Projected effects of climate change in precipitation and streamflows of the Upper Orinoco River Basin from global and regional climate models

The Orinoco River Basin has an area of approximately 1.0 × 10⁶ km² and a discharge of 3.5 × 10⁴ m³ s⁻¹, making it the third biggest basin in South America, with around 70 and 30% continental territory belonging to Venezuela and Colombia, respectively. This river basin also has the third highest discharge in the world (López et al. 2012). The basin is divided into three parts: the Alto Orinoco, located in the south, with the Ventuari river as the main tributary; the Medio Orinoco, mainly located in the Colombian Orinoco region, with tributaries such as the Inírida, Vichada, Guaviare, Meta, and Arauca; and finally, the Bajo Orinoco, near the delta, with the Caroni and the Caura as the most important tributaries (Silva León 2005; Frappart et al. 2012). Furthermore, this river basin shares a border with the Amazon River Basin, and both basins are the main sources of Andean sediments and dissolved elements in the Atlantic Ocean (Meade 2008; Mora et al. 2020). The Orinoco River Basin exhibits tremendous ecological heterogeneity due to its location between the northern Andes, Caribbean Mountain ranges, the Guianese shield, and the development of the low relief areas known as Llanos in both countries (Warne et al. 2002; Frappart et al. 2012).

General circulation models (GCMs) have been used to represent and evaluate the possible impacts of increased atmospheric concentrations of greenhouse gases and anthropogenic aerosols (Edwards 2011; IPCC 2022). The Coupled Model Intercomparison Project (CMIP), in its fifth and sixth phases (CMIP5 and CMIP6), presents the most recent advances to contribute to the understanding of the climate system and its responses to human activities (Taylor et al. 2012; Eyring et al. 2016). However, these models have limitations due to their systematic biases, coarse spatial resolution, and lack of representation of climatic processes at regional scales (Arias et al. 2021; Builes-Jaramillo & Pántano 2021). On the other hand, regional climate models (RCMs) calculate mass and energy balance at a finer resolution than GCMs, a process referred to as dynamical downscaling (Giorgi 2019; Giorgi et al. 2021; Brêda et al. 2022). Some authors argue that RCMs add value to GCMs because they offer a finer representation of regional climatic processes (Falco et al. 2019; Llopart et al. 2020; Ciarlo’ et al. 2021), which are important for assessing hydrological impacts (Teutschbein & Seibert 2012; Lee et al. 2019). Although RCMs are valuable tools for regional impact studies of climate change, their uncertainties should be acknowledged beforehand (Hawkins & Sutton 2009; Foley 2010; Brêda et al. 2022). To address these uncertainties, diverse techniques of bias correction (Christensen et al. 2008; Teutschbein & Seibert 2012; Muerth et al. 2013) or/and ensemble approaches (Hagedorn et al. 2005; Semenov & Stratonovitch 2010; Díaz et al. 2021) have been used in data from RCMs and GCMs.

In the last few decades, several authors have stated that factors such as human intervention in catchments, the impact of low-frequency variability, and human-induced climate change are the primary drivers of alterations in the water cycle, as well as the magnitude and frequency of floods (Perreault et al. 1999; Zhang & Schilling 2006; Villarini et al. 2009). Recently, the Intergovernmental Panel on Climate Change's Sixth Assessment Report (AR6) (IPCC 2022) has also concluded that human activities play a crucial role in diagnosing and understanding climate change, particularly in terms of its impacts, adaptation, and vulnerability. To better understand hydrologic phenomena and assess the effects of climate change on the hydrological cycle, researchers have used rainfall–runoff models (Jiang et al. 2007; Moradkhani & Sorooshian 2009; Chawla & Mujumdar 2015; Amanambu et al. 2019; Chu et al. 2019; Zeng et al. 2020; Carvalho et al. 2022; Chathuranika et al. 2022; Zhang et al. 2022; Salas et al. 2023). For instance, Pimentel et al. (2021) used a coupled hydrological model to study the Orinoco region of Colombia, considering the rain runoff process, the river–floodplain interaction, and anthropic processes such as surface water extraction and groundwater extraction. They analyzed five different agro-industrial expansion scenarios and found that the Colombian Orinoco could face significant pressure on water resources, leading to critical changes in the water availability regime.

