Abstract
We use a Generalized Watershed Loading Function (GWLF) model to simulate streamflow in the Gualí River Basin. The model's performance is assessed using metrics such as percentage of bias (PBIAS), Nash–Sutcliffe efficiency (NSE), RMSE–observations standard deviation ratio (RSR), and Kling–Gupta efficiency (KGE) to indicate good performance. Furthermore, we analyze projections of precipitation and streamflow using several global climate models (GCMs) from the Coupled Model Intercomparison Project Phase 6 (CMIP6) and three shared socioeconomic pathways (SSP1-2.6, SSP2-4.5, SSP5-8.5). Despite the uncertainties and coarse resolution, our results show that increases in the mean streamflow and significantly decreasing trends in projected precipitation and streamflow are observed from 2015 to 2099 under the SSP5-8.5 scenario. Furthermore, our findings suggest an increase in long-term mid-flow and low-flow. Moreover, this work provides a methodological framework for hydrological modeling in small tropical river basins, by incorporating data from GCMs while raising concerns and caveats. This study offers valuable insights into the potential effects of climate change on streamflow in an Andean river basin characterized by volcanic activity and significant human impacts. The findings reported here provide useful information for future decisions related to water supply for the social, environmental, and productive sectors in the seven towns within the catchment.
HIGHLIGHTS
There are few studies of streamflow projections for tropical river basins with volcanic activity.
We provide new information about streamflow modeling using data from CMIP6 models.
The Gualí River Basin has a high population with different economic activities, so this research is important to water resources management there.
This work sheds light on challenges in hydrological processes in complex river basins.
INTRODUCTION
Economic development in Colombia has favored the Andean territory ever since the 16th century. This is why more than 70% of its economy and population (more than 30 million) is located in the Andes. Therefore, Andean river basins in Colombia are under a lot of pressure to attend to the needs of a growing population, in terms of water for irrigation, industry, agroindustry, public utilities, and so on. With more than 80,000 inhabitants, the Gualí River Basin meets the fresh water supply requirements for the residential population, distributed in seven towns and undertaking different economic activities. The main direct and indirect pressures on the water resources across the basin are related to the expansion of agricultural activities, as well as commercial and illegal logging (CORTOLIMA et al. 2014a). These activities generate consequences, such as decreased soil productivity, increased water source sedimentation, reduced water quality and quantity, and diminished representative vegetation, and the fauna related to these ecosystems.
Furthermore, the Gualí River poses great modeling challenges because it lies in the central Andes at the foothills of the Nevado del Ruiz volcano (known as ‘Cumanday’ in the pre-Hispanic era). The volcano erupted in 1985, altering the river channel due to the mudflow produced by the lahar of the eruption, which reached more than 100 km and transported more than 5–10 hm3 of melted water and soil components (Huggel et al. 2007; Marulanda Aguirre et al. 2016). In particular, river basins on the slope of active explosive volcanoes undergo geomorphic transformations in their valleys, which may alter the main channel depth or even generate new channels. The deposition of volcanoclastic material (tephra) may affect future infiltration and underwater storage and the changes induced in the vegetation cover may change the evapotranspiration patterns of the basin (Pierson & Major 2014; Martini et al. 2019; Malawani et al. 2021). The review by Pierson & Major (2014) mentioned that river basins within 10–20 km of a volcanic eruption may experience disruptions to their equilibrium state.
The hydrology of the Gualí River Basin is defined by the complex, interdependent, and delayed interaction between its climate, vegetation, and geology. Climatologically, the river basin is mostly influenced by the intertropical convergence zone (ITCZ) migration, at a seasonal timescale, and the El Niño–Southern Oscillation (ENSO), at an inter-annual timescale; as much as the Tropical Andes basins (Espinoza et al. 2020). These climatic drivers, mixed with an altitudinal range from 600 to 4,100 m above sea level (m.a.s.l), configure wind currents derived from temperature and pressure gradients. These facilitate orographic precipitation over the mountain's foothills (CORTOLIMA et al. 2014a), therefore configuring a diverse range of climatic zones, including super humid high Páramo, humid high Páramo, super humid cold, and humid temperate, among others. The vegetation in this area mainly comprises páramos, barren lands, and rocky outcrops, with natural pastures also present (CORTOLIMA et al. 2012; Mena et al. 2021).
