Figure  presents the conceptual model of drought propagation within the catchment: across different domains and along the stream. A meteorological drought could trigger droughts at the surface level, via the reduction of soil moisture content and streamflow (hydrological drought). Hydrological drought also propagates downstream along the river. Later in time, a reduction in recharge can lead to drought in the groundwater. This is a slow process, as the average response times of the groundwater table in the region are around 1 month (see ). The reduced available soil moisture (and in the case of deep-rooted plants reduced groundwater table) can also lead to a drought in vegetation.

In the following, we will present how we analyzed the drought propagation over the different domains based on the available datasets. We used the conceptual model as a roadmap to try to infer the relations of the different drought types at the different sites.

Data pre-processing and all analysis steps were performed using R statistical software (version 4.4.1) (see Sect. ). The study period covers 13 hydrological years from November 2008 to April 2021, chosen after data availability. An overview of the data used, the data sources and the official names of the measurement sites as called by the data provider – or in the case of the NDVI data, the coordinates used to extract mean NDVI values – are summarized in Table .

Daily precipitation measurements were obtained from the Berlin-Tegel station via the German Weather Service (Deutscher Wetterdienst, DWD) using the R package “rdwd” . The dataset is assigned DWD’s quality measure QN9, indicating that while it is complete, some parameters may still require correction. Given the short distances between stations within the Tegeler Fliess catchment area, Berlin-Tegel was used to represent precipitation across the entire study area. This approach was taken for the sake of simplicity, but also due to the lack of nearby measurement sites in close proximity to data records over the study period. Spatial distribution in precipitation across the study sites was assessed using the gridded CER v2 dataset , confirming low spatial variability and suggesting negligible heterogeneity in this lowland setting (see Fig. S7 in the Supplement).

Daily surface water and groundwater level measurements were obtained from two regional water management authorities (see Table ) depending on the station’s location. Groundwater level data for Luebars and Tegel are available at daily resolution and contain only minor gaps, whereas groundwater level measurements at Schildow are provided at weekly resolution. At Schildow, groundwater levels were typically recorded four to five times per month (115 out of 150 months), with three measurements available in 31 months, most likely due to temporary technical issues. The only larger gap of missing records occurs between November 2008 and February 2009 at the beginning of the study period. To harmonize all groundwater level time series to the same daily temporal resolution, the datasets with missing values were linearly interpolated (see Fig. S1).

To evaluate the potential impact of localized anthropogenic factors – specifically construction and land use change – on surface and groundwater dynamics, high-resolution historical satellite imagery from Google Earth was visually inspected for the study period. Despite the sites being located near urban/peri-urban settings, no significant construction activity, large-scale urban expansion, or extensive surface sealing was observed in the immediate vicinity of the groundwater wells at Schildow, Luebars, or Tegel.

Vegetation health was assessed using MODIS NDVI (product MOD13Q1, L3 validation stage) at a spatial resolution of 250 m and a temporal resolution of 16 d, accessed using the R package “MODIS Tools” . The 250 m resolution represents a compromise at the sub-catchment level. A key challenge is the mixed-pixel problem in peri-urban zones, where one cell may merge signals from infrastructure, vegetation, and agricultural land. and showed this can weaken the NDVI response in such areas. To represent vegetation health at each site, a point was selected near each station (coordinates given in Table ) and the mean NDVI was calculated within a 500 m buffer (see Fig. ). Sampling points were placed in locations dominated by vegetation to minimize the influence of mixed pixels. Land cover within a 500 m buffer around each station was assessed to account for the influence of local vegetation and surface characteristics on the NDVI signal. The primary classification relied on the European Environment Agency’s CLCplus Backbone 2021 dataset, a 10 m resolution raster supplemented by visual assessment using the HCMGIS plugin in QGIS, since automated classifiers can miss small-scale features. High-resolution Google Satellite imagery was used to verify field structures, vegetation patterns, and built-up areas at each site, ensuring a more accurate representation of factors influencing the NDVI signal.

