Managing water resources sustainably as well as minimizing the risks linked to water-related hazards such as droughts and floods requires a good understanding of the water cycle.
To simulate the hydrological system, hydrological models are developed, for example to evaluate the response of water availability for a specific policy intervention, or to anticipate the effect of climate change on water resources. Regional hydrological models can be useful for short to medium term planning, while global hydrological models play an important role in informing global policies and evaluating the water-related impacts associated with different socio-economic and climate related challenges in the longer term.
Human activities have a huge impact on the water system, for example by building dams for storage, diverting water for various economic uses, developing settlements, or by shifting natural ecosystems into agricultural areas. These changes alter the amount, quality, and timing of the water. Therefore, a good representation of human impacts on the water system is key to improving hydrological models. And while many global hydrological models now include a parameterization of human activities, we are still unsure how well these perform. But let’s first explore how human activities are included in several state-of-the-art global hydrological models.
Human activities in global hydrological models
Here, we analyse five different state-of-the-art and widely used global hydrological models models (H08, LPJmL, MATSIRO, PCR-GLOBWB, and WaterGAP2). These models include three main categories of human impact in their modelling routines:
- Water usage for various sectors, such as the domestic sector (e.g., drinking water and sanitation), industry, agriculture, and livestock.
- Dams and reservoirs, operated for electricity generation, irrigation purposes and flood control, all of which can affect the river discharge. Some models also consider evaporation from the reservoirs.
- Human-induced land use and land cover changes have a large impact on important hydrological processes, such as evaporation, infiltration, and runoff generation. One major example is the transformation of forest to agricultural land.
Comparing simulations with and without human impacts
To assess how well human impacts are represented in hydrological models, we need to assess how well the simulations match the real world. The lower the deviation from real measurements, the better the model performs. In particular, we consider how well the models perform with and without the inclusion of human impacts.
Model simulations
First, we run simulations of monthly discharge (0.5° × 0.5° spatial resolution) for the period 1971–2010 using the aforementioned hydrological models, both with and without human impacts. All runs used the GSWP3 climate dataset (i.e., historical temperature and precipitation input). The simulations were carried out using the modeling framework of phase 2a of the Inter-Sectoral Impact Model Intercomparison Project ISIMIP2a.
Real-world observations
To compare the simulations to the real world, we collected observations from the Global Runoff Data Centre (GRDC). To perform a robust analysis, we selected only those stations that meet the following criteria:
- a minimum of 5 years’ coverage during the period 1971– 2010 with a completeness of observations of ≥95%
- a minimum catchment area of 9000 km2
The first criteria is important to ensure that a sufficiently long comparison between model simulations and observations can be made, while the second criteria is important because the current cell size of the analyzed hydrological models is relatively coarse (0.5° × 0.5°; ∼50 km × 50 km at the equator) and small catchments are thus ill-represented in these models.
Out of 9051 gauging stations in the GRDC database, this resulted in 471 stations with a total catchment area covering 19.8% of the global land. For these stations the average length of observations is 32.8 years. 92 out of the 471 are gauge stations are located at the outlet of the catchments, while the other stations located are further upstream. 226 stations are located in managed catchments and 245 in (near-) natural catchments.
Here, we define (near-) natural catchments as catchments where less than 2% of the catchment is irrigated and the volume of water that can be stored in dams is less than 10% of the mean annual discharge. All other catchments are considered as managed catchments.
Evaluation criteria
We compared two aspects between the model simulations and the observations, the results of which are shown in Figure 2.
- Monthly average discharge is evaluated based on three characteristics:
- Dynamics: whether the observed discharge values recorded at the gauge station follow a similar trend i.e., when observed discharge increases, simulated discharge increases as well, and vice versa.
- Intra and inter-annual variability: how good models are able to capture the changes in monthly and annual discharge that takes place within a year or from year to year.
- Deviation with respect to long-term mean: whether average monthly values overall underestimate, overestimate, or are in line with the mean observed records.
- Hydrological extremes are unusual high- and low-river flows with a low probability of occurrence. These extremes are especially important to simulate because they are associated with floods and hydrological droughts. Here, we consider a high flow when this or a higher discharge is only seen 1% of the time, based on the previous 100 months. Similarly, we define low flow equivalent to extremely low discharges that are only seen 1% of the times based on previous 100 months. Both monthly average discharge and hydrological extremes were evaluated separately for (near-) natural catchments and managed catchments.
What did we find?
Overall, we find that the inclusion of human activities in the model runs significantly improves the simulation of discharges and hydrological extremes for most catchments. Let’s have a look in detail at the monthly average discharge and hydrological extremes.