The Intergovernmental Panel on Climate Change's Sixth Assessment Report (AR6) emphasized the crucial role of human activities in understanding the impacts of climate change, as well as adaptation and vulnerability (IPCC 2022). All over the world, there has been evidence of increasing temperatures and changes in precipitation and streamflow in recent decades (Tabari 2020; Shah et al. 2021; Mann & Gupta 2022; Brêda et al. 2023). Central and South America, having already been exposed to and impacted by climate change, face exacerbated challenges because of factors such as inequality, poverty, and deforestation. This situation intensifies the risks to ecosystems, water, agriculture, and migration. Significant impacts on ecosystems, water, and food are anticipated, particularly in the northwest South American region, including the Andes and the Gualí River Basin. Effective adaptation strategies require cross-scale policies, participatory approaches, and the integration of ecosystem-based and community-based adaptation. A lack of sufficient data and analysis of adaptation experiences remains a key obstacle to addressing water challenges in the region, highlighting the need for improved knowledge.
A previous study by Mena et al. (2021) investigated the unmet water demand under climate change scenarios in the Gualí River Basin. To this end, they integrated two modeling tools, as well as precipitation and temperature data from three general circulation models from the coupled model intercomparison project (in its fifth phase – CMIP5), for the climate change scenarios associated with the representative concentration pathways (RCP) denoted as RCP2.6 and RCP8.5. They found that, for the period covering 2011–2100, flow decreased between 5.80% and 9.56%, for RCP2.6, and between 2.18% and 6.86% for RCP8.5. Furthermore, they mentioned that the period 2013–2040 will be the one with the greatest reduction in flow.
The latter study requires further research, considering the fact that the IPCC's sixth assessment report provided several new GCMs from the sixth phase (CMIP6), that included shared socioeconomic pathway (SSP) scenarios. In this sense, the findings by Mena et al. (2021) required: (i) verification using the new generation of CMIP6 GCMs; (ii) acknowledgement of the limitations of hydrological simulations of climate change projections due to climate and hydrological modeling uncertainties (Banda et al. 2022); (iii) understanding the need to downscale techniques like dynamic (Giorgi 2019; Brêda et al. 2020) or statistical downscaling (Ebrahim et al. 2013) to tackle the limitations of GCMs when simulating climate at a regional level; and, finally, (iv) facing up to the uncertainties associated with data from GCMs when implementing techniques such as bias correction (Christensen et al. 2008; Teutschbein & Seibert 2012; Muerth et al. 2013) and/or ensemble approaches (Hagedorn et al. 2005; Semenov & Stratonovitch 2010; Díaz et al. 2021).
We analyzed the projected precipitation and streamflow in the Gualí River Basin by addressing the following questions:
- 1)
Does precipitation and simulated streamflow over the basin exhibit changes in trends and/or exceedance probabilities for the future scenarios SSP1-2.6, SSP2-4.5, and SSP5-8.5?
- 2)
What are the possible impacts on streamflow according to the projected precipitation by different SSP scenarios?
- 3)
What are the shortcomings associated with using CMIP6 data in small tropical basins?
Trying to answer these questions will provide further evidence on how to assess the use of future CMIP projections for hydrological modeling in small tropical mountainous basins vulnerable to climate change, acknowledging the limitations given by scaling and modeling uncertainties.
This work is organized as follows: Section 2 describes the region of study. Section 3 describes the observational and simulated datasets for precipitation. Section 4 presents the general methodology. The results and discussion are presented in Section 5. Conclusions are drawn and future work is suggested in Section 6.
STUDY REGION
(a) Location of the Gualí River Basin within the slopes and the lowlands of the Andes mountain range, (b) regional context of the basin location. Blue dots show the locations of precipitation gauges, yellow squares indicate the locations of temperature record stations, and black triangles are the locations of streamflow gauges.
(a) Location of the Gualí River Basin within the slopes and the lowlands of the Andes mountain range, (b) regional context of the basin location. Blue dots show the locations of precipitation gauges, yellow squares indicate the locations of temperature record stations, and black triangles are the locations of streamflow gauges.
DATA
Observational data
We used time series data, at a daily temporal resolution, for precipitation and streamflow. This was provided by the Institute of Hydrology, Meteorology, and Environmental Studies (IDEAM is its acronym in Spanish). In particular, we used 14 time series of precipitation records, two time series of temperature, and four time series of streamflow records for the common period 1988–2021 (41 years record length). These time series of streamflow and precipitation were used in the semi-distributed rainfall–runoff model (see Section 4.1). In this work, we only provide results for the Honda streamflow gauge, which is located at the output of the basin. Moreover, we used time series for both variables, considering less than 10% of missing data in the time series due to the lack of data in this river basin. In particular, to replace the missing data, we used the long-term monthly average of available historic data for the streamflow gauge station. The data are freely available at http://dhime.ideam.gov.co/webgis/home/. We also used a gridded precipitation product from the Global Precipitation Climatology Centre (GPCC) for the period 1988–2014, at a monthly temporal scale and a 0.25° × 0.25° spatial grid size. This dataset is available at https://psl.noaa.gov/data/gridded/data.gpcc.html.
