Input–Output Global Hybrid Analysis of Agricultural Primary Production (IO-GHAAPP) Database

Abstract

In many regions of the world, water consumption exceeds the limits of sustainable water use. A commonly used method to examine the relationship between global water consumption and production is input–output analysis. However, between approximately 70% and 90% of freshwater consumption occurs in agricultural primary production, which is often represented by only a small percentage of the total number of sectors in input–output databases. As a result, water-related assessments based on input–output analysis are limited in their accuracy and substance. In addition, the assessment of the impact of water consumption is usually carried out at the national level, which can further contribute to the imprecision of the results. Therefore, the primary objective of this work was to develop an approach to better assess water use and its impacts in input–output analysis. In order to achieve this objective, a novel approach was adopted by integrating a global spatial model of agricultural primary production (MapSPAM) into an existing input–output database via prorating. In addition, the utilisation of MapSPAM allowed the calculation of water environmental extensions with unprecedented accuracy. The resulting Input–Output Global Hybrid Analysis of Agricultural Primary Production (IO-GHAAPP) approach includes (1) a novel input–output database and (2) novel environmental extensions for freshwater consumption and scarcity. The IO-GHAAPP database consists of 150 categories and 164 regions, resulting in a total of 24,600 region–category combinations. Forty-two of the categories are dedicated to agricultural primary production (28%). In comparison, the source input–output data consist of 120 categories and 164 regions, resulting in a total of 19,680 region–category combinations, of which 14 are dedicated to agricultural primary production (12%). The Python code and IO-GHAAPP database are openly available via Zenodo. The IO-GHAAPP approach is presented in a comparative analysis of agricultural primary production, along with the associated water consumption and water footprint, at both the global level and for the United States and India. Both countries are among the most important in the world in terms of agricultural primary production as well as associated water consumption and water scarcity. Furthermore, the IO-GHAAPP approach is applied in a simple case study of Germany, which stands in contrast as one of the largest importers of agricultural primary production on a global scale. The results show that the IO-GHAAPP approach adds a valuable layer of information to the disaggregated input–output data, allowing crop-specific analyses to be carried out that would otherwise not be possible, e.g., for specific leguminous or beverage crops. The results are relevant to practitioners of input–output analysis who are concerned with the impacts of agricultural primary production and who need highly resolved data, as well as to policy-makers who rely on such studies. The demonstrated IO-GHAAPP approach could be extended to other externalities relevant to agricultural primary production, such as land use, soil degradation or pollution.

1. Introduction

Over the past millennia and especially since the second half of the 20th century, humanity’s footprint on planet Earth has grown exponentially and at an accelerating rate [1,2]. The resulting impacts are severe and threaten the current stable state of the Earth’s biosphere, in which modern societies have thrived and on which they depend [3,4,5,6,7].

One of the most-stressed components of the earth system is the hydrosphere, which includes both its fresh surface waters and groundwaters. In many regions of the world, these resources are overexploited by human activities, often to the point of exceeding local sustainability thresholds [8,9,10,11,12].

Between 70% and 90% of global anthropogenic freshwater consumption is used for agricultural primary production [13,14,15,16]. The unsustainable use and consumption of freshwater has far-reaching and often irreversible impacts on ecosystems, while also increasing the risk of water scarcity for hundreds of millions of people worldwide [17,18,19]. Given the magnitude of the environmental and socio–economic risks of large-scale water scarcity, freshwater has consistently been recognised as one of the most pressing global socio–economic challenges [20,21]. As a result, the United Nations 2030 Agenda for Sustainable Development includes a specific subgoal to “[…] ensure sustainable withdrawals and supply of freshwater to address water scarcity” [17,22]. Furthermore, water scarcity is closely linked to the sustainable development goals “Promote sustained, inclusive and sustainable economic growth […]”, “Ensure sustainable consumption and production patterns” and others.

In order to explore the relationship between water consumption and its impacts and economic production and consumption, numerous researchers have turned to environmentally extended multi-regional input–output analysis (referred to as input–output analysis henceforth), e.g., Ono [23], Lutter et al. [24], Ridoutt et al. [25], Lenzen et al. [26], Lenzen [27] and Bunsen et al. [12]. Input–output analysis is a method primarily for macroeconomic analyses. It is widely used in sustainability sciences to allocate the externalities of an economy to its actors based on their monetary interdependencies [28,29,30,31]. However, when it comes to water-related assessments, input–output analysis faces two challenges.

First, global input–output databases cover only a limited number of aggregated regions and categories. At the same time, the input mixes of the consuming categories are directly proportional to the output mixes of the producing categories. This limitation, also known as proportionality assumption, can affect the robustness of input–output analyses and has been the subject of extensive academic debate [31,32,33,34,35]. Schoer et al. (referencing Section 1.2 and Section 1.3) offer a concise literature review on proportionality and disaggregation in input–output analysis [36].

Water consumption and its impacts are particularly influenced by aggregation. As stated above, 70 to 90% of global water consumption takes place in agricultural primary production [13,14,15,16]. However, in the most commonly used multi-regional input–output databases, the agricultural primary production categories account for a much smaller proportion of the total number of categories (Table 1).

Second, satellite accounts of water consumption and its impacts are typically compiled using national crop production data, national average water consumption values per crop produced and national average characterisation factors, e.g., Lenzen et al. [26], Lutter et al. [24] and Bunsen et al. [12]. However, unlike the impacts of greenhouse gases, which are largely the same regardless of where they are emitted, water-related assessments must take into account regional aspects, as the impacts of water consumption unfold in a highly regional context [11,37,38]. For water consumption, data are often available at a higher level of resolution than the typical national classification used in input–output databases. In such cases, disaggregating the lower-resolution input–output database into the classification of the higher-resolution externalities data can improve the accuracy of assessments based on input–output analysis [33,39].

Table 1. Overview of common input–output databases for sustainability assessments, and the number of agricultural primary production categories included in each (selection).

