3.1. Characterization of the Studies Selected: Communities, Geographical Scales and Quantification Methods
The 39 papers have a strong focus on rural populations, especially on farmers and herders. This may result from the fact that there is a strong link between water availability and rural livelihoods, with 70% of global freshwater used for agriculture, and agriculture being important as a livelihood basis for rural communities (See
Table 3). This is particularly the case in Africa and Asia, which shows clearly from the geographical spreading of the studies (See
Figure 2).
Most of the modeling of human responses is done at the sub-national scale, on regional, state/province or community level. This makes sense for three reasons: (1) to understand the changes in water systems that can induce human responses it is important to understand how these changes affect specific types of land use, such as deltas or irrigation areas, (2) modeling (groups of) actors of too large areas would result in very large models. Large-scale models would therefore require the aggregation of individual actors into actor groups. (3) The detailed data collection through interviews as applied in most studies is resource-intensive and can only be done among a limited set of agents. Nevertheless, it would be interesting to explore how small-scale findings can be applied to larger scales, to further improve continental scale assessments of societal impacts of, for example, climate change.
Four categories of quantification methods were found, with one paper not fitting any of the categories and assigned the label ‘other’. We find that most studies (
n = 16) use statistical methods (category 1) to assess correlations between responses and explanatory factors. Fourteen studies simulated human decisions in response to changes in their environment at the level of individuals or actor groups, mostly agent-based models (category 2). Six studies used system dynamics models (category 3), and two studies used optimization methods (category 4). Quantification by means of models does not however necessarily imply simulation or optimization. For example, [
22] developed equations to identify equilibrium situations; whereas [
23] quantified thresholds at which pastoralists would switch between survival and performance strategies. This last paper is assigned the category ‘other’ since it did not fit any of the four categories that we distinguished.
Often system dynamics models and agent-based models are connected to other models that represent the biophysical system, such as hydrological or water allocation models. For example, Gies et al. [
24] used a combined hydrological and system dynamics model to assess changes in income and the population affected by migration in relation to a variety of water and land management measures such as water harvesting, water storage or irrigation efficiency. Gohari et al. [
25] used a combined hydrological, socio-economic and agricultural system dynamics model to assess the impacts of climate change and various adaptation strategies. Human responses are included as feedback mechanisms in which decisions are made to maximize utility at system level, and in which variations in utility affect water demand.
3.2. Human Responses
The selected 39 papers all quantify human responses as a result of changes in water availability (See
Table 4 for an overview). We only mention which types of response options were included in the analyses in the various studies. This does not imply that the authors found these responses to be the consequence of changes in water availability. These response options can be grouped into five categories. Short-term coping strategies (category 1), other than alternative income, was mentioned in 13 of the papers, and were found to consist of (1) reducing consumption [
26,
27], (2) obtaining additional income by selling assets [
26,
27,
28], borrowing money or products [
27,
28], or (3) receiving food aid, additional remittances and other types of community sharing or external assistance [
23,
29,
30,
31]. In addition, short-term coping could require buying fodder to replace natural vegetation [
26,
27]. Livelihood adaptation is the most frequently described response, mentioned in 30 papers. Adaptation is achieved through either or both income diversification (
n = 10) to be less dependent on income sources subject to water availability, and adaptation of agricultural practices (
n = 27). We considered these as two separate human response categories. Income diversification (category 2) can consist of a (partial) shift to another livelihood, such as combining pastoralism with cereal cultivation or charcoal making [
32,
33] or finding paid labor on or off the farm [
34]. Adaptation within the agricultural sector (category 3) is done in various ways, for example, by managing water and other inputs [
28,
35], improving water supply infrastructure and wells [
31,
36], by making well-informed choices on crops or crop varieties, livestock management strategies [
32,
37], and new ways of preparing land or planting crops [
31,
38,
39]. Migration (category 4) is rather frequently (12 papers) mentioned as a response option as well, either seasonally or for other shorter periods of time, such as return migration as well as permanently [
40]. Violent behavior and social tensions (category 5) were discussed as the results of water shortage in two papers only [
22,
41]. Many studies (
n = 18) found responses in several categories, which shows the variation in possible human responses related to reduced water availability, and highlights that people do not always choose a single response but make use of several options, combining short-term coping strategies with longer-term structural adjustments.