As outlined in the introduction, various indices exist for assessing different drought types. In this study, drought is defined as a period of at least one month during which water availability is significantly below average, resulting in a deficit in its respective domain: precipitation deficit (meteorological drought), reduced surface water levels (hydrological drought), or reduced groundwater levels (hydrogeological drought). To capture these aspects, we applied the SPI for meteorological drought, the Standardized Surface Water Level Index (SSWLI) as a proxy for hydrological drought, and the Standardized Groundwater Level Index (SGLI) for groundwater drought. Each of these indices was selected to provide a consistent, standardized assessment of drought conditions, facilitating direct comparison and enabling a comprehensive analysis of drought propagation. The following subsections describe the calculation procedure and application of each index.

After calculating the different drought indices, different statistical tests were applied to compare them with each other. A Spearman correlation test was performed to assess potential relationships between drought indices (SPI3, SSWLI3, SGLI3) and NDVI across all stations. Spearman's rank correlation is a non-parametric measure used to determine the strength and direction of association between two variables. It is particularly useful in this context as it does not assume a normal distribution and can detect monotonic relationships. This test was chosen to investigate whether lower Drought Index values correspond to lower NDVI values and vice versa.

For trend analysis, the Mann-Kendall test was applied to each time series to identify statistically significant trends in SPI3, SSWLI3 and SGLI3 over the 13-year study period. This non-parametric test was chosen due to its robustness in detecting monotonic trends (either increasing or decreasing) in time series data that are not normally distributed. The test was performed using the “Kendall” R package .

Additionally, drought occurrences were counted for each index and station. To assess how drought conditions propagate between compartments, propagation rates and conditional probabilities were calculated between meteorological (SPI3), hydrological (SSWLI3), and hydrogeological (SGLI3) droughts. Here, “upstream” and “downstream” refer to the sequence of compartments (compare Fig. ) and not the geographic location.

The Granger causality test was used to assess the time-lagged relationships between drought indices and vegetation stress. introduced the Granger causality test originally within the field of economics as a statistical method for determining whether one time series can predict another. Since then, the test has gained widespread application across various disciplines, including ecosystem science and hydrology, due to its computational simplicity . The test checks whether information from one time series (e.g. SPI) improves the prediction of another time series (e.g. NDVI) beyond what could be predicted from past values of the latter alone . The “lmtest” package and the “vars” package were used, with lag periods ranging from 1 to 6 months to allow for delayed vegetation responses to drought conditions. Granger causality was used to determine whether different drought indices predict each other or influence future NDVI values, providing insight into the interaction between drought propagation and ecosystem response.

In Fig.  the general statistics of the input data: precipitation, surface water and groundwater levels are shown. The precipitation patterns at the Tegel-DWD station averaged 45 mm month⁻¹, with occasional heavy rainfall events exceeding 100 mm. These heavy rainfall events are likely to contribute to localized water level surges as can be seen in Fig. . However, both the linear and non-parametric trend analyses indicate no significant long-term change in precipitation (R2=0.001, p=0.67, MK p=0.51), confirming overall stable rainfall conditions over the study period.

Comparing surface and groundwater levels, however, reveals the variation across sites. Surface water levels at Luebars fluctuated between 0.35 and 0.55 m, visually showing the greatest variability between stations. Despite these fluctuations, Luebars exhibits only a weak and statistically not significant decline in surface water levels (R2=0.02, p=0.08). In contrast, Schildow displays more stable surface water levels, averaging around 0.23 m, but shows a significant decreasing trend (R2=0.21, p<0.001, MKτ=-0.32) throughout the study period. The Tegel station had a similar mean surface water level, but greater variability, indicating possible site-specific differences in hydrological behavior. At Tegel, no consistent trend was detected (R2=0.006, p=0.36). Tegel and Luebars displayed the highest surface water level variability with Coefficient of Variation (CV) of 30.9 % and 33.9 % respectively, with Schildow being the lowest at 20.4 %. Higher CV values may indicate less stability in water-level conditions, potentially due to factors like localized hydro-climatic influences, different water management practices, etc.

Groundwater levels showed further variability between stations. Schildow recorded the highest median groundwater level at about 40 m a.s.l., while Tegel recorded the lowest, with a difference of 8 m along the river. These observations underline the variability of water availability in the study region, which is likely to be influenced by local environmental conditions and hydrological patterns specific to each station. The trend analyses confirm a consistent and significant groundwater decline across all stations (Schildow: R2=0.28, MKτ=-0.38; Luebars: R2=0.29, MKτ=-0.35; Tegel: R2=0.47, MKτ=-0.48; all p<0.05), indicating a depletion of subsurface storage which is not explained by precipitation trends. A full statistical summary is presented in Table S1 in the Supplement.