Global climate models
We used precipitation data provided by the Global Climate Models (GCMs) from CMIP6 experiments (Eyring et al. 2016), which are available from the Earth System Grid Federation web page ESFG (https://esgf-node.llnl.gov/projects/esgf-llnl) for the period 1988–2014 and for the 2015–2099 period, in each of the three SSP scenarios selected for this study, denoted as SSP1-2.6, SSP2-4.5, and SSP5-8.5 (https://www.ipcc.ch/report/ar6/syr/). SSP1-2.6 represents a very low greenhouse gas (GHG) emission declining to net zero by about 2070. SSP2-4.5 has an intermediate GHG emission, maintaining current levels until the middle of the century, and pathway SSP5-8.5 represents very high GHG emissions, doubling the current levels by 2050. Table 1 provides some general information about climate models that provide precipitation projections for the region of study and that exhibit a good performance, according to Almazroui et al. (2021), Arias et al. (2021) and Dias & Simões Reboita (2021). For further comparison among the datasets, they were interpolated to a 0.25° × 0.25° common grid using bilinear interpolation (Accadia et al. 2003).
Description of global climate models (GCMs) from the CMIP6 experiment
Model . | Long × Lat resolution . | Institute . | Reference . |
---|---|---|---|
FGOALS-g3 | 2.00 ° × 2.25 ° | Institute of Atmospheric Physics, Chinese Academy of Sciences | Arias et al. (2021) and Li (2019) |
EC-Earth3 | 0.70 ° × 0.70 ° | EC-EARTH Consortium | Arias et al. (2021) and EC-Earth Consortium (2019b) |
EC-Earth3-Veg | 0.70 ° × 0.70 ° | EC-EARTH Consortium | Arias et al. (2021) and EC-Earth Consortium (2019a) |
MPI-ESM1-2-HR | 0.93 ° × 0.93 ° | Max Planck Institute for Meteorology | Arias et al. (2021) and Jungclaus et al. (2019) |
CESM2-WACCM | 1.25 ° × 0.93 ° | National Center for Atmospheric Research, Climate and Global Dynamics Laboratory | Almazroui et al. (2021) and Danabasoglu (2019) |
MRI-ESM2-0 | 1.18 ° × 1.12 ° | Meteorological Research Institute | Almazroui et al. (2021) and Yukimoto et al. (2019) |
CanESM5 | 2.81 ° × 2.81 ° | Canadian Centre for Climate Modelling and Analysis, Environment and Climate Change Canada | Dias & Simões Reboita (2021) and Swart et al. (2019) |
CMCC-CM2-SR5 | 1.25 ° × 0.93 ° | Fondazione Centro Euro-Mediterraneo sui Cambiamenti Climatici | Dias & Simões Reboita (2021) and Lovato & Peano (2020) |
INM-CM4-8 | 2.00 ° × 1.50 ° | Institute for Numerical Mathematics | Dias & Simões Reboita (2021) and Volodin et al. (2019a) |
INM-CM5-0 | 2.00 ° × 1.50 ° | Institute for Numerical Mathematics | Dias & Simões Reboita (2021) and Volodin et al. (2019b) |
Model . | Long × Lat resolution . | Institute . | Reference . |
---|---|---|---|
FGOALS-g3 | 2.00 ° × 2.25 ° | Institute of Atmospheric Physics, Chinese Academy of Sciences | Arias et al. (2021) and Li (2019) |
EC-Earth3 | 0.70 ° × 0.70 ° | EC-EARTH Consortium | Arias et al. (2021) and EC-Earth Consortium (2019b) |
EC-Earth3-Veg | 0.70 ° × 0.70 ° | EC-EARTH Consortium | Arias et al. (2021) and EC-Earth Consortium (2019a) |
MPI-ESM1-2-HR | 0.93 ° × 0.93 ° | Max Planck Institute for Meteorology | Arias et al. (2021) and Jungclaus et al. (2019) |
CESM2-WACCM | 1.25 ° × 0.93 ° | National Center for Atmospheric Research, Climate and Global Dynamics Laboratory | Almazroui et al. (2021) and Danabasoglu (2019) |
MRI-ESM2-0 | 1.18 ° × 1.12 ° | Meteorological Research Institute | Almazroui et al. (2021) and Yukimoto et al. (2019) |
CanESM5 | 2.81 ° × 2.81 ° | Canadian Centre for Climate Modelling and Analysis, Environment and Climate Change Canada | Dias & Simões Reboita (2021) and Swart et al. (2019) |
CMCC-CM2-SR5 | 1.25 ° × 0.93 ° | Fondazione Centro Euro-Mediterraneo sui Cambiamenti Climatici | Dias & Simões Reboita (2021) and Lovato & Peano (2020) |
INM-CM4-8 | 2.00 ° × 1.50 ° | Institute for Numerical Mathematics | Dias & Simões Reboita (2021) and Volodin et al. (2019a) |
INM-CM5-0 | 2.00 ° × 1.50 ° | Institute for Numerical Mathematics | Dias & Simões Reboita (2021) and Volodin et al. (2019b) |
METHODS
Schema of the methodological framework for going from observations and simulations to the hydrological simulation and evaluation of different scenarios.