Database	Categories *	Author(s)	Comment
Eora	Varying. For most regions, only one category is included, while for a few regions, several categories are included.	Lenzen et al. [40,41]	Few regions in the Eora database have more than 14 categories related to agricultural primary production (as does GLORIA), while most regions have only one category related to agricultural primary production. However, consistent classification across all regions makes it difficult to automate processes (e.g., compiling satellite accounts).
Eora26	1 of 26	Lenzen et al. [40,41]	A single generic category Agriculture.
Exiobase	8 of 200	Stadler et al. [42]	(1) Paddy rice; (2) Wheat; (3) Cereal grains n.e.c.; (4) Vegetables, fruit, nuts; (5) Oil seeds; (6) Sugar cane, sugar beet; (7) Plant-based fibres; (8) Crops n.e.c.;
FABIO	64 of 130	Bruckner et al. [43]	FABIO features 64 categories of agricultural primary production. FABIO does not include non-food or non-agriculture categories.
GLORIA	14 of 120	Lenzen et al. [44,45]	(1) Wheat; (2) Maize; (3) Cereals n.e.c.; (4) Leguminous crops and oil seeds; (5) Rice; (6) Vegetables, roots, tubers; (7) Sugar beet and cane; (8) Tobacco; (9) Fibre crops; (10) Crops n.e.c.; (11) Grapes; (12) Fruits and nuts; (13) Beverage crops (coffee, tea, etc.); (14) Spices, aromatic, drug and pharmaceutical crops;

* The number of agricultural primary production categories per region and the total number of categories per region.

Table 1. Overview of common input–output databases for sustainability assessments, and the number of agricultural primary production categories included in each (selection).

Database	Categories *	Author(s)	Comment
Eora	Varying. For most regions, only one category is included, while for a few regions, several categories are included.	Lenzen et al. [40,41]	Few regions in the Eora database have more than 14 categories related to agricultural primary production (as does GLORIA), while most regions have only one category related to agricultural primary production. However, consistent classification across all regions makes it difficult to automate processes (e.g., compiling satellite accounts).
Eora26	1 of 26	Lenzen et al. [40,41]	A single generic category Agriculture.
Exiobase	8 of 200	Stadler et al. [42]	(1) Paddy rice; (2) Wheat; (3) Cereal grains n.e.c.; (4) Vegetables, fruit, nuts; (5) Oil seeds; (6) Sugar cane, sugar beet; (7) Plant-based fibres; (8) Crops n.e.c.;
FABIO	64 of 130	Bruckner et al. [43]	FABIO features 64 categories of agricultural primary production. FABIO does not include non-food or non-agriculture categories.
GLORIA	14 of 120	Lenzen et al. [44,45]	(1) Wheat; (2) Maize; (3) Cereals n.e.c.; (4) Leguminous crops and oil seeds; (5) Rice; (6) Vegetables, roots, tubers; (7) Sugar beet and cane; (8) Tobacco; (9) Fibre crops; (10) Crops n.e.c.; (11) Grapes; (12) Fruits and nuts; (13) Beverage crops (coffee, tea, etc.); (14) Spices, aromatic, drug and pharmaceutical crops;

* The number of agricultural primary production categories per region and the total number of categories per region.

The main objective of this study was to develop and demonstrate an improved approach to assess the water consumption and related impacts of agricultural primary production in input–output analyses. The agricultural primary production categories were given particular attention due to their importance in global water consumption and related impacts. To achieve this objective, the spatially highly resolved Spatial Crop Production Model (MapSPAM; Section 2.1.2) was integrated into the Global Resource Input–Output Assessment (GLORIA; Section 2.1.1) database. The resulting Input–Output Global Hybrid Assessment of Agricultural Primary Production (IO-GHAAPP) was deployed in a case study to assess the exceedance of safe operating spaces of watersheds worldwide as a result of German consumption. Achieving the objective involved the following main working steps:

1.

Integration of the MapSPAM datasets in the GLORIA database: Disaggregating the monetary input–output data from the GLORIA database into the classification of crops of agricultural primary production from the MapSPAM datasets via prorating.

2.

Compilation of water extensions using watershed-level data: Compilation of water extensions for the disaggregated input–output data based on watershed-level data for agricultural primary production, water consumption per agricultural primary production and characterisation factors.

In this study, the term prorating refers to the subdivision of the monetary transaction values in the input–output dataset into specific crops of agricultural primary production based on the monetary value of the crops produced by that category as given in the MapSPAM datasets (Section 2.3.2). Agricultural primary production is considered to be the process of producing agricultural goods by photosynthesis [46]. Water consumption refers to blue-water consumption, which is typically defined as “fresh surface and groundwater, in other words, the water in freshwater lakes, rivers and aquifers” [47]. The term category has been adapted for the categories in the GLORIA database in order to follow the native terminology of the GLORIA database. This study adapts the method of Bunsen et al. [12], which defines water footprint as the potential environmental impact of water consumption in relation to exceeding the safe operating space of a watershed.

Section 2 provides detailed information on the datasets, tools and methods used in this study. Section 3 presents the results, highlighting the differences between state-of-the-art approaches and the IO-GHAAPP approach. The results are discussed in Section 4, and the conclusions are presented in Section 5.

2. Data Sets, Methods and Tools

The methodology section is structured as follows: Section 2.1 gives an overview of the datasets used. Section 2.2 explains how these datasets were made compatible with each other. Section 2.3 describes both a state-of-the-art approach (Section 2.3.1) and the IO-GHAAPP approach (Section 2.3.2). Section 2.4 describes the methodology of a case study using the IO-GHAAPP approach. Section 2.5 summarises the tools used.

Figure 1 provides a comparative schematic overview between the state-of-the-art and the IO-GHAAPP approaches. The state-of-the-art approach uses off-the-shelf input–output data and national-level data to compile extensions for agricultural primary production, water consumption and water footprint. The IO-GHAAPP approach has a higher resolution of monetary input–output data and compiles extensions for agricultural primary production, water consumption and water footprint at the watershed level. In addition, Figure A1 provides a comprehensive overview of the IO-GHAAPP approach through a detailed flowchart.