From the set of studies analyzed, no clear differences between the types of human responses considered in the different types of methods can be found (see
Table 5). However, the methods do differ in the way human responses are considered. Studies using a systems dynamics approach focus, logically, at the system level, and do not consider human behavior explicitly, but studies using agent-based models do. The questions that studies, using system dynamics or agent-based models, seek to answer differ. Whereas agent-based models focus on understanding the various factors, including behavior, influencing decisions of actors, system dynamics models focus more frequently on understanding how the (water) system would be impacted because of certain responses. Although agent-based models could also be used to assess the wider system level impacts, and system dynamics models could also integrate knowledge on human behavior in system-level models, this may not often be done.
3.3. Theories and Approaches to Conceptualizing Human Responses to Changes in Water Availability
A variety of approaches were used to conceptualize human responses, which are summarized in
Table 6. We found that 24 of the studies analyzed used a deductive approach, which Locke [
42] defined as “moving from the general to the specific.” This implies that the research starts with a hypothesis based on existing theories or conceptual frameworks. Seventeen of the 24 studies used a distinctive underlying theory, of which almost half of the studies strictly followed the chosen theory throughout the study [
32]. The rest of the studies loosely referred to a theory, using it mainly as inspiration for their conceptual framework or model [
43]. In the analyzed studies, we identify three different types of theories for the conceptualization of human responses, including (1) economically-based utility maximization theories, (2) behavioral theories from the field of sociology, psychology and migration studies, and (3) theories based on vulnerability and resilience concepts.
The most frequently applied (
n = 8) theory is utility maximization, which originates from welfare economics, assuming that people make rational decisions that optimize their welfare. This theory accounts for the comparison of different individual responses. For example, Berhanu et al. [
32] developed a model for the analysis of pastoralist responses to long-term climatic variability based on pastoral household utility maximization, whereas the system dynamics model of Gohari et al. [
25] used income-maximization as the main determinant for agricultural land use decisions.
Individual behavioral theories developed in sociology or psychology were used in five of the analyzed studies. This ranges from commonly used theories, such as the theory of planned behavior, to theories predominantly applied to the field of migration. We found that only Hailegiorgis et al. [
44], Krömker et al. [
43], and Yazdanpanah et al. [
45] used commonly known behavioral theories in their studies, respectively, the protection motivation theory and the theory of planned behavior. The protection motivation theory states that peoples’ decisions to protect themselves is based on the (1) perceived severity of a threatening event, (2) the perceived probability of the occurrence, (3) the expectancy that the recommended behavior is effective and (4) the ability of the person to execute the recommended courses of action successfully [
46]. Hailegiorgis et al. [
44] used this theory as part of a framework that describes the socio-cognitive adaptive behavior of households, which helps to explain the subjective adaptive capacity of individuals to climate change. Krömker et al. [
43] indirectly used the theory by applying a psychological action model which is based on the protection motivation theory. The theory of planned behavior started as the theory of reasoned action to predict an individual’s intention to engage in a behavior, intending to explain the majority of individuals’ behavior [
47]. It links beliefs to behavior based on the (1) attitude of an individual towards a behavior, (2) the subjective norm based on the individual’s social network and other social norms, and (3) the perceived behavioral control, including capacity and autonomy in the choice to adapt. Yazdanpanah et al. (2014) built on the theory of planned behavior and expanded this with moral norms to analyze water conservation adaptation responses. Both these theories allow for the evaluation of a range of possible human responses.
