Climate, desertification, and local human development: evidence from 1564 regions around the world

Abstract

How global warming affects human development is a central question for economists as well as social scientists. While most of the literature has focused on the impact of weather on income, less is known on the relationship between climate and local human development. This paper considers shocks in precipitation, temperature, and an original measure of soil aridity to first exploit the association between climate warming and human development, and second, on its dimensions. We show that while precipitations do not have a significant long-term impact on human development growth, variations in temperature and potential evapotranspiration negatively affect two of the three determinants of the Human Development Index, namely life expectancy at birth and education. These results suggest that other climate indicators, such as the potential evapo-transpiration of the soil, should be considered in addition to the standard indicators, when evaluating the localized economic effects of climate change.

1 Introduction

Understanding the effects of human-induced climate changes on economic and social development is critical for achieving policy interventions to counterbalance economic losses of less developed countries, which will pay the highest price of a warmer and drier climate. However, climate does affect economic development in many ways. The economic literature has primarily focused on income and social disparities across countries. On the other hand, spatial disparities within countries have been given considerably less attention.

To this regard, a growing body of literature in the climate change discussion now focuses on the uneven economic development within countries. For instance, Diffenbaugh and Burke (2019) have investigated whether global warming has affected the recent evolution of income inequality between countries finding that anthropogenic climate changes have increased economic inequality between countries.

Nevertheless, climate warming might affect human development through different dimensions, often neglected in the economic literature. For instance, rising temperatures have been associated with higher lower infant health and higher infant mortality in more vulnerable areas of the world (Banerjee and Maharaj 2020; Geruso and Spears 2018). Moreover, weather shocks are negatively associated with children’s education (Pellerin 2000) mainly through an increase in child labor (Colmer 2021; Myers and Theytaz-Bergman 2017; Boutin 2014). Finally, global warming has also been associated with lowering income (Burke et al. 2015) in particular for developing countries relying on agriculture (Zhang et al. 2017). In the economic geography literature, Dall’Erba and Domínguez (2016) and Dall’Erba et al. (2021) have focused on the impact of climate change on the productivity and trade of US agriculture, while Conte et al. (2021), Castells-Quintana et al. (2021) and Bosetti et al. (2021) focus on the distribution of population across urban-rural areas, even in terms of local migration patterns. Finally, Pan et al. (2020) focus on the future joint impacts of climate and land use changes on the Yangtze river economic belt, in China.

In this paper, we first assess the relationship between climate variability and a measure of human development within a country and then explore the determinants of such a relationship. To investigate whether global warming has affected human development, we combine spatially disaggregated historical climatic variables such as precipitation, temperature, and potential evapotranspiration with subnational Human Development Indexes (henceforth, HDI).

Our research contributes to the literature in three main directions. The first and most important contribution of the paper consists of an investigation of the relationship between climate shocks and local human development rather than standard economic variables.

Recent economic literature finds remarkable differences between human development and income growth (Rodionov et al. 2018; Ranis et al. 2006). Recently a number of measures of social development which go beyond the mere economic indicators have been developed.¹ However, to the best of our knowledge, this is the first paper assessing the relationship between climate and human development, which besides the income dimension also depends upon education, and life expectancy, locally.

The second contribution is that we go one step further in investigating the effects of climate on human development, by assessing the relationship between each of the three dimensions composing the HDI. Therefore, we want to see to which extent local income, education, and life expectancy are correlated with weather shocks. To understand this, we estimate the relationship between our climate variables and each of the three dimensions that compose the HDI, namely life expectancy at birth, gross national income, and completed years of education, by taking as units of observation subnational administrative units for a period between 1990 and 2015.²

The third contribution relates to the climate indicator we employ. Most of the economic effects of global warming literature have focused on precipitation and temperature shocks. Here we instead use a soil-water availability index, the potential evapotranspiration (PET), that considers the combined effects of precipitation, humidity, temperature, and UV radiations, among other climate variables. The adoption of the PET allows us to weigh the impact of rainfall on the growing cycle of a plant the extent to which water can be retained by the soil (Harari and Ferrara 2018). Furthermore, we show that the nonlinear interaction of these two variables (i.e., the PET) plays a significant role in explaining long-run variations of the HDI. On the other hand, no effect is found in the short term.

We assemble a panel dataset covering about 1564 administrative units in 135 countries from 1990 to 2015. By combining subnational human development data from Jahan et al. (2017) with an original measure of soil aridification, which is obtained interacting precipitation, humidity, and temperature data from the gridded Climatic Research Unit (CRU) Time-series (TS). First, we estimate the relationship between annual and long-term variations in climate variables (i.e., precipitation, temperature, and PET) and human development using OLS. Our benchmark specification is conditional on subnational region fixed effects and country-specific time trends so that we identify changes in the Human Development Index relative to a region’s historical mean.

We find that: first, there is a significant local-level relationship between annual precipitation and temperature shocks, and human development. According to our most conservative specification, a one standard deviation shock in precipitation and temperature during year t is associated with an 11 and 8% point increase in HDI, respectively. Second, the relationship appears to be concave (i.e., considerable higher amounts of precipitation and higher temperatures are negatively associated with the local human development). Third, when evaluating the 25-year relationship between climate variations and local human development, rainfalls and temperature do not explain development significantly, while soil aridification affects the HDI.

Our estimates indicate that one standard deviation increase in PET over the period 1990–2015 is associated with a decrease of 27.7% points in the Human Development Index.

To assess the principal mechanisms underlying the relationship between climate and local human development, we further extend the analysis to investigate the relationship between climate and each of the determinants of the Human Development Index. In particular, we document a strong and positive relationship between standard climate variables (i.e., precipitation and temperature) and life expectancy and income at the local level, thus confirming the conclusions drawn in the relevant economic literature (Dell et al. 2012; Burke et al. 2015). However, our estimates show the opposite effects of potential evapotranspiration.

