Policy Research Working Paper 10885 Agriculture Production Potential of Groundwater Irrigation in Sub-Saharan Africa Bhavya Srivastava Ifeanyi N. Edochie Aparajita Goyal Andrew Dabalen Africa Region Office of the Chief Economist August 2024 Policy Research Working Paper 10885 Abstract Sub-Saharan Africa’s low agricultural productivity exac- 8,099 groundwater grids, the study models the production erbates rural poverty. An important investment, the gains from groundwater irrigation for rain-fed croplands. sustainable use of groundwater for irrigation, has the poten- Simulation results indicate that groundwater access could tial to increase agricultural productivity, but the region has increase output by 27.97 to 129.42 percent, contingent been much slower to adopt this irrigation method com- on crop and model conditions. This research facilitates the pared to other regions, despite abundant reserves. This study assessment of the transformative potential of groundwater uses a simulation-driven approach to examine the benefits irrigation and identifies areas in Sub-Saharan Africa where of sustainably utilizing groundwater for irrigation. By map- investments can yield significant returns without depleting ping data from 291,798 global agro-ecological zones to the groundwater table. This paper is a product of the Office of the Chief Economist, Africa Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at bs1088@georgetown.edu. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Agriculture Production Potential of Groundwater Irrigation in Sub-Saharan Africa* Bhavya Ifeanyi N. Aparajita Andrew Srivastava Edochie Goyal Dabalen * Corresponding author: Bhavya Srivastava, email: bs1088@georgetown.edu. Bhavya Srivastava, Depart- ment of Economics, Georgetown University, email: bs1088@georgetown.edu; Ifeanyi N. Edochie, Poverty and Equity Global Practice, World Bank, email: iedochie@worldbank.org; Aparajita Goyal, Africa Chief Economist’s Office, World Bank, email: agoyal3@worldbank.org; Andrew Dabalen, Africa Chief Economist’s Office, World Bank, email: adabalen@worldbank.org. The authors thank Garance Genicot, Vivek Grewal, Reshmaan Hussam, Aude-Sophie Rodella, Lutz Sager, Amal Talbi, Marcus Wijnen, and Esha Zaveri for their valuable comments and feedback. The authors also thank Ashok Anom Dule, Agartha Adubofuor, and Di- ana Jaganjac for excellent research assistance. Funding from the Africa Chief Economist Office, World Bank is gratefully acknowledged. The findings, interpretations, and conclusions expressed in this paper are en- tirely those of the authors. They do not necessarily represent the views of the World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. 1 1 Introduction Pervasive rural poverty in Sub-Saharan Africa is rooted in inadequate agricultural produc- tivity growth (Suri & Udry, 2022). While many developing countries in other regions of the world have effectively raised their agricultural productivity, Sub-Saharan Africa still lags behind. However, enhancing agricultural productivity in Sub-Saharan Africa would not only elevate the incomes of farm households, comprising more than half of the re- gion’s population but also reduce food costs for non-farming populations. This possibil- ity of reduced food costs is especially important because food security remains an ongo- ing problem for 30% of the region’s population (Pfister, Bayer, Koehler, & Hellweg, 2011). Consequently, greater food security would spur broader economic growth by stimulating demand for non-agricultural goods and services, and high productivity would liberate re- sources such as labor for expansion toward other economic sectors. Therefore, improving agricultural productivity in Sub-Saharan Africa remains a crucial strategy for alleviating poverty, fostering inclusive growth, and facilitating structural transformation within the region (Goyal & Nash, 2017). One untapped source of potential agricultural productivity and production growth is the sustainable use of groundwater for irrigation. Groundwater irrigation accounts for 1% of irrigated agricultural land in Africa, compared with 14% in Asia (Altchenko & Vill- holth, 2015). Although international policymakers may be wary of promoting groundwa- ter utilization, sustainable and moderate usage of Sub-Saharan Africa’s vast groundwater resources could significantly enhance agricultural productivity. In this paper, we are interested in measuring the potential of groundwater irrigation for agricultural productivity and production gains. To do so, we first provide an overview of Africa’s groundwater endowment and its relative abundance relative to other parts of the world. We then proceed to study the foregone agricultural productivity of African farmers who rely on rain rather than irrigation for their croplands. The existing literature from Sub-Saharan Africa (henceforth referred to as SSA) docu- ments a substantial increase in crop yields and farm income attributable to the adoption of irrigation technology. Canal irrigation increases farm profits in Rwanda by 43%–62% (Jones, Kondylis, Loeser, & Magruder, 2022). Gravity irrigation through rivers, reservoirs, and dams has also been thoroughly studied. In Ghana, research shows this form of irriga- tion increases crop yields of vegetables and cereals by 3.5–28.6 tons per hectare (Namara et al., 2011) and crop revenues per hectare per season by 300–1700 USD (Acheampong, Ozor, & Sekyi-Annan, 2014). In Malawi, researchers note a 350% increase in annual Maize pro- duction (Concern Universal, 2012) and an increase in agricultural production of 1.25 tons per household Dillon (2011). Further, this irrigation form has an internal rate of return of 14% for vegetable farming in SSA as a whole (Inocencio et al., 2007). More modern tech- nologies such as drip and sprinkler irrigation are associated with a 193–200 kg per hectare increase in Maize yields in Rwanda (Ngango & Hong, 2021).1 1 Refer to Table A2 for a comprehensive summary of the literature on the effects of irrigation on agricultural outcomes in Sub Saharan African countries. 2 However, the literature studying the impacts of groundwater irrigation on agricultural outcomes is scant. In one study in Mali, Burney, Woltering, Burke, Naylor, and Pasternak (2010) find the introduction of solar-powered irrigation pumps is associated with a 1.9 ton increase in the production of crops, which results in a 17% reduction in food insecurity in treatment villages. The scarcity of research is attributable to low rates of groundwater- irrigation adoption in the region. We contribute to filling this gap in the literature by simulating the potential gains in agricultural production through groundwater irrigation across the region. This study makes three important contributions. First, we create a novel database for SSA that combines agricultural data with a gran- ularity of 291,798 grids with aquifer data composed of 8,099 grids. This approach allows us to combine information on agricultural production, productivity, and climatic and soil conditions with information on groundwater availability for each agricultural grid. Second, we provide quantitative evidence on agricultural productivity gains associ- ated with groundwater access. Our simulation framework estimates the gains in agricul- tural production through groundwater irrigation on existing rain-fed croplands. We reach these estimates by determining the proportion of existing crop water deficits in agricul- tural grids that can be mitigated through groundwater irrigation. We find the median gains in production range between 27.97% and 129.42% depending on the crop of interest and methodology used to predict crop yield gains. Third, this is the first study to incorporate sustainability conditions on groundwater extraction. We do so by simulating production gains only in agriculture grids with a suf- ficiently low ratio of groundwater extraction to recharge. Furthermore, our analysis is confined to calculating gains solely from renewable groundwater sources within the re- gion. The remainder of this paper is organized as follows. Section 2 provides a brief back- ground on the need for groundwater irrigation in SSA. Section 3 details the data used for the simulation. Section 4 provides statistics on groundwater availability in the region and the current state of agriculture. Section 5 describes the methodology used for the simula- tion. Section 6 presents and discusses the results. Section 7 conducts several robustness checks. Section 8 discusses groundwater access costs, and section 9 concludes. 2 Background The majority of the population in SSA resides in rural areas, where the primary source of livelihood is rain-fed agriculture. Extreme weather conditions such as droughts, high temperatures, or delayed rainfall can impose severe hardships, especially for the rural population in agriculture-dependent areas (Khan, Kuate, Pongou, & Zhang, 2024). These hardships are further exacerbated by the lack of adaptive measures to climate change. However, SSA harbors a largely untapped reserve of groundwater aquifers, which could mitigate the effects of adverse rainfall events such as droughts. A report suggests 3 countries in this region could supply 130 liters of drinking water per person per day from groundwater aquifers, without exceeding 25% of the long-term average recharge, with most using less than 10% (WaterAid, 2022). In places experiencing groundwater stress, access to shallow groundwater aquifers could enable farmers to irrigate their fields for up to two drought years. Nonetheless, the extraction rate is only 1%–25% of the existing renewable groundwater resources (Wat- erAid, 2022). Therefore, unlike in India, where groundwater irrigation led to a significant decline in resources, tapping into groundwater could be a sustainable choice for SSA, pro- vided that measures are taken early in the process to prevent its over-exploitation. Although water scarcity is a growing issue in SSA, our analysis shows a large, mostly untapped abundance of groundwater resources. The main challenge lies in a lack of knowledge and initiative to utilize these groundwater resources, as well as a policy frame- work for preventing over-extraction. The cost of accessing groundwater may also present a barrier for Africans, given the region’s average GDP per capita of 1700 USD. To address this issue, our analysis focuses on local shallow groundwater aquifers, which are abundant and relatively easier to access, due to their proximity to the earth’s surface. SSA has the highest share of local shallow aquifers (out of its total endowment) globally, comprising 60.77%, closely followed by South Asia at 54.54%. 2.1 Groundwater Aquifers Groundwater is an integral part of meeting the global demand for water. Estimates sug- gest nearly 50% of drinking, 40% of irrigation, and over 30% of industrial global water demands are currently sourced from groundwater (UNESCO and UN-Water, 2022); (Mar- gat & van der Gun, 2013). Aquifers, or groundwater aquifers, store precipitation that seeps through the soil. The aquifer itself is the rock or sediment that collects and holds groundwater in empty spaces underground. Some of these aquifers are easy to access, whereas others lie under rocks that are harder to cut through. Following the definitions of the Water Global Practice at the World Bank, we broadly have four aquifer types that we use in our analysis. These categories are based on various geological and hydrogeological characteristics of aquifers. The four categories of aquifers (or aquifer systems) are complex aquifers, karstic aquifers, major alluvial aquifers, and shallow/local aquifers. A complex aquifer consists of several interconnected layers of permeable and imper- meable materials. These types of aquifers can have a high storage capacity and can provide a reliable source of water for agriculture. However, the complexity of the aquifer can make efficiently managing and extracting water difficult. A karstic aquifer is formed in soluble rock such as limestone or dolomite and can have many cracks, fissures, and caves that al- low the rapid movement of water. These types of aquifers can be suitable for agriculture because they can provide a steady supply of water. However, the rapid movement of wa- ter through the aquifer can also lead to contamination from pollutants. An alluvial aquifer 4 is formed from the accumulation of sediment deposited by rivers or streams. These types of aquifers can have a high storage capacity and can provide a reliable source of water for irrigation and crop production. They are often found in floodplains or along the course of major rivers. The local shallow aquifer is typically found near the surface2 (Figure 2b) and can be replenished relatively quickly by local precipitation or surface-water sources (Figure 2a). These aquifers may be suitable for small-scale irrigation and crop production. Because around 70% of African farmers practice subsistence agriculture, their abundance (Figure 3) and relatively lower costs of extraction make these aquifers most suited for farming. Therefore, we will restrict our analysis to local shallow aquifers. 