Simulation of structural and non-structural measures for climate change adaptation in water resources systems in the Capibaribe and Ipojuca River basins, Pernambuco state, Brazil Open Access

Some of the most significant impacts of climate change are the quantity and quality of water resources (Barlett & Dedekorkut-Howes 2022), and their direct influence on supply systems. Water supplies are expected to decrease, especially with regard to food supplies in arid and semiarid regions (Ribeiro Neto et al. 2014), leading to food insecurity, economic losses, and environmental degradation (Felipe et al. 2023). At the same time, variations in temperature and precipitation can stimulate the reproductive cycle of arboviruses and waterborne diseases (Kemajou 2022).

This complex and uncertain scenario presents many insecurities. Low resilience in water systems can lead to desertification, food shortages, and population migrations. However, an extreme event will only be recognized for its severity when it affects previously unaffected social classes or threatens everyday social customs (Zangalli Junior 2024). Notably, developing countries, such as Brazil, already have great difficulty combating water insecurity and meeting the demand for drinking water. The State of Pernambuco, for example, suffers from droughts in its countryside and water rationing in the Metropolitan Region of Recife (RMR) on the coast. Pernambuco has Brazil's lowest water availability per capita, with 88% of its territory characterized by a semiarid climate.

Even though Pernambuco has one of the lowest levels of water availability in the country, some municipalities register losses in the distribution system higher than 70% (SNIS 2020). Many factors can make it difficult to reduce this index: the disorderly growth of cities, lack of measurements on both the micro and macro levels, and weak loss control policies (ABES 2015). Furthermore, the lack of water itself (and intermittent supply) contributes to the failure to identify leaks and make possible repairs.

Policy- and strategy-based governance is necessary to mitigate the effects of climate change. This governance must include all actors in participatory management, where they collaborate to deal with information and risks (Timmerman et al. 2017; Hossain 2021; Barlett & Dedekorkut-Howes 2022). This strategy cannot be supported by only structural measures, but must also require a network of non-structural measures based on management policies.

Several measures to mitigate climate change effects can be found in the literature. Some examples are water transposition between basins, coupling of supply systems, construction of new reservoirs, increase in water quality protection, financial mitigation compensation, water tariffs (integration policies – investors and users), participatory management, and an integrated cooperation structure (Huntjens et al. 2012; Silva et al. 2015; Yin et al. 2017; Beran et al. 2018; Pinto et al. 2018).

Adaptive measures can be divided into two categories: (a) structural measures or ‘hard path’: focused on the growth of existing water infrastructure and increasing water supply, such as water transposition, reservoir construction, increased production of treated water, and new pipelines, among others; and (b) non-structural measures or ‘soft path’: focused on demand management and water conservation, such as incentive policies for water-saving equipment and rational water use, loss control in water supply systems, financial compensation mechanisms in cases of water use conflicts, and water usage charges, among others.

Adaptation initiatives usually focus on water-related structural measures, with fewer actions aimed at management (or non-structural) procedures. However, since the water crisis is already a reality, in addition to collecting water from increasingly distant sources, different initiatives come into play, such as water rationing, consumption reduction, financial incentives, and penalties; the development of water conservation plans; the replacement of hydrosanitary elements with energy-saving equipment; and so on. Applications of this nature are highlighted in Australia, California, Waterloo, and Barcelona (Leuck 2008; Imteaz et al. 2012; Mini et al. 2014; Waterloo 2014; Haque et al. 2015; Rath et al. 2016; Singh et al. 2016; Silva et al. 2017a; England 2018).

In general, the ‘hard path’ is most frequently chosen. This focuses almost exclusively on centralized infrastructure and decision-making: dams, reservoirs, supply networks, treatment stations, and water agencies. Concurrently, the ‘soft path’ may depend on centralized infrastructure, but complement it with extensive investments in decentralized facilities. It is anchored on efficient technologies and human capital, generally requiring less initial financial investment and greater collaboration between many different actors (Wolff & Gleick 2002).

This study aimed to evaluate the structural and non-structural adaptation measures in the Capibaribe and Ipojuca River basins in the state of Pernambuco, Brazil, to identify the best strategies to address climate change and ensure water security in areas already suffering water scarcity. Both basins are important for various economic sectors in the state of Pernambuco and have been the subject of studies, notably Ribeiro Neto et al. (2014) in the Capibaribe River basin, which also focused on water security. However, it did not analyze the perspective of which measures would be most effective in future climate change scenarios.