A potential concern is the possible endogeneity of the climate variables. Indeed regions that have experienced higher temperatures and increasing soil evapotranspiration might also be the poorest in the world. If this is the case, our baseline estimates would be biased. We address this potential issue by evaluating the heterogeneous impact of aridity per income class according to the classification proposed by the World Bank.³ The World Bank income classification allows us to estimate the relationship between climate and HDI among areas within a particular income class. This alleviates concerns about the endogeneity of the regressors. We find that temperature and PET increases significantly impact low-income regions. In contrast, no significant effect is found among high-income regions of the world. Once again, these results corroborate those obtained in the baseline procedure.

The remainder of the paper is organized as follows: in Sect. 2, we summarize the differences between precipitations and soil aridification, while in Sect. 3, we present the analytical framework to estimate the annual relationship between rainfall, PET, and human development. In Sect. 4, we present the data used and describe each of the dimensions composing the Human Development Index. Section 5 discusses the results of the short and long-run relationships between climate variables PET and HDI. In Sect. 6, we address which of the three dimensions composing the HDI is the most affected by the process of desertification. Finally, Sect. 7 concludes.

2 Precipitation and soil aridification

The economic literature on the climate impact on socioeconomic development has focused on precipitation and temperature shocks in Africa and South-East Asia as primary climate variables. Maccini and Yang (2009) and Kudamatsu et al. (2012) show that individuals who, during their childhood, were exposed to greater amounts of precipitation and shorter drought periods had positive long-term impacts in terms of health and living standards compared to their peers. Jayachandran (2006) is yet another example of how rainfall shocks explain variations in wages and migration waves in agriculture-intensive rural India. Although the studies mentioned above are only the most relevant and recent papers dealing with the relationship between rainfalls and socioeconomic well-being, a vast literature has confirmed the positivity and significance of such effect. Interestingly enough, more recently, the effect of precipitation and temperature on human development has been found to be nonlinear, prompting the existence of a tipping point where the sign of the relationship becomes negative.⁴ Nevertheless, these studies do account for other, more “important” factors such as education, the initial level of development, and public investment when modeling socioeconomic development.

According to current projections on weather changes in the upcoming decades, average rainfall at the global level is foreseen to rise (Cherlet et al. 2018). In particular, the increase in precipitation will be more pronounced at high and mid-latitudes and in tropical regions (Kirtman et al. 2013; Donat et al. 2016).

Given the results obtained in the existing empirical literature, one may be tempted to infer a positive effect of climate change. However, three elements need to be taken into account. First, the nonlinearity of the relationships between climate variables and economic development implies that the size of the forecasted climate change will define the sign of the impact at the local level. Second, although the climate is defined by geophysical interactions on a global scale, the socioeconomic impacts need to be analyzed at a local scale. Third precipitations alone do not capture actual soil water availability, depending on concurring factors such as land quality, solar radiations, temperature, air humidity, and wind speed.

As a result, the effective water availability of these areas is not certain, and a precise measure of land dryness is needed. We use a measure of Potential Evapo-Transpiration (PET), which considers the combination of two sources of soil water loss: soil surface evaporation⁵ and reference crop transpiration (Allen et al. 1998).⁶ Following Rind et al. (1990) and Cherlet et al. (2018), we assume PET as an indication of local aridity.

Fig. 1

Global distributions of changes (mm/month) in (a) Precipitation, and (b) PET, taken as the difference between the present day (1990–2015) and the historical average (1900–1980). In Panel A, lighter cells identify areas with average precipitations in present day (1990–2015) higher than historical average (1900–1980). In Panel B, darker cells identify areas with average PET in present day (1990–2015) higher than historical average (1900–1980). Secular data for precipitation and PET are retrieved from Fu et al. (2016)

To better visualize this crucial aspect, Fig. 1 shows that for the period 1990–2015, most areas of the world have experienced increased precipitation compared with their historical mean (Panel A). Nevertheless, zones that have experienced increases in precipitation have also been accompanied by higher evaporation levels (Panel B).⁷ In particular, during the historical period (1900–1980), the global average monthly precipitation was 1.82 mm, while the present-day average (1990–2015) is 1.85 mm, with an increase of 0.03 mm per month, or 1.65%. Conversely, during the historical period (1900–1980), the global average monthly PET was 1.83 mm, while the present-day average (1990–2015) is 1.88 mm, with an increase of 0.05 mm per month, or 2.73% (Fu et al. 2016). Along these lines, Malpede and Percoco (2021) examine the effects of human-induced desertification on economic growth and find over the period 1990–2015, aridification reduced the GDPs of African and Asian countries by 12% and 2.7%, respectively.

Fig. 2

Correlation between climate variables and subnational Human Development Index. Panel a shows the correlation between annual variations in precipitations (left) and PET (right) expressed in mm/year and annual variations in HDI. Panel b shows the correlation between long-term (1990–2015) variations in precipitations (left) and PET (right) and HDI

Moreover, Panel A of Fig. 2 shows the correlation between annual variations in precipitations (left) and PET (right) expressed in mm/year and annual variations in Subnational HDI. Panel B of Fig. 2 instead shows the correlation between long-term (1990–2015) variations in precipitations (left) and PET (right) and HDI. It emerges a clear negative relationship between the PET and the SHDI, whereas the relationship is less clear for precipitation.

In this paper, we aim to investigate the relationship between climate and local development. To the best of our knowledge, our research is the first to assess such (nonlinear) relationship at the global level with local data and to extend the analysis to consider the impact of all dimensions of HDI: education, life expectancy, and income (Fig. 3).

Fig. 3

Long-term change in the subnational Human Developemnt Index (1990–2015)

3 Methodology

In this section, we present the analytical framework for discussing the relationship between climate change and human development. Our methodology consists in estimating a short-run and a long-run relationship between climate and HDI with its components. The final dataset is composed of annual observations covering about 1564 administrative units in 135 countries starting in 1990 until 2015.

