Three stream sites were assessed in Belo Horizonte, the third-largest Brazilian metropolis (c. 5.5 million inhabitants) (Baleares, Primeiro de Maio, and Nossa Senhora da Piedade; Figure 1). The rehabilitation was conducted by the Drenurbs Program of the Belo Horizonte Municipal Government using Inter-American Development Bank financial support (60% of the US$ 14.53 million was spent in these 3 streams; BID, 2008). Before the rehabilitation interventions, the streams had deficient drainage infrastructure, human occupation of the streambank and floodplains; inputs of untreated sanitary sewage in the natural stream channels; disposal of garbage in streams, and deforestation and erosion on the streambank and hillside. The site interventions occurred between 2006 and 2007 and included: (1) sewage collection networks and sewage treatment; (2) improvement of stormwater drainage systems; (3) bank erosion control and stabilization via artificial structures like gabions and walls; (4) streambank stabilization using geotextiles and revegetation of riparian zones using woody tree species; (6) stream bed stabilization using fixed boulder clusters; (7) flood control systems using detention basins; (8) removal of houses from floodplains and riverbanks; and (9) creation of a protected area with riparian parkland (Figure 2). The pre-rehabilitation condition of each site differed concerning the degree of impact and the population affected, which required different solutions in terms of costs and populations relocated. For instance, a slum was present along Baleares stream, but Primero de Maio was bordered predominantly by degraded grassland and few houses. Because of local morphology, the Baleares parkland is steeper than the others and occupies less area (25–45 m wide, 21,900 m² total area). Primeiro de Maio and Nossa Senhora da Piedade parklands have similar morphology and occupy larger areas (50–100 m and 55–170 m wide and 33,500 m² and 57,200 m² in area for Primeiro de Maio and Nossa Senhora da Piedade respectively). Both also had structures installed to mitigate freshets (Table 1). All the parks encourage visitors because recreational (workout) equipment and walking paths were installed. Additionally, Primeiro de Maio and Nossa Senhora da Piedade have sport courts. Woody species were placed near the streams (∼10–15 m), but grass areas are maintained for public uses.

Additionally, three reference sites, located in urban protected areas and water-collection areas within the Belo Horizonte metropolitan area were selected as control sites (Mangabeiras Municipal Park, Rola Moça State Park, and Burle Marx Municipal Park). These three reference sites were also on headwater streams (3^rd order streams with < 2 km² total catchment area and 1–2 km long) with > 90% natural cover. They were chosen because they are the closest preserved areas to the rehabilitated streams and are in the best available ecological condition (Figure 3).

To assess changes in water quality, instream habitat, riparian conditions, macroinvertebrate multimetric indices (MMIs), and taxonomic composition of benthic macroinvertebrate assemblages in the rehabilitated stream sites, we used data collected between 2003 and 2011 so that each site was visited at least 4 times a year (twice in the dry season, and twice in the rainy season) (PBH 2012; Feio et al., 2015; Macedo and Magalhães Jr., 2020). Additionally, four sampling visits were carried out (two in the dry season, and two in the rainy season) in 2018–2019, in both rehabilitated and reference stream sites. This allowed separating our dataset into four groups: pre-rehabilitation (2003–2005), early post-rehabilitation (2008–2011), later post-rehabilitation (2018–2019), and reference sites (2018–2019). To assess citizen perceptions of the stream interventions, we used questionnaires applied to randomly selected households within 150 m of the 3 rehabilitated stream sites in 2008 (n = 215) (Medeiros, 2009; Macedo and Magalhães Jr, 2011), and in 2019 (n = 180).