To investigate the impact of droughts on the vegetation, satellite-based NDVI data is used. The NDVI timeseries for the three sites are shown in Fig. . The seasonal variation of the vegetation in a cool temperate climate is evident from the NDVI values at all stations, with pronounced peaks in the summer months and dips in the winter months. Throughout the study period, NDVI values predominantly fall within the moderate to dense vegetation classes, indicating generally healthy vegetation. This resilience is consistent across Schildow, Luebars and Tegel, with Luebars and Tegel showing a slight advantage in vegetation health with 110 and 101 months of dense vegetation respectively. Schildow showed 96 months of dense vegetation, still within a healthy range but slightly lower than the other stations. Despite occasional fluctuations, NDVI values show a slight greening trend over time, although statistical tests did not yield significant results. For example, the Mann-Kendall test for NDVI at Schildow showed a MKτ=0.066 and p=0.23, suggesting that the greening trend, while present, is weak and not statistically significant.

In particular, the early years of the study (2009–2011) show instances where vegetation fell into the Bare Soil or Rock/Build-Up classes, especially during the winter months. However, from 2014 these classes no longer occurred. This could indicate an increase in vegetation cover or an improvement in overall stability. Studies demonstrating higher growth rates in high-density regions with increasing air temperatures and decreasing precipitation provide evidence in favor of this, indicating that vegetation may profit from these circumstances in specific urban environments . Furthermore, the shows a rising trend in Berlin's winter air temperatures during the same time period, which highlights the possibility of more resilient and stable vegetation in milder winters. An even more plausible explanation however is the presence of long-lasting snow cover in the winter months. German weather service data shows snow cover in the winters of 2010 and 2011 at the Tegel station. While there are no direct DWD snow measurements at the stations Schildow and Luebars, the regional climate and landscape characteristics suggest that similar snow conditions likely prevailed at these sites as well. The proximity of the stations and the absence of significant topographic differences support this assumption.

Luebars and Tegel were particularly affected by an SSWLI3 drought between 2009 and 2010, which seemed to coincide with reduced vegetation health in the following winter months. However, this relationship was not consistently observed, as winter months with lower NDVI values sometimes occurred without preceding drought events. After 2018, all stations experienced significant periods of drought, as reflected in the SSWLI3 and SGLI3 indices. SSWLI3 and SGLI3 values frequently fell below -1, indicating moderate to severe drought conditions. Despite these persistent droughts, NDVI values remained in the moderate to dense vegetation classes, suggesting a degree of vegetation resilience in these ecosystems. However, Schildow and Tegel showed prolonged periods of stress where the vegetation took a little longer to recover, as indicated by longer periods in the sparse or moderate vegetation classes.

The NDVI data point to overall stable vegetation conditions and a weak evidence of long-term greening, despite recurring drought periods. These observations are further supported by statistical analyses.

Spearman correlation tests between NDVI and the drought indices (SSWLI3, SGLI3 and SPI3) showed no significant results, with correlation coefficients close to zero at all stations. This lack of strong correlation suggests that vegetation health, as represented by NDVI, may not directly reflect drought conditions in these ecosystems. However, Granger causality tests revealed more nuanced relationships. At Schildow, the combination of SPI3 and SSWLI3 was a significant predictor of NDVI with a lag of 1 month, with p≪0.05. SPI3 alone was also a significant predictor of NDVI at this station, although the combination of SPI3 and SSWLI3 performed better. At Luebars, SPI3 alone was found to be a significant predictor of NDVI with a lag of 1 month, while at Tegel, the combination of SPI3 and SSWLI3 also gave the best results, with p≪0.05.

While these data suggest some predictive relationships, it is important to note that satellite-based NDVI may be too coarse to adequately capture localized vegetation responses to drought. In-situ measures of vegetation health could provide more precise insights and supplement satellite-derived indices in these analyses.

In summary, while meteorological drought events, particularly after 2018, are clearly reflected in the SSWLI3 and SGLI3 indices, they do not appear to have a direct and immediate impact on vegetation health as measured by NDVI. Instead, ecosystems at these stations show a degree of resilience: vegetation is either not affected by or quickly recovers after periods of drought. The statistical tests reinforce the idea that while SPI3 and SSWLI3 may influence vegetation health, the overall relationship between drought indices and NDVI remains complex and non-linear across stations.

