	0.33	0.00025	−0.16	4.63E−07	−0.49	837.383
16	631AF00	300	3	631AF	2	5	5	2	3	−0.39	0.38485	−0.78	0.03624	−0.39	497.494
17	523AB00	750	3	523AB	3	6	3	2	2	0.26	0.06389	−0.08	0.08109	−0.34	542.282
18	370AC01	51,020,100	3	370AC	2	6	2	3	1	0.26	8.31E−11	−0.07	0.00189	−0.33	519.468
19	127AA01	26,910	3	127AA	3	5	1	2	1	0.47	3.94E−57	0.15	1.28E−17	−0.32	20,974.2
20	681AA03	400	3	681AA	3	5	3	2	1	−0.22	0.00105	−0.54	6.78E−09	−0.32	233.997

For each aquifer, Sen’s slope₁ (2004–2014) and Sen’s slope₂ (2015–2023/2024) are reported with their significance values (p₁, p₂). The Sen’s slope difference is defined as Sen’s slope₂ − Sen’s slope₁ = Sen’s slope diff. The full list of aquifers with negative post-2015 changes is available in Supplementary Table T8 (T8_BDLISA_negative_aquifers_trend_shift.csv).

On the other hand, Fig. 15 shows that the 20 aquifers with the strongest post-2015 improvements record Sen’s slope differences (2015–2023/2024 minus 2000–2014) between roughly + 0.27 and + 1.42. These positive changes are most evident in the north and southwest, particularly in Hauts-de-France, Nouvelle-Aquitaine, and Occitanie. A few aquifers in the northeast and central France also show modest positive changes. Many of these recovering aquifers are large, contiguous systems with broad recharge areas. Most improvements are statistically significant (p ≤ 0.05), but a few have a higher p-value in the post-2015 period, indicating confidence in the reversal to more positive trends. Table 9 details the BDLISA properties and attribution of these 20 aquifers.

Table 9.

Twenty aquifers with the strongest post-2015 increases (m yr⁻¹) in groundwater levels.

Order	Codeeh	Ordreabseh	Niveaueh	Incluseh	Etateh	Natureeh	Milieueh	Themeeh	Origineeh	Sen’s Slope 1	p1	Sen’s slope 2	p2	Sen’s slope diff	Area_km2
1	334AE01	15,030,100	3	334AE	X	7	1	2	1	−1.27	1.05E−15	0.15	0.5529	1.42	5637.64
2	366AA01	31,010,100	3	366AA	3	6	8	2	1	−0.97	2.23E−09	0.01	0.90126	0.97	61,855.8
3	113AL07	12,130	3	113AL	3	6	4	2	1	−1.08	4.33E−19	−0.18	0.03968	0.9	2068.15
5	107AC09	4050	3	107AC	3	5	1	2	1	−0.57	5.38E−35	0.28	5.33E−08	0.85	3512.02
4	516AF00	50	3	516AF	3	6	1	2	2	−0.63	1.14E−16	0.22	0.00015	0.85	27.0341
6	681AB02	200	3	681AB	2	5	3	4	1	−1.31	1.73E−23	−0.53	0.00018	0.79	120.863
7	358AE03	27,040,100	3	358AE	3	5	8	2	1	−0.72	2.84E−46	−0.06	0.16467	0.66	27,293.7
8	107AC07	4000	3	107AC	3	5	1	2	1	−0.45	1.43E−51	0.06	0.01942	0.51	4310.73
9	334AC05	15,020,100	3	334AC	1	5	1	2	1	−0.51	8.14E−19	−0.06	0.4872	0.45	7383.85
10	334AJ01	9,010,300	3	334AJ	3	5	1	2	1	−0.42	6.04E−63	0	0.89734	0.41	255.508
11	139AP14	37,510	3	139AP	3	5	3	2	1	−0.1	0.00043	0.3	2.27E−07	0.4	2866.75
12	356AB01	26,020,100	3	356AB	3	5	5	2	1	−0.41	8.38E−08	−0.01	0.92488	0.39	13,550.6
14	516AG00	50	3	516AG	1	6	1	2	2	0.07	7.17E−05	0.45	3.90E−20	0.38	7.87526
13	123AK03	23,610	3	123AK	1	5	1	2	1	−0.37	5.43E−07	0.01	0.89734	0.38	1650.96
15	143AI03	42,500	3	143AI	3	6	2	2	1	−0.05	0.22174	0.29	4.34E−21	0.33	18,168.7
16	121AF01	17,000	3	121AF	3	5	1	2	1	−0.03	3.34E−08	0.28	1.73E−33	0.3	369.27
17	121AB11	18,120	3	121AB	3	5	4	2	1	−0.39	6.79E−09	−0.09	0.02333	0.3	383.769
19	113AQ37	12,050	3	113AQ	3	5	4	2	1	−0.14	8.32E−23	0.15	4.28E−08	0.28	1075.99
18	141AK01	41,400	3	141AK	3	5	2	2	1	−0.09	0.09686	0.2	0.02186	0.28	2479.05
20	121BC01	18,800	3	121BC	2	5	4	2	1	−0.12	0.02579	0.14	0.15437	0.27	2300.12

For each aquifer, Sen’s slope₁ (2004–2014) and Sen’s slope₂ (2015–2023/2024) are reported with their significance values (p₁, p₂). The Sen’s slope difference is defined as Sen’s slope₂ − Sen’s slope₁ = Sen’s slope diff. The full list of aquifers with positive post-2015 changes is available in Supplementary Table T9 (T9_BDLISA_positive_aquifers_trend_shift.csv).