3 Data 3.1 Groundwater Data We employ a novel aquifer-typology dataset compiled by Rodella, Zaveri, and Bertone (2022). It consists of a spatially exhaustive list of 8,099 aquifer grids in SSA, along with information on the surface area they cover, the volume of water they hold, and their aquifer type. This dataset is up to date as of 2018. This dataset contains information on the four classes of aquifers discussed in the Back- ground section. The data are collected at the grid level (0.5 degree fishnet). For each grid, the data contain information on what proportion of the grid is major alluvial, complex, karstic, and local shallow, as well as the percentage of the grid cell that is not covered by the data for typology classification. This dataset does not contain information on replen- ishment (or recharge) rates of aquifers. In this analysis, we combine the global aquifer dataset provided by Rodella et al. (2022) to the official World Bank administrative boundaries (with a special focus on SSA). The groundwater data contain the geographic centroids of each aquifer. We assign each aquifer to the country containing its centroid. To ascertain the exact shape, we use Rodella et al. (2022) to obtain the shapefiles of the major aquifers. For aquifers whose shapefiles we were unable to find, we construct circular buffers around these centroids such that the area of the circle matches the surface area of the aquifer. We estimate aquifer statistics including the area of land covered by the aquifers, the volume of water in absolute terms, per capita, and as a proportion of the entire land area of the country to get a sense of which countries have the richest resources of shallow ground water. We complement the Rodella et al. (2022) dataset with data from the Africa Ground- water Atlas, a dataset curated by the British Geological Survey. The Africa Groundwater 2 In SSA, 70% of aquifers lie at a depth of 50 meters or less, 20% lie at a depth less than 25 meters, and 30% lie at a depth of less than 7 meters. This finding suggests groundwater extraction using dug wells is possible in the region, allowing for access to groundwater at lower costs. 5 Atlas provides aquifer productivity3 and aquifer-type information from 51 African coun- tries. Additionally, it has information on the recharge rates of groundwater, as well as groundwater depth maps across the African continent. We merge the groundwater datasets with the Global Agro-Ecological Zones (GAEZ) dataset to impute water-supply information at the GAEZ grid level. 3.2 Agricultural Productivity We use the GAEZ dataset to measure agricultural productivity in SSA. The GAEZ dataset is curated by FAO and IIASA (2020) and provides information on agricultural productivity and estimates production gaps across the world for the years 2000 and 2010 at a granular level of approximately a 9x9 kilometer (km) grid. For our analysis, we consider the data for 2010 only, because they are the latest data.4 The data are also disaggregated by crop type and irrigation type: rain-fed or irrigated using water from sources such as rivers, lakes, and/or groundwater. We have data on 291,798 GAEZ grids, and our subsequent analysis is conducted at the GAEZ grid level. In addition to productivity data, the GAEZ dataset offers insights into agro-climatic conditions such as soil conditions, temperature, precipitation during the agricultural grow- ing season, water availability, and land type in the area. We focused our study on crop- lands to determine whether shifting toward groundwater irrigation could lead to signif- icant gains in agricultural productivity. To answer this question, we examined yield and potential yield at the crop–grid cell–year level to identify productivity gaps by irrigation type. We restrict our analysis to cereals, namely, barley, millet, maize, rice, sorghum, and wheat. We focus on cereals for our analysis, because they are the primary source of energy for most households and the demand for cereals is expected to grow in the coming decades (OECD, 2016). Table 1 provides descriptive statistics for the GAEZ dataset, showing that despite the average harvested area of rain-fed croplands being higher than that of irrigated croplands (100 hectares per grid cell versus 20 hectares per grid cell), the mean production and yield of irrigated croplands are relatively higher. This finding suggests production and produc- tivity gains are possible through irrigation, which is also reflected in larger production gaps for rain-fed croplands than for irrigated croplands. 3 Aquifer productivity, measured in liters per second, refers to the ability of an aquifer to sustain a certain level of groundwater extraction through a well or a bore. It can be thought of as the water yielded from a well or bore that sits on a certain aquifer. 4 Note the irrigation landscapes of 2000 and 2010 are very similar for SSA, and thus, we opt to work with cross-sectional data for the latest year, which has a slightly higher rate of irrigation. 6 3.3 Other Data 3.3.1 Night Lights Data We employ night light luminosity (NTL) data constructed by Elvidge, Zhizhin, Ghosh, Hsu, and Taneja (2021) as a proxy for electrification. More specifically, we use the aver- age VIIRS (Visible Infrared Imaging Radiometer Suite) annual cloud-free composite VNL version 2 for the year 2015. The resolution of these data is 15 arc seconds (approximately 0.0327 km around the Equator). Thus, after extracting it for SSA, we aggregate the NTL measure at the level of each GAEZ grid. 3.3.2 Water-Usage Data Large variations exist in water-usage patterns within countries. We employ annual country- level freshwater-withdrawals data for agriculture, industry, and municipal (i.e., domestic) use.5 Annual freshwater withdrawals refer to total water withdrawals, not including wa- ter loss due to evaporation from storage basins. Withdrawals also include water from desalination plants in countries where they are a significant source. Using this data, we calibrate country-level thresholds needed for crop irrigation purposes. By subtracting the freshwater needed for sanitation and industrial production purposes from the total fresh- water consumption of the country, the thresholds represent the share of freshwater that is available for irrigation purposes. The agricultural water withdrawals is collected at the country level. As a result, we are unable to account for within-country variation in access to water for sanitation. 3.3.3 TerraClimate Data To incorporate climatic factors into our yield estimations, we utilize the dataset curated by Abatzoglou, Dobrowski, Parks, and Hegewisch (2018). This dataset provides monthly climatic information at a spatial resolution of approximately 4 km (1/24th degree). For each of the GAEZ grids, we extract data on mean monthly temperature, total monthly pre- cipitation, and monthly Palmer Drought Severity Index (PDSI) for 2010. We then calculate the quarterly averages of the temperature and PDSI, as well as the quarterly sums of the total precipitation. Additionally, we consider the squared values of the mean quarterly temperature and the total quarterly precipitation. 4 Descriptive Statistics 4.1 Groundwater Access and Availability in SSA The first exercise we undertake is to map the existing groundwater availability in SSA. 5 For access to the data, refer to: https://ourworldindata.org/water-use-stress#freshwater -withdrawals-by-country. 7 As discussed earlier, our aquifer dataset consists of four types of aquifers. In our sim- ulations, we focus on groundwater supply through local shallow aquifers, due to their lower costs and relatively higher recharge rates than other aquifer types. In Table 2, we present the per-capita concentration of local shallow water aquifers in the African context compared with other aquifer types. SSA has the highest share of local shallow aquifers among all regions worldwide (see Figure 1). Furthermore, we observe that grids predominantly composed of local shallow aquifers are distributed across the region (see Figure 3). Comparing these areas with recharge rates (see Figure 2a) reveals that regions with a high share of local shallow aquifers also coin- cide with high recharge rates, supporting the earlier observation of sustainable recharge in these aquifers. Similarly, comparing groundwater depths (see Figure 2b), we find that regions with low groundwater depth align with areas predominantly containing local shal- low aquifers. One potential concern regarding local shallow aquifers is their lower productivity rel- ative to major alluvial aquifers, which may make them unsuitable for irrigation purposes. Therefore, we map aquifer productivity across SSA (see Figure 2c) and find only a few re- gions have aquifer productivity below low to moderate levels. This observation suggests the aquifers in the region can, at the very least, support small-scale irrigation. Given the availability of local shallow aquifers and their ease of access, the possibility of using them sustainably for irrigation presents an opportunity for African farmers. 4.2 Agriculture in Sub-Saharan Africa This paper aims to explore the potential increase in agricultural production and produc- tivity that can be achieved through groundwater irrigation. Currently, only 6.4% of the GAEZ grids have irrigated croplands. However, by examining the descriptive statistics presented in Figure 4, we can observe that although the harvest area of irrigated grids has remained constant over the years, the productivity, measured in terms of crop yields, of these irrigated croplands is nearly three times higher, on average, than that of rain-fed croplands. To conduct our analysis, we focus on cereals in the year 2010. Figure 5 illustrates the distribution of existing rain-fed croplands in green, irrigated croplands in blue, and regions where cereals are not grown in grey. Our analysis primarily investigates the potential increase in production that can be achieved by implementing groundwater irrigation in the existing croplands. 5 Methodology Given the minimal adoption of groundwater irrigation in SSA, we undertake a simulation approach to our analysis, where we quantify the potential gains in agricultural production and productivity achievable through access to groundwater irrigation. 8 To do so, we analyze GAEZ grids that are (partially or fully) rain-fed croplands. Our objective is to assess whether groundwater irrigation can compensate for crop water deficits experienced within these GAEZ grids for rain-fed cultivation. Thus, note that in this paper, we do not shift the cropping landscape, but just change the irrigation landscape of SSA. We simulate the production gains by assuming that with adequate water supply, the yield of rain-fed croplands would match that of irrigated croplands either within the same grid, or of irrigated croplands from grids that are either neighbors or have similar geographic and climatic conditions. In other words, we assume that holding geographic and demographic factors constant, the gap in yields of rainfed and irrigated croplands is explained by water use. We restrict our sample to the year 2010 because the data from that year are the latest made available to us. 5.1 Yield Difference We start by estimating the yield difference (YLD) for crop c in grid g between irrigated and rain-fed croplands. In other words, this estimation represents the change in produc- tivity that is attributable to irrigating the land. We follow two methods to arrive at our yield-difference measure: (1) using a simple difference between crop yields of irrigated croplands and crop yields of rain-fed croplands, and (2) employing a Structural Ricardian model using Heckman selection (SRHS) to obtain predicted differences in yields by allow- ing irrigation choice to be endogenous. 5.1.1 Simple Difference (SD) In this approach, we consider the difference in crop yields of irrigated and rain-fed crop- lands. This approach is valid under the assumption that irrigated and rain-fed croplands are comparable within the same GAEZ grid cell given that the same controls apply, en- abling a ceteris paribus evaluation of productivity gains.6 Two cases are possible under the SD approach. 1. The GAEZ grid g has both rain-fed and irrigated croplands for crop c. In this case, the yield difference is calculated following equation (1). D Y LDcg = Ycg − Ycg (1) D and Y are the crop yields of the grids that only consist of rain-fed crop- Here, Ycg cg lands and the grids that adopted some level of irrigation practices, respectively. 