The objective was to observe whether land use and land cover changes and precipitation affected streamflow trends from 1985–2019. Based on the analyses of the data, it has been shown that streamflow is affected by changes in the river basin, but the results were not conclusive, showing various variations. From 2011 on, forest area remained stable, although it showed a slight downward trend when compared to the beginning of the historical series. This tendency to reduce forest areas, especially in Caatinga areas, was also observed by Dutra (2019) for the entire state of Bahia, Brazil. In 2019, this class accounted for most of the Ribeirão da Caveira River Basin (RCRB) (Fig. 4). One reason for this stagnation could be the economic instability in the region over the last decade.
Agriculture area in the basin showed a trend similar to that reported by Silva (2019) when evaluating a basin in the southern region of Bahia state. The author has identified, in the Buranhém River Basin, an increase in agricultural area, mainly in the 1990s, and a tendency of recovery in the forest area after 2006. Overall, it is possible to observe a significant reduction in forest areas in Bahia in recent decades (Santos et al. 2020).
Bare soil area increased over the analyzed period, reaching its maximum value in the last three years of the time-series. This class represents a relatively small fraction of the basin; however, it is an important indicator of human presence, working as a measure to estimate the increase in population and demand for natural resources. Exposed soils occur mainly in locations characterized by strong human presence, in addition to rocky formations and urban infrastructure, according to the classification of MapBiomas (Belmont 2018; Gavioli & Hossomi, 2020).
Even with stationary behavior, precipitation can influence streamflow rate (Mendes et al. 2018), mainly because precipitation, in certain situations, is the agent of greater impact on flows in a basin (Gupta et al. 2015). These authors also observed that the impacts of changes in land cover have been increasing in recent decades due to the acceleration in urbanization. In analyzing land use and streamflow rates, reduction in forest area and increases in bare land and agricultural areas may have had a negative effect on Qmin in the RCRB. The results contrast with those of Dias et al. (2015), in which increasing trends in streamflow rate were identified due to the replacement of forests by other land uses in Amazon basins. Nevertheless, the increased demand for water resources and changes in land use can lead to a reduction in streamflow rates over time (Ferreira et al. 2020; Andrade and Ribeiro 2020). It is important to highlight that changes in land cover significantly impact streamflow dynamics, as well as climatic factors (Yin et al. 2017).
The existence of a correlation between flow and precipitation (Pws and Pa) is common because, in certain situations, the hydrographic regime of a basin is directly influenced by precipitation (Shao et al. 2018). The inverse relationship between BSI and streamflow can be an indication of the influence of vegetation cover on the surface runoff of a watershed (Bart et al. 2021).
Overall, land use and land cover classes showed weaker correlation with hydrologic and climatic variables, indicating that changes in land use may not significantly influence streamflow rates in the RCRB. Yin et al. (2017) reported a different result when evaluating a basin located in zone transitioning from semi-arid and semi-humid climate, in which land use was an important factor for the hydrological cycle.
The fitted models showed relatively satisfactory results for Qmean and Qmax. Qmean were positively influenced by rainfall regime in the wettest semester and negatively influenced by the intensity of exposed soil in the area (BSI). These variables were significant by the t-test (p < 0.05). Ferreira et al. (2021) obtained satisfactory R²a values (> 0.49) when modeling land use and cover and precipitation data from the Santo Antônio River Basin, in Minas Gerais state. According to the authors, Pds played an active role in the dynamics of flow regime.
Maximum flow rate, Qmax, was negatively correlated with Pds and BSI within the RCRB, as shown by the model with the highest R²a (Table 4). Taking into account the significance of both variables, one explanation for the negative action of Pds is the increase in surface vegetation cover due to greater water availability for plants. Thus, the greater presence of vegetation cover can increase the intensity of rainfall interception and, consequently, reduce runoff (Bart et al. 2021).
Of the explanatory variables, Psc and BSI had the most common association for the set of fitted models. The presence of bare soil in the basin area due to urban expansion and degraded areas may be an indication of the influence of increased water demands in the basin on the tendency to reduce flow regime (Dibala et al. 2020).
The different classes of land use and occupation did not have a significant influence on streamflow rates. Albuquerque et al. (2018) obtained similar results when analyzing land use and flow regime in the Verde Grande River Basin, in Minas Gerais, noting that changes in soil surface had little influence on streamflow. Similar results were obtained by Gupta et al. (2015), stating that precipitation is the factor with the greatest impact on streamflow rates. On the other hand, Aredo et al. (2021) observed that changes in land use, particularly the expansion of built-ups and agriculture, significantly affected streamflow, and were able to both increase and reduce the average monthly streamflow in the rainy and dry periods, respectively.
According to the models with the highest adjusted R², forest biomass has no significant effect on streamflow rates (Table 4). These results differ from those reported by Bart et al. (2021), in which the reduction of forest biomass increased streamflow in the rainy season. A possible explanation for the low flow prediction ability from the biomass values is that the equation used was fitted using data collected from a region with climatic conditions different from those of RCRB.
The models with the best results, according to the Willmott Concordance Index (d), were those having Qmax and Qmean as response variables. There is no absolute recommendation for d values, but, in general, values close to one (1) indicate high agreement, and values close to zero indicate poor model agreement (Yeo et al. 2021). In this work, the selected models exhibited moderate d results, indicating the possibility of using these models to predict streamflow rates of basins located in regions with similar characteristics of those of the RCRB.
Although precipitation is the most active factor on streamflow rates in the RCRB, changes in land cover play a role in enhancing the effects of climate variability (Kim et al. 2013). Using modeling techniques to understand the relationship between streamflow rates and other variables to predict water productivity in river basins is an important tool for the management of natural resources (Hwang et al. 2018). Models for predicting water regimes using climatic and spectral variables can be an important aid in estimating flows for water resource managers (Ferreira et al. 2021). However, informed use is necessary due to the technology’s limitations with regard to understanding the processes occurring in the basin, although they have an important application from a predictive point of view.