The estimation of the effects of climate variations on the human development in the short run follows the same logic of Diffenbaugh and Burke (2019) and Burke et al. (2018, 2015):

$$\begin{aligned} \begin{aligned} \Delta Y_{i,c,t+1-t}&=\alpha \;Y_{i,c,t}+\beta _{1}\;P_{i,c,t}+\beta _{2}\;P_{i,c,t}^{2}+\beta _3\;PET_{i,c,t}+\delta _{1}\;T_{i,c,t}+\delta _{2}\;T_{i,c,t}^{2}+\\&\quad +\sigma _{t}+\omega _{i}+\rho _{c}\;\tau +\epsilon _{i,c,t} \end{aligned} \end{aligned}$$

(1)

We denote with $\Delta Y_{i,c,t+1-t}$ the change in the Human Development Index (HDI) of subnational region i, of country c, from year t to year $t+1$. $P_{i,c,t}$ indicates the average annual amount of precipitation of region i in country c at year t and is expressed in millimeters. To account for nonlinear relationship between precipitation and HDI, we include $P_{i,c,t}^{2}$ which indicates the average annual precipitation in millimeters of grid i in country c at time t squared. The variable $PET_{i,c,t}$ instead, indicates the annual potential evapo-transpiration of grid i in country c at year t and is expressed in millimeters. In addition, we also control for average annual mean surface temperature $T_{i,c,t}$ and its squared value denoted as $T_{i,c,t}^{2}$.

The model considers year fixed effects, denoted with $\sigma _{t}$, region fixed effects denoted with $\omega _{i}$, and country linear trends in order to account for country specific trends over time; this is denoted with $\rho _{c}\;\tau$. Equation (1) is estimated by using OLS. Moreover, to account for the possible spatial autocorrelation of climate variables, we use the spatial autocorrection method proposed by Conley (1999). To this regard, it is important to note that a growing number of climate economic studies use other forms of spatial dependence, such as models with the spatial lag of the regressors and weight matrices based on actual network flows. Examples are Dall’Erba et al. (2021), where the weight matrix (W) is based on trade flows, and Bae and Dall’erba (2021), where W relies on up- and downstream surface water flows. The major difference between using this approach and the spatial error autocorrelation (such as the Conley approach or any model with W*e) is the assumption that covariates are themselves leading to spillover effects. Given the Panels A and B of Fig. 1, temperature and precipitation themselves may be spatially autocorrelated.

In addition to the annual relationship, we estimate the long-term relationship between climate and human development. This models implies a cross-sectional equation and is detailed below:

$$\begin{aligned} \begin{aligned} \Delta {Y}_{i,c}&=\alpha +\beta _{1}\;{\overline{P}}_{i,c}+\beta _{2}\;{\overline{P}}_{i,c}^{2}+\beta 3\;{\overline{PET}}_{i,c}+\delta _{1}\;{\overline{T}}_{i,c}+\delta _{2}\;{\overline{T}}_{i,c}^{2}+\mu _{c}+\epsilon _{i,c} \end{aligned} \end{aligned}$$

(2)

Here we denote with $\Delta {Y}_{i,c}$ is the long-term change in the Human Development Index (HDI) of subnational region i, of country c, from year 1990 to year 2015. ${\overline{P}}_{i,c}$, ${\overline{PET}}_{i,c}$, and ${\overline{T}}_{i,c}$ indicate the 25-year average HDI, precipitation, PET and temperature at region i of country c from 1990 to 2015. As in Eq. 1, we also account for nonlinear long-term relationship between average precipitation and HDI and we include ${\overline{P}}_{i,c}^{2}$ indicating the squared average amount of precipitation over the period 1990–2015. Similarly, we include the squared average local temperature denoted as ${\overline{T}}_{i,c}^{2}$. These variables capture long-run changes in climate, attenuating year-to-year fluctuations and only isolating a 26-year trend. Finally, the model considers country fixed effects denoted with $\mu _{c}$. We estimate Eq. 2 via OLS.

Variable $\Delta {Y}_{i,c}$, capturing the 25-year change in the HDI, is constructed as follows:

$$\begin{aligned} \Delta Y_{i,c}={\overline{Y}}_{i,c,2015}-{\overline{Y}}_{i,c,1990} \end{aligned}$$

(3)

where the average term ${\overline{Y}}_{i,c,t}$ is defined as follows:

$${\overline{Y}}_{i,c,t}=\frac{1}{5}\sum _{K=1}^{5}Y_{i,c,t+1-k}$$

These are five-year averages of annual average HDI. In other words, the variable ${\overline{Y}}_{i,c,2015}$ is the average of the HDI during the period 2010–2015. Similarly, ${\overline{Y}}_{i,c,1990}$ is the average of the HDI during the period 1990–1995. Consequently, $\Delta Y_{i,c}$ captures long-run changes in the HDI, attenuating year-to-year fluctuations and only isolating a 25-year trend.

4 Data description and summary statistics

4.1 Climate data

Precipitation, potential evapotranspiration, and temperature data are provided by the gridded Climatic Research Unit (CRU) Time-series (TS) version 4.00. Original climate data are expressed at a monthly level and extent for the period 1901–2015. The data are provided on high-resolution (0.5 $^{\circ }\times 0.5^{\circ }$) grids. Precipitation and PET⁸ are expressed in millimeters (mm/month), while the surface temperature is expressed in ${^\circ }$C. Total precipitation amounts range from a minimum of zero millimeters per month to a maximum of 91 mm per month.

Table 1 Summary statistics, subnational human development and climate variables, panel data sample

Potential evapotranspiration ranges from a minimum of zero millimeters per month to a maximum of 24.5 mm per month. Finally, the annual mean surface temperature ranges from a minimum of – 20 $^\circ$C to a maximum of 37.7 $^\circ$C. A total of 36,172 observations were collected for the years 1990 to 2015. Table 1 reports summary statistics for weather variables and human development dimensions.