The methods are divided into (a) datasets (observed and simulated) of climatological variables (precipitation, air temperature, and relative humidity); (b) performance analysis of the GCMs; (c) bias correction; (d) estimation of evapotranspiration; (e) rainfall–runoff hydrological model; (f) trend analysis; (g) calculation of water demands; (h) preparation of the water allocation model; and (i) simulation of structural and non-structural measures in the basins.

The Capibaribe and Ipojuca rivers are two of the main watercourses in Pernambuco. Capibaribe River Basin reservoirs currently guarantee the water supply for the capital, Recife, and several other municipalities in its Metropolitan Region. A strong industrial presence in different segments characterizes the Ipojuca River basin (Pernambuco 2010a, b).

The average water consumption per capita is below 150 L/inhabitants/day for all municipalities in both basins, with an emphasis on Taquaritinga do Norte, which has the lowest water consumption per capita, approximately 30 L/inhabitants/day. The high loss rate during distribution can also be seen, ranging from 8.74 to 70.6%, while developed countries achieve indicators close to 5%. The National Basic Sanitation Plan (PLANSAB) specifies short-, medium-, and long-term goals for Brazil and its macro-regions. The plan is to have a loss rate of no higher than 31% for the national average and 33% for the Brazilian northeast by 2033 (Brasil 2014; ABES 2015; SNIS 2020).

Other alarming factors in the river basins under study are the low water service rate, which indicates that some of the population still lacks quality water in their taps, and the low basic sanitation coverage, which is also responsible for the high level of pollution in the rivers.

The datasets used for projections were composed of data generated by the Center for Weather Forecasting and Climate Studies of the National Institute for Space Research (CPTEC/INPE) (Chou et al. 2014a, b; Lyra et al. 2017) and made available on the PROJETA Platform alongside data from the Global Climate Model (GCM) HadGEM3 (Sheffield et al. 2006; Hempel et al. 2013). The data extracted from the models included monthly precipitation, average air temperature, and average relative humidity.

The PROJETA Platform is the result of work developed within the scope of the Climate Change Policy Program – PoMuC, an initiative of the Brazilian government coordinated by the Ministry of the Environment. Its objective is to process and convert climate data, allowing broad and unrestricted access to different users in a flexible way, regarding format, data volume, and various parameters.

Table 1 presents the global circulation models (GCMs), the projected scenarios, the model resolution, and the ‘realizations’ that were carried out. The realization number distinguishes members of an ensemble typically created by initiating a set of runs with various, yet equally plausible, initial conditions. Models with the prefix ‘Eta-’ have undergone dynamic downscaling using the Eta/CPTEC regional model (Chou et al. 2014a, b; Lyra et al. 2017). They have been made available on the Platform PROJETA, within the PoMuC's scope.

Performance-wise, if the NRMSE coefficient is below 10%, it indicates an excellent simulation. For NRMSE values between 10 and 20%, simulations are considered good. Values between 20 and 30% indicate satisfactory performance; and coefficients above 30% indicate results that are considered insufficient. Simulations are considered satisfactory for PB coefficient and monthly data when the PB is between ‒25 and +25% (Tian et al. 2016; Hadinia et al. 2017; Ba et al. 2018; Zolghadr-Asli et al. 2019).

Precipitation data were applied in model performance evaluations and bias corrections. These data were obtained from the hydrometeorological network of the National Water and Sanitation Agency (ANA) and the Pernambuco Water and Climate Agency (APAC). This dataset encompasses 74 stations, whose consistency was verified in Inocêncio et al. (2021) and has a timeframe of 1969–2010. Observed air temperature and relative humidity data were obtained from three weather stations operated by the National Institute of Meteorology (INMET).

MODHAC (self-calibrated hydrological model) (Lanna 1997) is a concentrated rainfall–runoff model whose input variables are average precipitation, potential evapotranspiration, and streamflow. Three reservoirs represent these processes and transform rainfall into runoff. The reservoirs simulate interception, evapotranspiration, and runoff generation, determining the volume of water that will infiltrate into the soil or flow onto the surface. The model has 14 parameters that can be automatically calibrated using four objective function options. MODHAC hydrological simulations performed well in several basins in northeastern Brazil's semiarid region (Lanna 1997). The process of calibration and validation of the hydrological model is described in Pernambuco (2010a, b), based on three stream gauges on the Capibaribe River and two on the Ipojuca River. The application of MODHAC in the semiarid region (Ribeiro Neto et al. 2014; Viraes & Cirilo 2019) encourages its use in similar areas, such as the Capibaribe and Ipojuca River basins. Moreover, the model requires little input data (precipitation and evapotranspiration) and can be simulated at a monthly time step (Ribeiro Neto et al. 2014).

