Fit for purpose? Rapid development of water allocation models using global data: Application for the Upper Niger Basin
Graphical abstract
Introduction
Shortage of water can have various economic and social consequences. Low river discharges can lead to poor water quality and reduced fish catch, unsatisfactory rain and irrigation water can result in failed harvests or reduced availability of grasslands for livestock, reduced water supply for households can have implications for human health. These impacts of water shortage can subsequently lead to loss of lives, of livelihoods as well as of socioeconomic opportunities and may result in humanitarian crises. Perceptions of underlying causes of water shortage and of an unequal distribution of impacts can further contribute to dissatisfaction and could trigger social unrest and violent conflict between different water users, different communities or larger regions (United Nations and World Bank, 2018). To respond effectively to emerging or on-going disasters or conflicts and avoid humanitarian, diplomatic or water management measures that have unintended consequences, it is important to have information regarding the current water resources situation and the effectiveness of possible risk-reducing actions. Also, the development and sharing of incorrect perceptions about water resources availability and distribution that could lead to competition and conflict can be prevented by timely producing and sharing data and information that is easily accessible to all. Especially in protracted conflict situations, using openly available data can contribute to neutrality and trust of the data and of the resulting analysis results.
Water resources models are commonly used to generate information on how water availability and shortage vary under the impacts of climatological variation, increased water use or increased river regulation. These models are typically based on information on land use and elevation, climate, and water use and are used to understand the impacts of possible future developments or of policy and management actions on water infrastructure, water users and ecosystems. At the same time, they inform decision-makers in their decisions concerning the use, the development or the protection of water resources. Several requirements can be identified for the model and resulting information: models and resulting information need to be 1) salient, 2) credible, and 3) legitimate (Van Voorn et al., 2016). Salience refers to the relevance of the model for the policy questions that need to be answered, credibility refers to the scientific quality of the model, and legitimacy refers to the extent to which stakeholder views and concerns have been considered in the process of model development.
Various authors discuss model choices in relation to the selection of the model concept (e.g. Addor and Melsen, 2019; Babel et al., 2019). What has received less attention is the approach to developing these models, and the opportunities of making use of the increasing availability of global scale data. Model-based information is currently mainly available either through detailed catchment studies or through coarse global level models. Both have limitations, which we discuss below. Therefore, in this paper we argue that an intermediate level is required at which global data are used for catchment level analysis.
Van Voorn et al., (Babel et al., 2019) suggest that often a trade-off needs to be made between the three model quality criteria, and found that stakeholders value salience over credibility. Such trade-offs are usually necessary due to time and budget constraints. Using global data for catchment level analysis can provide a suitable method to develop a model relatively quickly. A question that this paper seeks to help answer is to what extent credibility and salience are compromised if this approach is taken. An additional benefit of using global data is its replicability - the method, including the data processing scripts, can potentially be re-used, further reducing time and budget requirements.
In Fig. 1 we present the three levels: 1) catchment models based on local data, 2) catchment models based on global data, and 3) global hydrological models based on global data in relation to salience, credibility and resource constraints. Because our focus is on testing the credibility and salience of a model based on global data, based on a desk-study, we leave legitimacy out of consideration, which would have required consultation with stakeholders in relation to an actual policy question. However, we believe that using global data sets that are accessible to all is beneficial for model legitimacy.
Here we first discuss the strengths and weaknesses of the catchment analysis using local data, and of the global water resources models, to then make our case for the intermediate level of catchment level analysis using global data.
Catchment-scale water allocation models using locally collected data have been the traditional approach for water resources planning studies for many decades (see e.g. Loucks and Van Beek, 2005). Because of the local level of the study, models and data-processing could be tailored to the specific characteristics of the catchment and the information needs of stakeholders, thus potentially resulting in high salience and legitimacy. Moreover, to achieve this, these models usually have a high level of detail, and need high-quality locally measured data which improve the credibility of the model. These studies are generally commissioned by government actors and are useful for strategic long-term planning which requires detailed analysis of what type of policy actions are suitable under a range of possible future situations. The results will answer the question of what measures will be most effective to achieve objectives for the water resources system and will help actors to decide what measures to implement. However, developing such models may be time-consuming and resource-intensive, and the locally measured data for this type of high-resolution model is not always available or of unsatisfactory quality.
