Climate Change, Comparative Advantage and the Water

Download Climate Change, Comparative Advantage and the Water

Preview text

Climate Change, Comparative Advantage and the Water
Capability to Produce Agricultural Goods
Candau F. (E2S UPPA)*, Regnacq C. (BRGM), Schlick J. (RITM, Univ. Paris-Saclay)„
May 12, 2022
This article analyzes how climate change inuences the capabilities to export agricultural goods and the specialization of nations (e.g., comparative advantages) by altering farmers' capability to use available water. Our main contribution is methodological since we present the rst attempt to link precisely the micro-determinants of production to the macro-determinants governing the specialization of countries. We use a rich set of data both locally (at the crop level analyzing thousand elds that cover the Earth's surface) and at the global level (analyzing bilaterally the international trade of nations). At the local level, we estimate the elasticity of production to the thermal and hydrologic conditions (including blue and green water as well as groundwater storage) along with xed eects (at country-product and at the crop level) to control for omitted variables. At the global level, we use the predicted value of these elasticities to compute an indicator of the water capability to export agricultural goods, which is then used in a trade gravity equation to control for trade costs that also shape the specialization of countries. From these estimates, we nally build an indicator of comparative advantage in agricultural goods and analyze how these relative advantages are aected by climate change in 2050. We present unexpected results at rst sight, that are however in line with the Ricardian theory, such as cases where a deterioration of the local conditions to produce a good does not prevent an improvement in the comparative advantage to produce it (representing 32.51% of cases in our simulation), or the reverse, when the improvement of the local conditions happens simultaneously with a deterioration of the comparative advantages (representing 18.16% of cases in our simulation).
Forthcoming in World Development
JEL Classication: Q17, Q25, Q56, F18.
1 Introduction
Climate change will have a myriad of eects on the productions of agricultural goods (IPCC (2007)). The most obvious, is that by deteriorating the natural conditions of plant growth, it will lead to less production. A more subtle consequence is that climate change will aect dierently the relative costs of production of dierent crops (Costinot et al., 2016). Then, apparently unexpected results can arise, such as the production of goods that are not fully adapted to the new climate but which are possible because climate change will have even worse consequences on other outputs in this country
*Universite de Pau et des Pays de l'Adour, E2S UPPA, CNRS, TREE, Pau, France. Mail: [email protected] „We are particularly grateful to Ariel Dinar, Gwenolé Le Velly, José De Sousa, Jaime De Melo and Esther Delbourg for
insightful comments and to participants at various seminars and conferences and in particular at the University California Riverside, at the American Economic Association meeting (session on Water and Agriculture), at the 26th European Association of Environmental and Resource Economists conference, and nally at the CEE-M (U. Montpellier), SMART LERECO, and Public Economy Unit (INRAE-AgroParisTech) seminars.