The Mann–Kendall test was applied to the series of streamflow data provided by the MODHAC and to the precipitation data from the GCMs to identify trend changes, as well as to detect its starting point.

The LabSid AcquaNet 2013 allocation model was developed by the Decision Support Laboratory (LabSid 2013). Its interface uses geographic information system (GIS) technology and can seamlessly analyze complex water resource systems. LabSid AcquaNet 2013 has analysis tools capable of dealing with most problems related to water allocation in basins. In short, it is a streamflow network model for simulating river basins, in which the user can assemble networks with reservoirs, demands, and sections of channels, representing the problem under study in detail (LabSid 2013).

The model was built and assigned various stages, priorities, and objective functions to meet the demands and their respective priorities. The first step is constructing a system that constitutes the studied basins. The program has four distinct elements: (i) reservoir, (ii) passing node, (iii) demand, and (iv) link. Each component has a specific function and requires a set of information. Among them, priorities and objective functions will guide demand hierarchy and reservoir rules of operation.

Only reservoirs with an accumulation capacity equal to or greater than 5 hm³ were considered for modeling purposes. The 11 reservoirs considered can accumulate a total of approximately 908 hm³. The initial volume contemplated for all reservoirs was half the maximum, not to indicate either a favorable situation (full reservoir) or a critical problem of scarcity (empty reservoir). The target volume priority and target volume indicate which reservoir will have priority in the allocation model and what maximum volume should remain. The maximum volume was adopted as the target for all reservoirs except those whose primary purpose is flood control (where the target is to maintain half of the maximum). The natural reservoir inflow was obtained from the hydrological model, and the evaporation rate was obtained using Equation (6).

The model interprets demand as the streamflow necessary for a given use. Only five uses were considered to create the water allocation model: human and industrial supplies, water for livestock and other farm animals, irrigation, and energy generation. In line with the National Water Resources Policy (Brasil 1997) and the State Water Resources Policy (Pernambuco 1998), the demands for human and animal supply were prioritized over-irrigation, industrial, and energy generation. To create a correct model that reliably represents the Capibaribe and Ipojuca River basins, information from hydro-environmental plans (Pernambuco 2010a, b) was used, as described in the Supplementary Material. The assessments of allocation scenarios were then based on basin demand deficits: the smaller the deficit, the better the response of the measure or set of measures considered.

Finally, a return rate of 80% was considered, which is commonly used in sewage projects and in studies (Tucci 2008). It is worth noting that this Brazilian National System for Water and Sanitation Data (SNIS) supply data also include part of the industrial supply, as the volume reported is produced by the local sanitation company and consumed by various users (residences, commerce, and industries, and among others). In addition, the water grant data provided by APAC in April 2020 was considered for industrial demand data.

Updated BEDA values were taken from the 2017 IBGE Agricultural Census database (IBGE 2020). This census obtained the number of animals in the basins, by municipality. Considering each municipality's occupancy percentages, the demand was divided by the hydro-environmental plan's analysis units (AU). For water consumption by animals, the return rate was considered zero, i.e., all water is consumed, according to hydro-environmental plans (Pernambuco 2010a, b).

As with the industrial sector, demand calculations for energy generation were based on the water grant data provided by APAC in April 2020. Only one current water grant, 84,000 m³/day, was registered for the Ipojuca River Basin. A 100% return rate was considered for this demand. No power generation concessions were identified in the Capibaribe River basin.

The irrigation projection scenario considered the growth rate used in the hydro-environmental plans for the Capibaribe and Ipojuca Rivers, which was 0.5% and 1.0% per year, respectively. Equation (13) was then applied, replacing the population variables with the total irrigated area. The projection for livestock and other animal supply and human supply was calculated similarly using Equations (11)–(13), except for the input data being BEDA instead of population.

No adjustments are foreseen for industrial demands and energy generation in the future scenario. The former is integrated into human supply, as previously highlighted. As for the latter, it is understood that the current scenario already encompasses all energy generation opportunities within the basin.

The allocation scenarios were built based on six distinct control keys, namely: (a) population growth; (b) construction of new reservoirs; (c) water transfer; (d) control/reduction of losses in water distribution systems; (e) implementation of water conservation measures in buildings; and (f) variable allocation according to the drought index (SPI-24). The 24-month scale was chosen to better analyze some processes affected in the medium term, such as supply reservoirs. Notably, part of the basin has intermittent watercourses, with streamflow only during the rainy season. However, this study adopted a 24-month scale because most reservoirs are located on perennial rivers, which will suffer from streamflow reduction on a larger scale than intermittent rivers.