4.2 The local HDI and its determinants

The concept of human development was first introduced in 1990 with the first Human Development Report (Haq 1990). The latter introduced a new approach for expanding the richness of human life, rather than simply the richness of the economy in which human beings live. It is, therefore, an approach focused on people and their opportunities and choice (Haq 1990).

The Human Development Index (HDI) was constructed to have a numerical representation of human development. The HDI is a summary measure of average achievement in three critical dimensions of human development (rather than a single one as in the case of the GDP): (1) the health dimension, (2) the education dimension, and (3) the living standard dimension. The HDI is the geometric mean of normalized indices for each of the three dimensions. Each dimension composing the HDI is assessed by one or more specif variables. Specifically, the health dimension is assessed by life expectancy at birth; the education dimension is measured by the mean of years of schooling for adults aged 25 years and more and expected years of schooling for children of school entering age. Finally, the income dimension is measured by the logarithm of the gross national income per capita.⁹

Similarly to its original definition, the HDI is an average of the subnational values of the three dimensions outlined above. In its official version, these dimensions are measured with the following indicators: Education measured with the variables “Mean years of schooling of adults aged 25 and above” and “Expected years of schooling of children aged 6”; health measured with “Life expectancy at birth” and standard of living measured with “Gross National Income per capita (PPP, 2011 US$).” Our database comprises 1564 subnational administrative units for which human development data are available annually from 1990 to 2015. Details about the estimation procedure can be found in the Human Development Report (Jahan et al. 2017).

4.2.1 Life expectancy index (LIFEX)

The health dimension of the HDI is the life expectancy index at birth (LIFEX). This index is defined as the total number of years newborn children would live if subject to the mortality risks prevailing for the cross section of the population at the time of their birth. The LIFEX ranges from a minimum of 20 to a maximum of 85. The sources are Eurostat, GDL-AD, and UNDP. Additional sources were used to compute the LIFEX (those can be found in Appendix).

For high-income countries (HICs) and some middle-income countries (MICs), subnational values of LIFEX were based on data derived from national statistical offices and Eurostat. For most of the low- and middle-income countries (LMICs), data were derived from the GDL-AD.

Fig. 4

Long-term change in the life expectancy index (1990–2015)

Figure 4 shows the long-term change in the LIFEX at the subnational level over the period 1990–2015. Green areas have experienced an increase in life expectancy, with areas in North Africa and Eastern Africa, Argentina, China, and Turkey increasing their life expectancy up to 20 years during the 25 years of the analysis. This is the case of developing economies that have experienced large socioeconomic growth from 1990 (i.e., China, Turkey) or countries that before the 1990s were greatly affected by civil wars (North and Eastern Africa), and in light green areas where the increase in life expectancy is minimal. This is especially true for advanced economies with already higher life expectancy (i.e., a woman is expected to live more than 70 years). Finally, in red are the areas that experienced a large reduction in life expectancy between 1990 and 2015. This is especially the case of Lesotho, South Africa, and Mali, which have experienced civil wars causing large numbers of casualties.

4.2.2 Education index

The education dimension is computed using two different measures of schooling years. The first measure is the mean years of schooling of adults aged 25 and more (henceforth, MYS). It ranges from a minimum of zero years of education to a maximum of 15 years. The sources are Eurostat, the Global Data Lab,¹⁰ and UNDP. Additional sources were used to compute the LIFEX (those can be found in Appendix). The second measure is the expected education years of schooling (EYS) which consists of the number of years of schooling a child of school entrance age can expect to receive if prevailing patterns of age-specific enrollment rates persist throughout the child’s schooling life. It ranges from a minimum of zero years of education to a maximum of 18 years. The sources are Eurostat, GDL-AD, UNDP. Additional sources used to compute the education index can be found in Appendix. Lacking for HICs outside EU.

The Global Data Lab computes the average years of schooling for each subnational administrative unit by taking for each region the average number of years of education completed by adults aged 25 and over in the survey and census datasets.

To assess whether climate variations are negatively associated with individuals’ health, we use data on life expectancy at birth. These data are recorded at the administrative level for each country and are available starting from 1990. To assess whether climate variations are negatively associated with individuals’ level of education, we use data on the number of completed years of schooling for adults aged 25 and more. It ranges from a minimum of zero years of education to a maximum of 15 years. These data are recorded at the administrative level for each country and are available starting from 1990 (Smits and Permanyer 2019). To assess whether climate variations are negatively associated with the average individual income for each region, we use data on the natural logarithm of the gross national income per capita (LGNIc) also derived from Smits and Permanyer (2019).

Fig. 5

Long-term change in completed years of education (1990–2015)

Figure 5 shows the long-term change in the completed years of education at the subnational level over the period 1990–2015. Green areas have experienced large increases in education. Examples are North Africa and Middle-East, China, South-East Asia, and Turkey, in which individuals increased, on average, their completed education by up to 1 year, and in light green areas where the increase in education rate was minimal. Once again, this is especially the case of advanced economies with already high levels of years of completed education (i.e., an individual is expected to study more than eight years).

4.2.3 Standard of living

The standard of living dimension of the HDI is represented by the natural logarithm of the gross national income per capita (LGNIc). Smits and Permanyer (2019) define it as the (log of the) sum of the value added by all resident producers in a given administrative unit. LGNIc is based on purchasing power parity (PPP) and is expressed in 2011 USD. It ranges from 100 USD to a maximum of 75,000 USD. The sources are Eurostat, GDL-AD, UNDP. Additional sources used to compute the LGNIc can be found in Appendix. For high-income countries (HICs) and some middle-income countries (MICs), subnational values of LGNIc were based on data derived from national statistical offices and Eurostat. However, for most low- and middle-income countries (LMIc), data on the standard of living were not available. In such cases, Smits and Permanyer (2019) derived it from the GDL-AD. A detailed description of the computation of the LGINc is found in the Appendix.