The reference year for the assessment is 2010, which is the most-recent common year for all datasets.

2.1. Data Sets

2.1.1. Monetary Input–Output Data (GLORIA)

This study used release 055 of the GLORIA database [44,45]. GLORIA comprises 164 regions and 120 categories per region, resulting in a total of 19,680 region–category combinations. Its transaction matrix contains over 385 million elements in current USD 1000. GLORIA is currently the most-comprehensive generic input–output database for sustainability assessment with the highest number of categories dedicated to agricultural primary production among all available generic input–output databases for sustainability assessment (Table 1). Release 055 of the GLORIA database is provided as a homogeneous multi-regional supply-use-table (MR-SUT) with basic price valuation that excludes co-production from the supply blocks. The categories of the supply and use blocks are homogeneous, and the use blocks of the MR-SUT represent the corresponding environmentally extended multi-regional input–output table.

2.1.2. Agricultural Primary Production Produce and Monetary Value (MapSPAM)

Data on agricultural primary production, including produce and monetary values, were obtained from the MapSPAM datasets [48]. The MapSPAM datasets cover 41 specific crops or crop groups, including 27 food crops, 14 non-food crops and one additional “rest” group (42 in total), with a spatial resolution of 2160 rows × 4320 columns, resulting in 9,331,200 grid cells. The MapSPAM dataset was selected because its geo-referenced information allows for the application of spatially highly differentiated characterisation factors, e.g., on a watershed level. This is important because some crops, such as cotton, can be very water-intensive. However, in some cotton-producing countries, specific watersheds are very water scarce and therefore have high characterisation factors, e.g., in terms of exceeding local safe operating spaces. National average characterisation factors would not take this into account.

2.1.3. Water Consumption of Agricultural Primary Production

A dataset on blue-water consumption per tonne of agricultural primary produce from Pfister and Bayer [49] was used in this study. The dataset was chosen because it provides spatially explicit values at both watershed and national levels.

2.1.4. Characterisation Factors

Characterisation factors from Bunsen et al. [12] were used to assess the water footprint in terms of exceeding local safe operating spaces of watersheds. These characterisation factors reflect the extent to which water use in a particular watershed exceeds the watershed’s local safe operating space. As a result, the water footprint in this study can be interpreted as a water scarcity footprint in terms of exceeding the local safe operating spaces of the watersheds.

2.1.5. Regional Boundaries

The boundaries for the regions in the GLORIA database were determined using vector maps from the dataset Admin 0—Countries provided by Natural Earth Data [50].

2.2. Data Concordance

2.2.1. Regions

The concordance between the regional boundaries (Section 2.1.5) used to extract data from the MapSPAM datasets (Section 2.1.2) and the regions in the GLORIA database (Section 2.1.1) was manually mapped.

2.2.2. GLORIA Categories, MapSPAM Crops and FAO Crops

The MapSPAM documentation contains the concordance between MapSPAM crops and the Food and Agriculture Organization (FAO) of the United Nations crops also used by Pfister and Bayer (Section 2.1.3). The concordance between MapSPAM crops and the categories in the GLORIA database was manually mapped.

Table A1. Concordance between the GLORIA categories, MapSPAM crops and the FAO classification. All columns except GLORIA category are copied from the file 4-Methodology-Crops-of-MapSPAM-2005-2015-02-26.csv of the MapSPAM documentation [48]. The table serves as a complement to Section 2.2.2 and provides further details on the concordance between the crop datasets.

MapSPAM Short	MapSPAM Long	FAO Name	FAO Crop Code	Group	Food or Non-Food	GLORIA Category
whea	Wheat	wheat	15	cereals	food	001. Growing wheat
maiz	Maize	maize	56	cereals	food	002. Growing maize
barl	Barley	barley	44	cereals	food	003. Growing cereals n.e.c
pmil	Pearl millet	millet	79	cereals	food	003. Growing cereals n.e.c
smil	Small millet	millet	79	cereals	food	003. Growing cereals n.e.c
sorg	Sorghum	sorghum	83	cereals	food	003. Growing cereals n.e.c
ocer	Other cereals	other cereals ++	68, 71, 75, 89, 92, 94, 97, 101, 103, 108	cereals	food	003. Growing cereals n.e.c
bean	Bean	beans, dry	176	pulses	food	004. Growing leguminous crops and oil seeds
chic	Chickpea	chickpea	191	pulses	food	004. Growing leguminous crops and oil seeds
cowp	Cowpea	cowpea	195	pulses	food	004. Growing leguminous crops and oil seeds
pige	Pigeonpea	pigeon pea	197	pulses	food	004. Growing leguminous crops and oil seeds
lent	Lentil	lentils	201	pulses	food	004. Growing leguminous crops and oil seeds
opul	Other pulses	broad beans ++	181, 187, 203, 205, 210, 211	pulses	food	004. Growing leguminous crops and oil seeds
soyb	Soybean	soybean	236	oilcrops	food	004. Growing leguminous crops and oil seeds
grou	Groundnut	groundnut, with shell	242	oilcrops	food	004. Growing leguminous crops and oil seeds
cnut	Coconut	coconut	249	oilcrops	food	004. Growing leguminous crops and oil seeds
sunf	Sunflower	sunflower seed	267	oilcrops	non-food	004. Growing leguminous crops and oil seeds
rape	Rapeseed	rapeseed	270, 292	oilcrops	non-food	004. Growing leguminous crops and oil seeds
sesa	Sesameseed	sesame seed	289	oilcrops	non-food	004. Growing leguminous crops and oil seeds
ooil	Other oil crops	olives ++	260–339, not 311 and the other oilcrops above	oilcrops	non-food	004. Growing leguminous crops and oil seeds
rice	Rice	rice	27	cereals	food	005. Growing rice
pota	Potato	potato	116	roots & tubers or starchy roots	food	006. Growing vegetables, roots, tubers
swpo	Sweet potato	sweet potato	122	roots & tubers or starchy roots	food	006. Growing vegetables, roots, tubers
yams	Yams	yam	137	roots & tubers or starchy roots	food	006. Growing vegetables, roots, tubers
cass	Cassava	cassava	125	roots & tubers or starchy roots	food	006. Growing vegetables, roots, tubers
orts	Other roots	yautia ++	135, 136, 149	roots & tubers or starchy roots	food	006. Growing vegetables, roots, tubers
vege	Vegetables	cabbages and other brassicas ++	358–463	vegetables	food	006. Growing vegetables, roots, tubers
sugc	Sugarcane	sugar cane	156	sugar crops	non-food	007. Growing sugar beet and cane
sugb	Sugarbeet	sugarbeet	157	sugar crops	non-food	007. Growing sugar beet and cane
toba	Tobacco	tobacco leaves	826	stimulant	non-food	008. Growing tobacco
cott	Cotton	seed cotton	328	fibres	non-food	009. Growing fibre crops
ofib	Other fibre crops	other fibres ++	773–821	fibres	non-food	009. Growing fibre crops
rest	Rest of crops	all individual other crops (eg spices, tree nuts, other sugar crops, mate, rubber)	161, 216–234, 671,677–839		non-food	010. Growing crops n.e.c.
oilp	Oilpalm	palmoil	254	oilcrops	non-food	012. Growing fruits and nuts
bana	Banana	banana	486	fruits	food	012. Growing fruits and nuts
plnt	Plantain	plantain	489	fruits	food	012. Growing fruits and nuts
trof	Tropical fruit	oranges ++	490–512, 567–591, 600–603	fruits	food	012. Growing fruits and nuts
temf	Temperate fruit	apples ++	515–560, 592, 619	fruits	food	012. Growing fruits and nuts
acof	Arabica coffee	coffee	656	stimulant	non-food	013. Growing beverage crops coffee, tea etc
rcof	Robusta coffee	coffee	656	stimulant	non-food	013. Growing beverage crops coffee, tea etc
coco	Cocoa	cocoa	661	stimulant	non-food	013. Growing beverage crops coffee, tea etc
teas	Tea	tea	667	stimulant	non-food	013. Growing beverage crops coffee, tea etc

