![\"\"](\"https:\/\/www.vtei.cz\/wp-content\/uploads\/2026\/10\/Bernsteinova-fig-1-2.jpg\")
<\/a>\nFig. 1. Differences in the mean climatic characteristics of seven GCMs under four SSP scenarios for the 2050 projection period (2035\u20132064), relative to the reference period\u00a0(1981\u20132010), aggregated for the Dyje River basin. SSP scenarios are distinguished by colour [7]. The GCMs used in this study are numbered as follows: (1) CMCC-ESM2, (2)\u00a0CNRM-CM6-1-HR, (3) EC-EARTH3, (4) GFDL-ESM4, (5) MPI-ESM1-2-HR, (6) MRI-ESM2-0, and (7) TaiESM1<\/h6>\n
The\u00a0model ensemble shows a\u00a0high degree of\u00a0variability and no consistent spatial pattern of\u00a0precipitation change, even within\u00a0individual SSP scenarios. Although SSP1\u20132.6 is generally associated with relatively moderate warming and smaller changes in\u00a0precipitation regimes, whereas SSP5\u20138.5 represents very strong warming accompanied by major changes in\u00a0precipitation, no consistent spatial pattern of\u00a0change can\u00a0be identified across the\u00a0models within\u00a0the\u00a0individual SSP scenarios. Moreover, inter-model variability in\u00a0long-term means differs substantially between decades of\u00a0the\u00a0future period. To understand and objectively quantify the\u00a0uncertainty associated with climate change reliably, it is therefore essential to evaluate the\u00a0entire climate projection ensemble rather than\u00a0identify a\u00a0single central scenario.<\/span><\/p>\n Because of\u00a0the\u00a0coarse spatial resolution and systematic biases of\u00a0the\u00a0global models, their outputs were transformed using the\u00a0Advanced Delta Change (ADC) method, employing observed time series representing locally specific climatic conditions during the\u00a01981\u20132010 reference period. This approach incorporates changes in\u00a0both means and variability into the\u00a0transformation process, thereby accounting for shifts in\u00a0extremes that differ from shifts in\u00a0mean\u00a0values and compensating for systematic biases in\u00a0precipitation simulations\u00a0[8]. Mean\u00a0daily temperature and precipitation totals entered the\u00a0modelling process directly as boundary conditions, while reference evapotranspiration (ETo) was calculated using the\u00a0standard Penman\u2013Monteith equation.<\/span><\/p>\n The\u00a0Dyje River basin\u00a0digital twin\u00a0is based on the\u00a0fully distributed, physically based MIKE SHE modelling system (Fig.\u00a02), designed to simulate complex hydrological processes across water-management components\u00a0[9, 10]. At regional scale, spatial discretisation uses an\u00a0orthogonal grid with 500 \u00d7 500\u00a0m cells.<\/p>\n Data are exchanged between modules at a daily time step, although the simulation engines operate at finer temporal resolutions according to their internal dynamics. Surface runoff is represented using a two-dimensional diffusive-wave approximation dynamically coupled with channel flow. Processes in the unsaturated zone are described by the one-dimensional Richards equation, including the effects of macropores, while the saturated zone is represented by a three-dimensional approximation of the Boussinesq\u00a0equation. This close coupling allows two-way interaction between the\u00a0aquifer and river network to be modelled, including baseflow and bank infiltration. Evapotranspiration is also included, combining evaporation from interception and bare soil with transpiration. Vegetation parameters, such as leaf area index, were derived at a\u00a0daily time step from MODIS sensor data from the\u00a0Terra satellite\u00a0[11]. The\u00a0model also includes 17 reservoirs, whose fully dynamic simulation is based on the\u00a0applicable reservoir operating rules. Finally, water-use data derived from historical records in\u00a0water-use registers are incorporated.<\/p>\n The\u00a0reliability of\u00a0the\u00a0tool is underpinned by measured data from 52\u00a0stream-gauging stations. The\u00a0model was calibrated for the\u00a0hydrologically normal period 2000\u20132010 and successfully validated for the\u00a0subsequent decade, 2011\u20132020, which included the\u00a0extreme drought of\u00a02015\u20132018, confirming its robustness even under extreme conditions. Calibration targeted the\u00a0parameters with the\u00a0greatest uncertainty and sensitivity (saturated-zone transmissivity, runoff characteristics, vegetation and snowmelt), while soil parameters were set using the\u00a0authors\u2019 own measurements of\u00a0hydropedological properties, including laboratory determination of\u00a0retention curves from soil pits at representative sites in\u00a0the\u00a0Dyje River basin. Agreement between simulated and observed discharge was assessed using the\u00a0Nash\u2013Sutcliffe efficiency coefficient (NSE), while groundwater levels (GWL) in\u00a070 boreholes were assessed using root mean\u00a0square error (RMSE). The\u00a0final run, which optimised NSE and RMSE (Fig.