Fig. 6

Long-term change in the gross national income, GNI (1990–2015)

As for the other two dimensions of the HDI, Fig. 6 shows the long-term change in the GNI at the subnational level over the period 1990–2015. Green areas have experienced an increase in gross national income, with areas in China and South-East Asia increasing their GNI up to 300% points during the 25 years of the analysis, and in light green areas with lower increases in the GNI. Once again, this is the case of advanced economies with already higher GNI (i.e., North America, Europe, and Australia). Finally, in red are the areas that experienced large reductions in GNI between 1990 and 2015. This is especially the case of sub-Saharan African countries and Libya. Those countries have experienced large political instability, which adversely affected economic development.

4.2.4 Construction of the HDI

Having presented the three dimensions used to compute the HDI, we now describe the simple formula that defines the HDI.

The first step is the computation of each of the three-dimensional indices for the administrative unit i at year t. To do that the following formula is used:

$$\begin{aligned} \text {Dimension Index}_{i,t}=\frac{AV_{i,t}-\text {Min. Value}_{t}}{\text {Max. Value}_{t}-\text {Min. Value}_{t}} \end{aligned}$$

(4)

The minimum and maximum values are the so-called goalposts, which are used to ensure that the dimension indices’ values remain between 0 and 1 (see Table 1). For life expectancy at birth, the UNDP goalposts are 24 and 85, and for the standard of living, they are 100 and 75,000. For expected years of schooling, they are 0 and 18, and for mean years of schooling 0 and 15. To obtain the dimension index for education, the geometric mean of the separate indices for expected years of schooling and mean years of schooling is taken. To compute the HDI based on the three-dimensional indices, the geometric mean of the three indices is taken:¹¹

$$\begin{aligned} \text {SHDI}_{i,t}=\left( \text {Education}_{i,t}\cdot \text {Health}_{i,t}\cdot \text {Income}_{i,t}\right) ^{\frac{1}{3}} \end{aligned}$$

(5)

5 Baseline results

Table 2 Annual climate-HDI relationship

Results of the relationship between annual variations of climate variables and the local Human Development Index are reported in Table 2. The regressors of interest are Prec, defined as the average level of precipitation during the year t, Temp, defined as the average surface temperature in year t; and PET, defined as the average level of potential evapotranspiration during year t.

Higher values of the variable Prec correspond to higher rainfall in year t. Higher values of the variable Temp correspond to higher average temperatures in year t. Conversely, higher values of the variable PET correspond to lower “effective” water availability in year t. We also check for a nonlinear relationship between precipitation and human development by including the variable $\text {Prec}^{2}$, defined as the quadratic term of the average level of precipitation during the year t. Similarly, we include the variable $\text {Temp}^{2}$, defined as the quadratic term of the average surface temperature in year t. Equation 1 also includes region and time fixed effects and country-specific linear trends.

Column (1) of Table 2 considers the whole sample of subnational regions for the whole world and shows the contemporaneous relationship between precipitation and temperature and local human development. It indicates the positive relationship between annual variations in precipitation levels and average temperature, and human development. In particular, one standard deviation increase in precipitation during year t is associated with an 8% point increase in the Human Development Index in the same year.

Similarly, one standard deviation increase in average temperature during year t is associated with an increase of 8% points in the Human Development Index in the same year. Moreover, the relationship appears to be concave (i.e., considerable higher amounts of precipitation and higher temperatures are negatively associated with the local human development). These results do not statistically change when we include year-fixed effects (column 2) and country-specific time trends (column 3). In column 4, we include the contemporaneous effects of PET. This inclusion does not appear to affect the relationship between precipitation and temperature and local human development. However, the sign is negative, although non-significant for the PET. This is no surprise since annual variations in PET are not common as precipitations. Instead, the increasing PET is a process occurring over decades. For this reason, we expect a negative relationship between PET and HDI when estimating Eq. 2, which is reported in the next section.

Results of the relationship between long-term variations of climate variables and the local Human Development Index are reported in Table 3. This time, the regressors of interest are ${\overline{Pre}}$, indicating the average amount of precipitation of region i, in country c, over 25 years expressed in millimeters, ${\overline{Temp}}$, indicating the average temperature of region i in country c over the 25-year period expressed in degree Celsius, and ${\overline{PET}}$, indicating the average annual amount of PET of region i, in country c, over the 25 years expressed in millimeters.

Higher values of the variable ${\overline{Pre}}$ correspond to higher average rainfall over the period 1990–2015. Higher values of the variable ${\overline{Temp}}$ correspond to higher average temperatures over the period 1990–2015. Conversely, higher values of the variable ${\overline{PET}}$ correspond to lower “effective” water availability over the period 1990–2015. As in Eq. 1, we also check for a nonlinear relationship between precipitation and human development by including the variable ${\overline{Pre}}^2$, defined as the quadratic term of the average level of precipitation over the period 1990–2015. Similarly, we include the variable ${\overline{Temp}}$, defined as the quadratic term of the average surface temperature over the period 1990–2015. Equation 2 also includes region and time fixed effects and country-specific linear trends.

Column (1) of Table 3 considers the whole sample of subnational regions for the whole world and shows the 25-year relationship between precipitation and temperature and local human development. Unlike the results obtained for the short-run analysis, the estimates of the relationship between long-term changes in precipitation and human development lose significance. These results are qualitatively robust to the inclusion of year-fixed effects (column 2) and country-specific time trends (column 3). Column (4) shows a negative and significant relationship between long-term variations in PET and local human development. In particular, one standard deviation increase in PET over the period 1990–2015 is associated with a decrease of 17.9% points in the Human Development Index.