gives an overview of the concordance between the datasets. GLORIA categories representing a single MapSPAM crop include Growing wheat; Growing maize; Growing rice; Growing tobacco and Growing crops n.e.c. GLORIA categories representing several MapSPAM crops include Growing cereals n.e.c.; Growing leguminous crops and oil seeds; Growing vegetables, roots, tubers; Growing sugar beet and cane; Growing fibre crops; Growing fruits and nuts and Growing beverage crops, coffee, tea, etc. No corresponding MapSPAM crop existed for the GLORIA categories Growing grapes and Growing spices, aromatic, drug and pharmaceutical crops. The MapSPAM crops ‘pearl millet’ and ‘small millet’ were grouped into the generic group ‘millet’, while ‘Arabica coffee’ and ‘Robusta coffee’ were grouped into the generic group ‘coffee’. This grouping owes to the FAO classification was used in Pfister and Bayer (Section 2.1.3), which does not distinguish between these crops.

2.3. Approaches

Section 2.3.1 provides an overview of the state-of-the-art approach, and Section 2.3.2 provides an overview of the IO-GHAAPP approach. The state-of-the-art approach is used as a reference in the case study to demonstrate the differences in the water extensions compared to the IO-GHAAPP approach (Section 3.2).

2.3.1. State-of-the-Art Approach

Monetary Input–Output Data

The original GLORIA database as referenced in Section 2.1.1 was used.

Water Extensions

In the state-of-the-art approach, the water consumption and water footprint are calculated at a national level by multiplying the national agricultural primary production (aggregated from IO-GHAAPP approach extensions in Section 2.3.2) with national blue-water consumption values (Section 2.1.3) and national characterisation factors (Section 2.1.4).

2.3.2. IO-GHAAPP Approach

The IO-GHAAPP approach disaggregates the monetary input–output data from the GLORIA database into the crop classification of agricultural primary production from the MapSPAM datasets via prorating. This allows for (1) the disaggregation of the monetary input–output data in the GLORIA database (Section 2.3.2) and (2) the calculation of high-resolution regional satellite accounts for water consumption and water footprint (Section 2.3.2). To link the data in the digital raster graphic files (GeoTIFF) from the MapSPAM dataset to the regions in the GLORIA database, the blue-water consumption values (Section 2.1.3), the characterisation factors (Section 2.1.4), and country boundaries (Section 2.1.5) were all intersected, resulting in a spatial template with over 55,000 geometries (Figure A6). For each geometry, the template was assigned information on the corresponding country, GLORIA region, blue-water consumption and a characterisation factor. The Python Rasterstats module [51] was then used to extract the production in tonnes, area harvested in hectares and the monetary value of harvested food and non-food crops in international dollars per hectare from the raster graphic files in the MapSPAM dataset based on the vector geometries in the template. Figure A1 provides a comprehensive overview of the IO-GHAAPP approach through a detailed flowchart.

Disaggregating the Input–Output Data via Prorating

The method used to disaggregate the input–output data has been described in various works, e.g., by the UN Statistics Division [52], and is also known as input–output-based hybrid analysis. Hybrid analysis is a common method used in input–output analysis to disaggregate data from a lower level of aggregation to a higher level of aggregation. In input–output-based hybrid analysis, the original output (GLORIA) is allocated to several sub-outputs (MapSPAM) based on shares of the the sub-outputs in the original output (Figure 2). For example, suppose there is an output fibre crops in the original input–output table, and the higher-resolution dataset contains the fibre crops jute and cotton, both of which are aggregated in the output fibre crops of the original input–output table. Suppose that jute has a share of 0.05 in the monetary value of the output of the fibre crops category and that cotton has a share of 0.95 in the monetary value of the output of the fibre crops category. In this case, 5% and 95% of the transaction value in the input–output table would be allocated to the subdivided outputs jute and cotton, respectively. This process is also known as prorating.