Water quality was evaluated in-situ for temperature (^oC), pH, turbidity (NTU) and total dissolved solids (mg/L), using a multiparameter probe (YSI - 10 - Yellow Springs, Ohio). In addition, 2 L of water was collected during each stream site-visit, iced, and analyzed in the laboratory for dissolved oxygen (mg/L), biochemical oxygen demand (mg/L), total phosphorus (mg/L), nitrate (mg/L), and Escherichia coli (MPN/100 ml) (APHA 2005). These parameters were compared against national and state water quality criteria (Brazil, 2005; Minas Gerais, 2008; Table 2). Additionally, these 9 parameters were integrated into the Water Quality Index (WQI) used in the state of Minas Gerais, which classifies water status as very bad (< 25), bad (25–50), fair (50–70), good (70–90) and very good (90–100). To calculate the WQI, we used the weighted product of each evaluated parameter (Table 2) and the value of the parameter obtained through the specific average quality curve where each parameter result has a scale between 0 and 100 (e.g., 20% dissolved oxygen saturation is 12; pH of 6 is 52; turbidity of 10 is 76) (Supplementary Material S1).

To assess the physical habitat and riparian zone conditions, we applied a rapid assessment protocol (Callisto et al., 2002). This protocol allows a rapid assessment through visual observations of the ecomorphological conditions in the stream channel and the riparian zone of sites (Supplementary Material S2). The assessment was based on a set of environmental parameters scored from zero to four (referring to the use and occupation of the riparian zone and the apparent characteristics of the water); and from zero to five (flow and substrate conditions important for aquatic biota). The protocol is synthesized in a final score that reflects the level of conservation of the ecomorphological conditions of each stream site, where 0–40 represents impaired sites, 41–60 altered sites, and > 61 reference condition sites (Callisto et al., 2002; Feio et al., 2015; França et al., 2019).

We collected three bottom subsamples with a kick-net sampler (30-cm opening, 0.09 m² area, 500-µm mesh) in each stream site, preferably one subsample in pebbles/gravel, another in fine sediments (sands and silts), and a third in leaf deposits. The sub-samples were individually placed in plastic bags and fixed with 70% alcohol. In the laboratory, we washed the subsamples with tap water, sorted the benthic macroinvertebrates, and identified individuals to family, except to suborder for Hydracarina, class for Bivalvia, and subclass for Oligochaeta. The organisms were identified under a stereomicroscope, by using taxonomic keys (Fernández and Domínguez, 2001; Costa et al., 2006; Merritt et al., 2008; Mugnai et al., 2010; Hamada et al., 2014), and the three subsamples were combined into a single composite sample for each site visit. Benthic macroinvertebrates were deposited in the Reference Collection of Instituto de Ciências Biológicas, Universidade Federal de Minas Gerais.

The biological data of each stream were used to calculate four biological metrics that constitute a multimeric index (MMI; Macedo et al., 2016): Ephemeroptera-Plecoptera-Trichoptera (EPT) taxa richness; Average Tolerance Score per Taxon (ASPT), Percentage of Predator individuals, and Percentage of Odonata individuals. These metrics were standardized to the same scale (0–100) considering all sites together and then the metrics were divided by their 95th percentile and multiplied by 100. Values above 100 were considered super optimal and were scored as 100. The final MMI was calculated as the average of its standardized biologic metrics.

To evaluate citizen perceptions of the rehabilitated stream sites, questionnaires composed of multi-option objective questions were applied in loco at 180 randomly selected households (60 questionnaires, one for each home within 120 m of each of the three stream sites) in July 2019. The questions were guided to measure the use of linear parks, population satisfaction with the rehabilitation interventions, and questions related to the social dynamics concentrated in the study areas. The questionnaire was structured with qualitative and quantitative questions, and finalized after a pre-test survey (i.e., application of a questionnaire in its preliminary version in only few households to adjust it). The respondents did not have access to the options, instead the interviewers marked the answers within the pre-existing options, or in the option “other” where the answer was described. After tabulation, responses were classified (e.g., “great” and “good” were grouped in “great or good”). When necessary, multi-option answers were entered (i.e., an interviewer could mark one or more options). Answers also were post-processed because some of the “other” answers could be categorized during data tabulation. We compared those results with an earlier 2008 survey based on 215 questionnaires also applied randomly near the three rehabilitated sites (Medeiros, 2009; Macedo and Magalhães Jr, 2011). In both dates, we used Belo Horizonte’s official address database to randomly select our homes. We used systematic repositioning procedures to change a sampled household when a household presented a refusal. We considered all the households within 120 m as our study population, so we can infer results to 1,490 and 1,550 homes in 2008 and 2019, respectively (PBH, 2012). To maintain temporal comparability, an adaptation of the questionnaire applied in 2008 (Supplementary Material S3) was used in 2019 (Supplementary Material S4). The questions were organized to assess general interviewee profile (i.e., sex, age, and education), perceptions concerning the positive and negative points (multiple answers allowed), solution of environmental problems, general satisfaction with the interventions, and preference for riparian park versus stream burial by an elevated road. The latter has been a common practice in Belo Horizonte, where approximately 200 km of streams have been covered by roads (∼30% of the city’s streams; PBH, 2022). Riparian stream parks were a new idea before 2008 in Brazil.