Each key presents two possible scenarios, except for key E, which presents three possibilities. This scenario has the key ‘off’ (represented by the color gray) and at least another one ‘on’ (represented by the color green). Table 2 summarizes the description of all keys and possibilities.

Table 2

Keys to the simulation scenarios of structural and non-structural measures in the basins

Key .	Code .	Description .
A		The scenario assumes that the population, irrigated area, animal husbandry, and so on did not expand or reduce; it considers the basin according to the base period
		The scenario considers population, irrigated area, and animal husbandry expansion or reduction
B		The scenario in which no new reservoirs are built
		The scenario considers that the planned reservoir (Engenho Maranhão Dam) for the Ipojuca Basin is built
C		The scenario considers that there will be no transfer to the basins
		The scenario considers the projects to transpose water from the São Francisco River to the studied basins as completed
D		The scenario predicts that no interventions will be carried out in the distribution network, and the same loss rate will continue
		The scenario foresees improvements in the system, meeting the target for the loss rate set out in the Sanitation Plan
E		The scenario does not foresee policies that encourage water conservation in buildings
		The scenario foresees incentives through public policies for water conservation in buildings, expecting a 5% reduction in industrial water consumption and 10% in residential buildings
		The scenario foresees a minimum consumption of 100 L/inhabitant/day. (94% of municipalities have consumption below 100 L/inhabitant/day; an average of 70 L/inhabitant/day)
F		The scenario in which no changes are foreseen in the granting instrument
		The scenario in which a reduction based on the SPI-24 drought indicator is applied depends on the basin conditions (mild, moderate, or severe drought)

Key .	Code .	Description .
A		The scenario assumes that the population, irrigated area, animal husbandry, and so on did not expand or reduce; it considers the basin according to the base period
		The scenario considers population, irrigated area, and animal husbandry expansion or reduction
B		The scenario in which no new reservoirs are built
		The scenario considers that the planned reservoir (Engenho Maranhão Dam) for the Ipojuca Basin is built
C		The scenario considers that there will be no transfer to the basins
		The scenario considers the projects to transpose water from the São Francisco River to the studied basins as completed
D		The scenario predicts that no interventions will be carried out in the distribution network, and the same loss rate will continue
		The scenario foresees improvements in the system, meeting the target for the loss rate set out in the Sanitation Plan
E		The scenario does not foresee policies that encourage water conservation in buildings
		The scenario foresees incentives through public policies for water conservation in buildings, expecting a 5% reduction in industrial water consumption and 10% in residential buildings
		The scenario foresees a minimum consumption of 100 L/inhabitant/day. (94% of municipalities have consumption below 100 L/inhabitant/day; an average of 70 L/inhabitant/day)
F		The scenario in which no changes are foreseen in the granting instrument
		The scenario in which a reduction based on the SPI-24 drought indicator is applied depends on the basin conditions (mild, moderate, or severe drought)

View Large

There are 96 possible scenarios; however, only ten allocation scenarios were simulated using data from the five GCMs and the two climate scenarios (rcp4.5 and rcp8.5). Table 3 presents the simulated allocation scenarios, the corresponding periods (base or future) applied, and whether the applied measure increased supply or reduced/increased demands.

Table 3

Simulated allocation scenarios

View Large

Scenario does not present any management measures in the basins and does not consider future demands (population, irrigation, animal husbandry, and industrial). Thus, it uses the 2018 consumption and population data from the SNIS (SNIS 2020) and data on irrigated areas and animal husbandry from the 2017 Agricultural Census (IBGE 2020). This scenario is the only one projected for the base period, as it represents the current situation of the basins. Scenario is identical to scenario , except that it considers the projection of demands.

Scenario not only considers the projection of demands (population, irrigation, animal husbandry, and industrial) but also anticipates the construction of the Engenho Maranhão reservoir in the Rio Ipojuca Basin. Scenario assumes the completion of the São Francisco River transfer (PISF from the Portuguese acronym) through the Agreste and Alto Capibaribe pipelines. These two scenarios correspond to ‘hard path’ measures for water resource management, foreseeing an increase in supply.

Another point to consider is that the São Francisco River Basin has a lower risk for water allocation because it transfers not only the imposition of the priority system (human supply) but also the monetary and volumetric risk (Silva et al. 2017b). It is worth mentioning that this project is currently in the testing phase. Other water transfers are underway to serve the basins, but they were considered because they are transfers within the state of Pernambuco itself.