The migration-focused push-pull theories and dread-threat theory use similar behavioral insights, such as the individual’s motivation to change and expectancy of the behavior, however, they are mainly developed to explain migration. Lee’s [
48] push-pull theory conceptualized the motivation of an individual to migrate by factors associated with the area of origin, factors associated with the area of destination, intervening obstacles and personal factors. The push-pull theory is still one of the most used theories to explain humans’ decision to migrate. Hassani-Mahmooei and Parris [
49] used this migration decision theory to model the migration dynamics of inhabitants of Bangladesh. Dread-threat theory originates from psychology and is related to the Intergovernmental Panel on Climate Change’s risk framework. It explores how the local perception of threat affects the decision-making about remaining in place, migrating or both. The perceived uncertainty of the threat is combined with the perceived impact of the threat; for example, an unpleasant death for future generations at risk. Collman et al. [
50] further developed the theory based on anthropological and journalistic accounts and translated this qualitative theory to a partly quantitative model to evaluate the attractiveness of threat response strategies. The two theories from the field of migration focus on the decision whether to migrate. Alternative human responses are not considered.
Four other papers used resilience or vulnerability theory as part of a socio-ecological framework to describe human responses. Unlike the utility maximization and individual behavioral theories, these are not well-defined theories to simulate responses. Typically, these papers relate components related to resilience, such as exposure, vulnerability, and adaptive capacity of individuals to human responses. For example, Martin et al. [
51] developed a framework to provide insights into the vulnerability of pastoral households in relation to droughts, which was based on, amongst others, vulnerability and livelihood security concepts, whereas Kansiime [
52] employed a framework that combined social and ecological approaches to analyze resilient behavior. In addition, Bradley and Grainger [
23] assessed resilient behavior to develop a social resilience model to describe the switch from performance strategies to survival strategies. Khanian et al. [
40] developed a conceptual model based on the relation between mechanisms of place attachment and adaptive capacity on migration decisions. Most of these papers do not use a single theory but combine theories with information from literature or their own insights.
The remaining seven of the 24 deductive studies did not explicitly apply a human response theory but used either qualitative or quantitative information from previous studies as a starting point for a hypothesis of human responses. The information could be part of an existing model, as demonstrated by Boone et al. [
29], who used the household model DECUMA to simulate decision-making or based on literature as shown by Desta and Coppock [
33] who used historic behavior of the Maasai to explore behavior of the Borana society.
As an alternative to a purely deductive approach, participatory methods were used to develop a conceptual framework or model through inductive methods, in which rules and cause–effect relations are derived from observations. We found seven studies that used participatory methods. A method was defined as participatory when there was active communication with a stakeholder group, ranging from focus group discussions and participatory modeling to virtual experiments. The execution of only a survey is not considered a participatory approach. For example, Bommel et al. [
53] demonstrated an approach in which an agent-based model was designed together with livestock farmers through several participatory workshops. In addition, Dieguez et al. [
54] used local knowledge as the basis for their agent-based model using a series of workshops to define decision-making strategies, which became a decisional sub-model in the Decision-Support System SequiaBasalto. Most participatory (design) methods use local data and information to develop a model that simulates decision-making in the same local or regional context. This relates to an indicative approach in which generic lessons can be drawn based on local data without using an existing theory.
Aside from studies that used a human response theory or inductive participatory approach, eights studies were categorized as not using a specifically defined human response approach or theory. These are primarily econometric or statistical studies using results of questionnaires or census data to derive relations between human responses and other variables. These studies do not specify a human response hypothesis based on theories or literature but carry out an analysis to answer the main research questions based on available (census) datasets or questionnaire results, sometimes implicitly informed by conceptual models and cause–effect relations from previous literature. For example, Ashraf [
28] evaluated farmers’ coping and adaptation behavior towards drought through a survey, without mentioning a specific hypothesis. In addition, Bai et al. [
26] used questionnaires, consisting of household census and adaptation measures, to define the most adopted adaptation measures and to relate adaptation to technical efficiency of livestock production. Esquivel-Hernández et al. [
41] conducted a spatial distribution and temporal analysis of water conflicts, making the link between conflicts and hydro-meteorological events. These studies can be considered as inductive, as these studies were not designed to test an existing theory, but to find correlations and patterns in behavior.