relatively to what occurs elsewhere, leading producers to focus their investments and resources on these less aected agricultural products. In case of a positive shock, the same mechanism holds, not all the favoured products by climate change are going to be produced, but only those with the lowest relative costs of production in comparison with foreigner competitors. This is at least what can be expected in a globalized world according to the theory of comparative advantage of Ricardo (1817): the specialization of countries is based on a comparison of their relative productivity dierences. In this paper, we analyze to what extent the utilization of an essential factor of production, water, explains the specialization of countries on dierent agricultural products. Then, we study how climate change, by aecting this resource, can destabilize the current comparative advantages of all the countries in
the world.1
We dene the water capability to produce as all the conditions related to water that enable to grow a plant. We thus build an indicator that includes the availability of all sources of water (such as river's runo, groundwater and precipitation), taking into account the water use competition between the dierent crops (as well as the competition with municipal and industrial consumption). We also consider the climatic constraints that impact the eciency of the available water in the production of agricultural goods. Indeed the evapotranspiration of plants (depending of their location), and the
temperature requirement of each crops determine whether there is enough water to sustain production.2
Finally, we build an indicator that measures the cost of using water for a particular crop relatively to all other production that are possible according to the climate and hydrologic conditions.
We focus on this water capability to produce, for two reasons.The rst and the most obvious is that the utilization of water is an essential component of the agricultural production and thus worth
of interest regarding the essential role of agriculture for human development.3 The second reason is
that many researches have been done by hydrologists and agronomists on water, enabling to have an accurate measure of the current water capability to produce at a very disaggregated spatial level and for dierent agricultural products all over the world. We then build an indicator of water capability at the scale of a 30 arc-minute worldwide grid or approximately 55 kilometers at the equator. This spatial disaggregation at the grid scale, which is then ner than a regional or even a communal scale, is important to observe where the specialization of countries comes from. Hence our detailed analysis enables to locate the development and crisis related to water for the production of agricultural inside each country. Furthermore, the product disaggregation of the water capability to produce dierent goods is also critical for our research question, because it enables to determine the water productivity
of a location for all agricultural activities, and not just those in which it is currently employed. This
allows to study how dierent lands can be converted to develop new types of products in front of climate change.
Why does it matter to take into account international trade in this analysis of the eect of climate change? After all, why exchange is so important, and not only internal food production? As stated earlier, from the theoretical framework from which the current analysis is based, namely the comparative advantage of Ricardo (1817), it is the process of international exchange that drives the specialization of nations. Such an emphasis on the importance of trade integration to explain the
1We focus here on how climate change by increasing water scarcity has a signicant negative eect on agriculture
production and then specialization. Obviously, climate change also aects agriculture via other channels, for instance, by enhancing the probability of oods, it would equally damage crops (e.g. Bronstert, 2003; Su, 2020). We do not consider here the eects of climate change on ooding, which would require a dierent analysis that is left for future research.
2More concretely, data on precipitation, runo and evapotranspiration come from the AQUAMAPS data of the FAO,
which are merged with data of the Global Agro-ecological Zones v3.0 (GAEZ) to take into account the thermal regime (including the crop suitability of the local thermal regime), the soil quality, the terrain slope, the share of land covered with building and other natural coverage (lake, forest or ice). These variables are then used to explain the agriculture production (available from MIRCA2000) to build our new indicator of the local water capability to produce agricultural goods.
3According to Bairoch (1973) and to a more recent literature in economic growth (see Ashraf and Galor, 2011) success-
ful productivity growth in agriculture has been the source of early structural transformation leading to industrialization, urbanization and development in most of today's high income countries.