Scenario , like all scenarios, forecasts demand projections and reduction of losses by water supply systems, with losses limited to 33%. Municipalities already meeting this condition maintain their current status. Scenario anticipates the adoption of water conservation measures in buildings, aiming for a 10% reduction in water consumption. Scenarios and represent water demand management measures in the basins, as does scenario , which establishes operation/permit rules for drought cases indicated by the SPI-24 drought index.

The principle of this key is that when activated (), the allocation will vary according to the standard precipitation index (SPI) proposed by McKee et al. (1993), thus assessing the effects of drought on runoff and water reserves. The time scale used was 24 months. The two basins were divided into two AUts to obtain the drought indices. One unit refers to the basin's upper part (UA1 and UA2), and the other refers to the basin's lower part (UA3 and UA4).

For each analysis unit, the average precipitation was calculated, and then the drought indices were obtained following the methodology of McKee et al. (1993). The classification adopted is the same as that of the original formulation, and for each dry season, a deflator was applied and incorporated into the allocations, as shown in Table 4.

The drought index was calculated considering the last 24 months. If it pointed to a dry season, a deflator was applied to a group of demands, as outlined in Table 4. If it indicated a rainy season, no deflator was applied. Therefore, upon detecting a drought period, users would be subjected to water rationing, thus avoiding the case where water reserves reach a critical state before more drastic measures are taken and stricter rationing is implemented, as typically occurs.

Scenario is the only scenario that foresees increased water supply to reduce water scarcity experienced in municipalities with a per capita water indicator of less than 100 L/inhabitant/day.

Finally, scenario encompasses all of the aforementioned measures: construction of reservoirs, completion of water transfer, reduction of losses in water supply systems, water conservation in buildings, and operation/allocation rules, in addition to projecting population growth. In contrast, scenario foresees increased demand in municipalities facing water stress instead of considering water conservation measures in buildings.

The HadGEM3 model has six equally possible runs. The performance tests were therefore applied to select the best run. In parallel, tests were also conducted on the ‘Eta-’ models. The ‘Eta-’ and HadGEM3 models were analyzed according to the baseline period presented in Table 1. Table 5 displays the performance test results.

The performance tests revealed that r1 was the best realization of the HadGEM3 model for the study basins, with NRMSE = 16.58% and PB = −15.79%, considered good by NRMSE (between 10 and 20%) and satisfactory by PB (between −25 and +25%), according to Zolghadr-Asli et al. (2019). It is worth noting that all realizations of the HadGEM3 model exhibited acceptable performances for PB and good results for NRMSE. Among the ‘Eta-’ models, only the Eta-MIROC5 showed satisfactory results for PB. However, this same model was the only one that presented an NRMSE value greater than 25%. From this point on, realization r1 will be used to represent the HadGEM3 model. All ‘Eta-’ models will be used in the following steps of the study.

The worst-case scenario regarding water availability predicts a reduction in the minimum annual rainfall of up to 82% (Eta-CanESM2, RCP8.5 scenario), in line with the results of Silveira et al. (2016). More favorable scenarios for water availability forecast an increase in the median annual precipitation by 266.9% (Eta-BESM, rcp4.5 scenario). Additionally, the Eta-BESM model predicts very different outcomes than other models, indicating a significant increase in rainfall. Gomes et al. (2020) also found increased precipitation during the dry season when simulating Eta-BESM-OA in the Madeira River basin; however, all scenarios indicated a reduction in precipitation during the rainy season.

The Eta-CanESM2 model, RCP8.5 scenario (the most unfavorable model), projects a decrease of 78% in the minimum annual precipitation. The Eta-BESM, rcp4.5 scenario suggests an increase in the median annual rainfall by 239.5%. The variations in precipitation by model and scenario for the Capibaribe and Ipojuca River basins are shown in Table 6.

Thus, in the Capibaribe River basin, the projected annual precipitation ranges from a minimum of 73 mm (Eta-CanESM2, RCP8.5 scenario) to a maximum of 3,239 mm (Eta-BESM, rcp4.5 scenario). The same scenarios in the Ipojuca River basin indicate a range of 100–3,190 mm.

The CanESM2 model also proved to show the most significant reduction in precipitation among the models analyzed by Silveira et al. (2016). The same study found that the HadGEM2-ES model did not represent the seasonality of precipitation in the São Francisco River Basin very well. Conversely, the results presented here show a different behavior when comparing Eta-HadGEM2-ES with other regional models (‘Eta-’ models).