2. The grid cell exclusively comprises rain-fed croplands for crop c. In this case, the 6 Note we cannot spatially identify which portion of the grid is rain-fed versus irrigated; this means all the information that we have on controls, such as soil conditions, other inputs in agricultural production, and so on, is at the GAEZ grid level. 9 yield difference is calculated following equation (2). ¯cg − Ycg Y LDcg = Y D (2) ¯ refers to the mean yield of irrigated croplands that are either the immediate Here, Y neighbors, neighbors within 25 kms, or the national average. Because, for a majority of the grid cells, we had to use equation (2), we have difficulty justifying the validity of the assumption regarding ceteris paribus comparisons of rain-fed and irrigated croplands. Thus, we turn to the second method. 5.1.2 Structural Ricardian Model Using Heckman Selection Another way to estimate yield differences is to run regression models of irrigation choice on agricultural productivity. However, evidence suggests irrigation choice is non-random and endogenous, and therefore, running a simple ordinary least squares model with con- trols would result in biased estimates. As Table 3 shows, grids with some form of irrigation differ from those comprising only rain-fed croplands. Notable distinctions include higher crop water deficits in irrigated croplands and their prevalence in rain-deficient terrains at higher elevations than rain-fed croplands. Therefore, farmers from specific regions and with specific characteristics may choose to follow irrigation practices for their croplands, necessitating tackling this issue to ensure unbiased estimates. A Heckman selection model has been used in the literature to address the selection into the choice to irrigate. Mendelsohn, Nordhaus, and Shaw (1994) were the first to apply a Structural Ricardian model to study the effects of climate change on agricultural rev- enue. This model, keeping factor prices fixed, estimates the impacts of climate and other inputs on net agricultural revenue. To address the irrigation choice being endogenous, Ku- rukulasuriya, Kala, and Mendelsohn (2011) extended the model to account for Heckman selection. We follow Kurukulasuriya et al. (2011) to estimate irrigation adoption (I equals 1 when the grid has any irrigation and 0 when it only has rain-fed croplands) for each crop separately in the first stage, as specified in equation (3). Ig = Xg ∗ β + µ(1,g) (3) Here, X includes climatic factors as well other factors, such as the slope of the terrain and soil moisture. We also control for the country and first administrative division that the grid lies in. To avoid arbitrary selection of controls, we apply a LASSO model for variable selection from a set of candidate indicators. Whereas most of our controls are at the grid level, we include the harvest area of irrigated croplands at the crop-grid level. In the second stage, we estimate conditional yield functions for grids that were irri- gated versus not. For all available exogenous variables Z and ZD (where Z, ZD ⊂ X), we estimate equations (4) and (5). Additionally, note that Z and ZD include the inverse mills ratio. 10 Yg = Zg ∗ γ + µ(2,g) if Ig = 1 (4) D YgD = Zg ∗ γ D + µ(3,g) if Ig = 0 (5) µ(1,g) ∼ N (0, 1) µ(2,g) ∼ N (0, ϕ(2,g) ) µ(3,g) ∼ N (0, ϕ(3,g) ) corr(µ(1,g) , µ(2,g) ) = ρ(2,g) corr(µ(1,g) , µ(3,g) ) = ρ(3,g) Here, I is the latent variable determining the irrigation status of the grid, and as before, YgD and Yg are the crop yields of the grids that only comprise of rain-fed croplands and the grids that adopted some level of irrigation practices, respectively. We allow the error terms to be correlated, such that µ(1,g) and µ(2,g) and µ(1,g) and µ(3,g) are jointly normally distributed independently of X and Z with an expectation of 0. This approach is consistent with the model proposed by Heckman (1979). Put simply, this model first predicts the exogenous factors that influence irrigation adoption, and then partitions the data, for each crop, into grids that had any irrigation adoption (I = 1) and grids that did not have any irrigation adoption (I = 0). For both, we separately estimate the yield gains given the existing set of climatic controls, while allow- ing for the errors in the first stage and each equation of the second stage to be correlated, because climatic controls drive the selection into irrigation choice, which is then associated with crop yield. Note that unlike the model proposed by Kurukulasuriya et al. (2011), our model is at the grid level instead of the farm level. The primary limitation of this exercise is that in our data, irrigation adoption is a fraction rather than a dummy; however, we restrict our irrigation adoption variable to either take the value of 0 or 1. Thus, despite having more information on the level (or share) of irrigation adoption within the grid, our model uses a more simplified approach to follow the Heckman framework. While using farm-level data, Kurukulasuriya et al. (2011) have a binary irrigation-adoption variable and therefore are not forced to restrict their data in any way. We run our model for each of the crops separately. Because we are interested in using these estimates to calculate yield differences be- tween irrigated and rain-fed croplands, we use the explained part of the model to obtain predicted conditional crop yield values and then compute the mean difference (YLD) at the “Administrative Level 1”,7 as shown in equation (6). ˆ D c,a(g) ˆ c,a(g) − Y Y LDcg = Y (6) 7 Administrative Level 1 refers to the largest subnational administrative unit of a country. It usually com- prises of states or provinces, depending on the country. 11 ˆD ˆ and Y Here, a(g ) represents the ”Administrative Level 1” that the grid is mapped to. Y ˆ and Y are the admin 1 level averages of predicted Y ˆ D , respectively. 5.2 Simulating Production Gains We then use the YLD obtained from each of the two methods described above to simulate production gains. However, we only calculate these production gains under three key conditions. First, the mean aquifer productivity or yield (=η ) for the GAEZ grid must be at or above the low to moderate threshold. Adequate aquifer productivity is essential for fulfilling irrigation needs, and this condition ensures its consideration. Second, the ratio of groundwater ex- traction to recharge, ν , must remain below 70%.8 This condition is crucial to ensure gains are computed only for regions with sustainable access to groundwater, addressing con- cerns regarding current or future over-exploitation of this resource. To set the threshold, we rely on the guidelines set in Chapter 22.2.1 by the Ground Water Resource Estima- tion Committee (2015), Government of India. Third, to account for ease in accessibility of groundwater, we consider aquifers that have a depth of 7 meters or less (ζ ). For more de- tails on the parameters used, refer to Appendix 9. We summarize all the parameters used in Table A1. We calculate gains in production using equation (7). QGAcg = HARcg ∗ Y LDcg ∗ Share of Irrigable Areacg (7) Here, QGA is the production gain for crop c in GAEZ grid cell g . HAR is the harvest area of rain-fed croplands, and Y LD refers to the yield difference between irrigated crop- lands and rain-fed croplands. Share of Irrigable Area denotes the fraction of the rain-fed area within the grid where the crop’s water demand can be sufficiently met with the crop’s water supply.9 It equals one if the crop’s water supply is greater than or equal to the crop’s water demand; otherwise, it is set as the crop’s water demand divided by the crop’s water supply. This variable can be thought of as a discount on the achievable production gain, where we scale down the gain in production to consider only the share of the grid where groundwater availability can fill in existing crop water deficits or the extent to which the existing crop water deficits can be filled using groundwater. 5.2.1 Estimating Water Supply We overlay water-volume gridded data with gridded agricultural data (referred to as GAEZ grids). Subsequently, we estimate the volume of local shallow groundwater in each GAEZ grid g using equation (8). 8 To compute this ratio, we divided the total water use by the mean recharge rate of each grid. For each country, we divided total water use across grids within the country weighted by the GAEZ grid’s area and multiplied by, ϕ, the share of water use that is attributable to groundwater use (=67.5%) (Rodella et al., 2022)). 9 In the water-supply calculations, we only consider groundwater availability after accounting for sanita- tion needs. 12 N Intersection area(g,q) V LS g = ( V GW q ∗ ∗ P oLS q ) ∗ (1 − SDUg ) (8) T otal areaq ) q =1 Here, V LS refers to the volume of local shallow groundwater in m3 that is available in GAEZ grid cell g . q ranges from 1 to N , representing the set of aquifer grids intersect- ing with GAEZ grid g . V GW is the total volume of groundwater across all aquifer types available in aquifer grid q . We multiply V GW by the share of the aquifer area that inter- sects with GAEZ grid g and the share of groundwater that is attributable to local shallow aquifers, P oLS . We then multiply the total volume of local shallow water by (1 - SDU ), where SDU is the share of groundwater needed for sanitation and drinking usage. To estimate a crop’s potential share of local shallow groundwater, we multiply V LS by the share of the crop’s harvest area relative to the grid’s total area. 5.2.2 Estimating Water Demand To estimate the water demand for each crop in each grid, we sum the evapotranspiration of the crop and the crop’s water deficit during the growing season. For grids where either of these variables is missing, we impute the crop water demand or crop’s water requirement based on the mean crop water demand of immediate neighbors, neighbors within 25 kms of the grid’s center, or the national average crop water requirement for that crop. 5.3 Measuring Income Gains Understanding the income potential of the simulated gains in production is important for agricultural policy engagements with governments in Africa. To estimate income gains, we simply apply producer prices to GAEZ grids with estimated production gains and then compare the total gains to the 1.90 USD poverty line. We employ a back-of-the-envelope approach as described below. We first source geo-located food product price data (Development Data Group, 2021) via the World Bank Micro-data Library API. Our API query returns local currency (nomi- nal) prices for all African countries with production gains for each crop and each market nom ). This allows us to increase the number of market-level between 2007 and 2015 (πcm prices available within each country and smooth over sudden price shocks. We apply the corresponding national monthly consumer price index (CPI) and PPP conversion factors from the World Bank’s Poverty and Inequality Platform (PIP) (Aron et al., 2023) to convert to real monthly average prices in 2010 USD for each crop in each market. nom πcm real πcm = (9) cpi ∗ ppp We then perform a spatial merge of the grids with production gains to the nearest market from the food producer price database. Consequently, the prices that apply to specific crop production gains in each grid are those from the nearest markets in the food 13 price database. Coincidentally, the size of the GAEZ grids and the sparsity of the market geo-location data prevent us from finding multiple markets within any grids. Finally, we compute expected per capita income gains by taking the product of crop prices and the simulated production gains for each crop-grid. We aggregate this for all grids (G) within a country to estimate an expected change in income for individuals at the poverty line. That is, for a representative farmer at the 1.90 USD extreme poverty line (2011 ppp) within a specific country, we compute the expected percentage income growth (E (θ)) from all crops combined. G real g =1 (πcg )(QGAcg ) E (θc ) = (10) G E (θc ) E (δθc ) = (11) AEP L Here, AEP L represents the annualized extreme poverty line (i.e., 1.90 * 365 USD), and therefore E (δθc ) represents the expected change in income for farmers around the poverty line from adopting groundwater irrigation to reduce the crop water deficit. 