The stations along the Tegeler Fliess – Schildow, Luebars and Tegel – showed unique responses to drought stress over the study period, despite their geographical proximity. While the Tegel surface water system experienced moderate to severe droughts from 2015 to 2020, the groundwater system faced more sustained stress. Schildow experienced more severe surface water droughts but had relatively stable groundwater levels. Luebars showed the greatest variability, with severe groundwater drought but more resilient surface water conditions.

At all stations, surface water droughts manifested and receded more quickly than groundwater droughts, which showed a delayed but longer-lasting response. For example, during the 2008–2010 and 2018–2020 droughts in Luebars, there was a decline in surface water levels followed by a lasting strain on groundwater resources, suggesting that it takes time for groundwater levels to recover once they have been depleted. Similarly, in Schildow, surface water droughts occurred rapidly with declining precipitation, but groundwater droughts developed more gradually, suggesting that these systems absorb cumulative meteorological impacts over time. The main differences between these sites are primarily driven by their unique local environments, with the Tegel station situated in a peri-urban landscape that is influenced by a nearby densely developed urban area. In such an environment, the great amount of sealed surfaces put a lot of stress on groundwater resources via disturbed recharge processes . Visual assessment of historical satellite imagery confirmed that no large-scale urban expansion or significant changes in surface sealing occurred directly near the wells during the study period (2008–2021). This increases confidence that the most severe groundwater drought impacts observed at Tegel are predominantly a response to hydrological stress.

While Tegel experienced the most severe groundwater stress, Schildow was more vulnerable to surface water droughts. Note that this could be an artefact from the methodology, as at the most upstream station river levels are the most sensitive to external influences. However the first half of the timeseries show similar drought dynamics, which indicates consistency in the methodology across the stations. Schildow is located in an open agricultural area, which implies a more direct connection with the atmospheric conditions. This resulted in a much faster hydrological drought dynamics, with droughts setting in and recovering faster than at the other sites. These fast recoveries however are not reflected in the groundwater: here, this site suffers the longest drought periods, with slightly mitigated impacts. These findings imply that drought monitoring programs should be tailored to the specific vulnerability of each site to ensure that both surface and groundwater resources are properly managed.

These results align with those of , who observed that groundwater droughts often persist after meteorological droughts and surface water droughts have ended. The propagation of drought, from meteorological to hydrological drought via soil moisture was shown in Berlin on a catchment level by . Our results show that this propagation is observable on a smaller scale, but emphasizes the importance of local factors in relation to drought resilience. Furthermore, the prolonged period of groundwater depletion in Tegel is comparable to the patterns observed by in Spain, where groundwater systems were highly vulnerable to prolonged drought conditions. The downward trend in SPI3 observed in highlights the increasing vulnerability to meteorological droughts across Europe due to changing precipitation patterns under climate change and could become relevant for our study site in the future, even though no significant trend in SPI3 was observed during our study period.

It is also important to note, that the three sites are located in hydrogeologically very different settings. Most notably, the aquifer at the Luebars site is an isolated local aquifer body, with limited recharge connections to the atmosphere. The Schildow site also features a thick isolating layer above the aquifer, which contributes to the observed higher resilience. This emphasizes the importance of recharge processes on groundwater resources of the region, and shows the importance to further investigate these processes in the region.

Despite the occurrence of severe and prolonged droughts, particularly from 2018 to 2020, vegetation along the Tegeler Fliess showed notable resilience. NDVI values generally remained within the moderate to dense vegetation classes, indicating that vegetation was not strongly impacted during periods of hydrological and hydrogeological drought. The heterogeneous land cover at the sites likely contributed to the inconsistent vegetation responses observed across stations. Applying a remotely sensed index like MODIS NDVI to a catchment as small and heterogenous as the Tegeler Fliess raises the question of scale. Precedent exist for using MODIS at this level: effectively tracked forest drought stress across Germany by pairing NDVI with drought indices, and used similar indicators to investigate vegetation lag times across broad climate areas. The key difference, however, lies in the spatial uniformity of those study areas. In transitional, peri-urban landscapes like the Tegeler Fliess, the 250 m resolution represents a coarse tool. One averaged MODIS pixel may blend the spectral signature of a dense forest, irrigated suburban lawn, and sealed surfaces. This is the classic mixed-pixel problem and it constitutes a fundamental limitation in this context, suggesting that the findings from the NDVI analysis should be interpreted with caution.