The results from the HadGEM3 model were closest to the average. In the observed period, the average annual precipitation for the basins was 724 and 792 mm/year, while the forecast for the period 2051–2080 was 789 and 880 mm/year, representing an increase of approximately 9 and 11% in the Capibaribe and Ipojuca basins, respectively. Meanwhile, the temperature increased by an average of 1.8 °C in the model projections, contributing to elevated basin evapotranspiration, as per Supplementary Material. This behavior agrees with the findings of Ribeiro Neto et al. (2014) in the Capibaribe River Basin, using the Eta-CPTEC and HadCM3 models.

The increase in maximum annual evapotranspiration is predicted by all models, including the most optimistic one (Eta-BESM, rcp4.5 scenario): 1% for the Capibaribe River basin and 1.7% for the Ipojuca River basin. In contrast, the Eta-CanESM2 model, RCP8.5 scenario, predicts an increase of 21.4% and 21.9% for the Capibaribe and Ipojuca River basins, respectively. These latter numbers represent absolute values of 1,045–1,248 mm/year for the Capibaribe Basin and 1,052–1,251 mm/year for the Ipojuca Basin. The variations in evapotranspiration by model and scenario for the Capibaribe River and Ipojuca River basins are shown in Table 7.

Regarding the evapotranspiration results, there is no unanimity in the literature. Gomes et al. (2020) found a reduction in evapotranspiration in all Eta-BESM-OA scenarios.

In the Mann–Kendall test, non-stationarity is identified when the null hypothesis is rejected, meaning when the Z-value is not within the bilateral interval values of −1.96 to 1.96 for a significance level of 0.05. The observed and historical series of precipitation and streamflow data did not show significant trend changes, meaning that the null hypothesis H0 was not rejected.

In general, the models did not show trend changes, other than Eta-CanESM2, which did not reject the null hypothesis for precipitation in all sub-catchments in the rcp8.5 scenario and which also reflected a trend change in the streamflow in most sub-catchments. Notably, all changes recorded in precipitation were identified in the decade of the 2050s, between 2055 and 2057.

The Eta-CanESM2 model, rcp8.5 scenario, showed the greatest trend changes, with all sub-catchments showing changes in precipitation and most reproducing the effects in their streamflow. All sub-basins of the Ipojuca River and Jucazinho in the Capibaribe showed an increase in maximum streamflow, ranging from 7% at the mouth (Ipojuca) to 410% in Belo Jardim (Ipojuca). The sub-basins of the Capibaribe, except for Jucazinho, showed reductions in average streamflow ranging from 78 to 85% and maximum streamflow varying from 65 to 82%. Generally, the sub-basins closer to the source showed increases in average and maximum streamflow (Jucazinho, Sanharó, and Belo Jardim), while those closer to the outlet showed reductions in average streamflow, with the difference being that the Capibaribe sub-basins also had reduced maximum streamflow, whereas streamflow in the Ipojuca River sub-basins increased.

In the rcp4.5 scenario, changes were identified in only two sub-basins of the Capibaribe: Cursaí and Complemento da UA4. In both cases, there was an increase in the maximum streamflow of 220% to 281% (Complemento da UA4) and a reduction in average streamflow from 16 to 26% in Cursaí.

In the BESM-ETA model, three sub-basins with trend changes in flows were identified in the rcp8.5 scenario, all in sub-basins of the Capibaribe River: Várzea do Una, Goitá, and Complemento da UA4. In all three, a reduction in the average streamflow of about 30% and a decrease in the maximum streamflow of 10–16% were observed. The worst results were found in the Complemento da UA4 sub-basin, where maximum streamflow decreased from 23.806 to 17.942 m³/s, and average streamflow decreased from 21.272 to 14.259 m³/s. The other models did not show significant trend changes at a significance level of 0.05.

Before running the allocation model simulations for future climate scenarios (rcp4.5 and rcp8.5), the allocation model was first run with the corrected PHA baseline data. The inflow into the reservoirs was calculated by the MODHAC using precipitation and evapotranspiration. The results generated by AcquaNet were summarized and condensed into three perspectives: accumulated deficit volume, frequency of reservoirs below target volume, and minimum supplied flow. The accumulated deficit volume corresponds to the total demand volume that was not met over the 30 simulated years. The frequency of reservoirs below the target volume indicates how long the reservoirs failed to maintain the target volume, which is the maximum reservoir volume in most cases. The minimum supplied flow is the system's lowest flow over the 30 simulated years to meet all demands.

Table 8 presents the accumulated deficit volume, frequency of reservoirs below target volume, and minimum flow of the allocation model, reproduced with baseline data.