The spatial template (Section 2.3.2) was used to extract the harvested area and production value per area harvested for the 42 crops in the MapSPAM dataset. The area harvested was extracted from the MapSPAM GeoTIFFs spam2010v2r0 global harvested area, while the value of production per area harvested was obtained separately for food crops (VP_FO_AR) and non-food crops (VP_NF_AR) from the MapSPAM GeoTIFFs spam2010v2r0 global value production aggregated. The harvested area was then multiplied by the monetary value per harvested area for food crops or non-food crops, giving an approximation of the total value of each crop in that geometry in international USD. The next step was to aggregate the monetary production value of each crop in the MapSPAM dataset for over 55,000 geometries into the 164 regions of the GLORIA database.

Prorating vectors were constructed for each GLORIA region and category by dividing the monetary production value of each MapSPAM crop in a GLORIA region and category by the sum of the values of all crops in that GLORIA region and category. For example, if the monetary production value of cotton and jute in a GLORIA region and category are int. USD 5 and int. USD 1, respectively, their respective prorating factors for the given region and category in question would be $5÷6=0.83$ and $1÷6=0.17$. The sum of all values per prorating vector is always 1.

For 178 region–category combinations, representing 7.8% of the region–category combinations pertaining to agricultural primary production, no corresponding monetary production value could be calculated from the MapSPAM database. Examples of such combinations are 005. Afghanistan—013. Growing beverage crops coffee, tea etc., 047. Denmark—009. Growing fibre crops, and 129. Qatar—003. Growing cereals n.e.c. This indicates that either the corresponding monetary value or the area harvested was zero in the MapSPAM dataset. However, in some cases, the corresponding GLORIA value was not zero, indicating discrepancies between the two data sources. In cases where no corresponding monetary production value could be calculated from the MapSPAM dataset, setting the GLORIA value to zero was not a viable option as it would have changed the overall totals in the GLORIA database. Manually assigning GLORIA values to a specific MapSPAM crop would have required subjective value judgments, which could have led to incorrect assignment decisions. Instead, assigning the average prorating vector of the corresponding sub-region allowed an objective and automated way of dealing with data discrepancies while maintaining the overall integrity of the GLORIA database. This approach is also more transparent and reproducible than manually assigning GLORIA values to a MapSPAM crop. The Natural Earth Data (Section 2.1.5) dataset provides several classifications of sub-regions, including those defined by the World Bank and the United Nations.

Extensions (Satellite Accounts)

The compilation of the environmental extensions includes agricultural primary production in $\mathrm{metric}\mathrm{tonnes}$, blue-water consumption of agricultural primary production in ${\mathrm{million}\mathrm{m}}^3$ and water footprint of agricultural primary production in ${\mathrm{million}\mathrm{m}}_{\mathrm{weighted}}^3$. First, the agricultural primary production was extracted from the MapSPAM file spam2010v2r0_global_prod.geotiff using the spatial template (Section 2.3.2). Next, the agricultural primary production within each geometry was multiplied by the blue-water consumption of the corresponding crop per tonne of production (Section 2.1.3) to obtain the total blue-water consumption of the crops. To determine the blue-water consumption of agricultural primary production, the crops in Pfister and Bayer [49] were matched to the corresponding MapSPAM crops (Section 2.2.1) and their mean values were calculated. Next, the total blue-water consumption of the agricultural primary production in each geometry was multiplied by the corresponding characterisation factors (Section 2.1.4) to calculate the water footprint. In a final step, the water extents were aggregated to the resolution of the input–output data in the IO-GHAAPP approach.

2.4. Case Study

A case study was conducted using the IO-GHAAPP approach, in which the method of Bunsen et al. [11] was adapted to assess the relationship between production and consumption and the exceedance of safe operating spaces of watersheds worldwide. The case study was conducted for Germany and the GLORIA category Growing leguminous crops and oil seeds, which includes the MapSPAM crops Bean, Chickpea, Cowpea, Pigeonpea, Lentil, Other pulses, Soybean, Groundnut, Coconut, Sunflower, Rapeseed, Sesameseed and Other oil crops. This GLORIA category was chosen because the MapSPAM model has particularly good data for disaggregating this category. The first step of the case study was to calculate the production-based inventory for Germany (Section 2.4). Subsequently, the contributions of the GLORIA regions and the MapSPAM crops to the production-based inventory of Germany were allocated to the watersheds of the producing regions. This allocation was consumption-weighted, taking into account the share of the watersheds in the total water consumption of a MapSPAM crop within the GLORIA regions.

The calculation of the production-based inventory follows the conventional steps outlined, for example, by Bunsen and Finkbeiner [31]. The matrix T contains information about the outputs (production) and inputs (consumption) of the categories. The vector x contains the total output of these categories—that is, both the output consumed by other categories and the output consumed through final demand. A matrix $L={(I-A)}^{-1}$, also known as the Leontief inverse, is computed to determine the inputs required by the categories to produce one unit of their output across an infinite number of production layers. The matrix I is an identity matrix with the same dimensions as T, while the technical coefficient matrix A is given by $A=T{x}^{-1}$ and specifies the inputs required by each category to produce one unit of output. The externalities associated with a unit of output of the categories are given by a vector $f=q{x}^{-1}$, where q is a vector containing the total externalities associated with the output of the categories. The matrix F contains multipliers that indicate how much of a particular externality is associated with a unit of output in each category across an infinite number of production layers. This matrix is given by $F=\widehat{f}{(I-A)}^{-1}=\widehat{f}L$, where the vector $\widehat{f}$ is the diagonalisation of f. By summing the rows of F multiplied by a given final demand, the production-based inventory associated with that final demand is obtained.

2.5. Tools

All calculations in this study were performed using Python with the NumPy [53], Pandas [54] and GeoPandas [55] packages. Results plots were generated with Matplotlib [56]. Some computations were run on the High-Performance Computing Cluster of Technische Universität Berlin. The IO-GHAAPP approach dataset presented in this study is openly available via Zenodo at https://doi.org/10.5281/zenodo.7835200.