6 Results Table 4 presents the simulation results we obtained using the two different approaches— SD and SRHS—when all three conditions are met. Each row in the tables corresponds to a unique combination of model and crop choice. The mean gains in production are much higher than the median gains, suggesting the distribution of production gains is right-skewed. Therefore, we use median production gains to discuss our results. Using the two approaches, our simulated median gains in production for different crops range between 29.57% and 33.28% for barley, 76.35% for millet, 107.96 and 123.26% for maize, 108.09% and 114.64% for rice, 89.49% and 129.42% for sorghum, and 27.97% and 53.08% for wheat. Refer to the map of production gains in Figure 6 and Figure 7 for the SD and SRHS approaches, respectively. Note we were not able to run simulations for millet using the SRHS approach, due to very few grids being irrigated. With the exception of barley, the simulated gains using SRHS are higher than when using the SD approach. These results suggest large production gains can be realized through groundwater irrigation in SSA, especially in the production of maize and rice. For a country-wise overview of our results, refer to the median pro- duction gains by country in Figure 9 and Figure 11 for maize and rice, respectively. For the results using the SRHS approach, refer to Figure 8 and Figure 10 for maize and rice, respectively. Furthermore, note that 24.91% of all GAEZ grids are existing croplands. Among these croplands, 21.96% satisfy all three conditions considered in our simulations. Since we do lose a substantial portion of grids by imposing a depth constraint of 7m, we check for 14 the robustness of our results by considering a depth of 50m. This increases the fraction of croplands satisfying our conditions to 62.05%; the high fraction of croplands satisfying our conditions provides further evidence for the feasibility and potential for sustainable groundwater irrigation in Sub-Saharan Africa. Our results remain consistent with this change in the depth constraint (refer to Table A3. Refer to the map of production gains in Figure A1 and Figure A2 for the SD and SRHS approaches, respectively. Figure A4 and Figure A6 plot the country-wise gains for maize and rice, respectively, using the SD approach. For the SRHS approach results, refer to Figure A3 and Figure A5. Our findings align with those in existing literature. For the Indian context, Duflo and Pande (2007) found surface water irrigation boosts cereal crop production by 34%-53%. Similarly, in Rwanda, Jones et al. (2022) observed a 43%-62% increase in annual cash prof- its. Dyer and Shapiro (2023) also reported a 41% increase in net revenues from fruits and vegetables through irrigation pumps in Kenya. Our results for most of the crops fall within the range of gains reported in these studies, but vary by crop type. 6.1 Income Gains Groundwater irrigation is one of several input factors that are determinants of agricultural productivity. Small to modest simulated income growth should be expected from filling these crop water deficits. Our modelling approach does not account for the possibility of water transportation arrangements from areas with low water deficits to areas with large water deficits. Consequently, we simulate expected income growth (over the annualized 1.90 USD International Poverty Line) ranging from 0% to 3.1% across African countries with simulated productivity gains. The gains appear to be highest in the Sahel countries with Togo (3.1%), Benin (2.4%) Guinea (2.4%), and Nigeria (1.0%). These results are driven by large deposits of avail- able groundwater in areas where crop water deficits exist and are thereby correlated with pre-existing low levels of crop production. It should also be noted that because we are un- able to estimate the number of farmers in these specific areas, we use World Bank national estimates of number of people in agriculture. Hence, our results are likely quite conserva- tive in terms of per-farmer earnings gains corresponding to filling in existing crop water deficits. 7 Robustness Checks 7.1 Considering a Different Combination of Conditions We test the robustness of our results by relaxing each of the three conditions. Table 5 and Table 6 present the simulation results we obtained using SD and SRHS, respectively. Each row in the tables corresponds to a unique combination of conditions that were imposed while running the simulations. The first row for each crop corresponds to the results in Table 4 where all three conditions hold. 15 Using the SD approach, our simulated median gains in production for different crops range between 29.57% and 47.96% for barley, 71.87% and 76.35% for millet, 89.38% and 107.99% for maize, 105.64% and 108.09% for rice, 59.99% and 89.49% for sorghum, and 27.97% and 38.96% for wheat. The corresponding gains using the SRHS approach are be- tween 33.28% and 41.87% for barley, 106.64% and 123.26% for maize, 112.56% and 114.64% for rice, 93.38% and 129.42% for sorghum, and 53.08% and 60.11% for wheat. Our results are more or less consistent despite considering a combination of conditions imposed. 7.2 Considering a Different Dataset for Water Supply Our groundwater supply measure comes from Rodella et al. (2022). We test if our results remain robust to using a different dataset for water supply, especially because ground- water volume computations tend to be probabilistic in nature. We use the Groundwater Storage Map in the digital groundwater maps of Africa10 collected by MacDonald, Bonsor, ´ Dochartaigh, and Taylor (2012) to obtain a measure of groundwater storage. O In our main simulations, we rely on the volumes of groundwater associated with the local shallow groundwater aquifer classification. However, the classification system we used—including the “local shallow aquifer” classification—is not consistent with hydro- geological classifications of groundwater aquifers. Thus, we consider the subset of aquifers that closely resemble the characteristics of local shallow aquifers. These aquifers include the following: • Unconsolidated Sedimentary (U): These aquifers consist of loose sediments, such as sand and gravel, which allow water to flow more easily. • Consolidated Sedimentary Fracture (CSF): These aquifers are composed of consoli- dated sedimentary rocks that have fractures or cracks, which enhance water storage and flow. • Consolidated Sedimentary Intergranular (CSI): These aquifers are made up of consol- idated sedimentary rocks with interconnected pore spaces, allowing water to move through the rock matrix. • Consolidated Sedimentary Intergranular / Fracture (CSIF): These aquifers are a com- bination of consolidated sedimentary rocks with both intergranular pore spaces and fractures, providing multiple pathways for water movement. • Basement (B): These aquifers are located in the underlying bedrock, typically con- sisting of hard and fractured rocks that can store and transmit water. These hydrogeological aquifer types are considered closest to the definition of local shallow aquifers because they share similar characteristics, such as their geological compo- sition and hydrogeological properties, which make them suitable for groundwater storage and extraction in the context of the study. 10 See https://www2.bgs.ac.uk/groundwater/international/africangroundwater/ mapsDownload.html. 16 To construct this measure of water supply, we first consider the modal aquifer type associated with each BGS storage grid. We then overlap the GAEZ grids with BGS storage grids. Because each grid can be mapped to multiple aquifers, we sum the volume for the BGS storage grids mapped to each GAEZ grid that has a (modal) aquifer type that is either U, CSF, CSI, CSIF, or B. Our results using this measure of water supply can be found in Table 8 and Table 9 for the SD and SRHS approaches, respectively. Using the SD approach, our simulated median gains in production for different crops range between 85.47% and 92.95% for barley, 59.57% and 75.11% for millet, 112.38% and 125.64% for maize, 124.41% to 127.22% for rice, 88.25% and 100.49% for sorghum, and 45.92% and 53.93% for wheat. The corresponding gains using the SRHS approach are be- tween 52.87% and 68.80% for barley, 148.02% and 158.14% for maize, 126.88% and 129.00% for rice, 148.48% and 170.86% for sorghum, and 71.02% and 80.75% for wheat. These re- sults are much higher than those found using the data from Rodella et al. (2022). We believe these differences arise due to the mapping of local shallow aquifers not being perfect be- tween the two datasets. However, because our main results are on the conservative side, we believe we can safely say large gains in production are possible through groundwater irrigation. 8 Discussion on Costs We have so far discussed the potential for increased production through groundwater irri- gation. However, its high initial cost can be a barrier to adoption, which might explain the low prevalence of this method among farmers who are credit constrained. The cost of accessing groundwater is twofold: the fixed cost of groundwater explo- ration and well construction, and the variable cost of regular operation and maintenance. Fixed costs cover the detection of groundwater resources, water-quality testing, trans- portation of materials, pump purchase and drilling costs. Variable costs refer to the fre- quent operational and maintenance costs of running groundwater pumps. Given the sub- stantial investment, some of the benefits previously discussed may not be realized. One approach to mitigating the fixed costs of groundwater irrigation involves the implemen- tation of treadle pumps, capable of accessing groundwater up to 7 meters deep (Burney, Naylor, & Postel, 2013). In Sub-Saharan Africa (SSA), Burney et al. (2013) note that treadle pumps are priced between USD 100 and USD 300, significantly lower than earlier estimates by Xenarios and Pavelic (2013). Furthermore, Giordano, de Fraiture, Weight, and van der Bliek (2012) suggest that introducing treadle pumps in SSA could yield annual net revenues of up to USD 22 billion and benefit 185 million individuals. Therefore, to consider gains that can be cost effective, our analysis focuses on local shallow aquifers that are less than 7 meters deep and therefore easier and cheaper to access. A recent study found irrigation pumps for pumping from surface water sources pay for themselves within three years in 17 Kenya (Dyer & Shapiro, 2023). Generalizing this finding to the context of groundwater irrigation pumps, the study suggests that although high costs may dissuade farmers from using groundwater irrigation, interventions providing information about the costs and benefits of these technologies could enhance productivity through increased adoption. It must be noted, however, that the costs substantially increase beyond 8 m groundwater depth (Sekhri, 2014). Research on groundwater costs across 10 SSA countries indicates the total cost for a pump with a depth of 40-100 meters ranges between USD 6,000 and USD 23,200, on average (Xenarios & Pavelic, 2013). With the expansion of input markets in SSA, these costs could have been substantially reduced over the past decade, and that is why we present some estimates on gains from groundwater access up to 50 m deep in our Appendix. Another constraint to groundwater access stems from the lack of access to the electric- ity grid. A proposed solution that reduces the variable cost of electricity is the provision of photovoltaic- (or solar-) powered drip irrigation pumps. Although these pumps may help reduce the variable cost, regulation of their use is crucial to prevent over-exploitation of groundwater resources. Finally, subsidizing the fixed costs of groundwater irrigation tech- nologies (while regulating sustainable use) is another avenue of inquiry for policymakers and future research. 