Assessment of human impact on water resources

The human impact is an essential determinant of hydrological behaviours in most of the societally relevant catchments all over the world. Human impact is playing a significant role on fresh water resources and in some regions is actually conditioning water resources availability. Experience suggests that human impact assessment is an essential prerequisite to ensure the sustainable design of water resources systems.

There is the widespread perception that the whole globe is significantly suffering from human impact, therefore determining a globally critical situation for climate and water resources management. Actually, most of the globe surface is not significantly human impacted, if one excludes the potential effect of human induced climate change, whose hydrological implications are the subject of a vivid and to some extent controversial debate. However, the most important catchments for society are typically those where humans live and exploit resources and these are of course human impacted, to a different extent.

Human impact affects environment in several ways (see Figure 1 and Figure 2). A list of potential consequences is is given here. There is the general perception that the human impact always induces negative consequences. However, the opposite actually happens: most of the consequences of the human impact are positive, as humans make an effort to shape the environment to meet their needs. However, there undoubtedly are negative consequences from the human impact. Unexpected consequences are concerning, especially those originated by non-linear effects that may give rise to the so-called tipping points.

Figure 1 and Figure 2: example of human impact over a catchment. Source of Figure 1:

Human impact should be adequately considered when planning the exploitation of water resources and mitigation of natural hazards. Therefore, it should be adequately considered in hydrology. Actually, human impact is a key element that is considered by integrated water resources management, but scientists are today dedicating much attention to improving our understanding and interpretation through models of human impact.

1. Human interventions impacting water resources

The human impact on freshwater resources can be classified into the following two main categories:

  • Human impact on freshwater availability;
  • Human impact on freshwater quality.

The above impacts can be direct or indirect, and take place at several different spatial and temporal scales. Examples of human interventions that may affect freshwater resources quantity and quality are:

  • Human induced climate change at the local and global level;
  • Regulation of river flow and water storage in lakes;
  • River trainining, including river diversions and river tunnelling;
  • Fresh water withdrawal;
  • Land-use change, including agriculture, drainage, afforestation and deforestation.

The impact of humans on water resources is the subject of a vivid debate within the scientific community. While it is agreed that water is essential for life and humans need more and more water, while at the same their action is compromising the quantity and quality of freshwater, it is not fully clear what are the most concerning impacts and therefore it is not clear what mitigation actions should be adopted with higher priority. To effectively and sustainably design water resources systems it is essential to identify, at the global and regional level, the main triggers of human impact.

2. Human impact identification and assessment

Human impact on water resources needs to be identified by coupling perceptional and quantitative assessment: perceptional assessment is based on the identification of hard and soft facts that may have impacted freshwater availability and quality. It provides the support to further and quantitative analysis. Quantitative assessment is based on data analysis and modeling. It provides the detailed information that is needed for engineering design. Quantitative assessment, like any other hydrological analysis, is affected by uncertainty. Perceptional assessment helps to reduce uncertainty and provide support to communication to society.

2.1 Perceptional assessment of human impact

Perceptional assessment is based on the identification of hard and soft facts that may have conditioned freshwater. It is sometimes not easy to discern among hard and soft facts. The scientific community is still divided on impact assessment of certain human induced changes. The distinction between hard and soft facts depends on the local situation and challenges. It may be useful to define hard facts as the direct effect of humans on freshwater through engineering works or massive land-use changes.

According to the above premise, hard facts may include:

  • Demographic expansion;
  • River diversion;
  • River damming;
  • River training;
  • Groundwater withdrawals;
  • Extended and substantial land-use change, including urbanization and deforestation;
  • Local climate change in climate-sensitive areas.

Soft facts may include:

  • Global climate change;
  • Local land-use change with reduced impact on freshwater resources;
  • Changes of social and environmental conditions, including agricultural changes;
  • Status of the water distribution systems.

The above facts should be assigned proper priority in order to transparently identify the reasons of concern. Stakeholders involvement, according to local policies and laws, is an essential requirement to set agreed basis for the design of mitigation actions.

2.2 Quantitative assessment of human impact

Quantitative assessment of human impact is usually carried out through:

  • Analysis of historical data, which is often based on trend analysis;
  • Modeling.

Application of the above approaches is discussed here below.