The scenarios showed reductions of up to 85% in low water flows in over 50% of the Colombian Orinoco Basin area. In the most extreme scenarios, the Meta, Vichada, and Guaviare rivers could experience reductions of 95, 98, and 50% in low water flows, respectively. However, this study only focused on effects of changes in land cover on the streamflow and did not consider the possible effects of precipitation changes under climate change scenarios. It is worth noting that the Upper Orinoco in Colombia connects the Andes Mountain range and the Llanos plains, which are shared by Colombia and Venezuela (Meade 2008; Frappart et al. 2014). This region must balance deforestation, human pressure, and conservation efforts after the peace treaty in Colombia (Clerici et al. 2020; Rodríguez-de Francisco et al. 2021).

Hence, we analyzed the future projected precipitation and streamflow in the upper Orinoco River Basin addressing the following questions: (1) Does monthly precipitation over the basin exhibit changes in trends for the future scenarios RCP2.6 and RCP8.5?; (2) Do the simulated monthly streamflow over the basin exhibit changes in terms of trends and exceedance probabilities for the future scenarios RCP2.6 and RCP8.5?; (3) Is there an agreement for future projected precipitation provided by the different global and RCMs for the RCP2.6 and RCP8.5 scenarios?; and (4) What are the possible impacts on the streamflow according to projected precipitation by different RCP scenarios?

We used time series at monthly temporal resolution for precipitation and streamflow provided by the Institute of Hydrology, Meteorology, and Environmental Studies (IDEAM for its Spanish acronym). Specifically, we used 20 time series with precipitation records and 3 time series with streamflow records for the common period 1990–2015 (26 years record length). We use time series for both variables considering less than 10% missing data in the time series. The missing values were replaced with the 26-year average of the existing monthly data. To clarify, any missing data point for a given month i in year j represented by X_i,j was replaced with the monthly average ⟨X_i,j⟩ calculated over a 26-year period for each precipitation and streamflow gauge, where ⟨·⟩ denotes the daily average. Then, these time series of streamflows and precipitation are used in the semi-distributed rainfall–runoff model (see Section 4.2). In this work, we will show results (Section 5) only for the streamflow gauge Guayare, which is located at the outlet of the basin. Data are freely available at https://dhime.ideam.gov.co/.

In addition, we use the GLCNMO dataset, which was developed by the International Steering Committee for Global Mapping (ISCGM) in collaboration with the Geospatial Information Authority of Japan (GSI), Chiba University, and NGIAs of respective countries and regions. In this work, we use the third version that uses MODIS data 2013 (Terra & Aqua) combined with remote sensing technology (Kobayashi et al. 2017).

We use precipitation provided by the global climate models from the CMIP5 and CMIP6 experiments (Eyring et al. 2016) as well as RCMs from the CORDEX-CORE experiment, which are available in the Earth System Grid Federation web page ESFG-https://esgf-node.llnl.gov/projects/esgf-llnl, for the 1990–2015 period and for two future forcing scenarios. In this way, we used the Representative Concentration Pathways (RCPs) RCP2.6 and RCP8.5 from CMIP5 and CORDEX-CORE, as well as the Shared Socioeconomic Pathways (SSPs) SSP1-2.6 and SSP5-8.5 from CMIP6. Since RCP2.6 and RCP8.5 are comparable to SSP1-2.6 and SSP5-8.5 in terms of radioactive forcing, respectively (Riahi et al. 2017), hereafter we use the terminology RCP2.6 and RCP8.5 for all of them. Table 1 shows some general information about climate models that provide precipitation projections for the region of study and that exhibit a good performance over northern South America according to Arias et al. (2021). Moreover, we use the regionalization of precipitation for the CMIP5 models provided by IDEAM (hereafter referred to as RegIDEAM) which is available at http://www.ideam.gov.co/web/tiempo-y-clima/escenarios-cambio-climatico.