specialization of nations is not solely at the heart of the modern Ricardian theory (e.g. Eaton and Kortum (2002), Costinot et al. (2012)) but it is a common feature of all analysis in international
economics.4 Moreover the relative cost of production inside a country is not enough to determine its
specialization, international trade costs also matter. For instance, the relative productivity advantage due to water in one country, can be outweighted by the high trade costs to export agricultural goods from this country (relatively to other countries). In brief, locational (dis)advantage also explains the specialization of nations. From that stand point, it would be illusive to analyze the specialization of countries without taking into account international trade. In this paper, we use international trade ows, to measure trade costs (via a trade gravity equation) and then to reveal the comparative advantage of countries (via the Balassa (1965) index, aptly named the Revealed Comparative Advantage
index, hereafter RCA).5 Then, by studying how climate change aects the water capability to export agricultural goods, we are able to predict how the agricultural RCA can evolve.6 Furthermore, by
considering the implications for the world development, analyzing climate change in an open economy also enables to study the conditions under which international trade can mitigate the consequences of climate change via agricultural reallocation of production between countries.
Our ndings concerning the production of agricultural goods measure the importance of the thermal and hydrologic conditions and the essential role of groundwater storage. Indeed, our estimates show that having insucient renewable water resources available to fulll the needs of plants may induce a minimum loss of capability to produce agricultural goods by approximately 4%. Yet, such a loss may be attenuated by pumping water from the underground as we nd that an increase of groundwater stock by 1% could potentially help to increase this capability by approximately 0.03% in the localities with insucient renewable water resources. When simulating the eects of climate change, we nd that 64.2% of agricultural lands may experience a decrease in their capabilities to produce. The most vulnerable countries to climate change will likely bear the most of these negative changes with a total
loss of almost 17% of agricultural lands.7 Regarding the international trade of agricultural products, we nd that the water elasticity to export8 is the smallest for the most vulnerable countries. This
nding that exports of the most vulnerable countries are less sensitive to the water conditions seems to be a good new, indicating a specialization in the production of goods that are less intensive in water. However, these specializations seem to be insucient to cope with climate change as we nd that these countries could experience a drop of their exports by 14 % partly due from a strong decrease in their capabilities to use water for producing agricultural goods at the local level (-17 %) and partly due to a
relatively low access to international markets (less than 10% of global trade in 2050).9 This indicates
a relative increase in the cost to produce agricultural goods and an increase in the marginalization of these countries to the world exchange. This may be particularly problematic for the development of these countries where exports of agricultural goods may have spillover eects in terms of productivity and explain the agricultural-demand-led industrialization (Adelman, 1995; De Pineres, 1999; Bustos
4In the neo-classical analysis of comparative advantage based on factor endowments (the Heckscher-Ohlin model),
openness leads countries to specialize their production in goods that use the most intensively the most aboundant factor. According to the new theories of trade, openness fosters a specialization in the production of goods which have the largest domestic market (the so-called home market eect, see Costinot et al. (2019)) or drives the resource toward the most productive rms (Melitz, 2003).
5The RCA is an index used in international economics for calculating the relative advantage or disadvantage of a
certain country in a certain class of goods as evidenced by trade ows. See French (2017) who shows why this index is appropriate to uncover countries' fundamental patterns of comparative advantage.
6In what follows, we often use the term comparative advantage instead of revealed comparative advantage or
RCA because it is more telling and convenient but all our analysis of comparative advantages is based on the RCA.
7In our analysis, we distinguish 4 group of countries depending upon their vulnerability to climate change (using an
index developped by the FERDI).
8The water elasticity of trade is a measure of how sensitive the export from one country to another is to its water
capability to produce agricultural goods. The term strong elasticity means that changes in the water capability to produce have a relatively strong eect on exports.
9We also present a substantial reallocation of production at the benet of the less vulnerable countries that could
experience a rise of their export by around +72% (with a strong increase of their comparative advantages: +59.5% on average) despite a drop of their average capabilities to use water for producing agricultural goods at the local level.