9 Conclusions Rural poverty in Sub-Saharan Africa is closely linked to the lack of substantial growth in agricultural productivity. Enhancing agriculture productivity could stimulate wider eco- nomic growth, promote inclusive development, structural transformation, and poverty re- duction. Agricultural productivity growth worldwide has been driven by investments in high return rural public goods, improved policies, and institutions. However, Sub-Saharan Africa significantly under-invests in these areas. This paper explores one untapped growth source—the sustainable use of groundwater for irrigation. We measure potential agricul- tural production gains in Sub-Saharan Africa through groundwater irrigation. To do so, we construct a novel dataset by merging gridded agricultural data with local shallow aquifer data and their corresponding details. We then use our dataset to simulate the gains in production that can be achieved if we use groundwater irrigation to fill in the existing crop water deficits that persist in the region’s rain-fed croplands. While run- ning our simulations, we impose conditions during our simulations to only compute gains in regions with sufficient renewable groundwater. We also limit our analysis to aquifers suitable for small-scale irrigation and accessible at low depths to ensure cost effectiveness. Our findings show the median production gains range from 27.97% to 129.42%, de- pending on the crop type and the method used to predict crop yield gains. The most pronounced gains are for maize and rice crops. Our results remain consistent, even with changes in the method and dataset used to measure groundwater availability. Most impor- tantly, our analysis enables policymakers to delineate regions likely to experience agricul- 18 tural production gains through groundwater irrigation. Knowledge of where agricultural production gains can be realized could facilitate strategic investment decisions regarding the deployment of irrigation technology in specific regions. One limitation of our study is that we were unable to causally estimate production gains due to the low rates of groundwater irrigation adoption in the region, as captured in household surveys, which would have helped us obtain more precise estimates of the net benefits of groundwater irrigation. We leave this exercise for future research work. 19 References Abatzoglou, J. T., Dobrowski, S. Z., Parks, S. A., & Hegewisch, K. C. (2018, January). Terraclimate, a high-resolution global dataset of monthly climate and climatic water balance from 1958–2015. Scientific Data, 5. doi: https://doi.org/10.1038/sdata.2017 .191 Acheampong, E., Ozor, N., & Sekyi-Annan, E. (2014, 06). Development of small dams and their impact on livelihoods: Cases from northern ghana. African journal of agricultural research, 9, 1867-1877. doi: 10.5897/AJAR2014.8610 Adeoti, A. I., Barry, B., Namara, R. E., Kamara, A. B., & Titiati, A. (2007). Treadle pump irrigation and poverty in Ghana (IWMI Research Reports No. 44520). International Wa- ter Management Institute. Retrieved from https://ideas.repec.org/p/ags/ iwmirr/44520.html doi: 10.22004/ag.econ.44520 Altchenko, Y., & Villholth, K. G. (2015). Mapping irrigation potential from renewable groundwater in africa – a quantitative hydrological approach. Hydrology and Earth System Sciences, 19(2), 1055–1067. Retrieved from https://hess.copernicus .org/articles/19/1055/2015/ doi: 10.5194/hess-19-1055-2015 Aron, D. V., Aguilar, R. A. C., Diaz-Bonilla, C., Betran, M. G. F., Foster, E. M., Fujs, T. H., . . . others (2023). Poverty and inequality platform (pip). Aseyehegn, K., Yirga, C., & Rajan, S. (2012, 01). Effect of small-scale irrigation on the income of rural farm households: The case of laelay maichew district, central tigray, ethiopia. Journal of Agricultural Sciences, 7. doi: 10.4038/jas.v7i1.4066 Balana, B. B., Bizimana, J.-C., Richardson, J. W., Lefore, N., Adimassu, Z., & Herbst, B. K. (2020). Economic and food security effects of small-scale irrigation technologies in northern ghana. Water Resources and Economics, 29, 100141. Retrieved from https://www.sciencedirect.com/science/article/pii/ S2212428418300744 doi: https://doi.org/10.1016/j.wre.2019.03.001 Briceno-Garmendia, C. M., & Foster, V. (2009, January). Africa’s infrastructure : a time for transformation (English) (Africa development forum Washington, D.C.) The World Bank. Retrieved from http://documents.worldbank.org/ curated/en/246961468003355256/Africas-infrastructure-a-time -for-transformation Burney, J., Naylor, R., & Postel, S. L. (2013, 07). The case for distributed irrigation as a development priority in sub-saharan africa. Proceedings of the National Academy of Sci- ences of the United States of America, 110(30), 12513-7. doi: 10.1073/pnas.1203597110 Burney, J., Woltering, L., Burke, M., Naylor, R., & Pasternak, D. (2010, 02). Solar-powered drip irrigation enhances food security in the sudano-sahel. Proceedings of the National Academy of Sciences of the United States of America, 107, 1848-53. doi: 10.1073/pnas .0909678107 Chazovachii, B. (2012). The impact of small scale irrigation schemes on rural livelihoods: the case of panganai irrigation scheme bikita district zimbabwe. Journal of Sustainable Development in Africa, 14(4), 217–231. 20 Concern Universal. (2012). A study of the economic and social impact of three irrigation schemes in dedza and ntcheu districts (Report). Retrieved from https://cepa.rmportal .net/Library/climate-change/A%20Study%20of%20the%20Economic% 20and%20Social%20Impact%20of%20Three%20Irrigation%20Schemes% 20in%20Dedza%20and%20Ntcheu%20Districts.pdf/view Connor, D., Comas, J., Macpherson, H.-G., & Mateos, L. (2008). Impact of small-holder irrigation on the agricultural production, food supply and economic prosperity of a representative village beside the senegal river, mauritania. Agricultural Systems, 96(1-3), 1-15. Retrieved from https://EconPapers.repec.org/RePEc:eee: agisys:v:96:y:2008:i:1-3:p:1-15 Dauda, T. O., Asiribo, O. E., Oladele, S. A., Saka, J. O., & Salahu, B. (2009, 05). An assess- ment of the roles of irrigation farming in the millennium development goals. African Journal of Agricultural Research, 4. Development Data Group. (2021). Monthly food price estimates by product and market. World Bank, Development Data Group. Retrieved from https://microdata .worldbank.org/index.php/catalog/study/WLD 2021 RTFP v02 M doi: 10.48529/2ZH0-JF55 Dillon, A. (2008). Access to irrigation and the escape from poverty: evidence from northern Mali (IWMI Working Papers No. H043847). International Water Management Institute. Retrieved from https://ideas.repec.org/p/iwt/worppr/h043847.html Dillon, A. (2011). The Effect of Irrigation on Poverty Reduction, Asset Accumula- tion, and Informal Insurance: Evidence from Northern Mali. World Development, 39(12), 2165-2175. Retrieved from https://ideas.repec.org/a/eee/wdevel/ v39y2011i12p2165-2175.html doi: 10.1016/j.worlddev.2011.0 Duflo, E., & Pande, R. (2007). Dams. The Quarterly Journal of Economics, 122(2), 601- 646. Retrieved from https://EconPapers.repec.org/RePEc:oup:qjecon: v:122:y:2007:i:2:p:601-646. Dyer, J., & Shapiro, J. (2023). Pumps, prosperity and household power: Experimental evi- dence on irrigation pumps and smallholder farmers in kenya. Journal of Development Economics, 163, 103034. Retrieved from https://www.sciencedirect.com/ science/article/pii/S0304387822001766 doi: https://doi.org/10.1016/ j.jdeveco.2022.103034 Elvidge, C. D., Zhizhin, M., Ghosh, T., Hsu, F.-C., & Taneja, J. (2021). Annual time series of global viirs nighttime lights derived from monthly averages: 2012 to 2019. Remote Sensing, 13(5). Retrieved from https://www.mdpi.com/2072-4292/13/5/922 doi: 10.3390/rs13050922 Ersado, L. (2005, January). Small-scale irrigation dams, agricultural production, and health - theory and evidence from Ethiopia (Policy Research Working Paper Series No. 3494). The World Bank. Retrieved from https://ideas.repec.org/p/wbk/wbrwps/ 3494.html FAO and IIASA. (2020). Global agro ecological zones version 4 (gaez v4). Retrieved 2023-03-01, 21 from http://www.fao.org/gaez/ Fuglie, K., & Rada, N. (2013, 02). Resources, policies, and agricultural productivity in sub-saharan africa. SSRN Electronic Journal. doi: 10.2139/ssrn.2266459 Gadedjisso-Tossou, A., Avell´ ¨ an, T., & Schutze, N. (2020, 01). Impact of irrigation strategies on maize (zea mays l.) production in the savannah region of northern togo (west africa). Water SA, 46. doi: 10.17159/wsa/2020.v46.i1.7894 Gebregziabher, G., Namara, R., & Holden, S. (2009, 08). Poverty reduction with irriga- tion investment: An empirical case study from tigray, ethiopia. Agricultural Water Management, 96, 1837-1843. doi: 10.1016/j.agwat.2009.08.004 Gebregziabher, G., Villholth, K. G., Hanjra, M. A., Yirga, M., & Namara, R. E. (2013, October). Cost-benefit analysis and ideas for cost sharing of groundwater irriga- tion: evidence from north-eastern Ethiopia. Water International, 38(6), 852-863. Re- trieved from https://ideas.repec.org/a/taf/rwinxx/v38y2013i6p852 -863.html doi: 10.1080/02508060.2014.847 Giordano, M., de Fraiture, C., Weight, E., & van der Bliek, J. (2012). Water for wealth and food security: Supporting farmer-driveninvestments in agricultural water manage- ment (Tech. Rep.). Colombo, Sri Lanka: International Water Management Insti- tute. Retrieved from https://www.iwmi.cgiar.org/Publications/Other/ Reports/PDF/Water for wealth and food security.pdf esultats] (No. Goyal, A., & Nash, J. (2017). Reaping Richer Returns [Obtenir de meilleurs r´ 25996). The World Bank Group. Retrieved from https://ideas.repec.org/b/ wbk/wbpubs/25996.html Ground Water Resource Estimation Committee. (2015). GEC 2015 Detailed Guidelines. Retrieved 2023-18-12, from https://cgwb.gov.in/sites/default/files/ inline-files/2020-11-17 Detailed Guidelines GEC2015.pdf Heckman, J. J. (1979, January). Sample Selection Bias as a Specification Error. Econometrica, 47(1), 153-161. Retrieved from https://ideas.repec.org/a/ecm/emetrp/ v47y1979i1p153-61.html Inocencio, A., Kikuchi, M., Tonosaki, M., Maruyama, A., Merrey, D., Sally, H., & de Jong, I. (2007). Costs and performance of irrigation projects: A comparison of sub-saharan africa and other developing regions (IWMI Research Reports No. H036214). International Wa- ter Management Institute. Retrieved from https://EconPapers.repec.org/ RePEc:iwt:rerpts:h036214 Jambo, Y., Alemu, A., & Tasew, W. (2021, 07). Impact of small-scale irrigation on household food security: evidence from ethiopia. Agriculture Food Security, 10. doi: 10.1186/ s40066-021-00294-w Jones, M., Kondylis, F., Loeser, J., & Magruder, J. (2022, July). Factor market fail- ures and the adoption of irrigation in rwanda. American Economic Review, 112(7), 2316-52. Retrieved from https://www.aeaweb.org/articles?id=10.1257/ aer.20210059 doi: 10.1257/aer.20210059 Khan, A. M., Kuate, L., Pongou, R., & Zhang, F. (2024, June). Weather, Water, and 22 Work : Climatic Water Variability and Labor MarketOutcomes in Sub-Saharan Africa (Pol- icy Research Working Paper Series No. 10823). The World Bank. Retrieved from https://ideas.repec.org/p/wbk/wbrwps/10823.html Kurukulasuriya, P., Kala, N., & Mendelsohn, R. (2011). Adaptation And Climate Change Impacts: A Structural Ricardian Model Of Irrigation And Farm Income In Africa. Climate Change Economics (CCE), 2(02), 149-174. Retrieved from https://ideas .repec.org/a/wsi/ccexxx/v02y2011i02ns2010007811000255.html doi: 10.1142/S2010007811000255 ´ Dochartaigh, B. , & Taylor, R. G. (2012). Quantitative MacDonald, A. M., Bonsor, H. C., O maps of groundwater resources in Africa. Environmental Research Letters, 7(2), 153- 161. Retrieved from https://iopscience.iop.org/article/10.1088/1748 -9326/7/2/024009 doi: 10.1088/1748-9326/7/2/024009 Margat, J., & van der Gun, J. (2013). Groundwater around the world: A geographic synopsis. doi: 10.1201/b13977 Maru, H., Haileslassie, A., & Zeleke, T. (2023, 05). Impacts of small-scale irrigation on farmers’ livelihood: Evidence from the drought prone areas of upper awash sub- basin, ethiopia. Heliyon, 9, e16354. doi: 10.1016/j.heliyon.2023.e16354 Mendelsohn, R., Nordhaus, W., & Shaw, D. (1994). The impact of global warm- ing on agriculture: A ricardian analysis. American Economic Review, 84(4), 753- 71. Retrieved from https://EconPapers.repec.org/RePEc:aea:aecrev:v: 84:y:1994:i:4:p:753-71 Mupaso, N., Manzungu, E., Mutambara, J., & Hanyani-Mlambo, B. (2014, 10). The impact of irrigation technology on the financial and economic performance of smallholder irrigation in zimbabwe. Irrigation and Drainage, 63. doi: 10.1002/ird.1820 Namara, R., Awuni, J., Barry, B., Giordano, M., Hope, L., Owusu, E., & Forkuor, G. (2011). Smallholder shallow groundwater irrigation development in the upper east region of ghana. doi: 10.5337/2011.214 Ngango, J., & Hong, S. (2021). Adoption of small-scale irrigation technologies and its impact on land productivity: Evidence from rwanda. Journal of Integrative Agri- culture, 20(8), 2302-2312. Retrieved from https://www.sciencedirect.com/ science/article/pii/S2095311920634177 doi: https://doi.org/10.1016/ S2095-3119(20)63417-7 Nkhata, R. (2014). Does irrigation have an impact on food security and poverty: Evidence from bwanje valley irrigation scheme in malawi:.. Retrieved from https://api .semanticscholar.org/CorpusID:182505023 OECD. (2016). Agriculture in sub-saharan africa: Prospects and challenges for the next decade. Retrieved from https://www.oecd-ilibrary.org/content/ component/agr outlook-2016-5-en doi: https://doi.org/https://doi.org/ 10.1787/agr outlook-2016-5-en Owusu, V. (2016). The economics of small-scale private pump irrigation and agricultural productivity in ghana. Journal of Developing Areas, 50(1), 289-304. 23 Retrieved from https://EconPapers.repec.org/RePEc:jda:journl:vol .50:year:2016:issue1:pp:289-304 Pfister, S., Bayer, P., Koehler, A., & Hellweg, S. (2011). Projected water consumption in future global agriculture: Scenarios and related impacts. Science of The Total Environ- ment, 409(20), 4206-4216. Retrieved from https://www.sciencedirect.com/ science/article/pii/S0048969711007376 doi: https://doi.org/10.1016/ j.scitotenv.2011.07.019 Rodella, A.