Overall, the three rehabilitated stream sites showed progressive improvement in water quality over time. Most parameters within national Brazilian and Minas Gerais state legal limits for class 2 rivers (Brazil, 2005; Minas Gerais, 2008) were achieved in post-rehabilitation early and late phases, except for total dissolved solids, pH, water temperature, WQI, and Echerichia coli (Figure 4). Total phosphorus, dissolved oxygen, biochemical oxygen demand, turbidity, and WQI showed improvements, with most of the values in the later post-rehabilitation phase meeting class 2 water quality criteria (meaning that the waters were safe for human recreational primary contact). Nitrate concentrations were reduced, but were within the legal limits before the interventions. Temperature and pH remained somewhat higher than in the reference stream sites. All three reference stream sites met the federal and state water quality limits. The later post-rehabilitation stream sites had levels of total phosphorus, dissolved oxygen, biochemical oxygen demand, turbidity, and pH that were not statistically different from those in the reference stream sites. However, WQI scores differed significantly between later post-intervention stream sites versus reference stream sites, mainly because of higher levels of Escherichia coli and total dissolved solids.

Instream physical habitat and riparian zone conditions indicated improvement following rehabilitation (Figure 5A). Prior to rehabilitation, scores were concentrated in the impaired zone (< 40 points) and in the early and late post-rehabilitation periods, they were within the altered range (40–60), but showing progressive improvements. However, the scores were significantly lower than those of the reference stream sites (generally close to 80 points).

Similar to the water quality and physical habitat results, the MMI results indicated progressive improvement over time (Figure 5B) and MMI and physical habitat scores were highly correlated with each other (Pearson’s r = 0.96). Before rehabilitation, MMI scores were very low (< 0.2), but improved significantly in both post-rehabilitation phases. However, assemblage condition remained significantly below that in the reference sites.

The taxonomic composition of the benthic macroinvertebrate assemblages in the three rehabilitated sites differed from the reference sites in all sampling periods. The Bray-Curtis index scores indicated that all possible pairings differed significantly (Table 3) and the nMDS results show a noticeable shift in late post-rehabilitation towards the reference sites (Figure 6).

All urban riparian parks surroundings had similar profile about sex, age and education pattern (Table 4). More positive aspects than negative ones were reflected in the 2008 and 2019 questionnaire results (Figures 7A,B). As expected, some of the positive aspects were more evident soon after the interventions, such as cleaning the river (60 % vs. 5% in 2008 and 2019, respectively, X ² = 131.282; df = 1, p < 0.0001) or completing the park (51% in 2019 and 18% in 2008, X ² = 50.931; df = 1, p < 0.0001). Other aspects were mentioned only at one time period, such as improving road access by 25% in 2008 and removing the slum, which was answered by 9% in 2008, or creating a space for community use, which was remembered only in 2019 for 18%. Other positive aspects have similar values over time, with emphasis on the creation of leisure areas (45%; X ² = 0.46; df = 1, p = 0.831), improvement of aesthetic aspects (15 % and 20%; X ² = 1.64; df = 1, p = 0.2), decreased flood risk (2 % and 3%; X ² = 1.64; df = 1, p = 0.2, and improved real estate value (4% and 1.6%; X ² = 2.134; df = 1, p = 0.144).