Schema of the methodological framework for going from observations and simulations to the hydrological simulation and evaluation of different scenarios.
Hydrological model
Hydro-BID includes the Generalized Watershed Loading Function (GWLF) model, which is used to simulate the rainfall–runoff process in basins over a wide spatial range, using the attributes of soil type and land use (Wu & Lin 2015; Qi et al. 2019). GWLF takes into account snow melt, potential evapotranspiration, runoff, percolation, flow, and routing (Haith & Shoemaker 1987; Moreda & Miralles-Wilhelm Raúl Muñoz Castillo 2014). The summary equations of the GWLF models are shown in Table 2. Details about Hydro-BID are available on the web page (https://hydrobidlac.org/).
Components of the GWLF model (USAID 2018)
Component . | Equation . |
---|---|
Snow melt | ![]() ![]() ![]() ![]() |
Potential evapotranspiration | ![]() ![]() ![]() ![]() ![]() |
Potential evapotranspiration adjusted to land use and cover | ![]() ![]() ![]() ![]() |
Runoff | ![]() ![]() ![]() ![]() ![]() ![]() |
Percolation | ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Flow | ![]() ![]() ![]() ![]() |
Component . | Equation . |
---|---|
Snow melt | ![]() ![]() ![]() ![]() |
Potential evapotranspiration | ![]() ![]() ![]() ![]() ![]() |
Potential evapotranspiration adjusted to land use and cover | ![]() ![]() ![]() ![]() |
Runoff | ![]() ![]() ![]() ![]() ![]() ![]() |
Percolation | ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Flow | ![]() ![]() ![]() ![]() |
Wu & Lin (2015) recommended a set range of values for the calibration process of the GWLF model (Table 3).
Recommended ranges for the GWLF model parameters
Parameter . | Recommended ranges of values . |
---|---|
Curve number (CN) | 0.8–1.2 |
Available water content (AWC) | 0.2–1.2 |
Recession coefficient (r) | 0.001–0.750 |
Seepage coefficient (S) | 0.005–0.100 |
Growth season ET factor | 0.5–1.5 |
Dormant season ET factor | 0.5–1.5 |
Impervious cover percentage | 1 |
Temperature threshold | 1 |
Melt factor | 1 |
Parameter . | Recommended ranges of values . |
---|---|
Curve number (CN) | 0.8–1.2 |
Available water content (AWC) | 0.2–1.2 |
Recession coefficient (r) | 0.001–0.750 |
Seepage coefficient (S) | 0.005–0.100 |
Growth season ET factor | 0.5–1.5 |
Dormant season ET factor | 0.5–1.5 |
Impervious cover percentage | 1 |
Temperature threshold | 1 |
Melt factor | 1 |
Bias correction for the GCMs’ precipitation and performance criteria for the rainfall–runoff model
We used Taylor diagrams (Taylor 2001) to compare the historical observations of precipitation with the one simulated by the GCMs over the Gualí River Basin, for the different SSP scenarios over the period 1988–2014. In this sense, Taylor diagrams were used to quantify the degree of correspondence between the observations and simulations, in terms of the Pearson correlation coefficient, the root-mean-square error (RMSE), and the standard deviation. This analysis was used to determine whether it was necessary to implement a bias correction for the simulated precipitation provided by the GCMs. From this perspective, bias correction is a common procedure that is intended to correct systematic biases in climate models (Teutschbein & Seibert 2012; Chathuranika et al. 2022).
For hydrological modeling considering climate change scenarios, in particular, bias correction is used to reduce the uncertainty before the use of the GCM data for hydrological simulations (Hagedorn et al. 2005). When bias correction was required, we used simple empirical quantile mapping between observed and simulated precipitation by the GCMs. In general, quantile mapping is a technique for correcting extreme values and modifying the shape of distributions by finding a statistical relationship between the quantiles of the two distributions; this method provides comparable results to others based on the probability distribution functions (Maraun 2013; Dekens et al. 2017; Prasanna 2018).
Regarding the performance criteria for the rainfall–runoff model, we used the percentage bias (PBIAS), the Nash–Sutcliffe efficiency (NSE), and the ratio of the RMSE to the standard deviation of measured data (RSR), as recommended by Moriasi et al. (2007), which stated that, for streamflow, a model simulation can be judged as satisfactory if NSE > 0.50, RSR < 0.70, and PBIAS ± 25%. Moreover, we used the Kling–Gupta efficiency (KGE) (Gupta et al. 2009), for which other studies suggested that KGE > 0.5 is high enough to indicate a suitable and favorable performance of a hydrological model (Wu et al. 2022).