Taken together, these results suggest that while precipitations are an essential determinant of human development in the short run, when considering long-run climate variations, the effect of climate on HDI is likely to work through soil aridification.

We also implement a second model to assess the long-term relationship between climate and human development as a robustness check of the estimates. In this model, we construct the change in the Human Development Index from 1990 to 2015 for each region i of country c, and estimate the relationships between long-term variations in climate variables and variations of the HDI. Results of this further check are reported in Table 7 in the Appendix.

Table 3 Climate and the HDI per income classes

5.1 Spatial heterogeneity

Further, we assess the heterogeneous impact of weather shocks on human development on different income classes. The estimation of the relationship between weather and human development among areas within a particular income class also reduces possible concerns on the endogeneity of the treatment.

We group countries following the income classification proposed by the World Bank. In its classification, the World Bank defines countries with an initial GDP per capita of 4000 USD or below as low-income countries. Countries with an initial GDP per capita between 4000 USD and 8000 USD compose the Lower-middle-income group. The third group of upper-middle-income countries is composed of those countries with a GDP per capita between 8000 USD and 12,000 USD. Finally, high-income countries with a GDP per capita of 12,000 USD or above compose the fourth group. We consider the GDP per capita in 1990 for the classification of the countries within each group.¹²

Table 4 Long-term climate-HDI relationship

Results of this robustness check are shown in Table 3, which indicates that soil aridification significantly impacts the human development of low- and lower-middle-income areas. In contrast, no significant effect is found among high-income regions. This last result is remarkable and needs to be addressed in more detail. One possible explanation is that it might result from the use of irrigated water or other agricultural innovations. However, our empirical model does not allow us to assess the mechanisms behind this relationship (Table 4).

6 Evidence on health, education, and income

Having established the significant adverse effects of soil aridification on within-country human development, we proceed to investigate which of the three dimensions composing the HDI constitutes the main driver. Several studies provide valuable insights on the effects of climate variation on income (Zhang et al. 2017; Burke et al. 2015) and individual health, focusing primarily on infant mortality (Banerjee and Maharaj 2020) and education attainment (Colmer 2021).

We estimate Eqs. 1 and 2 where $\Delta Y$ this time denotes the annual and 25-year change in the Human Development Index for region i of country c, respectively.

Table 5 Annual relationship between climate and HDI determinants

Table 6 Long-term relationship between climate and HDI determinants

Tables 5 and 6 report results of the contemporaneous and long-terms relationship between climate and local life expectancy, completed years of education and gross national income. Column (1) of Table 5 considers the whole sample of subnational regions for the whole world and includes year fixed effects and country-specific time trends. It shows the contemporaneous relationship between precipitation and temperature and life expectancy expressed in years. A strong and positive relationship emerged between the annual variations in precipitation and expected years of life. The same conclusions are shown for temperature.

In particular, one standard deviation increase in precipitation during year t is associated with an increase of 0.4 years in life expectancy in the same year. Similarly, one standard deviation increase in average temperature during year t is associated with an increase of 0.3 years in life expectancy in the same year. The concavity of the relationship is confirmed (i.e., considerable higher amounts of precipitation and higher temperatures are negatively associated with the life expectancy at the local level).

These results do not statistically change when we also include the variable PET (column 2). Column (3) of Table 5 shows the contemporaneous relationship between precipitation and temperature and completed years of education expressed in years. This time, it appears that short-term climate variations do not affect education. These results do not statistically change when we also include the variable PET (column 4). Column (5) of Table 5 shows the contemporaneous relationship between precipitation and temperature and local income expressed as the natural logarithm of the gross national income.

Once again, a strong and positive relationship emerges between the annual variations in precipitation and local income level. On the other hand, annual temperature variations do not seem to be related to variations in the GNI. In particular, one standard deviation increase in precipitation during year t is associated with an increase of 0.4 years in life expectancy in the same year. Similarly, one standard deviation increase in average temperature during year t is associated with an increase of 0.3 years in life expectancy in the same year. The concavity of the relationship is confirmed (i.e., considerable higher amounts of precipitation are negatively associated with income at the local level). These results do not statistically change when we also include the variable PET (column 6).

Taken together, these results show a strong positive relationship between standard climate variables (i.e., precipitation and temperature) and life expectancy and income at the local level, thus confirming the conclusions drawn in the relevant economic literature (Dell et al. 2012; Burke et al. 2015). However, the opposite effects of the potential evapotranspiration of the soil emerge (column 2).

Concerning the relationship between long-term variations of climate variables and determinants of the subnational human development, Table 6 shows a negative and significant relationship between PET and life expectancy and completed years of education, while a negative but less robust relationship with our measure of local income.

As for the relationship between climate and HDI, we perform a robustness check of the estimates of the relationship between climate variables and each determinant of the HDI. We construct the change in the life expectancy, year of education, and GNI from 1990 to 2015 for each region i of country c, and estimate the relationships between long-term variations in climate variables and variations of the three determinants of the HDI. Results of this further check are reported in Table 8 in the Appendix and reinforce the conclusions drawn in this article.

7 Conclusions

Recent economic literature has addressed the issue of the impacts of climate change on economic development, focusing mainly on standard economic activities.

This paper investigates the effects of climate change on human development. Our results indicate that, while precipitations and temperature are associated with a higher Human Development Index at the subnational level, the inclusion of the potential evapotranspiration of the land partly offset the benefits of higher precipitation levels. We also show that these effects are more pronounced for economies that extensively rely on the agricultural sector. Finally, we argue that the two principal mechanisms through which climate impacts human development are income and life expectancy, suggesting that the channels through which desertification impacts human development are the reduction in agricultural production and the associated malnutrition exposing children to higher mortality rates. Our conclusions shed light on the practical economic impacts of climate variations.