Where high-quality local data is lacking, global hydrological models can offer an alternative. Data sets, amongst others on climate, weather, soil, geology and topography, that cover large regions, or the entire world have increasingly become available over the past couple of decades. The availability of such (near) global data sets has enabled the development of global hydrological models (GHMs). Global hydrological models have long development and run-times, but increasingly model results and underlying data are directly available free of charge online, for example through the eartH2Observe Water Cycle Integrator (WCI) (Plymouth Marine Laboratory, 2020). In such cases, required resources to use the information are very low and accessibility is high. The level of detail of GHMs is low, compared to catchment models. GHMs are gridded models in which all water balance computations take place at the grid cell level, with a typical spatial resolution of 30 arc minutes, which is around 50 × 50 km at the equator. See Bierkens et al. (2015) for an overview of various GHMs. Over time, GHMs have developed into increasingly finer resolutions (down to 5 × 5 km), increasing their applicability at the local level (e.g. PCR-GLOBWB 2.0 (Wetlands International, 2019), WaterGap 3 (Verzano et al., 2012) and the W3 (version 2) model (van Dijk et al., 2018). Many of these GHMs also take water demand and river regulation by reservoirs into account. Despite the increasing resolution and possibilities to include water demand and regulation, GHMs follow a single approach for all catchments and are not designed to be tailored to the specific needs and interests of stakeholders in a specific catchment. Also, most GHMs do not offer functionalities to easily test different reservoir operation rules or water demand changes. This means that salience of these models is likely to be unsatisfactory for specific catchment level questions. Moreover, the GHMs are generally not calibrated for specific catchments. As a result, the representation of the local hydrology is unknown, and credibility may be inadequate. GHMs are very suitable to quickly identify where and when water shortage-related risks could occur, and to get a general indication of the causes of water shortage and possible solution options. They are less suitable to analyze catchment-specific variations in water use, water regulation or options to explore different water management strategies and thus might not meet decision-makers’ needs.
Because of the limitations of GHMs regarding salience and credibility, and of models based on local data regarding resource requirements, we present here an intermediate level in which catchment models are developed using online global data. In this paper, we focus on this intermediate level, which has been much less applied and researched than the other two levels. Existing basin scale applications of global data tend to focus only on the biophysical components (e.g. Gusev et al., 2006; Conway and Mahé, 2009, Andersson et al., 2017). Liersch et al. (2019) developed a model of the Niger Basin that includes water abstractions for irrigation and reservoir operation. In that study, only the rainfall-runoff component was done using global data. Additional local data and information were used for the water allocation and demand component. Pechlivanidis and Arheimer (2015) developed a hydrological model for India, only based on global datasets (including reservoirs, agricultural and irrigation datasets), and calibrated the irrigation parameters that regulate water demand and abstraction. Siderius et al. (2018) applied a GHM (LPJmL), including irrigation abstractions, to the Rufiji river basin in Eastern Africa, and changes to the input (rainfall), model schematization and parameters were required to improve model performance from generally poor to reasonable or good. The use of solely open data to develop a local (basin level) water resources model has not yet been extensively tested. In this paper, we extend the use of open data for catchment level analysis with water use and water regulation. For this purpose, we develop and evaluate a combined model, in which a gridded rainfall runoff model provides input to a network-based water allocation model, both based solely on global open data sets that are available online. We evaluate the credibility of the combined model in two ways, by considering both standard hydrological performance metrics, and dedicated ‘fit for purpose’ indicators. We apply this approach to the Upper Niger Basin in Mali. The Inner Niger Delta that depends on inflowing water from the Upper Niger Basin is prone to violent conflict that may have a link with natural resource availability. Decisions need to be taken on a new dam and expansion of irrigated areas in the Upper Niger Basin, that are likely to alter the natural flooding regime of the wetland area and will subsequently impact the ecosystem services that form the basis for the livelihoods of its inhabitants. Information on how upstream interventions will impact the livelihoods of the people in the Inner Niger Delta and the conflicts over natural resources is important to into account in decision on theses interventions. The water resources in the catchment have been studied using local data, a recent example is the study by Liersch et al. (2019). We therefore use that study as comparison material for the global data-based model presented in this study.