Signature indices for streamflow projection analysis
Once we achieved a calibrated rainfall–runoff model, we simulated runoff using the precipitation from GCMs for historic and future periods, using the scenarios SSP1-2.6, SSP2-4.5, and SSP5-8.5. Furthermore, we quantified trends using Kendall's non-parametric test (Kendall 1975) for a monotonic trend, using the Theil–Sen method (Sen 1968) to estimate the slope, confidence intervals, and significance test at a significance level of 0.05. To simulate streamflow, we quantified changes in the FDC for simulations with historical observations and simulations with GCM runs, for the periods 2015–2047, 2048–2073, and 2074–2099, as established by Chathuranika et al. (2022).






RESULTS AND DISCUSSION
Performance of the rainfall–runoff model
Precipitation and discharge for the period 1988–2004. The precipitation ensembled with the time series within the river basin is shown in cyan; observed discharge in the output station is shown in black; and the discharge simulation with satisfactory performance, according to Moriasi et al. (2007), is shown in blue.
Precipitation and discharge for the period 1988–2004. The precipitation ensembled with the time series within the river basin is shown in cyan; observed discharge in the output station is shown in black; and the discharge simulation with satisfactory performance, according to Moriasi et al. (2007), is shown in blue.
Annual cycle and flow duration curves for the observed (black) and simulated (blue) discharges. (a) Annual cycle for the calibration period 1989–1999. (b) Annual cycle for the validation period 2000–2004. (c) Curve of probability of exceedance for the calibration period 1989–1999. (d) Curve of probability of exceedance for the validation period 2000–2004.
Annual cycle and flow duration curves for the observed (black) and simulated (blue) discharges. (a) Annual cycle for the calibration period 1989–1999. (b) Annual cycle for the validation period 2000–2004. (c) Curve of probability of exceedance for the calibration period 1989–1999. (d) Curve of probability of exceedance for the validation period 2000–2004.
On the need for bias correction in the precipitation provided by GCMs for streamflow simulations
Taylor diagram for monthly precipitation in the Gualí River Basin, according to different GCMs in the period 1988–2014. Data without bias correction are in red. Data with bias correction are in blue. The GPCC dataset is represented as a black dot.
Taylor diagram for monthly precipitation in the Gualí River Basin, according to different GCMs in the period 1988–2014. Data without bias correction are in red. Data with bias correction are in blue. The GPCC dataset is represented as a black dot.
For this reason, Figure 5 shows the Taylor diagram for precipitation over the historic period (1988–2014), after bias correction (blue markers), using simple empirical quantile mapping (Maraun 2013). In general, after bias correction, precipitation exhibits clustering among GCM models with standard deviations ranging from 80 to 120 mm/month, showing a reduction compared with the non-bias-corrected data. The correlation coefficients for all models remain below 0.40, indicating that GCMs do not exhibit a notable improvement, in terms of correlations between models and the reference. Correlations between observations and CMIP6 simulations were reported in the range 0.3–0.4, for the tropical Andes, in the work by Arias et al. (2021). Furthermore, the centered root-mean-square (RMS) errors after bias correction decrease to a range between 100 and 150. It is important to note that, despite the low correlations between bias-corrected data and the observations, the hydrological impact assessment of CMIP6 projections has been carried out in other regions (Thalli Mani et al. 2022; Acharki et al. 2023; Rudraswamy et al. 2023).
Taylor diagram for monthly streamflow in the Gualí River Basin, for the historic period 1988–2005, according to different GCMs, from simulations using precipitation with bias correction.
Taylor diagram for monthly streamflow in the Gualí River Basin, for the historic period 1988–2005, according to different GCMs, from simulations using precipitation with bias correction.
Changes in projected precipitation and streamflow for different climate change scenarios
Boxplots of precipitation and flow. (a) Boxplot of precipitation for each climate scenario for the full period 2015–2099. (b) Boxplot of precipitation for each climate scenario in three different periods: 2015–2047, 2048–2073, and 2074–2099. (c) Boxplot of streamflow for each climate scenario for the full period 2015–2099. (d) Boxplot of streamflow for each climate scenario in three different periods: 2015–2047, 2048–2073, and 2074–2099. Axis labels 2.6, 4.5, and 8.5 correspond to the SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios, respectively.
Boxplots of precipitation and flow. (a) Boxplot of precipitation for each climate scenario for the full period 2015–2099. (b) Boxplot of precipitation for each climate scenario in three different periods: 2015–2047, 2048–2073, and 2074–2099. (c) Boxplot of streamflow for each climate scenario for the full period 2015–2099. (d) Boxplot of streamflow for each climate scenario in three different periods: 2015–2047, 2048–2073, and 2074–2099. Axis labels 2.6, 4.5, and 8.5 correspond to the SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios, respectively.