To validate the MODIS-based NDVI results, we also analysed NDVI based on Landsat data, which offer higher spatial resolution (30 m) compared to MODIS (250 m). However, the Landsat record contained many temporal gaps due to cloud cover and limited revisit frequency, while MODIS provided more consistent temporal coverage. For the most part, both datasets aligned well (see Fig. S4), except for a systematic discrepancy during winter months, which is interesting but not relevant to this study’s objectives.

Granger causality tests showed that SPI3 and SSWLI3 influenced NDVI at several stations, particularly at Schildow and Tegel. However, SGLI3 did not show a significant relationship with NDVI. These results suggest that precipitation and surface water played a more critical role in maintaining vegetation than groundwater. The weak correlation between drought indices and NDVI indicates resilience within the ecosystem, shaped by multiple factors, including vegetation type, soil moisture, groundwater depth, and land management practices. In the regions around Schildow and Luebars, the NDVI remained stable despite hydrological and groundwater drought conditions. The structural complexity of the land cover likely fostered a more nuanced ecosystem response compared to more uniform environments. The small distance to the groundwater table near the stream allowed deep rooted plants to use groundwater resources directly as a buffer during drought periods with low soil moisture (see Fig. ). While this connection could weaken as groundwater levels decrease, no such severance occurred during the study period. Regional-scale studies however warn of a general decreasing trend in groundwater levels , therefore the investigation of the groundwater-vegetation relationship stays relevant for future research.

While water infrastructure like trench networks near the agricultural fields in Luebars are present, these structures were primarily designed for drainage and to accelerate surface runoff. Without active retention measures, these networks do not increase resilience and may even exacerbate vulnerability to hydrological extremes. Therefore, the observed resilience of the local vegetation is primarily dependent on the shallow groundwater availability rather than surface water redistribution. Monitoring of vegetation health should go beyond the use of NDVI, as this index alone may not capture the extent of water stress, particularly in agricultural areas. Adaptive management strategies should therefore focus on balancing the needs of natural ecosystems and agricultural activities, which can further exacerbate water depletion in times of drought.

The divergence between drought indices and NDVI observed in this study is consistent with findings from similar regions, where vegetation resilience or human intervention (e.g. irrigation) mitigated the effects of drought . This suggests that in peri-urban catchments such as the Tegeler Fliess, the response of vegetation to drought is more complex and may require more detailed, long-term studies that consider both natural and anthropogenic factors.

Although historical satellite imagery suggested that localized land use changes did not considerably influence the observed dynamics, the analysis lacked precise local data on specific anthropogenic activities, such as subsurface pumping rates, development permits, or changes in storm water infrastructure. The absence of such data limited the quantification of human impacts on water level dynamics. Furthermore, the classification of NDVI values into vegetation classes based on may not fully capture the complexity of vegetation dynamics, as any fixed classification scheme simplifies the continuous and context-dependent nature of vegetation responses.

The linear interpolation applied to harmonize groundwater level time series introduced a potential source of uncertainty, particularly at Schildow, where measurements were recorded at weekly and not daily resolution. However, because drought indices were calculated on a monthly basis, the influence of short-term interpolation on index values is expected to be minimal. The only substantial gap at the beginning of the study period (November 2008 to February 2009) falls outside the main drought events discussed and does not affect their interpretation.

The vegetation resilience observed across the stations suggests that the natural ecosystems in the Tegeler Fliess area remained relatively robust during periods of water scarcity. The study period ending in April 2021 may not have been long enough to capture the delayed vegetation responses to the prolonged drought conditions of 2018–2020. The time frame of our study was limited by the availability of complete data sets. Even though some measurements date as far back as 1945, the surface water level station at Luebars began recording in 2008 and precipitation measurements were discontinued at the site in 2021. This is a key limitation, as the study period of 13 years may not fully capture longer-term climate cycles that influence drought patterns.

Most of the discussed limitations could be addressed by increasing the amount of in-situ data collected on the water-vegetation interactions, showing that site-specific monitoring is a core element of understanding the impacts of climate change in such settings (and how the declining number of such stations is an issue; ).