3. Results

Section 3.1 and Section 3.2 provide overviews of the monetary transactions and water extensions in the IO-GHAAPP approach, respectively. These overviews are presented for the global scale as well as for the United States of America (USA) and India, both of which are ranked among the top five regions in terms of the monetary value of production (2nd and 3rd, respectively), production volume (4th and 3rd, respectively), water consumption (4th and 1st, respectively), and water footprint (5th and 1st, respectively) of agricultural primary production. Given the importance of both the USA and India in relation to these four dimensions, both countries were considered good examples to demonstrate the additional layer of information in the IO-GAAPP approach. In addition, Section 3.3 provides a comparison of the water consumption and water footprint of the state-of-the-art and the IO-GHAAPP approaches for the 15 regions with the highest water consumption. Finally, Section 3.4 presents the findings of the IO-GHAAPP case study on the exceedance of safe operating spaces of watersheds globally.

3.1. IO-GHAAPP Approach—Transaction Matrix and Final Demand

The IO-GHAAPP approach consists of 150 categories and 164 regions, resulting in a total of 24,600 region–category combinations. Forty-two of the categories are dedicated to agricultural primary production, amounting to 28% of the categories. In comparison, the GLORIA databases consist of 120 categories and 164 regions, resulting in a total of 19,680 region–category combinations. Fourteen of these categories are dedicated to agricultural primary production, which represents 12% of all categories.

3.1.1. Disaggregated Monetary Input–Output Data

The total monetary production volume in the GLORIA database amounts to USD 62.23 trillion, out of which USD 1.54 trillion is attributed to agricultural primary production (2.47%). While these totals remain constant, the monetary production volumes of the agricultural primary production GLORIA categories are further disaggregated based on the MapSPAM model (Section 2.3.2). Figure 3 as well as the bullet list below show the GLORIA categories for which several MapSPAM crops were available, along with the corresponding MapSPAM crops and the estimated monetary share within the GLORIA category at the global level.

• Growing cereals n.e.c.: Barley (35%), Sorghum (27%), Other cereals (21%), Pearl millet (10%) and Small millet (6%)
• Growing leguminous crops and oil seeds: Rapeseed (35%), Sunflower (24%), Other oil crops (24%), Sunflower (24%), Sesameseed (7%), Soybean (6%) and Coconut (1%)
• Growing vegetables, roots, tubers: Vegetables (64%), Potato (24%), Sweet potato (6%), Cassava (3%), Yams (1%) and Other roots (1%)
• Growing sugar beet and cane: Sugarcane (88%) and Sugarbeet (12%)
• Growing fibre crops: Cotton (90%) and Other fibre crops (10%)
• Growing fruits and nuts: Temperate fruit (42%), Tropical fruit (30%), Oilpalm (24%), Banana (3%) and Plantain (1%)
• Growing beverage crops coffee, tea, etc.: Arabica coffee (44%), Tea (39%), Robusta coffee (14%) and Cocoa (3%)

The share of the monetary value of agricultural primary production in most regions differs from the global pattern. This is illustrated by the examples of the USA and India.

• United States of America

The main differences between the USA mix (Figure A2) and the global mix (Figure 3) are the much larger share of the GLORIA category Maize and much smaller share of the GLORIA category Rice in the USA mix. In addition, the USA mix has a larger share in the GLORIA category Leguminous crops and oil seeds, with a different subdivision compared to the global mix. The MapSPAM crop Sunflower (36%) is more relevant than Rapeseed (32%), and Sesame seed has no relevance. In the GLORIA category Sugar beet and cane, the MapSPAM crop Sugar beet (81%) is more relevant than Sugar cane (18%), which is the opposite in the global mix. The GLORIA category Beverage crops has no relevance in the USA mix.

• India

The main differences between the Indian mix (Figure A3) and the global mix (Figure 3) are that the GLORIA category Rice has a larger share, while the GLORIA category Maize has a smaller share. In the GLORIA category Sugar beet and cane, only the MapSPAM crop Sugar cane is relevant, and in the GLORIA category Fibre crops, only the MapSPAM crop Cotton is relevant. In addition, the GLORIA category Leguminous Crops and Oilseeds has a higher share of the MapSPAM crop Rape Seed (57%) than the global mix, followed by the MapSPAM crop Sesame Seed (18%). In contrast to the global mix, the GLORIA category Vegetables, roots and tubers is dominated by the MapSPAM crops Vegetables (57%) and Potatoes (40%). Furthermore, the GLORIA category Fruit and nuts is dominated by the MapSPAM crop Tropical fruit (69%) rather than Temperate fruit (28%).

3.2. Extensions (Satellite Accounts)—Production Volume, Water Consumption and Water Footprint

Figure 4 illustrates the global production volumes of MapSPAM as well as the corresponding global water consumption and water footprint. The most-relevant GLORIA categories in terms of production volume are the GLORIA categories Sugar beet and cane (Sugarcane: 88%; Sugarbeet: 12%), Vegetables, roots and tubers (Vegetables: 64%; Potato 24%; Sweet potato: 6.4%), Fruit and nuts (Temperate fruit: 42%; Tropical fruit: 30%; Oilpalm: 24%), Maize, Wheat and Rice. The MapSPAM crops Soybean (46%) and Barley (48%) make the most-significant contributions to the GLORIA categories Leguminous crops and oil seeds and Cereals n.e.c., respectively. In terms of water consumption and water footprint, the relevance of the GLORIA categories Vegetables, roots and tubers, Maize and especially Sugar beet and cane decreases. On the other hand, the importance of the GLORIA categories Wheat, Leguminous crops and oil seeds and especially Fibre crops increases. In the GLORIA category Leguminous crops and oil seeds, the MapSPAM crop Rapeseed dominates in terms of production volume, while the MapSPAM crop Sunflower is the most important in terms of water consumption. The MapSPAM crop Groundnut has the highest relevance in terms of water footprint within this GLORIA category. In the GLORIA category Fruits and nuts, the MapSPAM crop Oilpalm has a lower relevance in terms of water consumption and especially water footprint compared to its production volume. Analogous to the breakdown of the monetary input–output data outlined in Section 3.1, the regional production volumes, water consumption and water footprints differ from the global mix. To illustrate this, the regional mixes of the USA (Figure A4) and India (Figure A5) are examined.