\u00a03), achieved NSE > 0.5 at 78\u00a0% of\u00a0stations. Calibration emphasised agreement across the\u00a0full flow-duration curve, with particular attention to low flows. Lower calibration efficiency in\u00a0the\u00a0Svitava catchment resulted from the\u00a0limitations of\u00a0spatial schematisation in\u00a0representing highly complex hydrogeological conditions.<\/p>\n The design of the adaptation measure comprised several successive phases (Fig. 4) and was based on an official methodology [12]. The first step was to define and conceptualise the effect of the adaptation measure in the physical reality of the catchment and identify the processes directly affected by it. Suitable locations for targeting the proposed measure then had to be identified objectively. The next step was the so-called \u201ctranslation\u201d of the physical effects of the measure into the MIKE SHE model-parameter framework, while ensuring that the measure remained functional and logically consistent with the nonlinear feedback relationships among parameters in the four-dimensional\u00a0simulation environment (3D space and time) of\u00a0the\u00a0digital twin. These phases tend to be time-consuming and involve an\u00a0iterative process of\u00a0testing the\u00a0effects of\u00a0the\u00a0proposed measure and reconciling them with expectations based on expert judgement, measured data or a\u00a0literature review. In\u00a0the\u00a0next phase, the\u00a0model incorporating the\u00a0measure is run under 26 climate scenarios, and the\u00a0results of\u00a0the\u00a0individual runs are compared with the\u00a0reference-state model and with future projections without adaptation measures.<\/p>\n The aim of the measure termed \u201chydrologically optimal agriculture\u201d was to enhance natural infiltration, soil-water retention and baseflow. Its spatial targeting was based on an analysis of ecosystem vulnerability, which identified the greatest risk of declining runoff in highland areas [10]. As the modelling results demonstrated the low\u00a0effectiveness of\u00a0isolated measures \u2013 increasing soil-water retention where it was constrained by a\u00a0moisture deficit primarily increased evapotranspiration \u2013 a\u00a0complex set of\u00a0measures termed \u201chydrologically optimal agriculture\u201d was designed. The\u00a0concept is theoretically grounded in\u00a0the\u00a0available literature and the\u00a0interdisciplinary expert judgement of\u00a0the\u00a0CzechGlobe team, and integrates synergistic measures including no-till farming\u00a0[13], biochar application\u00a0[14], surface mulching\u00a0[15], year-round soil cover using cover crops\u00a0[16], targeted support for soil biota to restore macropores\u00a0[17], and modifications to land-drainage systems to slow episodic runoff\u00a0[18]. The\u00a0design also prioritises crops with lower transpiration requirements\u00a0[19]; under Czech conditions, millet may be one such crop. Translating these modifications into the\u00a0logic of\u00a0the\u00a0simulation tool involved defining a\u00a0new shape for soil-water retention curves, increasing saturated hydraulic conductivity and enhancing vertical flow through the\u00a0macropore system. Rapid runoff was slowed in\u00a0the\u00a0model by increasing Manning\u2019s\u00a0roughness and enhancing macropore flow. Changes in\u00a0vegetation cover were represented by modifying leaf area index (LAI) and rooting depth, enabling the\u00a0digital twin\u00a0to simulate the\u00a0hydrological response of\u00a0the\u00a0catchment in\u00a0detail.<\/p>\n The\u00a0modelling results indicate that, by around 2050, climate change can\u00a0be expected to reduce mean\u00a0and low flows throughout the\u00a0Dyje River basin\u00a0relative to the\u00a0reference period. Analysis of\u00a0flow-duration curves for mean\u00a0daily discharges, evaluated over the\u00a0P25\u2013P75 percentile range (Fig.\u00a05<\/em>), provides a\u00a0comprehensive view of\u00a0future development. A\u00a0substantial decline is expected across the\u00a0full flow range (up to 20\u00a0% in\u00a0highland areas and 40\u00a0% in\u00a0lowland areas), except at the\u00a0highest flows, where optimistic climate scenarios (predominantly SSP1\u20132.6) instead simulate a\u00a0slight increase. Variability in\u00a0the\u00a0decline in\u00a0discharge is greater in\u00a0the\u00a0lowland catchment.<\/p>\nRiver basin\u00a0digital twin<\/h3>\n
<\/a>\nFig. 2. MIKE SHE model structure, showing the processes represented in the model (left) and spatially distributed input-parameter layers (right)<\/h6>\n
<\/a>\nFig. 3. Calibration results for mean daily discharge: spatial distribution of NSE values (left) and multicriteria selection of the final model run based on minimising RMSE for\u00a0simulated and observed groundwater levels (GWL) and maximising NSE for mean daily discharge (right)<\/h6>\n
Conceptualisation of\u00a0adaptation measures<\/h3>\n
<\/a>\nFig. 4. Process for designing a complex adaptation measure and assessing its impact<\/h6>\n
RESULTS<\/h2>\n
<\/a>\nFig. 5. Effects of climate change by 2050 on flow duration curves for mean daily discharge, including the P25\u2013P75 uncertainty interval of the climate scenarios, in\u00a0the\u00a0highland catchment (left) and lowland catchment (right)<\/h6>\n