et al., 2016).10
Therefore, our analysis going from the micro geographical scale of the crop to the macro analysis of countries specialization enables to determine where the deterioration of the local conditions could lead to a deterioration of the comparative advantages (e.g. for cassava in Africa), as well as the reverse, namely the identication of locations where the improvement of the local conditions could lead to an improvement of the comparative advantages (this represents approximately half of the cases). We also present less intuitive results, such as cases where the deterioration of the local conditions to produce a good leads to an improvement in the comparative advantages to trade it (representing 32.5 % of cases, e.g. potatoes in the mediteranean countries), or when the improvement in the conditions to produce a good leads to a deterioration of the comparative advantages to trade it (representing 18.2 % of cases, e.g. rice in India).
Our analysis is related to dierent elds. Our concept of the water capability to produce agricultural goods is related to an emerging array of research on the link between water scarcity and the supply of crops. For instance, Vallino et al. (2020) propose to use an indicator of management of domestic and transboundary water resources to approximate the economic water scarcity and show that this indicator explains the agricultural productivity of countries. Interestingly they nd that their indicator is not always associated to high country's income or to the hydrological water scarcity. Rosa et al. (2020) also propose an indicator of agricultural economic water scarcity, dened as the lack of irrigation due to limited institutional and economic capacity. They identify agricultural economic water scarce lands where investments in sustainable irrigation have the possibility to increase food production (e.g. in Sub-Saharan Africa). In comparison, our study does not analyze the role of institutions and governance of water, but better captures the hydrologic constraints and physical opportunities to produce agricutural goods.
Hydrologists have also developed models of the determinants of agricultural production linked to the water availability at the local level, such as the LPJmL (Lund-Potsdam-Jena managed Land, developed by Bondeau et al. (2007)), a dynamic global vegetation model designed to simulate vegetation composition and distribution for natural and agricultural ecosystems. However, such type of models have not only the major drawback of being very dicult to handle due to the complexity of the
computation,11 but also often failed to analyze the economic consequences such as the impact of the
economic water scarcity of the agricultural specialization of nations. Economists on the other side, investigate in details how comparative advantages evolve (Costinot et al., 2016, Coniglio et al., 2021), but neglect the detailed hydrological information at the local level to understand the role of water in the production of agricultural goods. For instance, Murphy (2009) and Afkhami et al. (2018) capture the availability of water with annual runos and groundwater recharges in the exporting country. These indicators are problematic because the sole water endowment aggregated at the country level is misleading as some regions inside a nation can have sucient water resources but not the temperatures
or soil quality required to produce.12 Our value added is thus to propose an indicator richer than the
one used by economists, but simpler and more transparent than those presented by hydrologists in order to understand how the specialization of nations in dierent agricultural goods.
Our contribution can also be discussed at the light of a large literature on virtual water trade,
10There are many researches and debates on the conditions under which exports of agricultural goods stimulate growth.
Bustos et al. (2016), for instance, show the introduction of genetically engineered soybean seeds in Brazil in the context of a high level of trade openess has led to industrial growth in this country because this new technology was strongly labor-saving.
11The use of these models are challenging for non-hydrologist scientists who may consider them as black boxes.
Furthermore, these models also turn-out to be sometimes inadequate for economic studies as key parameters are generally estimated through country xed eects (which induces major concerns about endogeneity bias when this indicator is used as an explanatory variable in a regression that also introduces xed eects).
12In fact, authors implicitly assume that a surplus of water in one location can be transferred to another part of
country where water is missing, but unlike other types of production factors such as capital, water resources cannot always be moved easily from a water-abundant region where production is not possible to a water-scarce one where other conditions to produce are located.

which measured the volume of water used in production of goods that are traded at the global level. Debaere (2014) nds that relatively water abundant countries export more water intensive products. Delbourg and Dinar (2020) demonstrate that arid countries use trade in order to alleviate their problem of water scarcity. They also show that some countries with abundant factors (land and labor) use water with less eciency. A lively debate has however being at the center of this literature, since numerous authors have found opposite results, such as the fact that some water scarce countries
actually export water-intensive crops (and vice versa).13 Vallino et al. (2021) improve these analyses
by weighting the global virtual water trade with a new composite water scarcity index that combines physical and economic water scarcity. They nd that almost half of water volumes traded comes from countries that are worse-o than their partners regarding their composite water scarcity and their economic wealth. While this literature considers that governance, institutions and economic development at the local level can explain the conict between the comparative (dis)advantage of countries and their specialization in agricultural goods, we pursue another explanation by considering that the unequal access to international markets for exporters from dierent countries can contradict the comparative advantage dened at the local/national level. By considering this, our article proposes a new contribution to the literature on the interaction between specialization and trade costs. To date, the most signicant ndings in this literature have been theoretical. Venables and Limao (2002) rst propose a theoretical model to demonstrate that the equilibrium pattern of specialization involves a
trade-o between comparative production costs and comparative transport costs.14 Deardor (2014)
also shows that this trade-o matters and develops a concept of local comparative advantage (dened as autarky prices in comparison to nearby countries) to explain the specialization of countries. While some empirical analysis have been done from these theoretical foundations (e.g. Harrigan, 2010), we propose here a new methodology to reconciliate the local comparative advantage (linked to water) to international trade costs in order to determine the comparative advantage at the national level.
Finally, the consequence of climate change on agricultural trade have been at the core of many studies in the last decade (Huang et al. (2011), Costinot et al. (2016), Gouel and Laborde (2021)). The conclusion is optimistic, international trade can mitigate the consequences of climate change via agricultural reallocation of production inside and between nations. Climate change will also induce yield changes and large price movements fostering incentive to adjustments. In comparison with this literature, we propose a very dierent methodology which is based on estimations and not on numerical simulations of Computable General Equilibrium models (CGE). We also provide a new analysis of the hydrological conditions at the source of the comparative advantage of nations. Our results also dier and are less optimistic, for example, we present many cases where the reallocation of production inside nations fails to sustain the comparative advantage of countries.
The reminding part of this article includes the following sections. In Section 2, data on precipitation, runo, groundwater storage, evapotranspiration and the thermal regime are used to explain the agricultural production at the local level. Then, the predicted values of this estimation are at the basis of our new indicator of the local water capability to produce agricultural goods. In Section 3,
this indicator is used to explain bilateral trade between countries.15 We also study how this water
13See Fraiture et al. (2004) and Kumar and Singh (2005) for early critical analyses of virtual water trade. See
also Ramirez-Vallejo and Rogers (2004) who show that virtual water trade ows are independent of water resource endowments in contradiction with one standard theory of international trade (the Heckscher-Ohlin Theory). Verma et al. (2009) quantify and critically analyze inter-state virtual water ows in India. Han et al. (2021) show that virtual water trade intensies the water scarcity in Northwest China.
14The literature called the New Economy Geography summarized in Fujita et al. (2001) and Candau (2008) has
analyzed how increasing returns and transport costs interact to explain the agglomeration of activities, and then indirectly explains how trade costs aect the trade patterns, however few analysis have been done on comparative advantage (see however Ricci, 1999).
15Trade decisions are obviously complex and not completely determined by changes in production patterns, they
depend on prices, policies, and changes occurring in other nations. To take into account all these elements, and then to isolate the role of our water variable, we use the best practice in the literature of international trade (see Head and Mayer (2014) for a survey), that consists to explain trade ows via a gravity equation with bilateral xed eects to control for all the bilateral relationships between partners (e.g. trade agreements), country eects and product eects that control