Negative aspects were pointed out by fewer residents (< 15%), however with lower values 10 years after the rehabilitation, indicating good results from the interventions. Local citizen perceptions improved regarding increased security 13 %–3% in 2008 and 2019, respectively (X ² = 16.35; df = 1, p < 0.0001). Some aspects showed no differences between the 2 years: 3 % and 1.6% considered that road system worsened (2008 and 2019, respectively; X ² = 0.566; df = 1, p = 0.452) and people circulation were a negative aspect to 1.3 % and 3%; (X ² = 1.634; df = 1, p = 0.201). Other aspects were only mentioned after the interventions, such as the impact of construction disturbances (8%), the removal of neighbors and relatives (7%), and that drainage and sewage system problems had not been solved (5%).

In general, the acceptance of the interventions was and remains very high in 2019, with more than 83% approval by citizens, while 5% disapproved (X ² = 166.23; df = 2, p < 0.0001). In addition, 65% of respondents said that the environmental problems were solved, 30% partially, which reinforces social acceptance (X ² = 75.57; df = 2, p < 0.0001) (Figure 8). Finally, the survey results show that the linear stream park was preferred by 58% in 2008, but it was preferred by 90% versus stream burial in 2019, showing there has been a significant increase in people who prefer the rehabilitated stream after 10 years of rehabilitation interventions (X ² = 51.07; df = 1, p < 0.0001; Figure 9).

The implementation of sewage networks, garbage collection, streambank stabilization, and riparian vegetation recovery improved water quality to the degree that most indicators met most Brazilian national and state standards for Class 2 waters (Brazil, 2005; Minas Gerais, 2008). Those standards are deemed sufficient to protect aquatic communities, primary water-recreational activities, irrigation, aquaculture, and drinking water supply after conventional treatment. Reduced organic loads resulting from the rehabilitation interventions are verified by improved water quality indicator scores (Wantzen et al., 2019), including total phosphorus, nitrate, dissolved oxygen, turbidity, biochemical oxygen demand, and WQI. Four of those parameter values (pH, total phosphorus, turbidity, biochemical oxygen demand) did not differ significantly from those at reference sites (two of which are springs used for drinking water), indicating recovery of water quality for those parameters. However, total dissolved solids and water temperature, despite improvement, still exceed the desired limits. Water temperature and total dissolved solids concentrations are linked to surface runoff from the entire catchment, which is difficult to control without stormwater collection and treatment (Paul and Meyer, 2008; Loflen et al., 2016). Because of sewage collection, Escherichia coli and total dissolved solids now are related to surface runoff, which was not controlled in the rehabilitated sites, and which therefore lowered WQI scores. Achieving legal standards is a useful goal for rehabilitation projects (Hughes et al., 2014b); therefore, because Brazil lacks standards for physical habitat and aquatic assemblages (Feio et al., 2021), achieving water quality goals is a relevant outcome.

The Escherichia coli values were lower in early post-rehabilitation than in late post-rehabilitation, although they exceeded water quality standards in both cases as well as in pre-rehabilitation. Because Escherichia coli occur in the lower gastrointestinal tract of warm-blooded organisms, it would be useful to investigate whether the contamination in the latter post-rehabilitation phase is associated with increased incidence of animals attracted to the parks (Soller et al., 2010).

Several studies point to the importance of stream and riparian habitat structure in the ecological rehabilitation of streams (Bernhardt et al., 2005; Paul and Meyer, 2008; Hughes et al., 2014b). Our physical habitat data indicate that the interventions only resulted in improvement from impaired to altered—but well below reference condition (Callisto et al., 2002). It is noteworthy that some parameters related to the surroundings or the hydrological regime could not undergo major recovery because of the characteristics of consolidated urban areas (Hughes et al., 2014b; Loflen et al., 2016) and the markedly greater time usually required for recovering physical-habitat and flow-regime (which was not considered) than for water quality recovery (Hughes et al., 2014b).