Simulated river discharge compared to observations
Include human impacts:
Performance metrics:
Hydrological model:
This figure shows the various evaluation criteria used for the 471 discharge stations used. Toggle "include human impacts" to see how the performance of the models changes when human impact is (not) included. Note that 1.0 is the optimal for all scores. Higher or lower scores indicate lower agreement between the observations and simulations.,
FIG 2 / Adapted from Veldkamp et al. (2018)
Monthly average discharge
The inclusion of human impacts parameters leads to a significantly better performance (i.e., better match with observations) in 41% to 72% of the land area, depending on the model. In general, the highest improvements are in catchments with a higher level of human transformation.
Figure 2 shows that, while performance increases in almost all regions, a decrease is observed in some other regions. An overall strong performance is observed in Latin America, southern Africa, and the northwest US. There are only a limited number of stations for which the inclusion of human impact parameters leads to a significant decrease in overall hydrological performance for the majority of global hydrological models or where no, or limited changes occur. This is mostly in near-natural areas, such as the Amazonas.
Yet, for most global hydrological models the positive effects outweigh the negative effects. Specifically, WaterGAP2 performance improves mostly in all global regions, while MATSIRO performance increases substantially in Europe and Latin America due to the inclusion of human impact parameters. H08 and LPJmL perform best in the USA, particularly the East Coast, while PCR-GLOBWB performance improves in the US East Coast, Latin America and Europe.
Despite the inclusion of the human impact parameters, overall performance of global hydrological models is still better in (near-)natural catchments, because the human impact is extremely complex and thus, despite best efforts, remains difficult to capture in a large-scale hydrological model.
Hydrological extremes
Overall, hydrological extremes are reasonably well represented in global hydrological models when human impact parameters are included. The simulation of high-flows is significantly improved (i.e., potential floods) across 35% to 77% of land area, depending on the model. For low-flows the simulations improve significantly across 39% to 80% of land area.
The inclusion of human impact parameters in general decreases the value of extreme high flows. For low flows the results are more mixed. This is likely because human activities have a larger impact on flows during droughts, as the percentage of water abstractions and of volumes managed for reservoir operations represent a higher share of the total flow.
However, despite the overall improvement, there are remarkable differences in the way the five global hydrological models simulate hydrological extremes. Geographically, models are less able to represent these events in high latitudes, particularly in the east and west coast of North America, Canada, and Eurasia.
Pathways to improve hydrological models
While we show that the inclusion of human impact indicators substantially improves model performance, global hydrological models are still far from perfect, especially so in managed catchments. A large number of improvements could lead to a better representation of the real world, of which several examples are listed below:
- the improvement of primary climate forcing data such as precipitation and temperature. Expanding the global coverage of these data, as well as ensuring harmonization and data quality protocols will likely have a highly positive effect on the performance of models.
- enhancement of the calculation of evapotranspiration. This is a key hydrological process, which is affected by climate data, soils, vegetation, irrigation timing and quantity, among others. Its calculation has a large impact on the simulation of water flows and the replenishment of groundwater aquifers.
- the improvement of the capacity of hydrological models to “split” between the water that flows as surface, subsurface and for those replenishing aquifers. The soil processes that determine this split depend on multiple biophysical factors that can vary substantially from one location to another, and therefore increases the difficulties of models to provide accurate results.
- gains in the calculation of snow accumulation and melt. These processes are not well captured in some global hydrological models, but can have a large effect on the simulation of the inter- and intra-annual flows, especially in mountainous or high-latitude catchments.
Limitations and future studies
There are several limitations to this study that could be addressed in the future. Most notably, the study was conducted with data that covers only 19.8% of the total global land. For large parts of the world, including sub-Saharan Africa (except for southern Africa), the Middle East, South Asia, and Southeast Asia, the lack of data impedes evaluating the performance of global hydrological models. Given that some of these areas are also among the most populated ones, it becomes crucial to improve the data collection and sharing programs in these parts of the world.
Moreover, future studies should look more specifically into how individual human impact parameters are included and the effect on model performance. While this is a computationally expensive process (i.e., each individual parameter setting needs to be tested and assessed) this would likely help to improve and global hydrological models.
Finally, this study only investigated monthly discharge values at a relatively coarse spatial temporal scale, while water demand and water-related risks can change daily and are often very location-specific. Therefore, this study could benefit from a higher spatial temporal resolution. Fortunately, the global hydrological modeling community is making substantial progress to increase the temporal and spatial resolution of their models, which might also help to make global models more suited for national planning. Nevertheless, for this study the resolution was kept “coarser” for consistency purposes with the data availability as part of ISIMIP program.
Acknowledgements
This summary article was written in collaboration with the Editorial team of ISIpedia project.
References
Cover image: Eric Vance for EPA
Affiliations
1 International Institute for Applied Systems Analysis, Laxenburg, Austria
2 Department of Physical Geography, Utrecht University, the Netherlands