We simulate runoff using precipitation data from GCMs and RCMs for historic and future periods using the RCP2.6 and RCP8.5 scenarios. Furthermore, we quantify 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. Moreover, we quantify changes in the probability of exceedance curves of the streamflow (commonly referred to as Flow Duration Curve or FDC) between historical observations and simulations for the periods 2020s (2021–2047), 2050s (2048–2073), and 2080s (2074–2099), as was recently used by Chathuranika et al. (2022).

We use the Soil Moisture Method (hereafter SMM) presented by Yates et al. (2005), which computes the water changes within the soil moisture zone using a 2-storage soil water accounting scheme that considers precipitation, runoff, and actual evapotranspiration (Ev), while using potential evapotranspiration (PET) to drive the extraction of water from the soil moisture.

We use the SMM model incorporated as a semi-distributed rainfall–runoff model in the software Water Evaluation and Planning – WEAP (https://www.weap21.org/). Then, for the rainfall–runoff model implementation, we define three catchments or sub-areas in the river basin, which are denoted as C1, C2, and C3 (see Figure 1), which have the types of Land Covers shown by percentage in Table 2. For each catchment, we characterize the types of land cover and assign the parameters of the model, to establish the representative time series of precipitation (C1, C2, and C3 have 16, 1, and 3 rain gauges, respectively), and the streamflow gauges at the output of each sub-area (see Figure 1). In particular, the parameters related to surface runoff (RRF and K_c) are assigned for each type of land cover, whereas the parameters associated with sub-surface runoff, interflow, and deep percolation (Dw, f, Sw, k_s, k_d, z₁, z₂), are selected as a unique value for all catchments and land covers because of the lack of sub-surface runoff, interflow, and deep percolation observational data. Moreover, we use random sampling to obtain the values of the parameters (Dw, f, Sw, k_s, k_d, z₁, z₂) while achieving optimization of the simulation performance metrics PBIAS, NSE, and RSR provided by Moriasi et al. (2007).

In addition, the SMM model performs well in both the calibration and validation phases, effectively capturing the annual cycle variations of observed streamflow as well as its flow duration curve. Although the performance metrics suggest that the SMM model exhibits a slightly lower performance for the calibration period compared to the validation one, the overall results remain satisfactory for the scope of this work (Wen et al. 2024).

Hence, due to the differences among GCMs and RCMs models, we use the Linear Scaling method (Teutschbein & Seibert 2012; Chathuranika et al. 2022) to correct systematic biases in climate models. Furthermore, we compute ensembles for each group of models (CORDEX, CMIP5, and CMIP6) to reduce the uncertainty before the use of these time series for hydrological simulations (Hagedorn et al. 2005). Hereafter, we refer to the ensembles for each group of data as Ens_CORDEX for the CORDEX-CORE ensemble, Ens CMIP5 for the CMIP5 ensemble, Ens_CMIP6 for the CMIP6 ensemble, and Ens_RegIDEAM for the regionalization of the CMIP5 models carried out by IDEAM.

It is important to note that while RCMs and GCMs may not demonstrate high performance according to the metrics of Taylor's diagram, hydrological impact assessments using global and regional projections have been conducted in other regions (Acharki et al. 2023; Fang et al. 2023; Rudraswamy et al. 2023; Salas et al. 2023). Furthermore, there is no clear added value from a specific group of global or RCMs (CMIP5, CMIP6, CORDEX-CORE). Therefore, we employ bias correction using the linear scaling method and an ensemble approach to address inconsistencies among simulations, ensuring that the bias-adjusted climate data closely resemble observed climate signals. This facilitates the interpretation of trends and supports further analysis (Semenov & Stratonovitch 2010; Muerth et al. 2013; Teng et al. 2015; Díaz et al. 2021).