Our results point to three policy implications: (1) the effect of climate change needs to be analyzed at the local level, where also adaptation and mitigation policies need to be implemented; (2) climate is not only a matter of changes in the atmosphere and on the ground, with soil aridification affecting significantly socioeconomic well-being in the long run; (3) in some cases, climate change will affect human health not necessarily by affecting income. Finally, we find no significant effect among high-income regions. One possible explanation is that it might be the result of the use of irrigated water or other agricultural innovations. Future research could focus on the different mechanisms explaining the heterogeneous effects of climate shocks on human development for different classes of income. Furthermore, future efforts need to try to detect which time lag (e.g., 2, 3 or 4 years) might be the most relevant to explain how human development is affected by climate shocks.

Finally, future research is expected to be directed toward the local channels of transmission driving the impact of soil aridification on development, and in this respect, the analysis of the impact on the composition of agricultural production and new crop and technology adoption to cope with terrain degradation may prove to be interesting from a research perspective and valuable for policy making.

8 Figures and tables

The graph on the left hand side shows the negative relationship between potential evapotranspiration and human development for the African continent. Results are obtained considering year-to-year variations from 1990 to 2015. The graph on the right hand side is obtained considering the long-term change over the 25-year period. This represents the long-term relationship between PET and human development.

Based on these data, on the other hand we do not see a strong correlation between variations in precipitations and human development.

Appendices

A.1 Data appendix

1.1 A.1.1 Construction of the HDI

The subnational Human Development Index used in this study are a translation of the UNDP’s official HDI and GDI (hdr.undp.org) to the subnational level. Data are publicly available through the Global Data Lab Project (Smits and Permanyer 2019).

The HDI is an average of the subnational values of three dimensions: education, health and standard of living. In its official version defined at the national level, these dimensions are measured with the following indicators: Education measured with the variables “Mean years of schooling of adults aged 25 and above” and “Expected years of schooling of children aged 6”; health measured with “Life expectancy at birth” and standard of living measured with “Gross National Income per capita (PPP, 2011 US$).” Details about the estimation procedure can be found in the Human Development Report (Jahan et al. 2017).

To construct the HDI for the period 1990–2018, Smits and Permanyer (2019) computed the subnational variation in these indicators and applied it to their national values derived from the UNDP website. For computing the subnational values, basically two different data sources were used: indicators derived from the Area Database of the Global Data Lab and indicator data obtained from statistical offices. The indicators derived from the GDL Area Database are the major source of data for low- and middle-income countries (LMICs) and those obtained from statistical offices for high-income countries (HICs).

Because life expectancy and gross national income per capita (GNIc) are not readily available in the household surveys and census datasets from which the GDL Area Database is constructed, their subnational values for LMICs had to be estimated. For life expectancy, this was done on the basis of data on under-five mortality and for GNIc on the basis of household wealth. To measure household wealth, the International Wealth Index (IWI) was used, an indicator of household’s standard of living based on asset ownership, housing quality and access to public services (Smits and Steendijk 2015).

For years for which subnational data for one or more indicators was missing, the values of these variables were estimated by linear interpolation between the preceding and succeeding year for which this information was available. If interpolation was not possible, the subnational values of the nearest year were used. The subnational variation obtained in this way was subsequently applied to the UNDP national values, so that for each year in the period 1990–2017 the (population weighted) mean of the subnational values is in line with the UNDP values. Further information on the construction of the HDI Database can be found in Smits and Permanyer (2019); Permanyer and Smits (2020).

The Global Data Lab

The Global Data Lab provides since 2016 freely downloadable subnational development indicators for low- and middle-income countries (LMICs) through its Area Database.¹³ These indicators are constructed by aggregation from representative survey and census datasets. The major data sources used by GDL for this purpose are Demographic and Health Surveys,¹⁴ UNICEF Multiple Indicator Cluster Surveys,¹⁵ and datasets from population censuses distributed by IPUMS International.¹⁶ These sources provide large samples, often 50,000 to 100,000 or more respondents, containing information on all household members. For LMICs for which these sources are not available, GDL uses other country-specific surveys, such as the Afrobarometer or the Americas barometer surveys,¹⁷ which include only adults instead of complete households. For most LMICs, GDL-AD provides the two indicators needed for creating the educational index, mean years of schooling and expected years of schooling. However, the indicators needed for the health and income dimensions are usually not available in the required form in household survey and census datasets. The subnational values of these indicators for LMICs are therefore estimated using data on child mortality and household wealth that is derived from GDL-AD.

Education and LIFEX

When only educational attainment data were available, the HDI is computed considering the corresponding years of schooling. In case of missing data, LIFEX was estimated using information on child mortality. This is the case for most of the LMICs. Given that household surveys and censuses generally do not contain information on LEXP, subnational values of this indicator were for these countries estimated based on information on under-5 mortality (U5M). To estimate LEXP on the basis of U5M, Smits and Steendijk (2015) constructed a regression model that explained the variation in national LEXP derived the UNDP database on the basis of national U5M scores derived from GDL-AD.

Standard of Living

In case of missing data, LGNIc was estimated on the basis of IWI scores. Indeed, for most LMICs, data on standard of living were not available. Therefore Smits and Permanyer (2019) derived it from the GDL-AD. Given that household surveys and censuses for LMICs often do not contain information on income and, if they do, this information is not very reliable in poor areas, subnational values of LGNIc were estimated based on household wealth. For this purpose, the International Wealth Index (IWI) was used, which measures household wealth on the basis of information on asset ownership, housing quality and access to public services. The IWI scale runs from 0 to 100, with 0 meaning ownership of none of the assets and bad quality housing and services and 100 indicating ownership of all assets and best quality housing and services. To estimate LGNIc for the subnational regions on the basis of IWI, a regression model was constructed that explained the variation in national LGNIc derived from the UNDP database on the basis of national IWI scores derived from GDL-AD.