capability to produce agricultural goods inuences dierently the exports of dierent groups of countries distinguished by their degree of vulnerability to climate disruption (and endowment in capital). In Section 4, we build an index of the revealed comparative advantage that takes into account this variable of water and all the determinants of trade (including trade costs). All this analysis enables to follow the eect of climate change on water conditions at the local level, and then on production and trade to analyze the specialization of countries. Finally, Section 5 concludes with some avenues for future research.

2 Agricultural production and water at the crop level, a world analysis

The agricultural production depends on water but obviously depends on many other determinants, that

need to be taken into account to isolate the productive eect of water on agricultural specialization.

Some determinants are related to water and are often dened at the micro-spatial level (e.g. at

the crop level where the water is available). Other determinants that depend, for instance, on the

agricultural technology and/or on the global market access of producers are dened at the national

and/or even at the international level. Our analysis is thus divided in two parts. In this section we

lead a substantial investigation at the crop level to measure all the micro-constraints encounter by
agricultural to produce. These micro-level constraints refer to the capability of producers to transform

the available water resources into agricultural goods at the local level, they depend on temperature,

soil quality, and so on. At this micro-spatial level we measure the relative costs of using water for a

particular product relatively to all other products that can be produced (namely in the spirit of the

Ricardian theory of comparative advantage). In the next section, we use a trade gravity equation to

capability capture all the determinants that limit the

to export. Among these constraints, trade costs

between countries,16 and dierences in technology are taken into account.