Both physical habitat and water quality rehabilitation occurred to a greater degree than biological rehabilitation—which did not approach reference conditions–as has been reported by others (Charbonneau and Resh, 1992; Selvakumar et al., 2010; Kail et al., 2015; Macedo and Magalhães Jr., 2020). We observed improved macroinvertebrate assemblage composition and MMI scores over time, but we cannot state whether further improvement in biological conditions are possible because the urban catchments and migration/colonization barriers limit biological processes and MMI scores (Bond and Lake, 2003; Angold et al., 2006). Nonetheless, we found that use of an MMI, which combines and quantifies multiple measures of assemblage condition into a single indicator, provided a useful and easily communicated tool for expressing biological monitoring results over time as reported by Vadas et al. (2022). In addition, MMI scores closely tracked the physical habitat index scores (Figure 5), suggesting that further improvements in the latter may be more effective than further improvements in water quality. Our results also were similar to several other studies dealing with urban rehabilitation projects in Global North countries, which showed varied biological results, with only 5%–20% showing significant biological improvements (Al-Zankana et al., 2020). Those results were not always revealed by richness and diversity indices, rather, they were shown by relative abundances or taxonomic composition (Palmer et al., 2014; Kail et al., 2015). This expectation of varied responses reinforces the need to use multimetric indices in the evaluation of rehabilitation projects, because they present the response of multiple attributes of assemblages (Callisto et al., 2022). Regarding our family level identification, in the Neotropics, it is usually the highest possible taxonomic resolution to be achieved for many taxa, because of the huge biodiversity and limited taxonomic knowledge for most groups (Helson and Williams, 2013). Even when identification to a finer level is possible, it is often too work-intensive to be used in biomonitoring programs (Martins et al., 2022). In fact, many studies have shown that family level identification is able to attain the same accuracy as lower taxonomic level identifications in both biomonitoring and ecological studies (Bailey et al., 2001; Buss and Vitorino, 2010; Silva et al., 2016).

Although ecological conditions were not restored to natural conditions, we found that the interventions, especially the installation of riparian parks and sewage treatment, were viewed positively by citizens neighboring the stream sites. Furthermore, the results indicated that those people favored those options over stream burial, which is widely employed in large Brazilian cities. This indicates that those rehabilitation options can provide important ecosystem services, in addition to improving ecological and biological conditions (Ranta et al., 2021). Because the three stream sites had similar sociodemographic characteristics, we analyzed the data from the surroundings of the three urban parks together.

Despite being widely neglected in the evaluation of rehabilitation projects (Guida-Johnson and Zuleta, 2017), citizen perceptions are vitally important because they reveal taxpayer support or opposition regarding intervention operations over time (Bernhardt and Palmer, 2007). Furthermore, citizen support for ecosystem services stimulates more interventions of particular types. In Global North countries, wastewater infrastructure has been widely implemented (da Silva and Porto, 2021); but in Brazil industrial and municipal wastewater pollution is widespread in urban areas. Therefore, wastewater treatment was initially the most pertinent intervention for the local Belo Horizonte population, but subsequently parks became the preferred intervention.

Rehabilitated streams also showed improvement in the provision of ecosystem services in urban areas (Ranta et al., 2021). The creation of parks provides public use for recreation, which clearly provides cultural services related to mental and physical health and environmental education (Zhang et al., 2017). In addition, the rehabilitation of the floodplain-streambed system improves infiltration conditions, aquifer recharge and sediment and flood control (Mould and Fryirs, 2018), a fact perceived by the local citizen population. The current water quality could provide the service of public supply through simple treatment, even if in practice the quantity of water does not allow the feasibility of an action like this.

Our study reinforces the need for long-term monitoring of rehabilitation actions, which despite its importance, represents a major gap in rehabilitation projects (Hughes et al., 2014a,b; Al-Zankana et al., 2020; dos Reis Oliveira et al., 2020). It is also important to note that legislation has been a major motivator for environmental rehabilitation globally (Feio et al., 2021). Brazil has environmental laws that protect riparian vegetation (Brazil, 2012) and chemical and physical water quality standards for rivers, streams and lakes (Brazil, 2005), but neither mandates the rehabilitation of freshwater biological conditions. Considering the high monetary value of urban land, rehabilitation interventions are a great challenge, so legal instruments that encourage protecting, rehabilitating, and monitoring streams are essential (da Silva and Porto, 2021).