Methods

Study area and data

Groundwater level records from 8093 piezometers across metropolitan France were obtained from the Bureau de Recherches Géologiques et Minières (BRGM) national monitoring network. The dataset spans 1974–2024 and encompasses diverse hydrogeological settings and measurement frequencies, from weekly observations in early installations to daily records in recent years. Missing observations are common due to variability in monitoring continuity. Records from piezometers located outside metropolitan France were excluded before analysis, and for the analysis, we considered the period from 2000 onward to focus on more recent conditions.

To address the prevalent missing data, a gap-sensitive interpolation procedure was applied. Months with no recorded measurements were only infilled if both the preceding and following months contained at least one valid observation. In such cases, the missing monthly value was estimated as the mean of the preceding and following months’ averages. Gaps longer than two consecutive months were not interpolated to avoid uncertainty, and time series segments separated by such long gaps were treated as independent continuous records.

For trend analysis, only piezometers with a final observation in 2023 or 2024 were retained, ensuring that results reflected current conditions. Continuous records shorter than five years were excluded to ensure reliable slope estimation. For each aquifer, we first subtracted the overall mean of the time series for each piezometer within the aquifer, then calculated the mean value across piezometers for each time step to produce a representative time series for the aquifer. This aggregated series was used to calculate Sen’s slope and perform the Mann–Kendall test. At the city scale, median Sen’s slopes were calculated from all intersecting piezometers within the city boundary. Beyond these measures, no additional transformations or preprocessing steps were applied.

Data analysis

The Theil–Sen slope estimator⁴⁸ is a non-parametric method used to quantify the monotonic trend in a time series. It is robust to outliers and performs well with irregular time intervals, making it well-suited for hydrological applications. For a time series of observations () taken at times , the slope for each pair of points is given by (Eq. 1):

1

The overall slope is the median of all values, where positive indicates an upward trend, negative a downward trend, and a slope not significantly different from zero indicates no discernible trend. Slopes were computed using the theilslopes function from SciPy (v1.16.1), which returns the point estimate and a 95% confidence interval. Time intervals were measured in seconds and converted to units of change per year. The statistical significance of trends was assessed using the Mann–Kendall (MK) rank correlation test^49,50, a non-parametric method for detecting monotonic changes in time series data. The MK test statistic is calculated as (Eq. 2):

2

where the sign function (Eq. 3) is:

3

When ties occur in the data, the variance of (Eq. 4) is adjusted as:

4

where ​is the size of the p-th group of tied values and is the number of tie groups. The standardized test statistic (Eq. 5) is:

5

The standardized test statistic Z is used to evaluate the significance of the trend. To implement this, Kendall’s tau correlation coefficients which measure the strength of the monotonic trend, and their associated p-values (derived from ) were computed using the kendalltau function from SciPy (v1.16.1). For each time series, the two-sided p-value returned by kendalltau was used to test the null hypothesis (Kendall’s tau = 0) of no monotonic association between groundwater level and time, and trends were classified as statistically significant when p ≤ 0.05. Theil–Sen slope provides the corresponding estimate of trend magnitude (m yr⁻¹).

For city-scale trends, we first estimated a Theil–Sen slope for each piezometer within a given city. The city-scale slope (Eq. 6) was then defined as the median of these slopes:

6

where ​ is the Sen’s slope of the piezometer , and ​ is the set of piezometers in the city .

For aquifer-scale series, we derived composite time series in two steps (Eqs. 7–8). First, for each piezometer , we removed its long-term offset by subtracting the mean of its time series .

7

where ​, , denote the set of observation times, the mean offset, and the offset-corrected groundwater level series, respectively, for the piezometer .

Second, the offset-corrected records of piezometers in the aquifer were averaged at each time step to obtain the composite aquifer series

8

where denotes the set of piezometers within the aquifer observed at the time . were then subjected to Theil–Sen’s slope estimation and Mann–Kendall significance testing in the same manner as individual piezometer series. All computations were performed in Python (v3.11.13), and all plots were produced using Matplotlib (v3.10.5) and Seaborn (v0.13.2).

Experimental design

Two complementary analyses were undertaken to address the study objectives. The first quantified long-term groundwater trends across metropolitan France using continuous piezometer time series segments ending between 2023 and 2024, each with a minimum length of five years to ensure statistical robustness. Theil–Sen slope estimates were computed for each piezometer and visualised both as individual observation points and as aggregated values at the aquifer scale. The second analysis examined temporal shifts in groundwater behavior by partitioning the dataset into two intervals (2004–2014 and 2015–2023/2024) and estimating Sen’s slope separately for each. This periodization was chosen to capture recent changes in groundwater dynamics. Differences in slope between the two intervals were also summarised at the aquifer scale and mapped alongside piezometer-level changes, allowing identification of the aquifers and regions experiencing the most pronounced deterioration or recovery.

Conclusions

Our national-scale analysis of groundwater trends in metropolitan France reveals a geographic divide. Southern, southwestern, and Mediterranean aquifers exhibit widespread and often accelerating declines, while northern systems show modest recoveries. This high-resolution assessment, providing the first detailed evaluation of trends since 2015, identifies critical vulnerability hotspots and offers a vital benchmark. These findings can directly inform sustainable extraction limits, target recharge efforts, and strengthen national drought resilience planning in the face of climate change.

Author contributions

Ahmad Ali Abin: Conceptualization, Supervision, Methodology, Review & editing, Validation. Seyed Soroush Majd: Methodology , Writing original Draft, Analysis, Data Curation. Hamideh Rafiee: Methodology, Writing original Draft, Analysis, Data Curation.

Funding

Not applicable.

Data availability

Data supporting the findings of this study can be downloaded from https://hubeau.eaufrance.fr/page/api-piezometrie.

Declarations

Competing interests

The authors declare no competing interests.