Figure 7(c) shows that the simulated streamflow in each of the scenarios, for the period 2015–2099, is approximately 46 m3/s, and this overestimates the mean reference value, with greater variability in the data distribution for all scenarios. In addition, the SSP2-4.5 scenario shows higher atypical values, while the SSP5-8.5 scenario shows none. Figure 7(d) shows the simulated streamflow for each climate scenario in three different periods (2015–2047, 2048–2073, and 2074–2099). The periods 2015–2047 and 2048–2073 show an increase in the mean simulated streamflow, in relation to the reference, ranging from 8% to 10%, whereas the period 2074–2099 exhibits a 5% decrease in the mean streamflow for the SSP5-8.5 scenario, in relation to SSP1-2.6 and SSP2-4.5.
A larger number of atypical values are observed for the SSP2-4.5 scenario in the 2015–2047 period, while the SSP1-2.6 scenario shows atypical values in the 2015–2047 and 2074–2099 periods, being the only scenario with atypical values for this latter period. The SSP5-8.5 scenario does not show any atypical values in any of the periods, and greater variability is also observed between the maximum values of the scenarios for all periods, with values between 46 and 56 m3/s, showing greater difference in the first quartile between the SSP1-2.6 and SSP5-8.5 scenarios in the 2015–2047 and 2074–2099 periods. Our results contrast with the findings of Mena et al. (2021), who used three CMIP5 models and concluded that, for the period covering 2011–2100, the available flow is expected to decrease by 6%–10% for RCP 2.6, and by 2%–7% for RCP 8.5, with the greatest reduction in flow occurring during 2013–2040.
Trends in precipitation and streamflow for different climate change scenarios
Future projection of mean annual precipitation (upper) and discharges (lower) in the Gualí River Basin for the period 2020–2099. The trend is drawn from the ensemble for all the CMIP6 models used. The blue area in the background is the intermember spread of the ensemble.
Future projection of mean annual precipitation (upper) and discharges (lower) in the Gualí River Basin for the period 2020–2099. The trend is drawn from the ensemble for all the CMIP6 models used. The blue area in the background is the intermember spread of the ensemble.
Signature indices for different climate change scenarios
Flow duration curves for streamflow. The reference curve from IDEAM for the historic period (1988–2015) is shown in black. The ensemble of simulated streamflow using precipitation from GCMs for the historic period (1988–2015) is shown in orange. The simulated streamflow for each GCM for the historic period (1988–2015) is shown in gray. The ensemble of projected streamflow (2015–2099), using projected precipitation from GCMs for the scenarios SSP1-2.6, SSP2-4.5, and SSP5-8.5, is shown in blue, green, and red, respectively. The mean discharge for the IDEAM reference, over the historic period, is shown as a dashed line.
Flow duration curves for streamflow. The reference curve from IDEAM for the historic period (1988–2015) is shown in black. The ensemble of simulated streamflow using precipitation from GCMs for the historic period (1988–2015) is shown in orange. The simulated streamflow for each GCM for the historic period (1988–2015) is shown in gray. The ensemble of projected streamflow (2015–2099), using projected precipitation from GCMs for the scenarios SSP1-2.6, SSP2-4.5, and SSP5-8.5, is shown in blue, green, and red, respectively. The mean discharge for the IDEAM reference, over the historic period, is shown as a dashed line.
Table 4 shows six signature indices to quantify the change in the runoff for three climate change scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) for the 2015–2099 period. The percentage bias in the mean values (BiasRR) indicates that the mean streamflow for future periods is higher than observed, with a range of 7.6%–9.3%. Moreover, BiasRR tends to increase in each SSP scenario, being higher for the SSP5-8.5 scenario (9.31%). Furthermore, the percentage bias in mid-range flow levels (BiasFMM) shows that simulated runoff is higher than the reference for the analyzed period, in the range 17.1%–20.5%, the highest value being for SSP5-8.5 (20.58%).