Section snippets
Study area
The study area consists of the entire Upper Niger catchment including the Inner Niger Delta (IND), until the border between the Tombouctou and Gao regions in Mali (Fig. 2a). The catchment is located in Guinea, Ivory Coast, Burkina Faso, and Mali and is around 380,000 km2 in size. The headwaters of the main Niger river branches are located in the Daro Massif in Guinea, while the main tributary, the Bani, originates in southern Mali and Ivory Coast. The Bani enters the Niger river at the upper
Hydrological performance
Table 7, Table 8 present the KGE and NSE performance indices for the wflow_sbm model with the adjusted parameters, for discharge simulation at a daily time step and the resampled discharge simulations to a monthly time step. Fig. 3a, Fig. 3ba and b presents a comparison between the simulated and observed monthly discharges for simulations with MSWEP and CHIRPS, respectively. The performance indices were computed for the simulation with MSWEP, for the period 1980–2012, and for the simulation
Discussion
In this paper, we present a new method to quickly construct water resources models based solely on open data that can be used for rapid assessments where time is of the essence or high-quality local data is unavailable, and which is suitable for the evaluation of alternative interventions in land and water management. Considering the combined results of the hydrological performance metrics and the fit for purpose indicators, we argue that our model based on open data with minimal tuning is of
Conclusions
In this paper, we presented a new method to quickly construct water resources models based solely on open data, and which is suitable for the evaluation of alternative interventions in land and water management, especially in situations where decision-making is required to maintain or reestablish cooperation, stability and peace. This type of model presents an intermediate level between coarse global models and detailed models based on local data. This modelling level can provide results within
Software availability
Name of software: wflow; Developer: Deltares Contact information: [email protected] Year first available: 2014.
Required hardware and software: 64-bit Windows/Python or Linux/Python.
Program language: Python 3.6+
Program size: 2.5 Gb (conda environment).
Availability and cost: wflow is released as free open source code under the GNU General Public license version 3 (https://github.com/openstreams/wflow). Free license for executable.
Name of software: RIBASIM (River Basin Simulation Model).
Developer:
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
The research for this paper was done in the framework of the Water, Peace and Security (WPS) partnership, supported by the Netherlands Ministry of Foreign Affairs, grant number 4000003751. The specific component focusing on the use of global data for rapid water allocation model development, which is the topic of this paper, received additional support from Deltares' research program on Water and Societal Stability. We thank two anonymous reviewers for their constructive comments, which helped
References (57)
- et al.
Decision-making in model construction: unveiling habits
Environ. Model. Software
(2019) - et al.
Google Earth Engine: Planetary-scale geospatial analysis for everyone
Rem. Sens. Environ.
(2017) - et al.
A software framework for construction of process-based stochastic spatio-temporal models and data assimilation
Environ. Model. Software
(2010) - et al.
Water resources planning in the upper Niger river basin: are there gaps between water demand and supply?
J. Hydrol.: Reg. Stud.
(2019) - et al.
Going local: evaluating and regionalizing a global hydrological model’s simulation of river flows in a medium-sized east african basin
J. Hydrol.: Reg. Stud.
(2018) Low flow hydrology: a review
J. Hydrol.
(2001)- et al.
A checklist for model credibility, salience, and legitimacy to improve information transfer in environmental policy assessments
Environ. Model. Software
(2016) - et al.
Legacy, rather than adequacy, drives the selection of hydrological models
Water Resour. Res.
(2019) - et al.
The Niger river basin: a vision for sustainable management (English)
Directions for Development
(2005) - et al.
Globcover 2009
ESA Living Planet Symposium 27 June - 2 July 2010 Bergen, Norway
(2010)