-S., Zaveri, E. D., & Bertone, F. (2022). The hidden wealth of nations : The economics of groundwater in times of climate change (english) (Report). World Bank Group. Retrieved from http://documents.worldbank.org/curated/en/ 099257006142358468/IDU0fb2550de013100434708d920a3e3bec6afb1 Schonberger, S., & Wijnen, M. (2018). Assessment of groundwater challenges opportunities in support of sustainable development in sub-saharan africa (Report). World Bank Group. Retrieved from https://documents1.worldbank.org/curated/en/ 420291533931251279/pdf/Assessment-of-groundwater-challenges -and-opportunities-in-Sub-Saharan-Africa.pdf Sekhri, S. (2014). Wells, water, and welfare: The impact of access to groundwater on rural poverty and conflict. American Economic Journal: Applied Economics, 6(3), 76– 102. Retrieved 2024-05-10, from http://www.jstor.org/stable/43189490 Suri, T., & Udry, C. (2022, February). Agricultural technology in africa. Journal of Economic Perspectives, 36(1), 33-56. Retrieved from https://www.aeaweb.org/articles ?id=10.1257/jep.36.1.33 doi: 10.1257/jep.36.1.33 Svendsen, M., Ewing, M., & Msangi, S. (2009). Measuring irrigation performance in Africa: (IFPRI discussion papers No. 894). International Food Policy Research Institute (IF- PRI). Retrieved from https://ideas.repec.org/p/fpr/ifprid/894.html Tesfaye, A., Bogale, A., Namara, R., & Bacha, D. (2008, 06). The impact of small-scale irriga- tion on household food security: The case of filtino and godino irrigation schemes in ethiopia. Irrigation and Drainage Systems, 22, 145-158. doi: 10.1007/s10795-008-9047-5 Tillie, P., Louhichi, K., & Gomez Y Paloma, S. (2018, July). Does small-irrigation boost smallholder farming? Evidence from Niger (2018 Conference, July 28-August 2, 2018, Vancouver, British Columbia No. 277088). International Association of Agricul- tural Economists. Retrieved from https://ideas.repec.org/p/ags/iaae18/ 277088.html doi: 10.22004/ag.econ.277088 e, F. Torou, B. M., Favreau, G., Barbier, B., Pavelic, P., Illou, M., & Sidib´ (2013, July). Constraints and opportunities for groundwater irrigation arising from hy- drologic shifts in the Iullemmeden Basin, south-western Niger. Water International, 38(4), 465-479. Retrieved from https://ideas.repec.org/a/taf/rwinxx/ v38y2013i4p465-479.html doi: 10.1080/02508060.2013.817 UNESCO and UN-Water. (2022). The united nations world water development report 2022: Groundwater: Making the invisible visible (Report). WaterAid. (2022). Groundwater: the world’s neglected defence against climate change (Re- 24 port). Retrieved from https://www.wateraid.org/in/sites/g/files/ jkxoof336/files/2022-04/Groundwater%20The%20world%E2%80%99s% 20neglected%20defence%20against%20climate%20change.pdf Xenarios, S., & Pavelic, P. (2013, 01). Assessing and forecasting ground- water development costs in Sub-Saharan Africa. Water SA, 39, 00 - 00. Retrieved from http://www.scielo.org.za/scielo.php?script=sci arttext&pid=S1816-79502013000400010&nrm=iso Xie, H., You, L., & Takeshima, H. (2017). Invest in small-scale irrigated agriculture: A na- tional assessment on potential to expand small-scale irrigation in nigeria. Agricultural Water Management, 193, 251-264. Retrieved from https://www.sciencedirect .com/science/article/pii/S0378377417302846 doi: https://doi.org/10 .1016/j.agwat.2017.08.020 Xie, H., You, L., Wielgosz, B., & Ringler, C. (2014). Estimating the potential for expand- ing smallholder irrigation in Sub-Saharan Africa. Agricultural Water Management, 131(C), 183-193. Retrieved from https://ideas.repec.org/a/eee/agiwat/ v131y2014icp183-193.html doi: 10.1016/j.agwat.2013.08.0 You, L., Ringler, C., Wood-Sichra, U., Robertson, R., Wood, S., Zhu, T., . . . Sun, Y. (2011). What is the irrigation potential for africa? a combined biophysical and socioeco- nomic approach. Food Policy, 36(6), 770-782. Retrieved from https://ideas .repec.org/a/eee/jfpoli/v36y2011i6p770-782.html doi: 10.1016/j .foodpol.2011.09 You, L., Xie, H., Wood-Sichra, U., Guo, Z., & Wang, L. (2014, 08). Irrigation potential and investment return in kenya. Food Policy, 47, 34–45. doi: 10.1016/j.foodpol.2014.04 .006 You, L. Z. (2008, June). Africa : Irrigation investment needs in sub-saharan africa (World Bank Publications - Reports No. 7870). The World Bank Group. Retrieved from https://ideas.repec.org/p/wbk/wboper/7870.html 25 Tables Table 1: Summary Statistics for Sub-Saharan Africa Irrigated Rain-fed Variable Cropland Cropland Difference p-value (1) (2) (3) (4) Harvested Area (1000 Ha) 0.00 0.09 -0.08 0.00*** Production (1000 T) 0.57 0.23 0.33 0.00*** Yield (T/Ha) 14.70 5.00 9.70 0.00*** Production Gap (1000 T) 0.57 0.63 -0.06 0.00*** Notes. In this table we compare agricultural outcomes across rain-fed croplands and irrigated crop- lands. Columns (1) and (2) compute the averages of the outcomes for rain-fed- and irrigated- croplands, respectively. Column (3) indicates the difference in their means, whereas column (4) denotes whether these mean differences are statistically significant. *p<0.1 **p<0.05 ***p<0.01. Table 2: Aquifer Types Across SSA Volume Per Capita Aquifer Share of Area Volume Per Day Type (%) (Billion Liters) (Liters) (1) (2) (3) (4) Local Shallow 60.77 1698302 4530 Karstic 4.97 138873 370 Major Alluvial 8.16 228088 608 Complex 26.10 729357 1945 Notes. This table presents descriptive statistics on the four aquifer types in Sub-Saharan Africa. Column (1) specifies the aquifer type. Column (2) is the mean share of the aquifer grid area that is covered by the aquifer type. Column (3) presents the corresponding total groundwater volume. Column (4) presents the total groundwater volume available per person per day. To compute the volume per capita per day, we divided the volume by the total population of the country that each groundwater aquifer grid lies in and then summed up the volume per-capita measure. Subsequently, we divided the measure by 365 days to obtain an estimate of groundwater availability per person per day. We use the World Bank’s curated groundwater aquifer dataset to construct this table. 26 Table 3: T-Test Results and Mean Differences Only Any Variable Rainfed Irrigated Difference p-value (1) (2) (3) (4) Growing Season: Longest Consecutive Dry Days 85.718 90.157 + 0.000*** Growing Cycle: Crop Water Deficit (mm) 33.741 54.269 + 0.000*** Terrain Median Altitude 733.693 834.656 + 0.000*** Terrain Median Slope 4.101 4.678 + 0.000*** Nutrient Availability, Low Inputs 7.523 8.314 + 0.000*** Nutrient Retention Capacity 8.823 9.408 + 0.000*** Water Collection Sites 1.27 1.74 + 0.000*** No of Growing Season Days 197.36 155.923 - 0.000*** No. of Rain Days 158.776 114.575 - 0.000*** Grid Suitability: MS-VS 8179.999 5450.374 - 0.000*** Crop Actual Evapotranspiration (mm) 259.967 216.966 - 0.000*** Growing Cycle: Accumulated Temp (C / Day) 3183.938 3164.676 - 0.000*** Moisture Constraint in Growth 9158.173 6337.78 - 0.000*** Temp Winter 31.59 30.287 - 0.000*** Temp Spring 32.04 30.884 - 0.000*** Temp Fall 28.711 27.075 - 0.000*** Temp Summer 31.73 28.919 - 0.000*** Prec Winter 969.649 929.788 - 0.000*** Prec Spring 1057.05 898.431 - 0.000*** Prec Fall 1276.33 736.766 - 0.000*** Prec Summer 1028.897 645.531 - 0.000*** PDSI Spring 0.165 0.311 + 0.000*** PDSI Fall -0.425 -0.29 + 0.000*** PDSI Summer -0.223 0.248 + 0.000*** PDSI Winter 0.215 0.116 - 0.000*** Rooting Condition, High Inputs 9.105 8.704 - 0.000*** Rooting Condition, Low Inputs 9.107 8.706 - 0.000*** Root Oxygen Availability, High Inputs 10.017 10.017 + 0.794 Root Oxygen Availability, Low Inputs 9.391 9.389 - 0.539 Presence of Salinity and Sodicity, High Inputs 9.884 9.817 - 0.000*** Presence of Salinity and Sodicity, Low Inputs 9.886 9.82 - 0.000*** Presence of Lime and Gypsum, High Inputs 9.993 9.956 - 0.000*** Presence of Lime and Gypsum, Low Inputs 9.995 9.958 - 0.000*** Workability, High Inputs 9.149 8.775 - 0.000*** Workability, Low Inputs 8.704 8.115 - 0.000*** Annual P-PET Ratio 69.273 50.969 - 0.000*** N 905150 84259 Notes. In this table we compare agro-climatic factors and conditions across croplands that were solely rain-fed and croplands that practiced any level of irrigation. Columns (1) and (2) compute the averages of the outcomes for rain- fed- and irrigated- croplands, respectively. Column (3) indicates the sign difference in their means, whereas column (4) denotes whether these mean differences are statistically significant. *p<0.1 **p<0.05 ***p<0.01. 27 Table 4: Simulation Results Using World Bank Measure of GW Volume Grids Median Mean Satisfying Grids with Gains in Gains in Constraints Production Production Production Model Crop (%) Gains (%) (%) (%) (1) (2) (3) (4) (5) (6) SD Barley 5.47 2.75 29.57 101.40 SD Millet 5.47 4.31 76.35 102.88 SD Maize 5.47 43.45 107.96 184.96 SD Wetland Rice 5.47 41.41 108.09 126.08 SD Sorghum 5.47 18.22 89.49 112.43 SD Wheat 5.47 8.05 27.97 46.87 SRHS Barley 5.47 4.68 33.28 65.23 SRHS Millet 5.47 0.00 SRHS Maize 5.47 57.93 123.26 221.50 SRHS Wetland Rice 5.47 44.04 114.64 135.14 SRHS Sorghum 5.47 40.36 129.42 242.49 SRHS Wheat 5.47 14.02 53.08 102.22 Notes. In this table we present the results of the SD approach using the groundwa- ter supply or volume measure using the aquifer dataset constructed by World Bank. Each row represents one simulation. Column (1) specifies the choice of model. Col- umn (2) specifies the crop for which the analysis was run. Column (3) presents the fraction of all croplands where the constraints were met. Column (4) presents the share of croplands where we see production gains, as compared to the all the crop- lands where the constraints were met. Columns (5) and (6) present the median and mean of the simulated gains in production. All the results presented imposed con- straints on groundwater depth ≤ 7m, an extraction recharge ratio ≤ 0.7, and aquifer productivity ≥ 0.05. 28 Table 5: Simulation Results Using SD Approach and World Bank Measure of GW Volume Extraction Grids Median Mean GW Recharge Aquifer Satisfying Grids with Gains in Gains in Depth Ratio Productivity Constraints Production Production Production Crop Constraint Constraint Constraint (%) Gains (%) (%) (%) (1) (2) (3) (4) (5) (6) (7) (8) Barley 7.00 0.70 0.05 5.47 2.75 29.57 101.40 Barley 7.00 0.70 5.47 2.75 29.57 101.40 Barley 7.00 0.05 5.48 2.74 29.57 101.40 Barley 0.70 0.05 20.98 2.94 47.91 98.42 Barley 7.00 5.48 2.74 29.57 101.40 Barley 0.70 20.98 2.95 47.96 98.47 Barley 0.05 24.91 2.48 47.91 98.42 Barley 24.91 2.48 47.96 98.47 Millet 7.00 0.70 0.05 5.47 4.31 76.35 102.88 Millet 7.00 0.70 5.47 4.31 76.35 102.88 Millet 7.00 0.05 5.48 4.31 76.35 102.88 Millet 0.70 0.05 20.98 7.04 71.87 105.05 Millet 7.00 5.48 4.31 76.35 102.88 Millet 0.70 20.98 7.04 71.87 105.05 Millet 0.05 24.91 5.93 71.87 105.05 Millet 24.91 5.93 71.87 105.05 Maize 7.00 0.70 0.05 5.47 43.45 107.96 184.96 Maize 7.00 0.70 5.47 43.45 107.96 184.96 Maize 7.00 0.05 5.48 43.39 107.99 185.02 Maize 0.70 0.05 20.98 30.56 89.39 158.64 Maize 7.00 5.48 43.39 107.99 185.02 Maize 0.70 20.98 30.56 89.38 158.63 Maize 0.05 24.91 25.75 89.50 158.90 Maize 24.91 25.75 89.49 158.89 Wetland Rice 7.00 0.70 0.05 5.47 41.41 108.09 126.08 Wetland Rice 7.00 0.70 5.47 41.41 108.09 126.08 Wetland Rice 7.00 0.05 5.48 41.34 108.09 126.08 Wetland Rice 0.70 0.05 20.98 24.76 105.64 133.09 Wetland Rice 7.00 5.48 41.34 108.09 126.08 Wetland Rice 0.70 20.98 24.76 105.64 133.09 Wetland Rice 0.05 24.91 20.85 105.64 133.09 Wetland Rice 24.91 20.85 105.64 133.09 Sorghum 7.00 0.70 0.05 5.47 18.22 89.49 112.43 Sorghum 7.00 0.70 5.47 18.22 89.49 112.43 Sorghum 7.00 0.05 5.48 18.24 88.61 112.10 Sorghum 0.70 0.05 20.98 15.72 59.99 103.19 Sorghum 7.00 5.48 18.24 88.61 112.10 Sorghum 0.70 20.98 15.72 59.99 103.19 Sorghum 0.05 24.91 14.08 63.90 107.25 Sorghum 24.91 14.08 63.90 107.24 Wheat 7.00 0.70 0.05 5.47 8.05 27.97 46.87 Wheat 7.00 0.70 5.47 8.05 27.97 46.87 Wheat 7.00 0.05 5.48 8.14 28.24 48.22 Wheat 0.70 0.05 20.98 8.83 36.39 61.23 Wheat 7.00 5.48 8.14 28.24 48.22 Wheat 0.70 20.98 8.83 36.39 61.24 Wheat 0.05 24.91 7.81 38.93 69.25 Wheat 24.91 7.81 38.96 69.26 Notes. In this table we present the results of the SD approach using the groundwater supply or volume measure using the aquifer dataset constructed by World Bank. Each row represents one simulation. Column (1) specifies the crop for which the analysis was run. Columns (2)-(4) specify whether or not constraints were imposed; if yes, then what the thresholds were. Column (5) presents the fraction of all croplands where the constraints were met. Column (6) presents the share of croplands where we see production gains, as compared to the all the croplands where the constraints were met. Columns (7) and (8) present the median and mean of the simulated gains in production. 