Our findings agree with previous studies indicating changes in streamflow extremes in several South American tropical river basins over the past four decades, characterized by drying trends and accelerated flow rates attributed to factors such as decreasing rainfall, increased agricultural water use, and deforestation (Saurral et al. 2017; de Jong et al. 2021; Chagas et al. 2022). Additionally, the Orinoco River Basin is situated in a region where historical precipitation trends have shown a decrease (Carmona & Poveda 2014; Mesa et al. 2021; Builes-Jaramillo et al. 2024), and projected precipitation decline is robust under the highest emission scenario, both in terms of direction and magnitude (Almazroui et al. 2021). Furthermore, our results are consistent with other studies indicating a drier outlook for the Orinoco River Basin throughout the 21st century (Brêda et al. 2023; Correa et al. 2024).

The percent bias in the slope of the mid-segment of the FDC and the percent bias in the high-segment volumes (BiasFDCmidslope and BiasFHV, respectively) suggest that these indicators are lower for future simulated flows than for the reference ones in the periods 2021–2047 and 2048–2073 for both RCPs, whereas both indices show an increase for the period 2074–2099 for the two RCPs. These results suggest a gradual increase in the mid and high flows in the period 2074–2099. Finally, the differences in long-term baseflow and the differences in the flows at the 95% percentile (BiasFLV and DiffQ95, respectively) agree on an increase in the baseflow for all periods and RCP scenarios.

Previous works have studied the climate change impacts on river discharges through hydrological signatures as a crucial aspect for understanding the overall impact on river systems (Casper et al. 2012; Wang et al. 2020; Fatehifar et al. 2021; Salas et al. 2023). For the Orinoco River Basin, although other works have mentioned that climate change has the potential to significantly impact river discharges (Asadieh & Krakauer 2017; Brêda et al. 2023), there are no previous works using the signature indices approach proposed by Casper et al. (2012). Notwithstanding, other indices for hydrologic alteration have been used to study the characteristics of ecosystems and their maintenance within a basin (Mathews & Richter 2007; Pérez-Sánchez et al. 2020; Liu et al. 2024). Mathews & Richter (2007) argue that these types of streamflow indices provide information for describing the ways in which an organism experiences river flow variability. In this sense, low flows determine the amount and characteristics (e.g., temperature, flow velocity, connectivity, etc.) of aquatic habitat that is available for most of the year whereas high flows are associated with floods that play a critical role in a river ecosystem (e.g., sediment transport, refresh water quality, remove vegetation, etc.). In this sense, further research is needed in the Orinoco River Basin to improve our understanding about the association between alterations in the streamflow and its biodiversity and ecosystem services (Frappart et al. 2014; Lasso et al. 2016). Moreover, Pimentel et al. (2021) found that agro-industrial expansion and cover land changes in the Orinoco could increase pressure on water resources, leading to critical changes in the water availability regime. Hence, further research is also needed to advance the comprehension of the conjoined effect of those processes and climate change, which is out of the scope of this work.

Our results regarding the Upper Orinoco River indicate that precipitation decreases at a rate of 3.54 mm/decade (for RCP2.6) and at a rate of 23.9 mm/decade (for RCP8.5). Hence, these suggest a reduction of precipitation around ∼1.0% (for RCP2.6) and ∼6.0% (for RCP8.5) at the end of the century in relation to the long-term mean precipitation over the river basin (∼3, 100 mm yr⁻¹). Streamflow decreases at a rate of 1.225 m³ yr⁻¹ (for RCP2.6) and decrease at a higher rate of 5.969 m³ yr⁻¹ (for RCP8.5). Hence, these suggest a reduction of streamflow of around 1.5% (for RCP2.6) and 7.0% (for RCP8.5) at the end of the century in relation to the long-term mean runoff of the river basin (∼6,200 m³ s⁻¹). All streamflow simulations show a clear difference in the FDCs in the range of 0.2–0.4 which is not observed in the data. This finding in the simulated FDCs may be explained by a systematic bias in the climate models that could not be corrected by the bias correction process. These features require further research if these models are to be used to analyze streamflow in the range of 0.2–0.4 of the FDCs.