2.3 Quantitative assessment of human impact through historical data analysis

Historical data analysis is a frequently used approach to assess human impact. It has been recently criticized because the representativity of past data to assess changes in environmental variables has been widely questioned, for the presence of errors and gaps in observations and the reduced spatial density of monitoring stations. In several cases questions arise on the homogeneity of data, for example, about whether the series is equally reliable throughout its length. For this reason, in some fields like climatic research the use of models is a frequently used approach. In hydrology and water resources management past data analysis is usually preferred for the applied character of the related assessments. When design variables are to be determined in practice, data analysis is frequently preferred as a more reliable approach with respect to model application, in particular when uncertainty of models cannot be convincingly assessed.

Several approaches can be applied to assess change, and therefore human impact, in environmental data. The most used approach is trend analysis of a historical time series. If the trend can be assumed to be linear, trend analysis can be undertaken through regression analysis, as described in Trend estimation. If the trends have other shapes than linear, trend testing can be done by non-parametric methods, which are discussed here.

Linear trend estimation is essentially based on fitting a linear straight line interpolating the data. Therefore, a given time series y(t) is assumed to be decomposed in a straight line plus a noise:

y(t) = a + bt + ε(t)

where a and b are the intercept and slope of the trend line, respectively, and ε(t) is a randomly distributed error whose statistical properties are assumed not to change in time. If the slope b is significantly positive or negative, then a trend in the data exists which provides an indication of the related change. This approach is indicated for the estimation of changes that area gradually occurring in time, while it is not representative of sudden changes like river diversions, river damming or the perturbation given by other infrastructures.

To estimate a and b and the statistical behaviors of ε(t), which can be considered as parameter of the above linear model, the least squares approach is usually applied.This method minimizes the sum of the squared values of ε(t), as computed by the difference between the y(t) values and the available and corresponding observations:

min ∑(y(t) - a - bt)2

where y(t) is the observation. By using proper inference methods a confidence band can be computed for the value of the slope. In statistics, confidence bands or confidence intervals potentially include the unobservable true parameter of interest. How frequently the estimated confidence band contains the true parameter if the experiment is repeated on different time series is called the confidence level. Therefore the confidence level provides an estimate of the probability that the computed slope is actually different from zero. The confidence level is the complement of the level of significance, namely, a 95% confidence interval reflects a significance level of 0.05.

More generally, given the availability of a hypothesis testing procedure that can test the null hypothesis b = 0 against the alternative that b ≠ 0, then a confidence interval with confidence level γ = 1 − α can be defined as the interval around 0 containing any b value for which the corresponding null hypothesis b = 0 is not rejected at significance level α.

Computation of the confidence interval is affected by the assumptions on ε(t) and in particular the assumptions related to its probability distribution. Usually the errors are assumed to follow a normal distribution and to be uncorrelated. These assumptions heavily impact the estimation of the confidence interval. In particular, the presence of correlation widens the width of the confidence interval significantly. Therefore, the question often arises on the possible presence of correlation in environmental variables, and in particular the present of short correlation extended over long time ranges, which corresponds to the presence of long term cycles. If the considered time series is affected by such long term periodicities, then distinguishing between a linear trend and a long term cycle may be difficult. For instance, characterising the current trend that is observed in the global temperature is challenging if one accepts the idea that the climate system may be affected by such long term cycles.

An idea of the variability of environmental data and related tendencies and cycles may be given by Figure 3 and Figure 4, which display the progress of a trend line estimated in a moving window starting from 1920 and encompassing 50 observations of annual maxima (Figure 3) and annual minima (Figure 4) of the Po River daily flows at Pontelagoscuro. The considered time series is extended from 1920 to 2009 and therefore provides the opportunity of testing how the slope of the regression line changed along the observation period. Even if 50 years is a long period, which is usually considered extended enough to allow a reliable trend estimation, one can see that the slope of the regression line is continuously changing, therefore proving that natural fluctuations lead to considerable changes in the dynamics of the river flow which cannot be easily interpreted through a linear trend.

Figure 3. Regression line estimated along a moving window encompassing 50 years of annual maxima of the Po River at Pontelagoscuro. Increasing and decreasing slopes are depicted in red and blue, respectively.

Figure 3. Regression line estimated along a moving window encompassing 50 years of annual minima of the Po River at Pontelagoscuro. Increasing and decreasing slopes are depicted in blue and red, respectively.