A.2 Robusteness of estimates

We also implement a second model to assess the long-term relationship between climate and human development. This models implies a cross-sectional equation and is detailed below:

$$\begin{aligned} \begin{aligned} \Delta Y_{i,c}&=\alpha +\beta _{1}\;P_{i,c}+\beta _{2}\;P_{i,c}^{2}+\beta 3\;PET_{i,c}+\delta _{1}\;T_{i,c}+\delta _{2}\;T_{i,c}^{2}+\mu _{c}+\epsilon _{i,c} \end{aligned} \end{aligned}$$

(6)

Here we denote with $\Delta Y_{i,c}$ the change in the Human Development Index from 1990 to 2015 for region i of country c. $\Delta P_{i,c}$ indicates the change in average annual amount of precipitation of region i in country c over the 25-year period and is expressed in millimeters. As in Eq. 1 we also account for nonlinear relationship between change precipitation and HDI and we include $\Delta P_{i,c}^{2}$ indicating the change in the average annual amount of precipitation in millimeters of grid i in country c squared. The variable $\Delta PET_{i,c}$ instead, indicates the change in annual potential evotranspiration of grid i in country c from 1990 to 2015 and is expressed in millimeters. In addition, we also control for the 25-year change of annual mean surface temperature $\Delta T_{i,c}$. Finally, the model considers country fixed effects denoted with $\mu _{c}$. We estimate Eq. (6) via OLS.

We construct these variables capturing the 25-year change as follows:

$$\begin{aligned}&\Delta Y_{i,c}={\overline{Y}}_{i,c,2015}-{\overline{Y}}_{i,c,1990} \end{aligned}$$

(7)

$$\begin{aligned}&\Delta P_{i,c}={\overline{P}}_{i,c,2015}-{\overline{P}}_{i,c,1990} \end{aligned}$$

(8)

$$\begin{aligned}&\Delta PET_{i,c}={\overline{PET}}_{i,c,2015}-{\overline{PET}}_{i,c,1990} \end{aligned}$$

(9)

$$\begin{aligned}&\Delta T_{i,c}={\overline{T}}_{i,c,2015}-{\overline{T}}_{i,c,1990} \end{aligned}$$

(10)

The average terms, ${\overline{Y}}_{i,c,t}$, ${\overline{P}}_{i,c,t}$, ${\overline{PET}}_{i,c,t}$, and ${\overline{T}}_{i,c,t}$ are defined as follows:

${\overline{Y}}_{i,c,t}=\frac{1}{5}\sum\limits _{K=1}^{5}Y_{i,c,t+1-k}$, ${\overline{P}}_{i,c,t}=\frac{1}{5}\sum\limits _{K=1}^{5}P_{i,c,t+1-k}$

${\overline{PET}}_{i,c,t}=\frac{1}{5}\sum \limits_{K=1}^{5}PET_{i,c,t+1-k}$, ${\overline{T}}_{i,c,t}=\frac{1}{5}\sum \limits_{K=1}^{5}T_{i,c,t+1-k}$

where ${Y}_{i,c,2015}$, ${P}_{i,c,2015}$, ${PET}_{i,c,2015}$, and ${T}_{i,c,2015}$ indicate the annual average HDI, the annual average precipitation, PET and temperature at region i of country c in year 2015. These are five-year averages of annual average HDI, precipitations, PET and temperatures. Similarly ${Y}_{i,c,1990}$, ${P}_{i,c,1990}$, ${PET}_{i,c,1990}$, and ${T}_{i,c,1990}$ indicate the annual average HDI, the annual average precipitation, PET and temperature at region i of country c in year 1990. These are five-year averages of annual average HDI, precipitations, PET and temperatures. As a result, Eqs. 7, 8, 9, and 10 measure changes in average HDI, precipitations, PET, and average temperature over 25 years as these are differences between year 2015 and 1990. These variables capture long-run changes in climate, attenuating year-to-year fluctuations and only isolating a 25-year trend.

Results of the relationship between long-term variations of climate variables and the local Human Development Index are reported in Table 7. This time, the regressors of interest are $\Delta \text {Prec}$, indicating the change in the average annual amount of precipitation of region i, in country c, over 25 years expressed in millimeters, $\Delta \text {Temp}$, indicating the change in average annual temperature of region i in country c over the 25-year period expressed in degree Celsius, and $\Delta \text {PET}$, indicating the change in the average annual amount of PET of region i, in country c, over the 25 years expressed in millimeters. Higher values of the variable $\Delta \text {Prec}$ correspond to higher average annual rainfall over the period 1990–2015. Higher values of the variable $\Delta \text {Temp}$ correspond to higher average temperatures over the period 1990–2015. Conversely, higher values of the variable $\Delta \text {PET}$ correspond to lower “effective” water availability over the period 1990–2015. As in Eq. 1, we also check for a nonlinear relationship between precipitation and human development by including the variable $\Delta \text {Prec}^{2}$, defined as the quadratic term of the average level of precipitation over the period 1990–2015. Similarly, we include the variable $\Delta \text {Temp}$, defined as the quadratic term of the average surface temperature over the period 1990–2015. Equation 6 also includes region and time fixed effects and country-specific linear trends.

Column (1) of Table 7 considers the whole sample of subnational regions for the whole world and shows the 25-year relationship between precipitation and temperature and local human development. Unlike the results obtained for the short-run analysis, the estimates of the relationship between long-term changes in precipitation and human development lose significance. These results are qualitatively robust to the inclusion of year-fixed effects (column 2) and country-specific time trends (column 3). Column (4) shows a negative and significant relationship between long-term variations in PET and local human development. In particular, one standard deviation increase in PET over the period 1990–2015 is associated with a decrease of 27.7% points in the Human Development Index.

Taken together, these results suggest that while precipitations are an essential determinant of human development in the short run when considering long-run climate variations, the effect of climate on HDI is likely to work through soil aridification.

Table 7 Robustness: long-term local climate-HDI relationship

Table 8 Long-term relationship between climate and HDI determinants