• United States of America

The USA mix (Figure A4) shows a significant difference from the global mix (Figure 4) in terms of production, with almost half of the total production volume contributed by the GLORIA category Maize. However, this GLORIA category is much less relevant in terms of water consumption and even less so in terms of water footprint. The second most-relevant GLORIA category in terms of production volume is Leguminous crops and oil seeds, with Soybean and Bean accounting for the highest shares within this GLORIA category. Although the GLORIA category Sugar beet and cane has a comparatively high share in production volume, it has a lower share in water consumption and water footprint. On the other hand, the GLORIA category Fibre crops and the MapSPAM crop Cotton, which are not very relevant in terms of mass, are highly relevant in terms of water consumption and even more so in terms of water footprint.

• India

The India mix (Figure A5) differs significantly from the global mix (Figure 4) in terms of production, with the GLORIA category Sugar beet and cane and the MapSPAM crop Sugarcane contributing more than a third of the total production volume. However, this GLORIA category is less relevant in terms of water consumption and even less so in terms of water footprint. In particular, the GLORIA category Fibre crops and the MapSPAM crop Cotton become more relevant in terms of water consumption and water footprint. Compared to production volume, the GLORIA categories Leguminous crops and oil seeds become more relevant in terms of water consumption and water footprint. Although the MapSPAM crop Soybean has the highest relevance within the GLORIA category Leguminous crops and oil seeds in terms of production volume, the MapSPAM crop Groundnut is of similar relevance in terms of water consumption and of much higher relevance in terms of water footprint.

3.3. Water Consumption and Water Footprint—IO-GHAAPP versus State-of-the-Art

In the native GLORIA satellite accounts, agricultural primary production accounts for 89.6% of water consumption. In the IO-GHAAPP approach satellite accounts, agricultural primary production accounts for 91.5% of water consumption. Although these values are similar, there are significant regional differences in the regions’ water consumption and especially in their water footprints (Figure 5).

• Blue-Water Consumption

Figure 5 illustrates that all regions show higher water consumption values when the state-of-the-art approach is applied, with some (such as Pakistan) showing small deviations of around 5% from the IO-GHAAPP approach and others showing larger deviations of 20% or more, such as India, China and the USA.

• Water Footprint

Furthermore, it can be seen from Figure 5 that there is no consistent pattern when it comes to estimating different water footprints between the state-of-the-art and the IO-GHAAPP approaches. For GLORIA regions such as India, China, Pakistan, USA, Morocco and Saudi Arabia, the water footprint is higher when the IO-GHAAPP approach is applied. On the other hand, for regions such as Iran, Egypt, Spain, Uzbekistan, Turkmenistan, Turkey, Syria, Mexico and Iraq, the water footprint is higher when the state-of-the-art approach is applied. In particular, for Egypt and Turkmenistan, the water footprint is several times higher when using the state-of-the-art approach than when using the IO-GHAAPP approach.

This is probably due to the fact that both Egypt and Turkmenistan have watersheds in which the exceedance of local safe operating spaces is extremely high. This is despite the fact that Bunsen et al. [12] took into account outliers and set the minimum and maximum values for the magnitude of the exceedance of local safe operating spaces to the 0.05 and 0.95 quantile, respectively. It seems that when applying the IO-GHAAPP approach, the use of local characterisation factors leads to significantly higher water footprint values than the national average characterisation factors in the state-of-the-art approach.

3.4. IO-GHAAPP Case Study

The GLORIA category of Leguminous crops and oil seeds contributes 12%, 26% and 13%, respectively, to the production volume, water consumption and water footprint of the German production-based inventory. Figure 6 and Figure 7 showcase the global contributions of the MapSPAM crops Rapeseed and Soybean from the GLORIA category Leguminous crops and oil seeds of the German production-based water consumption inventory and the global water footprint hotspots of these crops.

• Rapeseed

The GLORIA regions Germany (DEU), France (FRA), India (IND), Rest of Asia-Pacific (XAS) and Australia (AUS) have the highest contributions to the production-based water consumption inventory of Germany resulting from the production of Rapeseed. In the case of the production-based water footprint inventory, the same regions, in the order of India (IND), France (FRA), Australia (AUS), Germany (DEU) and Rest of Asia-Pacific (XAS), have the largest contributions resulting from the production of Rapeseed.

• Soybean

The GLORIA regions United States of America (USA), Brazil (BRA), Rest of Asia-Pacific (XAS), India (IND) and Argentina (ARG) have the highest contributions to the production-based water consumption inventory of Germany resulting from the production of Soybean. In the case of the production-based water footprint inventory, the same regions, in the order of United States of America (USA), India (IND), Rest of Asia-Pacific (XAS), Iran (IRN) and Ukraine (UKR) have the largest contributions resulting from the production of Soybean.

4. Discussion

As introduced in Section 1, previous studies have emphasised the importance of improving the resolution of input–output data [28] and have highlighted the potential benefits of disaggregation in achieving this goal [33]. This section discusses the results of integrating the MapSPAM model into the GLORIA database (Section 4.1), as well as the IO-GHAAPP approach to compiling satellites (Section 4.2). It also reflects on the case study (Section 4.3) and discusses limitations of the IO-GHAAPP approach and future work (Section 4.4).

4.1. Integration of the MapSPAM Model into the GLORIA Database

The integration of the MapSPAM model into the GLORIA database adds an additional layer of information beyond the existing fourteen categories of agricultural primary production. The added informative value is demonstrated when analysing the monetary subdivision of the newly added MapSPAM crops within the GLORIA categories of agricultural primary production at both global and country levels (Section 3.1.1). The analyses reveal information that would otherwise be indistinguishable, such as the different proportions of the MapSPAM crops within the GLORIA categories of the global mix (Figure 3) and local mixes (Figure A2 and Figure A3).