29 Table 6: Simulation Results Using SRHS Approach and World Bank Measure of GW Vol- ume Extraction Grids Median Mean GW Recharge Aquifer Satisfying Grids with Gains in Gains in Depth Ratio Productivity Constraints Production Production Production Crop Constraint Constraint Constraint (%) Gains (%) (%) (%) (1) (2) (3) (4) (5) (6) (7) (8) Barley 7.00 0.70 0.05 5.47 4.68 33.28 65.23 Barley 7.00 0.70 5.47 4.68 33.28 65.23 Barley 7.00 0.05 5.48 4.67 33.28 65.23 Barley 0.70 0.05 20.98 4.33 41.85 68.83 Barley 7.00 5.48 4.67 33.28 65.23 Barley 0.70 20.98 4.33 41.87 68.89 Barley 0.05 24.91 3.64 41.85 68.83 Barley 24.91 3.65 41.87 68.89 Millet 7.00 0.70 0.05 5.47 0.00 Millet 7.00 0.70 5.47 0.00 Millet 7.00 0.05 5.48 0.00 Millet 0.70 0.05 20.98 0.00 Millet 7.00 5.48 0.00 Millet 0.70 20.98 0.00 Millet 0.05 24.91 0.00 Millet 24.91 0.00 Maize 7.00 0.70 0.05 5.47 57.93 123.26 221.50 Maize 7.00 0.70 5.47 57.93 123.26 221.50 Maize 7.00 0.05 5.48 57.84 123.26 221.52 Maize 0.70 0.05 20.98 42.72 106.65 246.21 Maize 7.00 5.48 57.84 123.26 221.52 Maize 0.70 20.98 42.73 106.64 246.20 Maize 0.05 24.91 35.99 106.67 246.28 Maize 24.91 35.99 106.67 246.26 Wetland Rice 7.00 0.70 0.05 5.47 44.04 114.64 135.14 Wetland Rice 7.00 0.70 5.47 44.04 114.64 135.14 Wetland Rice 7.00 0.05 5.48 43.96 114.64 135.14 Wetland Rice 0.70 0.05 20.98 26.44 112.56 136.85 Wetland Rice 7.00 5.48 43.96 114.64 135.14 Wetland Rice 0.70 20.98 26.44 112.56 136.85 Wetland Rice 0.05 24.91 22.27 112.56 136.85 Wetland Rice 24.91 22.27 112.56 136.85 Sorghum 7.00 0.70 0.05 5.47 40.36 129.42 242.49 Sorghum 7.00 0.70 5.47 40.36 129.42 242.49 Sorghum 7.00 0.05 5.48 40.35 129.23 242.05 Sorghum 0.70 0.05 20.98 37.39 93.76 195.35 Sorghum 7.00 5.48 40.35 129.23 242.05 Sorghum 0.70 20.98 37.39 93.74 195.34 Sorghum 0.05 24.91 32.33 93.39 194.80 Sorghum 24.91 32.33 93.38 194.80 Wheat 7.00 0.70 0.05 5.47 14.02 53.08 102.22 Wheat 7.00 0.70 5.47 14.02 53.08 102.22 Wheat 7.00 0.05 5.48 14.09 53.80 102.49 Wheat 0.70 0.05 20.98 12.51 57.32 91.78 Wheat 7.00 5.48 14.09 53.80 102.49 Wheat 0.70 20.98 12.51 57.37 91.79 Wheat 0.05 24.91 10.91 60.10 94.95 Wheat 24.91 10.91 60.11 94.96 Notes. In this table we present the results of the SRHS approach using the groundwater supply or volume measure using the aquifer dataset constructed by World Bank. Each row represents one simulation. Column (1) specifies the crop for which the analysis was run. Columns (2)-(4) specify whether or not constraints were imposed; if yes, then what the thresholds were. Column (5) presents the fraction of all croplands where the constraints were met. Column (6) presents the share of croplands where we see production gains, as compared to the all the croplands where the constraints were met. Columns (7) and (8) present the median and mean of the simulated gains in production. 30 Table 8: Simulation Results Using SD Approach and BGS Measure of GW Volume Extraction Grids Median Mean GW Recharge Aquifer Satisfying Grids with Gains in Gains in Depth Ratio Productivity Constraints Production Production Production Crop Constraint Constraint Constraint (%) Gains (%) (%) (%) (1) (2) (3) (4) (5) (6) (7) (8) Barley 7.00 0.70 0.05 5.47 3.20 85.47 158.42 Barley 7.00 0.70 5.47 3.20 85.47 158.42 Barley 7.00 0.05 5.48 3.19 85.47 158.42 Barley 0.70 0.05 20.98 3.34 92.72 146.09 Barley 7.00 5.48 3.19 85.47 158.42 Barley 0.70 20.98 3.34 92.95 146.11 Barley 0.05 24.91 2.81 92.72 146.09 Barley 24.91 2.81 92.95 146.11 Millet 7.00 0.70 0.05 5.47 4.01 59.57 89.75 Millet 7.00 0.70 5.47 4.01 59.57 89.75 Millet 7.00 0.05 5.48 4.00 59.57 89.75 Millet 0.70 0.05 20.98 7.13 75.11 114.85 Millet 7.00 5.48 4.00 59.57 89.75 Millet 0.70 20.98 7.13 75.11 114.85 Millet 0.05 24.91 6.00 75.11 114.85 Millet 24.91 6.00 75.11 114.85 Maize 7.00 0.70 0.05 5.47 48.27 125.63 245.42 Maize 7.00 0.70 5.47 48.27 125.63 245.42 Maize 7.00 0.05 5.48 48.19 125.64 245.48 Maize 0.70 0.05 20.98 33.76 112.38 223.85 Maize 7.00 5.48 48.19 125.64 245.48 Maize 0.70 20.98 33.76 112.38 223.83 Maize 0.05 24.91 28.44 112.42 224.09 Maize 24.91 28.44 112.40 224.07 Wetland Rice 7.00 0.70 0.05 5.47 45.43 127.22 146.03 Wetland Rice 7.00 0.70 5.47 45.43 127.22 146.03 Wetland Rice 7.00 0.05 5.48 45.35 127.22 146.03 Wetland Rice 0.70 0.05 20.98 26.71 124.41 150.98 Wetland Rice 7.00 5.48 45.35 127.22 146.03 Wetland Rice 0.70 20.98 26.71 124.41 150.98 Wetland Rice 0.05 24.91 22.49 124.41 150.98 Wetland Rice 24.91 22.49 124.41 150.98 Sorghum 7.00 0.70 0.05 5.47 18.40 100.18 115.91 Sorghum 7.00 0.70 5.47 18.40 100.18 115.91 Sorghum 7.00 0.05 5.48 18.48 100.49 116.65 Sorghum 0.70 0.05 20.98 16.03 88.25 128.10 Sorghum 7.00 5.48 18.48 100.49 116.65 Sorghum 0.70 20.98 16.03 88.25 128.09 Sorghum 0.05 24.91 14.43 96.28 144.78 Sorghum 24.91 14.43 96.28 144.77 Wheat 7.00 0.70 0.05 5.47 8.46 45.92 66.40 Wheat 7.00 0.70 5.47 8.46 45.92 66.40 Wheat 7.00 0.05 5.48 8.54 46.09 67.56 Wheat 0.70 0.05 20.98 9.23 50.89 80.35 Wheat 7.00 5.48 8.54 46.09 67.56 Wheat 0.70 20.98 9.23 50.93 80.35 Wheat 0.05 24.91 8.14 53.92 87.59 Wheat 24.91 8.15 53.93 87.59 Notes. In this table we present the results of the SD approach using the groundwater supply or volume measure using the aquifer dataset constructed by BGS. Each row represents one simulation. Column (1) specifies the crop for which the analysis was run. Columns (2)-(4) specify whether or not constraints were imposed; if yes, then what the thresholds were. Column (5) presents the fraction of all croplands where the constraints were met. Column (6) presents the share of croplands where we see production gains, as compared to the all the croplands where the constraints were met. Columns (7) and (8) present the median and mean of the simulated gains in production. 31 Table 9: Simulation Results Using SRHS Approach and BGS Measure of GW Volume Extraction Grids Median Mean GW Recharge Aquifer Satisfying Grids with Gains in Gains in Depth Ratio Productivity Constraints Production Production Production Crop Constraint Constraint Constraint (%) Gains (%) (%) (%) (1) (2) (3) (4) (5) (6) (7) (8) Barley 7.00 0.70 0.05 5.47 5.05 52.87 100.25 Barley 7.00 0.70 5.47 5.05 52.87 100.25 Barley 7.00 0.05 5.48 5.04 52.87 100.25 Barley 0.70 0.05 20.98 4.68 68.80 101.21 Barley 7.00 5.48 5.04 52.87 100.25 Barley 0.70 20.98 4.68 68.80 101.26 Barley 0.05 24.91 3.94 68.80 101.21 Barley 24.91 3.94 68.80 101.26 Millet 7.00 0.70 0.05 5.47 0.00 Millet 7.00 0.70 5.47 0.00 Millet 7.00 0.05 5.48 0.00 Millet 0.70 0.05 20.98 0.00 Millet 7.00 5.48 0.00 Millet 0.70 20.98 0.00 Millet 0.05 24.91 0.00 Millet 24.91 0.00 Maize 7.00 0.70 0.05 5.47 65.07 158.09 286.43 Maize 7.00 0.70 5.47 65.07 158.09 286.43 Maize 7.00 0.05 5.48 64.97 158.14 286.44 Maize 0.70 0.05 20.98 48.09 148.02 358.88 Maize 7.00 5.48 64.97 158.14 286.44 Maize 0.70 20.98 48.09 148.02 358.85 Maize 0.05 24.91 40.51 148.09 358.91 Maize 24.91 40.51 148.07 358.89 Wetland Rice 7.00 0.70 0.05 5.47 48.72 129.00 155.34 Wetland Rice 7.00 0.70 5.47 48.72 129.00 155.34 Wetland Rice 7.00 0.05 5.48 48.65 129.00 155.34 Wetland Rice 0.70 0.05 20.98 28.81 126.88 156.57 Wetland Rice 7.00 5.48 48.65 129.00 155.34 Wetland Rice 0.70 20.98 28.80 126.88 156.57 Wetland Rice 0.05 24.91 24.26 126.88 156.57 Wetland Rice 24.91 24.25 126.88 156.57 Sorghum 7.00 0.70 0.05 5.47 43.03 148.48 291.29 Sorghum 7.00 0.70 5.47 43.03 148.48 291.29 Sorghum 7.00 0.05 5.48 43.07 148.61 291.26 Sorghum 0.70 0.05 20.98 43.18 167.81 309.64 Sorghum 7.00 5.48 43.07 148.61 291.26 Sorghum 0.70 20.98 43.18 167.80 309.62 Sorghum 0.05 24.91 37.30 170.86 312.03 Sorghum 24.91 37.30 170.86 312.02 Wheat 7.00 0.70 0.05 5.47 14.29 71.02 113.35 Wheat 7.00 0.70 5.47 14.29 71.02 113.35 Wheat 7.00 0.05 5.48 14.36 71.60 113.55 Wheat 0.70 0.05 20.98 12.84 77.83 105.62 Wheat 7.00 5.48 14.36 71.60 113.55 Wheat 0.70 20.98 12.84 77.88 105.63 Wheat 0.05 24.91 11.19 80.74 108.35 Wheat 24.91 11.19 80.75 108.36 Notes. In this table we present the results of the SRHS approach using the groundwater supply or volume measure using the aquifer dataset constructed by BGS. Each row represents one simulation. Column (1) specifies the crop for which the analysis was run. Columns (2)-(4) specify whether or not constraints were imposed; if yes, then what the thresholds were. Column (5) presents the fraction of all croplands where the constraints were met. Column (6) presents the share of croplands where we see production gains, as compared to the all the croplands where the constraints were met. Columns (7) and (8) present the median and mean of the simulated gains in production. 32 Table 7: Percentage Income Gains Per Capita of 1.90 USD International Poverty Line Country ISO-3 Code Gains (%) 1 Togo TGO 3.20 2 Benin BEN 2.42 3 Guinea GIN 2.39 4 Sierra Leone SLE 1.91 5 Cameroon CMR 0.92 6 Nigeria NGA 0.83 7 Liberia LBR 0.61 8 Tanzania TZA 0.32 9 Kenya KEN 0.24 10 Uganda UGA 0.15 11 ˆ d’Ivoire Cote CIV 0.13 12 Lesotho LSO 0.10 13 Central African Republic CAF 0.05 14 Guinea-Bissau GNB 0.04 15 Zambia ZMB 0.02 16 Angola AGO 0.02 17 Ethiopia ETH 0.02 18 Malawi MWI 0.01 19 Congo, Democratic Republic of COD 0.01 20 Mali MLI 0.01 21 Rwanda RWA 0.01 22 Gabon GAB 0.01 23 Chad TCD 0.01 24 Mozambique MOZ 0.00 25 Burundi BDI 0.00 26 Congo, Rep. of COG 0.00 27 Zimbabwe ZWE 0.00 28 Botswana BWA 0.00 29 Senegal SEN 0.00 30 Burkina Faso BFA 0.00 31 Mauritania MRT 0.00 32 Niger NER 0.00 33 Sudan SDN 0.00 34 Eritrea ERI 0.00 35 South Africa ZAF 0.00 36 eSwatini SWZ 0.00 37 South Sudan SSD 0.00 Notes. In this table we compute the expected percentage income growth from all six ce- reals in our analysis. Income growth is defined as an expected change in income, defined as the monetary value of the simulated production gains, for individuals or farmers at the poverty line. We use the results from the SRHS model (when all three constraints are imposed) for computing the monetary value of production gains. 33 Figures Figure 1: Percentage of Aquifer Type by Region Notes: In this figure, we illustrate the composition of each aquifer type (in terms of percentage) across the major regions of the world. Source: Rodella et al. (2022). 34 Figure 2: Map of Characteristics of Groundwater Aquifers in SSA (a) Recharge Rates (b) Depth to Groundwater (c) Aquifer Productivity Notes: In this figure, we map the characteristics of groundwater aquifers in Sub-Saharan Africa. Panel (a) maps the mean annual recharge rate (in mm) of aquifers. Panel (b) maps the depth to groundwater (in m). Panel (c) maps the mean aquifer yields or productivity of aquifers (in liters per second). A majority of the aquifers have high recharge rates, are at low depth and thereby easy to access, and have sufficiently high yields for irrigation purposes. The data are sourced from MacDonald et al. (2012). 35 Figure 3: Share of Local Shallow Aquifers in Sub-Saharan Africa Notes: In this figure, we map the composition of local shallow aquifers as compared to the total endowment of groundwater in Sub-Saharan Africa. Source: Rodella et al. (2022). 36 Figure 4: Change in Cropland Types over Decades Notes: In this figure, we consider the means of harvest area and crop yield at the grid level for the years 2000 and 2010, for irrigated and rain-fed croplands separately. The data are sourced from FAO and IIASA (2020). 37 Figure 5: Distribution of Croplands by Crop Notes: This figure maps the distribution of croplands (shown in green and blue) in Sub-Saharan Africa for each of the six cereals considered for this analysis. In blue, we highlight the regions that irrigate their croplands in 2010. The data are sourced from FAO and IIASA (2020). 38 Figure 6: Distribution of Production Gains (SD Model) by Crop Notes: This figure maps the distribution of gains in production across croplands in Sub-Saharan Africa for each of the six cereals considered for the simulation analysis using the SD model to compute yield gains. 39 Figure 7: Distribution of Production Gains (SRHS Model) by Crop Notes: This figure maps the distribution of gains in production across croplands in Sub-Saharan Africa for each of the six cereals considered for the simulation analysis using the SRHS model to compute yield gains. 40 Figure 8: Distribution of Production Gains in Maize by Country (SRHS Model) Notes: This bar plot shows the country-wise and country-wide median gains in production for maize using the SRHS model to compute yield gains. We only plot the gains for countries that cultivated maize and fulfilled the three conditions of sustainability, cost-effectiveness, and sufficient groundwater aquifer productivity. We omit the countries where we found the median gains to be zero: these are Chad, Ghana, Namibia, and The Gambia. 41 Figure 9: Distribution of Production Gains in Maize by Country (SD Model) Notes: This bar plot shows the country-wise and country-wide median gains in production for maize using the SD model to compute yield gains. We only plot the gains for countries that cultivated maize and fulfilled the three conditions of sustainability, cost-effectiveness, and sufficient groundwater aquifer productivity. We omit the countries where we found the median gains to be zero: these are Chad, Ethiopia, Kenya, Namibia, and South Sudan. 42 Figure 10: Distribution of Production Gains in Rice by Country (SRHS Model) Notes: This bar plot shows the country-wise and country-wide median gains in production for wetland rice using the SRHS model to compute yield gains. We only plot the gains for countries that cultivated wetland rice and fulfilled the three conditions of sustainability, cost-effectiveness, and sufficient groundwater aquifer productivity. We omit the countries where we found the median gains to be zero: these are Benin, Chad, Ghana, and The Gambia. 