2.4 Quantitative assessment of human impact through models

Models for assessing the human impact on water resources may be classified between:

  • Methods representing the human impact as an external forcing.
  • Methods embedding humans as an inherent component of the hydrological cycle.
2.2.1. Methods representing the human impact as an external forcing

These methods are based on the study of pristine hydrological systems to obtain an estimate of the desired design variable in absence of human impact. Then, humans are modelled through a separate approach, therefore determining a correction of the design variable to take human impact into account. A very simple example is the estimation of the flow duration curve (FDC) for a river affected by upstream water withdrawals. One may first estimate the flow duration curve in undisturbed conditions, and then estimate the amount of water withdrawal which is subsequently subtracted from the unimpacted FDC. This approach may ignore downstream effects of the water withdrawal, like for instance increased infiltration in the river bed given by the groundwater deficit originated by the withdrawal itself. Therefore, the actual human impact on the undisturbed flow duration curve may go beyond the water withdrawal alone. These effects, that may be difficult to estimate, are called "feedbacks" of the human activity into hydrological processes, which are affecting the dynamics of water systems. Therefore, there is an underlying assumption that the human and the water system evolve independently each other.

Representing the human impact as an external forcing is a simple solution that in many cases captures the relevant behaviors of the integrated human-water system. It has been used in most of the practical applications, to design water supply systems, urban drainage systems, flood mitigation actions and so forth. Its advantages are robustness and therefore reduced uncertainty with respect to more detailed approaches. The method can be applied to either deterministic or stochastic models. However, in the latter case a probabilistic interpretation of the human behaviours is needed if uncertainty in the human impact is to be taken into account.

2.2.2. Methods embedding humans an an inherent component of the hydrological cycle

These methods are based on the integration of the human dynamics into the models leading to the estimation of the design variables. Such embedding generally leads to an increased number of model parameters and therefore to an increased estimation variance, but allows for a possible integration of any interaction between water and humans, therefore providing the means to account for the relevant feedbacks. A simple example is given by a model that may estimate effective evapotranspiration as a function of the water withdrawals, for irrigation, therefore modifying the representation of the hydrological cycle depending on human impact.

A relevant question is how to represent the human system, which is very complex by its nature. As for the case of hydrological models, one may decide to adopt a deterministic versus a stochastic approach. The deterministic approach may be adopted by expliciting the dependence of some of the model variables on selected social dynamics, by ensuring feedbacks (otherwise one may fall back to the case of external forcing). A relevant question related to the deterministic approach is its representativity for the human dynamics. While hydrology is by its nature governed by physical deterministic equations, and uncertainty is determined by imprecise knowledge or imprecise representation of the system geometry or measured input and output variables, human dynamics are not governed by deterministic relationships. Therefore, the use of deterministic dynamics is even less justified for human systems with respect to hydrological systems. Therefore, the use of deterministic social-hydrological models turns out not to be useful for engineering design. Their application may serve for improving our understanding of the outcome that may result from assigned conditions, namely, to assess "what if" scenarios.

Describing the combined human and hydrological system as a physically-based stochastic approach presents the relevant opportunity of describing uncertainty. However, proper assumptions are needed to make sure that uncertainty is properly represented, and testing should be carried out to ensure that models are reliable. By no means the hydrological model should be combined with improper or unreliable representations of interlinked processes. In fact, while an incorrect external forcing does not impact the inherent uncertainty of the hydrological model, an integrated representation presents the relevant risk of making the hydrological analysis unreliable if the links and feedbacks are not properly described. Therefore uncertainty analysis, through statistical testing and model selection, is an essential tool to decide whether it is the case of combining water and human dynamics rather than treating humans an external forcing.

There is the general perception that estimating uncertainty for hydrological models is a challenging task. Actually, one only needs the identification of any relevant uncertainty source and running the simulation models for several different values of any uncertain external forcing and/or parameter, extracted from suitable probability distributions. Finally, one should add to the obtained simulations the structural error of the adopted modelling scheme, whose distribution must be properly assessed, preferably through model validation. A blueprint for uncertainty assessment is described by Montanari and Koutsoyiannis (2012). The theoretical part of the paper is essentially technical but the applications therein presented are conceptually simple.

which is affected by randomness, within a hydrological model. One should consider that the uncertainty of the resulting estimation may be inflated, for the random evolution of the human system that is superimposed to the random evolution of the hydrological system. Therefore, the opportunity of increasing the model complexity should be evaluated in light of the resulting uncertainty in the design variable.

Next Lecture;
Previous Lecture.

Last modified on April 2, 2017