However, information is added not only on a monetary level but also on the production volume of agricultural primary production and the associated water consumption and water footprint. The examples in Section 3.2 show how the added layer of information can provide a more comprehensive and nuanced picture of traded agricultural primary production and its associated water consumption and water footprint. Such information can potentially support policies and decision making that take into account more spatially nuanced information.

Overall, the results suggest that the integration of the MapSPAM model into the GLORIA database provides an informational advantage for several GLORIA categories, such as Leguminous crops and oil seeds, Fruits and nuts, Crops n.e.c. and Sugar beet and cane. However, some of the most-relevant GLORIA categories in terms of monetary value of agricultural primary product, such as Rice, Maize and Wheat, are not further disaggregated by the integration of the MapSPAM model into the GLORIA database. This is because these crops already have specific categories in the GLORIA database, and the integration of the MapSPAM model does not provide any additional informative value beyond the IO-GHAAPP approach in compiling water extensions.

4.2. IO-GHAAP Approach in Compiling Water Extensions

To the best of the authors’ knowledge, this study represents the first application of input–output analysis using water extensions based on sub-national water consumption values and characterisation factors. Previous studies have relied on national production values from the FAOSTAT database, which were then multiplied by national water consumption values from either Mekonnen and Hoekstra [47] or Pfister et al. [49]. Section 3.3 shows that the state-of-the-art and IO-GHAAPP approaches yield significant differences in water consumption and water footprint inventories. Future work could investigate the extent to which these results depend on the method of Bunsen et al. [12] or whether other impact assessments methods for water scarcity footprinting such as AWARE [37], WAVE+ [38] or WSI [57] would lead to similar effects. This could be the case if the approaches for calculating national average characterisation factors were different between these methods. As the focus of this study was on the integration of the MapSPAM model into the GLORIA database, a comparative analysis of different impact assessment models was beyond its scope.

4.3. Case Study

Numerous publications have used input–output analysis to examine the water consumption or water footprints of nations. Lutter et al. [24] and Bunsen et al. [12] allocated production-based inventories to watersheds worldwide. However, these studies were limited by sectoral (category) resolution and were not coupled with a spatial model of agricultural primary production. The results of the case study in Section 3.4 show how the IO-GHAAPP approach provides unprecedented spatial detail. Given the extreme importance of a few nations in global water consumption and water footprints, regional analysis at the sub-national level can improve the information generated by such assessments.

4.4. Limitations and Suggestions for Future Work

The integration of the MapSPAM model into the GLORIA database presents numerous benefits, yet informed usage is advised. It is important to note that homogeneity was assumed throughout the disaggregation process. The MapSPAM production mix (output) of a region was assumed to be identical in all consuming (input) regions. Although the aim of this study was to mitigate the implications of proportionality in input–output analysis, it is important to recognise that the IO-GHAAPP approach is not exempt from similar limitations. Overall, however, the IO-GHAAPP approach offers a new perspective not only in terms of disaggregating the monetary input–output data but also in its regionalised compilation of water consumption for agricultural primary production and its characterisation.

In addition, practitioners should note that only the GLORIA categories of agricultural primary production, which account for 70% to 90% of global water use, have been disaggregated (Section 1). Furthermore, the GLORIA categories of Grapes and Spices, aromatic, drug and pharmaceutical crops have no corresponding MapSPAM crops (Section 2.2.2). It is possible that these crops are included in the MapSPAM crop Rest of crops or Temperate fruits. However, Grapes and Spices, aromatic, drug and pharmaceutical crops are not known to contribute significantly to global water consumption or scarcity. Nevertheless, a higher level of differentiation would further improve the IO-GHAAPP approach.

The MapSPAM model distinguishes between irrigated and non-irrigated crops. However, the data presented by Pfister et al. [49] are only available for total crop production without distinguishing between the two proportions. The availability of water consumption data specifically for the irrigated portion of crops could further improve the accuracy of the IO-GHAAPP approach.

5. Conclusions

The main objective of this study was to improve water consumption and water footprint assessments in input–output analyses. To this end, the novel IO-GHAAPP approach was developed, posing two original contributions to the field of study: first, the integration of the global spatial model for agricultural primary production MapSPAM into the GLORIA database (Section 2.3.2) and second, the compilation of water environmental extensions (satellite accounts for water) based on regional crop production values, regional water consumption values and regional characterisation factors (Section 2.3.2).

Existing global generic input–output databases for sustainability assessments lack comprehensive sectoral information with regard to agricultural primary production. This limitation is significant considering that between 70% and 90% of freshwater consumption takes place in agricultural primary production. Consequently, the lack of detailed sectoral data related to agricultural primary production reduces the informative value of water consumption and water footprint assessments within input–output analysis. This issue is further exacerbated in state-of-the-art approaches, where water footprint values are often calculated based on national-level characterisation factors (as discussed in Section 2.3.1). The IO-GHAAPP database consists of 150 categories and 164 regions, resulting in a total of 24,600 region–category combinations. Forty-two of the categories are dedicated to agricultural primary production (28%). In comparison, the source input–output data consist of 120 categories and 164 regions, resulting in a total of 19,680 region–category combinations, of which 14 are dedicated to agricultural primary production (12%)

The novel information value of the IO-GHAAPP approach was demonstrated in a comparative overview between the distribution of monetary values, production volumes of agricultural primary production and the associated water consumption and water footprint compared to the state-of-the-art approach (Section 3.1). The comparison shows that the IO-GHAAPP approach provides unprecedented detail on agricultural primary production in a generic input–output database. In addition, the case study demonstrated the suitability of the IO-GHAAPP approach to assess the relationship between national consumption and global water use.

Overall, the improvements introduced by the IO-GHAAPP approach can contribute to the development of more region-specific policies aimed at mitigating the impacts of water consumption in agricultural primary production. The demonstration of the IO-GHAAPP approach may inspire future research on other externalities relevant to agricultural primary production in a similar way, such as land use, soil degradation, specific pollutants and others. For widespread practical application, the generation of comprehensive time series using the IO-GHAAPP approach would be a significant improvement.