43 Figure 11: Distribution of Production Gains in Rice by Country (SD Model) Notes: This bar plot shows the country-wise and country-wide median gains in production for wetland rice using the SD model to compute yield gains. We only plot the gains for countries that cultivated wetland rice and fulfilled the three con- ditions of sustainability, cost-effectiveness, and sufficient groundwater aquifer productivity. We omit the countries where we found the median gains to be zero: these are Benin, Chad, Ghana, and The Gambia. 44 Appendix: For Online Publication Appendix Tables Table A1: Summary of Parameters Used Measure Parameter Value Unit Source Aquifer productivity threshold η 0.5 l/s MacDonald et al. (2012) Groundwater extraction- ν 70 % Ground Water Resource Estima- recharge ratio threshold tion Committee (2015) Aquifer depth constraint ζ 7 m/s MacDonald et al. (2012) Share of water use that is at- ϕ 67.5 % Rodella et al. (2022) tributable to groundwater use Notes: This table presents the parameters used in the simulation, their values, and where they were sourced from. 45 Table A2: A Review of the Literature in SSA Study Region/Country Outcome Variable Estimate Type of Irrigation Crops Jones et al. Rwanda Profits (% growth) 43-62% HSI: gravity fed Horticulture (2022) canals (SW) Dyer and Kenya (Machakos, Net farm revenue (% 13% SSI: household hip Fruits and Shapiro Kiambu and increase as compared pumps (SW-rivers, Vegetables (2023) Embu) to control group GW-public mean) boreholes) Ngango Rwanda Avg. maize yields 193– 200 SSI: Drip, sprinkle Maize and Hong (increases by, kg/ha (SW) (2021) compared to non-adopters of irrigation) Aseyehegn, Ethiopia (Laelay Total Income (as 37.03% SSI: Micro-dams Vegetables Yirga, and Maichew district, compared to (SW) and Cereals Rajan Tigray region) irrigation non-users) (2012) Burney et Sudano–Sahel Produce per PVDI 1.9 SSI: PVDI pump Vegetables al. (2010) (Northern Benin) system (Tones/ (SW/ GW) month) Dillon Northern Mali Agricultural 1.25-1.89 SSI: Motor pumps Cereals (2008) production (SW) (increases by, tons per household) Dillon Northern Mali Yield (increase by 2.5- 3.8 SSI: Motor pumps Rice (2011) tones/ ha) (SW) Ersado Ethiopia (Tigray) Crop yield (% higher, 19.25% SSI: Dams (SW) Cereals and (2005) compared to more Vegetables distant villages from dams) Nkhata Malawi (Bwanje Annual agricultural 65%- SSI: (SW) Cereals and (2014) Valley) income (% more as 134% Legumes compared to non-participants) Tesfaye, Ethiopia (Ada Total value of Birr SSI: dams (SW) Vegetables Bogale, Liben district) production 11777, and Legumes Namara, (irrigation users, Birr 2255 and Bacha non-users) (2008) Maru, Ethiopia (Awash Annual on-farm ETB SSI: open channel Grains and Haileslassie, sub-basin) income (ETB, as 9024±2267 gravity-river water Cereals and Zeleke compared to (SW) (2023) non-irrigating farmers) Jambo, Ethiopia Mean incomes ETB ETB SSI: rivers, lake Vegetables, Alemu, and (Adamitulu Jido (irrigation users vs 102213 (SW) Cereals, and Tasew Komoblcha non-users) (users), Legumes (2021) district) ETB 7660 (non- users) 46 Study Region/Country Outcome Variable Estimate Type of Irrigation Crops Balana Ghana IRR (alt/base): Water 0.48, 0.46 SSI: watering can/ Vegetables et (Bihinaayili) can (SW), motor motor pump (SW) al. pump (SW) (2020) Ghana (Zanlerigu) IRR (alt/base): Water 0.51, 0.19 SSI: watering can/ Vegetables can (GW), motor motor pump (sha. pump (GW) GW) Adeoti, Ghana (Volta and Net income, TP $1443 SSI: TP (SW) Vegetables Barry, Ashanti) adopters vs. non (adop.), Namara, adopters $1050 Kamara, (non-ad.) and Titiati (2007) Connor, Mauritania (village Yield (kg/ha): Rice, 3972, SHI: Pumped, flood Cereals and Comas, beside Senegal Sorghum, Cowpea 2242, irrigation (SW) Legumes Macpher- river) 3017 son, and Mateos (2008) ** Gadedjisso- West Africa Yields (kg/ha), 100% 2200.4 Irrigation: Maize Tossou, (Northern Togo) FI (100% volume kg/ha micro-sprinkler Avell´an, irrigated water) and ¨ Schutze (2020) Tillie, Niger Farm income, 7% SSI: wells, reservoir, Cereals, Louhichi, country level (%, pumps (SW, GW, Legumes, and and Gomez increase by) RF) Vegetables Y Paloma (2018) ** Mupaso, Zimbabwe IRR Flood, Sprinkler, 42%, SHI: dam (SW) Cereals and Manzungu, (Chirumanzu Drip 38%, Legumes Mutam- District) 23% bara, and Hanyani- Mlambo (2014) Owusu Ghana VA /acre (US $): GW $11.18, SSI: motor pump/ Cereals and (2016) motor pump, SW $5.07, manual pump Vegetables motor pump, GW $12.86 (GW/SW) manual pump Gebregziabher,North-eastern IRR (Total for 0.29; SSI: Wells (GW) Vegetables Villholth, Ethiopia Raya-Kobo); IRR 0.47; 0.14 and Fruits Hanjra, (Raya-Kobo (Raya); IRR (Kobo) Yirga, and Valley) Namara (2013) Gebregziabher,Ethiopia (Tigray) Average income gain Birr SSI: micro-dam N.A. Namara, (Birr, per household 4000- (SW), shallow wells and Holden per year) 4500 (GW) (2009) 47 Study Region/Country Outcome Variable Estimate Type of Irrigation Crops Torou et al. South-western Net income (XOF) / XOF SSI: Boreholes, Vegetables (2013) Niger (Dantiandou season/ farm 40,000 wells, ponds (GW) Valley) Namara et Ghana (White Economy share US$1.1 SSI: Wells (Shallow Vegetables al. (2011) Volta Basin) (US$); US$/GW mn; GW) and other irrigating US$455 crops compound/yr. Dauda, South-west Avg. net profit per US$ 914 SSI: River (SW) Vegetables Asiribo, Nigeria farmer (2 months Oladele, period) Saka, and Salahu (2009) Acheampong Northern Ghana Net revenue of crop US$300- Small dams- Vegetables et al. (2014) (Upper East, yield per hectare per 1700 Pipe/wells/ Upper West) season canal/pump (GW/SW) Chazovachii Zimbabwe Average Yields per 60 kg- SSI: N.A. Vegetables (2012) (Panganai) Season per Farmer 1000 kg and Cereals (range of crops) L. You et al. AFR/ SSA IRR: AFR (LSI: SW), 6.61%, LSI: Dam-based Cereals, (2011) SSA (LSI: SW), AFR 5.68%, (SW)/ SSI: Run-off Vegetables, (SSI: GW) 28% (GW) and Fruits L. Z. You SSA BCR: Baseline, 8.9, 4.91, LSI: Dams existing Cereals, (2008) medium irrigation 3.68 (SW) Legumes, cost, high irrigation Vegetables, cost Fruits, and high value crops SSA BCR: Baseline, 1.9, 1.35 SSI: Rainwater, Cereals, medium irrigation small reservoir Legumes, cost (SW) Vegetables, Fruits, and high value crops L. You, Xie, Kenya IRR: SSI, LSI 17% - SSI: Surface run off, High-value Wood- 32%, rainfall (SW)/ LSI: crops Sichra, >12% Dam-based (SW) Guo, and Wang (2014) Svendsen, AFR/ SSA Irrigated output (% 37.7%, Irrigation (general) Cash and Ewing, and of total value of agri. 24.5% food crops Msangi output): AFR, SSA (2009) Fuglie and SSA Avg. farm yields (as 90% Irrigation (general) N.A. Rada (2013) compared to rainfed (higher) areas) 48 Study Region/Country Outcome Variable Estimate Type of Irrigation Crops Xie, You, SSA Net revenue US$ $14- $22 SHI: pumps, Vegetables, Wielgosz, billion/yr (irrigated reservoir, river and Cereals and Ringler area-20-30mn ha) (GW/SW) (2014) Xie, You, Nigeria Net revenue (mn $610- SSI: Rivers, stream Rice, Maize, and USD) per year, min.- $1353 (SW) and Takeshima max. Vegetables (2017) ** Briceno- SSA Economic rates of 22.20% Irrigation (general) N.A. Garmendia returns and Foster (2009) Inocencio et SSA EIRR (compared to 14% Irrigation: River, Vegetables, al. (2007) non-SSA countries dam, etc. (SW/GW) Fodder, etc Asia/ Latin America, rehab. projects) Concern Malawi (Dedza, Total irrigation MK Irrigation: gravity, Maize, Universal Ntcheu) benefits per 0.1 ha 6220, earth dams, river Tomatoes and (2012) (MK): Grain Maize, MK (SW) Cobs Tomatoes, Cobs 144562, MK 40000 Namara et Ghana Yield min-max (t/ha) 3.5 - 28.6 Irrigation: well, Cereals and al. (2011) t/ha river, reservoir Vegetables (SW/GW) Schonberger SSA Agricultural Yields ≥50% Irrigation (GW) N.A. and Wijnen (2018) Notes: SSA- Sub-Sharan Africa, AFR- Africa, BCR- Benefit cost ratio, IRR- Internal rate of return, EIRR- Eco- nomic internal rate of return, GW- Ground water, GWI - Ground water irrigation SW- Surface water, SSI- Small Scale Irrigation, LSI- Large scale irrigation, HSI- Hillside irrigation, RF- Rainfall, VA- Value added, mn- million, hectare- ha, N.A.- Not available, SHI- Smallholder irrigation, PVDI- Photovoltaic (or solar) powered drip irrigation, ETB- Ethiopian Birr, TP- Treadle pump, FI- Full irrigation, MK- Malawian Kwacha currency, XOF- West African CFA franc currency (1 XOF= US$0.0017), ** Simulation study. 49 Table A3: Simulation Results Using World Bank Measure of GW Volume Grids Median Mean Satisfying Grids with Gains in Gains in Constraints Production Production Production Model Crop (%) Gains (%) (%) (%) (1) (2) (3) (4) (5) (6) SD Barley 15.46 3.97 47.05 97.54 SD Millet 15.46 8.16 72.39 99.28 SD Maize 15.46 39.57 88.54 150.86 SD Wetland Rice 15.46 33.58 105.61 132.98 SD Sorghum 15.46 19.54 64.67 100.57 SD Wheat 15.46 11.13 33.31 54.98 SRHS Barley 15.46 5.84 41.79 68.73 SRHS Millet 15.46 0.00 SRHS Maize 15.46 54.87 106.32 216.84 SRHS Wetland Rice 15.46 35.86 112.55 136.81 SRHS Sorghum 15.46 45.30 113.01 210.35 SRHS Wheat 15.46 16.11 54.01 90.20 Notes. In this table we present the results of the SD approach using the groundwa- ter supply or volume measure using the aquifer dataset constructed by World Bank. Each row represents one simulation. Column (1) specifies the choice of model. Col- umn (2) specifies the crop for which the analysis was run. Column (3) presents the fraction of all croplands where the constraints were met. Column (4) presents the share of croplands where we see production gains, as compared to the all the croplands where the constraints were met. Columns (5) and (6) present the median and mean of the simulated gains in production. All the results presented imposed constraints on groundwater depth ≤ 50m, an extraction recharge ratio ≤ 0.7, and aquifer productivity ≥ 0.05. 50 Figure A1: Distribution of Production Gains (SD Model) by Crop Notes: This figure maps the distribution of gains in production across croplands in Sub-Saharan Africa for each of the six cereals considered for the simulation analysis using the SD model to compute yield gains. The results plotted fulfilled the three conditions of sustainability, cost-effectiveness (with a depth constraint of 50m), and sufficient groundwater aquifer productivity. Appendix Figures 51 Figure A2: Distribution of Production Gains (SRHS Model) by Crop Notes: This figure maps the distribution of gains in production across croplands in Sub-Saharan Africa for each of the six cereals considered for the simulation analysis using the SRHS model to compute yield gains. The results plotted fulfilled the three conditions of sustainability, cost-effectiveness (with a depth constraint of 50m), and sufficient groundwater aquifer productivity. 52 Figure A3: Distribution of Production Gains in Maize by Country (SRHS Model) Notes: This bar plot shows the country-wise and country-wide median gains in production for maize using the SRHS model to compute yield gains. We only plot the gains for countries that cultivated maize and fulfilled the three conditions of sustainability, cost-effectiveness (with a depth constraint of 50m), and sufficient groundwater aquifer productivity. We omit the countries where we found the median gains to be zero: these are Ghana, Namibia, The Gambia, and Somalia. 53 Figure A4: Distribution of Production Gains in Maize by Country (SD Model) Notes: This bar plot shows the country-wise and country-wide median gains in production for maize using the SD model to compute yield gains. We only plot the gains for countries that cultivated maize and fulfilled the three conditions of sustainability, cost-effectiveness (with a depth constraint of 50m), and sufficient groundwater aquifer productivity. We omit the countries where we found the median gains to be zero: these are Ethiopia, Namibia, Somalia, South Sudan, and Sudan. 54 Figure A5: Distribution of Production Gains in Rice by Country (SRHS Model) Notes: This bar plot shows the country-wise and country-wide median gains in production for wetland rice using the SRHS model to compute yield gains. We only plot the gains for countries that cultivated wetland rice and fulfilled the three conditions of sustainability, cost-effectiveness (with a depth constraint of 50m), and sufficient groundwater aquifer productivity. We omit the countries where we found the median gains to be zero: these are Ghana and The Gambia. 55 Figure A6: Distribution of Production Gains in Rice by Country (SD Model) Notes: This bar plot shows the country-wise and country-wide median gains in production for wetland rice using the SD model to compute yield gains. We only plot the gains for countries that cultivated wetland rice and fulfilled the three condi- tions of sustainability, cost-effectiveness (with a depth constraint of 50m), and sufficient groundwater aquifer productivity. We omit the countries where we found the median gains to be zero: these are Ghana and The Gambia. 56