2.1 Supply of crop at the micro-geographical level
To infer the impact of local water availability upon the capabilities to grow dierent type of crops, it
is possible to follow a rich literature17 that denes the behavioral rules of farmers regarding the crop
acreage choice subject to agronomic and climatic constraints. On that matter, dierent models have been proposed such as multicrop production models with multinomial logit acreage shares presented in Carpentier and Letort (2013) or the Ricardian model of trade in Costinot et al. (2016). However, we do not use a formal model from this literature, as we aim to develop an indicator that does not depend on the expected gross margins per hectare of crop, on the price of the good produced or on the intermediate goods used. Our aim is to determine the supply of crops related to water that depends on natural determinants such as precipitations, the surface of waters, groundwaters and the water requirement of the dierent crops. This point matters to alleviate an endogenous bias in our empirical investigation (due to reverse causality between production and export). Indeed since our indicator is used to explain exports, we need an index that is not based on the current production choice of farmers but on a hypothetical choice based on the water endowment of locations. In other words, we do not
compute an indicator on the types of products that are produced at a location l (depending of prices, fertilizers and so on), but rather on the types of products that could be produced in l given various
exogenous conditions. In that respect, our work is based upon the idea of capabilities to transform the available natural resources into valuable goods given local natural constraints that can be traded on the international markets.
for the competitiveness of exporters, demand eects, and changes occuring in other nations (what's the literature called the multilateral resistances).
16These trade costs, that can be decomposed in four components, called the four Ts by Spulber (2007), are trans-
action costs (due to customs, business practices, and legal environments), tari and non-tari trade barriers (including environmental regulation and anti-dumping practices), transport costs and time costs.
17See Carpentier et al. (2015) for a literature review.


Furthermore, we follow the core principle of these aforementioned models by analyzing the relative capability to produce agricultural goods instead of an absolute value. Thus, similar to Carpentier and
Letort (2013) or Costinot et al. (2016), we dene our local capability to use water for producing a
specic crop in comparison to the capabilities to use water to produce other crops and/or to fulll
non-agricultural needs.
Formally, we divide the world into gridded cells of 30 arc-minutes each representing localities l in which farmers can grow multiple crops, indexed by k. The categorization of these crops will follow the
international nomenclature of the Harmonized Commodity Description and Coding Systems (called
HS4 later in the text).18 In such a setting, the heterogeneity of farmers comes from the fact that
farmland does not match the cells such that multiple farmers may be within each locality l and the
farmland of one farmer may overlays multiple localities. In that respect, dierent behaviors may arise
within each locality leading to a certain diversity of crop acreage in l.
Following Carpentier et al. (2015), local constraints to grow a given crop k in locality l can be
decomposed into two broad categories: the land use choice variables and the acreage choice variables.
The land use choice describes the individual choice to produce agricultural production in a given
locality. Among the variables that determine this choice, there are, for instance, the soil quality, the
thermal regime, the hydrologic conditions. These variables are important determinants of the average
productivity of land for agricultural use in the broad sense. They are also useful to dene localities
l where agricultural production is not possible for any type of crop k. This last point is particularly
important once we consider climate change, since the land use choice to produce agricultural goods
can be hindered by a deterioration of these variables (IPCC, 2007).
Acreage choice describes the choice of farmers to produce a particular good among the dierent
types of crop k that can be grown in each locality l. This choice depends on a vector of variables, hereafter denoted Clk, that are dened at the crop level, such as the crop specic suitability of hydro-
logic conditions and of the thermal regime at the cell level. In other words, this vector of variables
encompasses all the agronomic factors that aect directly the capability of farmers to use water for
agricultural production which will be more precisely dene in the next section. This choice is also done
by a comparative analysis: the incentive to produce a particular product depends on its suitability to
the natural conditions compared to the suitability of all other goods that can be produced under these
conditions ( k Clk) (Carpentier and Letort, 2013; Costinot et al., 2016). This choice also depends on
the alternative uses of water (namely water not used for the agricultural production), that reduce the
water available to grow the crop k (Flörke et al. (2018)). These alternative uses (e.g. water for consumption in cities) are denoted Wlmun. The higher Wlmun, the smaller the share of a eld l allocated to any given crop k that can benet from water, and then the smaller the production at this location. To summarize, we aim to capture here the relative cost to use water for a particular k in comparison
with other use of water (agricultural and non-agricultural).
We combine these elements19 to build an indicator of the production of a given crop k in locality
l, Lkl , that may be interpreted as a measure of the local water capability to produce agricultural goods
such as:

Lkl = Land U sel × W mun C+lk Ck (1)



Acreage Choice

This expression (1) takes a similar form than the one presented in Costinot et al. (2016) (Equation 8) where farmers allocate their xed land inputs to multiple crops with land share assigned to each
18A list of all products we consider in our analysis can be found in table (4) in the Appendix and a detailed explanation
of these dierent classication can be found at
19See the data section 2.2 where we further expand the explanation and computation of these dierent variables.


crop being somewhat proportional to its relative productivity. Here, we solely substitute the concept of productivity by the one of capability.