Our results show agreement regarding that precipitation and streamflow will decrease according to all datasets, aligning with the conclusions of previous works (Asadieh & Krakauer 2017; Brêda et al. 2023; Builes-Jaramillo et al. 2024). The predicted reductions for the RCP8.5 scenario in the period 2021–2099 are approximately 190 mm and 477 m³ s⁻¹, respectively. Previous studies have shown that deforestation and agro-industrial expansion in the Colombian Orinoco region may cause reductions in the Guaviare River's low water flows by up to 50% (Pimentel et al. 2021). In contrast, our results suggest an increase in low flows associated with climate change of around 7%. The general agreement of the GCMs and RCMs on the trends of simulated streamflow is mostly because the bias correction and ensemble procedures bring simulations closer together.

Our findings show that, for the northern Orinoco, GCMs and RCMs exhibit good simulation of precipitation, with the best performance provided by the CORDEX regionalization of the NorESM1-M. In this sense, our results do not show significant added value from CMIP6 to CMIP5 in terms of simulating precipitation. We hypothesize that this is a region with few observational records, a high mountain range, and a lack of parametrizations for the RCMs and GCMs, which have always been a caveat of climate simulation results in basins like the Amazon and La Plata in South America (Gulizia & Camilloni 2015; Sörensson & Ruscica 2018; Builes-Jaramillo & Pántano 2021).

Further questions about the effect of climate change on precipitation and streamflow of the upper Orinoco River Basin (Guaviare River basin) deserve future work: (i) How are the effects of climate change in conjunction with the land cover change in the basin?; (ii) How are the changes in low-flow at finer timescales?; (iii) Where in the basin the effects of climate change are more pronounced in the Andean or in the Llanos portion of the river basin?; (iv) How those changes in river flows could affect the river sediment discharge, ecological features, and environmental services?; (v) How to improve the knowledge about sub-surface runoff, interflow, and deep percolation with remote sensing in non-instrumented basins like upper Orinoco River?

Thanks to IDEAM for precipitation and streamflow data records and the regionalization of the CMIP5 models for Colombia. Thanks to the Stockholm Environment Institute (SEI) for proving a free license to the software Water Evaluation and Planning (WEAP) and particularly to Angelica M. Moncada Aguirre for her technical instruction about WEAP previously to carry out this research. The authors thank the research groups and institutions that provide complete access to the climate model results through the Earth System Grid Federation (ESFG) nodes https://esgf.llnl.gov/. Thanks to Prof. Jhony Perez for drawing a schematic representation of the rainfall–runoff model used in this work. Thanks to the Global Land Cover by National Mapping Organizations (GLCNMO) for Land Cover information.

The work of H.D.S. is supported by the Institución Universitaria Colegio Mayor de Antioquia and the Instituto Tecnológico Metropolitano. The work of A.B.-J. is supported by the Institución Universitaria Colegio Mayor de Antioquia. The work by C.F.-V. and J.V. is 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 M.B. and Y.T. is supported by the project ‘Modelación de caudales mensuales en la Cuenca del Río Guaviare (Colombia) ante escenarios de cambio climático’ (Project FAI107). The work by K.E. and M.R. is part of SICA, hydroclimatology group (HidroSICA).

H.D.S., A.B.-J., J.V., and C.F.-V. contributed to conceptualization; methodology; formal analysis; investigation; writing – original draft; visualization; writing – review and editing. H.D.S and A.B.-J. did project administration and supervision. C.F.-V. and J.V. contributed to processing of climate and hydrological data. H.D.S, Y.T., and M.B. did data curation, calibration, and validation of the model, M.R. and K.E. did analysis and processing of land cover.

All relevant data are available from an online repository or repositories: GCMs: ESFG-https://esgf-node.llnl.gov/projects/esgf-llnl, Gauges: https://dhime.ideam.gov.co/, IDEAM Regionalization: http://www.ideam.gov.co/web/tiempo-y-clima/escenarios-cambioclimatico.

The authors declare there is no conflict.