All corrected base period data, except for Eta-MIROC5, showed significant accumulated deficit volumes (Mm³), even though some reservoirs never empty completely. These results align with the deficits indicated by the basins' hydro-environmental plans. The primary affected users are those in the Ipojuca River Basin: energy, irrigation throughout the basin, human supply, and livestock and farm animal supply in UA1.

Analysis of the information provided in the Ipojuca Basin Hydro-environmental Plan – Volume IV (Pernambuco 2010c) shows that the HadGEM3 model result is closest to the observed data, considering that the plan does not estimate water deficits in the Capibaribe River Basin. Furthermore, the PHA's water balance indicated a deficit of 45.31 Mm³/year for the Ipojuca River Basin (Pernambuco 2010c). Another relevant factor is that this model is the only one that demonstrates that the reservoirs reached their target volumes at some point in time, which aligns with the observed data.

The minimum flow provided in the Eta-HadGEM2-ES model, despite being close to most other models, still presents the highest accumulated deficit volume. This behavior is explained by lower flows recorded in UA1 (Ipojuca River Basin), resulting in up to 13 consecutive months with flows below the necessary demand, representing about 10% of the total time below the required demand. Some other models had similar minimum flows: Eta-BESM with a maximum of six consecutive months and 2.22% of the time below the required demand; Eta-CanESM2 with a maximum of eight consecutive months and 7.78% of the time below the required demand; and HadGEM3 with three consecutive months and 1.67% of the time below the required demand.

With the reduction in precipitation and increase in temperature, a decrease in average streamflow was expected, according to the results of Ribeiro Neto et al. (2014). Schuster et al. (2020) found similar responses when simulating HadGEM2-ES and CanESM2 in the Lagoa dos Patos Basin, but an increase in streamflow when simulating the MIROC5 model.

All scenarios of the allocation model were then simulated by applying the rcp4.5 and rcp8.5 scenario data from the GCMs. Thus, ten simulations were conducted for each GCM scenario. Table 9 presents the results for minimum flow in different scenarios and models.

Table 9

Minimum flow (m³/s) for the rcp4.5 and rcp.8.5 scenarios of the GCM for all reservoirs in the two basins

View Large

Scenarios and represent situations without implemented measures, but includes population projections and, consequently, all demands. It is observed that, except for the GCM Eta-CanESM2 in both scenarios, all other GCMs managed to keep up with the demand increase and provided higher flows throughout the period, ensuring more significant minimum flows.

The structural measure of building new reservoirs, indicated by key , shows few significant results, including negative results (Eta-BESM, rcp4.5) or null results (Eta-CanESM2, rcp4.5, and rcp8.5). The PISF, represented by key , showed the best individual results, improving supply almost threefold in Eta-CanESM2 (rcp4.5). On average, all other GCMs showed an improvement in supply by almost double.

For non-structural measures, a reduction in demand was detected in the scenario of loss reduction (represented by key ), in the application of water conservation measures (represented by key ), and in changes in the allocation instrument (represented by key ), which is reflected in the necessary minimum flow. Setting the minimum per capita consumption to 100 L/inhabitant/day (represented by key ), however, resulted in an increase in demand, which led to higher minimum flows, indicating that the basins would be able to slightly better meet the population's needs most of the time. By combining all actions, either through the conservation measures path () or through increasing minimum water consumption (), the basins had higher minimum flows, demonstrating that, most of the time, the system is able to offer more water.

The Eta-BESM model is the only one in which the reservoirs do not fail most of the time for most scenarios, being the model with the highest flows, as previously explained. The Eta-CanESM2 model has the lowest flows, and consequently, the reservoirs never reach their target volume, being near empty most of the time. It is necessary to analyze the existence and size of the accumulated deficit volumes in order to better understand the allocation scenario results. Table 10 presents the accumulated deficit volumes for all models and scenarios.

Table 10

Accumulated volume of deficits (Mm³) for the rcp4.5 and rcp.8.5 scenarios of the GCMs for all reservoirs in the two basins

View Large

The Eta-BESM GCM did not show accumulated deficits, revealing that the minimum flows obtained are equal to those necessary to meet all demands, with reservoirs primarily used to regulate flow.

The Eta-CanESM2 GCM had the worst results. If this model's projections materialize, combined with the lack of implementation of any structural or non-structural measures, the deficit volume could range from 4,494 to 1,3,085 Mm³.