Signature indices for monthly streamflow for the climate change scenarios SSP1-2.6, SSP2-4.5, and SSP5-8.5 for the period 2015–2099
. | SSP1-2.6 . | SSP2-4.5 . | SSP5-8.5 . |
---|---|---|---|
BiasRR | 7.64 | 9.06 | 9.31 |
BiasFMM | 17.10 | 20.16 | 20.58 |
BiasFDCmidslope | −41.55 | −31.89 | −40.18 |
BiasFLV | 122.26 | 99.31 | 59.98 |
BiasFHV | 337.89 | 345.41 | 334.28 |
DiffQ95 | 30.07 | 22.86 | 27.93 |
. | SSP1-2.6 . | SSP2-4.5 . | SSP5-8.5 . |
---|---|---|---|
BiasRR | 7.64 | 9.06 | 9.31 |
BiasFMM | 17.10 | 20.16 | 20.58 |
BiasFDCmidslope | −41.55 | −31.89 | −40.18 |
BiasFLV | 122.26 | 99.31 | 59.98 |
BiasFHV | 337.89 | 345.41 | 334.28 |
DiffQ95 | 30.07 | 22.86 | 27.93 |
The percentage bias in the slope of the mid-segment (BiasFDCmidslope) shows that, for all SSP scenarios, the streamflow projections exhibit lower values than the reference value. BiasFDCmidslope reduction ranges between 41.5% and 31.9%, with the highest values being for SSP2 and SSP5. Furthermore, for all SSP scenarios, the differences in long-term baseflow (BiasFLV) suggest that simulations are higher than observations (ranging between 60% and 122%). The percentage bias in high-segment volumes (BiasFHV) shows a bias higher than 330%, for all scenarios during the period. Lastly, the difference of FDC at the 95% percentile (DiffQ95) shows an increase for all scenarios (higher than 20%), the highest increases being for SSP1-2.6 and SSP5-8.5 (30% and 27.9%, respectively.)
DISCUSSION
For the projections under climate change scenarios, we found contrasting results, indicating that although the magnitude of the mean streamflow of the Gualí River is higher for the simulations than for the observations, the variability is greatly reduced (mainly due to the ensemble and bias correction approaches). For scenarios SSP1-2.6 and SSP2-4.5, there is an increased streamflow but no significant trends for precipitation or streamflow. For SSP5-8.5, although the mean streamflow increases compared with the observations, there is a reduction of streamflow of around 5% when compared with SSP1-2.6 and SSP2-4.5, for the 2074–2099 period. These results coincide with those provided by Mena et al. (2021), who, through CMIP5 models, reported reductions in streamflow of around 2% and 7%. Also for SSP5-8.5 in terms of trends, there is a coherent significantly decreasing trend for precipitation and streamflow (−29.023 mm/dec and −0.572 m3/dec, respectively).
Furthermore, our findings indicate that, if GCMs deviate from a good simulation for precipitation, these deviations can induce discrepancies in the simulated streamflow magnitude, notwithstanding that signals of a decreasing trend are observed for both precipitation and streamflow. In that sense, our results introduce the quantification of trends with CMIP6 data to support the panorama of climate change shown by Mena et al. (2021). The results have helped us to address the first question posed in this study. Despite the uncertainties and the lack of confidence in future projections derived from GCMs, for this particular geographical scale, our study reveals a consistent increase in mean streamflow for the basin.
To address the second research question in this study, in terms of signature indices, the streamflow projections exhibit an increase in the mean-, high-, and low-flow features. Moreover, it is remarkable that the projections suggest an increase in mid-flow and low-flow features (BiasFDCmidslope, BiasFMM, BiasFLV, BiasFHV), as well as an increase in the percentage bias in the mean values and the 95th quantile of the flows (BiasRR and DiffQ95). These findings provide useful information about the long-term mean streamflow and baseflow, which provides a continuous supply of water for human consumption, irrigation, and other human uses. In addition, it helps to maintain the health of aquatic and terrestrial ecosystems.
There are still some challenges regarding the calibration and validation of the rainfall–runoff modeling for the Gualí River Basin, which could be useful to know for similar applications in the region. In this sense, we found acceptable (but not high) performance metrics for the hydrological model, which can be explained because: (1) the Gualí River Basin is located in the Andes mountain range and processes such as orographic forcing, vegetation diversity, and tropical climates induce particularities, as stated by other authors (Canchala et al. 2020; Yeditha et al. 2022; Ghaderpour et al. 2023); (2) we hypothesize that the difficulties in the model representing the higher streamflow values can be related to the alterations in the infiltration regime across the basin, induced by the most recent explosive eruption of the Nevado del Ruíz in 1985. The eruption deposited volcanoclastic materials during the explosion and then lahar, which affected the Gualí River Basin due to its proximity to the volcano (within 10–20 km); (3) we lack data for the melt water and soil from the Nevado del Ruiz, to be able to improve model representation; and (4) the performance of GCMs at finer scales is not optimal (Banda et al. 2022); therefore, regional modeling, downscaling or bias correction (in our particular case, the basin size) and the coarse resolution of the GCMs may be sources of uncertainty that must not be disregarded.