2.2 Data and Empirical Strategy
In this section, we present the data and the dierent steps necessary to build our indicator of the local water capability to produce agricultural goods in the next section. Since this involve a complex methodology, we breakdown the procedure into two dierent steps which are dedicated to the vector
of variables Clk where we rst present the data and computation used to dene each variable included in Clk and in the second step, we estimate the weight of each of this variable through a structural estimate of equation (1) allowing us to build a set of values Clk for each locality l and crop k.

2.2.1 Data and Computation of the Variables in Clk
The vector of variables that dene the acreage choice concerning Clk is dened hereafter.
The crop specic thermal regime (Tlk) is the temperature constraint factor from the Global Agro-
ecological Zones v3.0 (GAEZ) to dene the suitability of each cell l for growing any specic crop k. Here, the monthly prole is not necessary as the variable given by GAEZ already accounts for the
adequation of temperatures in the growing period of each crop. This variable is expressed in percentage
and thus ranges from 0 to 1 (where 0 implies the thermal regime of the locality l being unsuitable for the crop k and 1, the locality l being perfectly suitable for growing the crop k).
The crop specic suitability of local renewable hydrologic supplies (RWlk) is computed as the
monthly average of the ratio between the local renewable water availability and the local crop wa-
ter requirement. On the one hand, the local renewable water availability in locality l for each month m is the sum of soil moisture dened as RlSmM (data taken from the C3S Soil Moisture developed within ESA's Climate Change Initiative Soil Moisture Project), surface water runo dened as RlSmR (data
taken from the Global Composite Runo Fields, CSRC-UNH and GRDC, 2002) and groundwater
recharge dened as RlGR (data taken from the Groundwater Resources of the World, WHYMAP
GWR). Finally, since ground and surface water resources can be interconnected, the simple sum may
induce some double counts. To avoid that, we follow the FAO guidelines (FAO (2003)) to calculate a
common water variable at the local level (C Wl) that will be subtracted from the calculation to avoid the double counting of the amount of water available in each cell l. On the other hand, the crop specic water needs (dened as Dlkm) can be approximated by the evapotranspiration in each cell l for the month m and the crop k using the monthly potential evapotranspiration from the Global map
of monthly reference evapotranspiration, AQUAMAPS-FAO which is multiplied by a crop coecient
given in the Chapter 6 of Allen et al. (1998) (for each crop, we build a Kc curve which calculates
the amount of water required depending of the growing stage of the plant, allowing to calculate an
accurate monthly water requirement for agricultural production). More formally, Dlkm = ck × P ETlm with ck, the crop coecient for k and P ETlm, the potential evapotranspiration in l for the month m. Thus, the computation of RWlk is as follow:

RWlk = M1k

1 Dlkm

RSM + RSR + RlGR − CWl




Where Mlk is the number of months of the growing period of the crop k in locality l. We divide
the adjusted groundwater recharge by this variable because data are only on annual basis such that
we transform these yearly values into monthly ones with the assumption that farmers may capture
the full yearly recharge to use it during the sole growing period.20 Interpretation of RWlk is relatively straightforward: if RWlk < 1, the amount of renewable water in locality l is insucient to fulll the needs of the crop k but if RWlk ≥ 1, then the crop k in locality l is not limited by water resources.

20We are thankfull to the anonymous referee for pointing out this aspect.