Overall, the measure that individually presents the best results is the PISF (represented by key ). However, when considering other elements, such as necessary investment, adaptive measures with almost zero implementation costs, and changes in the allocation instrument (represented by key ), simulations show median results that can become good when combined with other measures. Ribeiro Neto et al. (2014) also concluded that the combination of infrastructure development and improvement in water management policies can result in effective measures for climate change adaptation.

Among the non-structural measures, creating an incentive for water conservation in buildings proved to be the best option, with the highest returns. However, this measure depends on a broad educational and social campaign to change people's behavior and on financial support from governmental entities.

The Eta-CanESM2 model shows demand deficits over 70% of the time, with the maximum value reaching almost 20 m³/s in the rcp8.5 scenario, which occurs about 50% of the time. As expected, the Eta-BESM, which presents scenarios that have higher precipitation, indicates that demands are met 100% of the time. Except for Eta-CanESM2, all other scenarios did not show deficits for more than 50% of the time.

Demand deficits are reduced in all models and scenarios. However, the Eta-CanESM2 model, rcp8.5, continues to show deficits approximately 90% of the time. The frequency of non-compliance in all models, except for Eta-CanESM2, is about 10%, with a maximum flow of 2 m³/s.

In light of the uncertainties regarding water availability in future scenarios, in-depth study and analysis of measures that can ensure water security are essential. Among the models studied, the Eta-BESM showed results with the highest volume of water availability, while the Eta-CanES2M showed the lowest values. This discrepancy among the models highlights uncertainties that must be investigated in more detail. Nevertheless, it is important to mention the agreement between the outcomes in the other models. Furthermore, the Eta-CanES2M also showed a trend change in both analyzed scenarios (rcp4.5 and rcp8.5).

Based on the allocation model outputs, all adaptive measures are essential for the sustainable development of the river basins under study. However, it is understood that difficulties in implementation, especially in the allocation of financial resources, constrain the development of all measures.

Among the ‘hard path’ measures (structural), construction of new reservoirs should not be a priority, as they do not yield significant gains, especially when considering construction costs, with little overall supply gain in scenarios where the benefits are negligible.

On the other hand, the PISF appears to provide the most significant water security. Among all solutions, it offers the highest increase in supply, but it requires the most substantial investments. This measure is a priority, a goal of public policies to combat drought, aiming at improvements in several other basins, including those in different Brazilian states. The costs of implementing the PISF have already exceeded BRL 12 billion.

The scenario aiming to increase per capita consumption to at least 100 L/capita/day has shown that it may be possible most of the time, as demonstrated by the increased available flow. Implementing this scenario would require expanded production of treated water and improvements in pipelines, water treatment, and distribution.

Regarding the ‘soft path’ measures, changes in operational rules based on drought indices, reduction of loss rates, and water conservation measures in buildings presented satisfactory results very similar to one another, viewed from the perspective of available flow. However, from the deficit perspective, water conservation showed the best results, followed by loss reduction and operational rules. From a financial perspective, operational rules are the simplest and can be implemented immediately. It is important to note that several other operational rules could be analyzed, in addition to optimizing the best drought index, which could predict a different SPI for different regions of the basins. For example, SPI-12 for the upstream region (intermittent rivers) and SPI-24 for the downstream region (perennial rivers). In addition, the operational rules could be discussed for other users, particularly irrigation.

It is important to highlight that even though the implementation costs of structural measures are higher than those of non-structural measures, an in-depth cost-benefit analysis relating water security to the associated costs would be ideal.

Furthermore, several other scenarios can be envisioned that would consider various other measures, such as the implementation of other management instruments like charging and classification of water bodies, which could constitute relevant resources for basin conservation through riparian forest revitalization, promotion of sanitation, and the guarantee of investments for the improvement of irrigation technologies. Quantifying the benefits of these measures in water resources management would take considerable work, even though they are considered essential measures for the system.

Another relevant factor is that water quality should be emphasized, considering that the deficient water quality of the Capibaribe and Ipojuca Rivers is already well documented. The adaptive measures proposed here focus only on water quantity. Nevertheless, the literature points out several adaptive measures that focus on ensuring water quality in uncertain scenarios, such as reforestation, improvement in sewage treatment (including technological advances), and permeable pavements, among others. Significant financial resources need to be allocated to prevent water quality from being a prohibitive factor in the use of these water sources. Overall, the cost of adapting water supply systems is expected to be most significant in Latin America (Narain et al. 2011).

Public agencies must examine the ‘soft path’ measures (non-structural) more closely, as they require fewer financial resources and can yield satisfactory results in various scenarios. However, in some scenarios, these measures are not sufficient to meet all demands, and other solutions must be considered.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.