CONCLUSIONS AND FUTURE WORK
We simulated monthly streamflow in the Gualí River Basin (Colombia) using the GWLF model included with Hydro-BID software. To this end, we used a calibrated and validated hydrological model that met the literature's satisfactory performance criteria. Furthermore, we used precipitation simulations from ten GCMs and three shared socioeconomic pathways (SSPs) (denoted as SSP1-2.6, SSP2-4.5, and SSP5-8.5) and quantified the trends and signature indices for projected precipitation and streamflow in the period 2020–2099. The rainfall–runoff modeling process for this river basin is evidence that it is necessary to verify the coherence between observed precipitation and simulated precipitation by GCMs, over the historic period, before incorporating these data into future streamflow projections. In general, we found disagreement in the Taylor diagrams between the precipitation reference and precipitation provided by GCMs for the historic period. This disagreement could be explained in terms of the limitations of the GCMs, such as systematic biases, coarse spatial resolution, and lack of representation of climatic processes at regional scales (Falco et al. 2019; Llopart et al. 2020; Ciarlὸ et al. 2021).
Moreover, our results show the necessity of bias correction or ensemble approaches (Hagedorn et al. 2005; Muerth et al. 2013) in order to face upto the uncertainties of GCM projections and streamflow simulations. After the bias correction of precipitation from GCMs, we obtain streamflow simulation results that agree with the observations, in terms of magnitude and FDC for the historic period, which provide confidence in the streamflow simulation for future projections, especially for probabilities of exceedance above 0.20. For probabilities of exceedance below 0.20, the FDCs exhibit higher dispersion and an underestimation of simulations, in relation to the observed streamflow.
We would caution practitioners when using GCM datasets, even with the bias correction proposed by Meresa et al. (2022), when assessing precipitation extremes and flood risk with the CMIP dataset. These issues about geographical scale and the uncertainties when resolving convection in GCMs are common in hydrological analysis involving CMIP6 datasets, as stated by Martel et al. (2022), for catchments in the USA, and by Guo et al. (2022), for basins in the northern hemisphere where GCMs provide higher performances than in the tropics. Notwithstanding this, we note that our case study in South America provides valuable information to improve climate models which, in the absence of historical datasets of sufficient length from systematic and reliable observations, are suitable for certain purposes, such as risk management at a regional scale, the evaluation of environmental determinants, and supporting processes immersed in the structure of integrated water resources management (IWRM).
Further research is needed in light of these contrasting results. We propose some aspects that should be focused on: (1) explore the basin size/model resolution sensitivity in the Andean basin context; (2) generate a hydrological multi-model ensemble that, like the approach with GCMs, allows enhanced confidence intervals after the inclusion of climate change scenarios in water planning; (3) studying the infiltration variations over the basin due to continuous changes in the soil, as a result of explosive eruptions of the Nevado del Ruíz, is of utmost importance for the improvement of this variable in hydrological modeling; and (4) carry out hydrological modeling experiments considering land use and land cover changes in the river basin (Nama et al. 2022), in order to incorporate alterations that help policy makers and management support decision-making around water resources.
ACKNOWLEDGEMENTS
We would like to thank the Institute of Hydrology, Meteorology, and Environmental Studies (IDEAM is its acronym in Spanish) for providing precipitation and streamflow data records. Thanks also go to Banco Interamericano de Desarrollo (BID) for providing a free license to the software Hydro-BID. The authors thank the research groups and institutions that provide complete access to the global climate models from CMIP6 through the Earth System Grid Federation (ESFG) nodes https://esgf.llnl.gov/. We also thank the National Oceanic and Atmospheric Administration (NOAA), for providing the Global Precipitation Climatology Centre (GPCC) dataset, and the Instituto Geográfico Agustín Codazzi (IGAC) for the maps of Land Use and Land Cover in the Gualí River Basin (https://www.colombiaenmapas.gov.co/).
FUNDING
The work of H. D. Salas and A. Builes-Jaramillo was supported by the Institución Universitaria Colegio Mayor de Antioquia. The work by C. Florian-Vergara and J. Valencia was supported by the project: ‘Evaluación de los efectos del cambio climático y de los cambios en las coberturas del suelo y su incorporación en la modelación integral del recurso hídrico: caso de estudio cuenca hidrográfica del Río Gualí’ (Project FAI58). The work by D. Mena was supported by the Universidad Santo Tomás (Bogotá) and the work by J. C. Parra and J. C. Valdés was supported by the Politécnico Colombiano Jaime Isaza Cadavid.
AUTHOR CONTRIBUTIONS
H.D.S., A.B.J., C.F., and J.V. conceptualized the whole article, developed the methodology, and wrote the first version of the manuscript. H.D.S and A.B.J. administered the project and supervised the work. C.F. and J.V. processed climate and hydrological data. D.M. and J.C.P. reviewed the first version of the manuscript. H.D.S, C.F., A.B.J., and J.V. analyzed data and edited the final version of the manuscript.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.
CONFLICT OF INTEREST
The authors declare there is no conflict.