The supplemental quantity of non-renewable water (N RWlk) is dened as underground non-renewable
water that can be used by farmers in case of insucient renewable water. The groundwater storage,
noted RlGS is taken from the study of Gleeson et al. (2015) who estimate the volume and the spatial
distribution of modern groundwater storage with several methods. We choose to use the method that
matches recharge and water table in our calculation (we also test with another method that matches
recharge and porosity and found only very marginal changes). We then assume that this water storage
is only used when the renewable water in locality l is insucient to fulll the needs of crop k (pumping
non-renewable underground being often costlier than the surface water, this resource is generally used
in last resort, Siebert and Döll (2010); Wada et al. (2012)). In that respect,

N RWlk =

RlGS 0

if RWlk < 1 otherwise

2.2.2 Estimation of the Variables in Clk
We estimate a Log-linearization of Equation (1) that takes the following form:

ln Lkl = cL +θˆT ln Tlk + θˆRW ln RWlk + 1{RW k<1} θˆNRW ln RlGS + ηˆ +


+fok +εkl


ln Clk

ln land usel
Wlmun + k Clk

where the variables that dene Clk (namely Tlk, RWlk and RlGS ) are introduced additively. Following

Morais et al. (2018), we add a locality xed eect, noted fl that captures all variables of the land use
choice of Equation (1) and of the crop independant variables in the acreage choice (more formally,
fl = ln[land usel/ Wlmun + C¯l ]) and we also add a second set of xed eect at the country-product level (named fok) intended to capture the macroeconomic eects (national policies, international trade,

The coecient θˆT , θˆRW and θˆN RW represent the estimated elasticities of the crop thermal regime
suitability, the crop suitability of local renewable hydrologic supplies and the supplemental quantity of
non-renewable water respectively. When θˆT (θˆRW ) is higher, the supply is more sensitive to changes in the thermal regime (renewable water available). The indicator variable 1{RWlk<1} takes the value 1 if the quantity of renewable water is insucient to fulll the need of the crop k (RWlk < 1) and is factorized with the groundwater storage (RlGS ) to correspond to the variable N RWlk. Finally, the estimated parameter ηˆ captures potential heterogeneity of θˆRW in localities with a low quantity
of renewable water supplies (as such, this coecient is also factorized with the indicator variable
1{RW k<1}). Table (1) presents estimates of coecients θˆT , θˆRW , θˆN RW and ηˆ from the Equation (2). l

Column (1) explains the production of agricultural goods by considering the thermal and hydrologic conditions, the groundwater storage and our indicator of Insucient Renewable Water Resources, herefater denoted IRWR. Thermal and hydrological conditions have the expected sign and the groundwater variable has a positive signicant eect in locations where the renewable water resources are insucient (0,2362=0.296-0.0598). The lack of controls in this estimation is however problematic due to the bias of omitted variables. Column (2) introduces locality xed eects and presents similar results notably
concerning the suitability of the thermal regime and of water conditions.22 Surprisingly, the IRWR
21These xed eects are only used to control the estimation and are not used in the calculation in the rest of the
22The eect of groundwater storage is not estimated (and the interaction of this variable with IRWR is no longer
signicant) due to the colinearity with xed eects that are dened at the same locality level.


Table 1: Micro-geographical Level Capabilities Estimations

Thermal Regime Suitability
- θT

(1) Production
0.763∗∗∗ (0.0251)

(2) Production
1.123∗∗∗ (0.0208)

Renewable Water Suitability
- θRW

0.148∗∗∗ (0.00707)

0.283∗∗∗ (0.00984)

- log(RGl s)
Insucient Renewable Water Res.

-0.0598∗∗∗ (0.00569)
-0.218∗∗∗ (0.0287)

0.0899∗∗∗ (0.0342)

Insucient Renewable Water Res. × Groundwater - θNRW

0.296∗∗∗ (0.00911)

0.000812 (0.0122)

Localities FE Country-Product FE Observations Log-Likelihood R2 ajusted Standard errors in parentheses.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

0.0230 (0.0171)
No No 256378 -678966.2 0.0120

-0.0152 (0.0115)
Yes No 255446 -602176.6 0.366

(3) Production
0.863∗∗∗ (0.0358)
0.0203∗∗ (0.00884)
-0.0399∗ (0.0224)
0.0261∗∗∗ (0.00789)
0.0322∗∗∗ (0.00836)
Yes Yes 255336 -483199.0 0.748


Preparing to load PDF file. please wait...

0 of 0
Climate Change, Comparative Advantage and the Water