Main

Precipitation is paramount to ensure a productive agricultural system that can fulfil society’s demand for food1. Yet global precipitation patterns are likely to become more erratic as climate extremes and variability have increased in frequency and intensity in recent decades2,3. More than half of loss incidences in global crop production are attributed to climate extremes4. Whereas floods could lead to 4% anomaly in crop yields5, drought can cause up to 10% crop production loss and 8% yield loss6,7. Climate variability also accounts for more than one-third of the variability in global crop yields8. Thus, overlooking the risk of anomalous precipitation on global crop production undermines the resilience of the food systems, particularly in light of climate change.

Forests can provide stable moisture to their downwind regions, especially during dry seasons and climate extremes, such as droughts and heatwaves9,10,11. On average, forests evaporate more moisture than other land covers12 through several processes, such as interception from forest canopy and litter13, deeper root access to groundwater14 and higher available energy through lower albedo15. Changing forest cover to another non-forested land cover significantly reduces evaporation16,17, affects downwind precipitation18,19 and modulates onset of rainy seasons20,21. Likewise, deforestation leads to lower precipitation on downwind agriculture specifically22,23, which affects agricultural production and revenue24,25, although most studies on this topic mainly focus on the Amazon. Forests’ contribution to downwind precipitation and agriculture may vary between extratropical and tropical regions. Tropical forests significantly buffer the variability of downwind precipitation over a long term26, whereas extratropical forests are crucial during critical periods, such as through sustained evaporation during heatwaves27 and peak evaporation during the growing season28. Studying moisture flows from forests to agriculture globally helps broaden our understanding of the dependence of agriculture on forests and addresses the knowledge gap on the understudied extratropical forests.

Furthermore, given that agricultural crops are distributed globally through trade, a key question is whether the provision of precipitation from forests to agriculture may render crop importers to be indirectly dependent on forests. As countries are linked through the ever-increasingly complex trade network29, the international trade could expose importing countries to climate-induced variability and shocks of crop production in exporting countries30,31. A 5% caloric deficit in importing countries following a 10% reduction in crop export could affect the lives of millions of people living in poverty32. Hitherto, it remains unclear how moisture and crop trade flows combine to reveal an additional complexity of the direct and indirect dependence between countries. Meanwhile, transboundary moisture flows across countries have been quantified in earlier studies33,34,35,36, but they lack the forest-to-agriculture context that is relevant for investigating country interdependence through moisture and crops. Understanding the secondary dependence of crop importers on forests upwind of their crop exporters could help identify the vulnerability of global crop supply to changes of forest cover remote from agricultural areas.

In this study, we aim to understand how crop production and export depend on the provision of precipitation from upwind forests within and across countries. We do this by mapping the nationally recycled and transboundary moisture flows from upwind forested to downwind crop-producing and -exporting countries (M_nat,prod, M_nat,exp, M_transb,prod, M_transb,exp in Fig. 1) and classifying the combination of moisture and traded crop flows between countries into circular and cascading flows (MC_transb,circ, MC_transb,casc, MC_nat,casc in Fig. 1). In circular flows, two countries are interdependent, in which forested countries import crops from their downwind crop producers and exporters, whereas cascading flows involve three countries, in which crop importers depend indirectly on forests that are located upwind of their crop producers and exporters. We harmonized different land-cover datasets to produce a map of forest-dominated areas (‘forests’) and agriculture-dominated areas (‘agricultural areas’) globally. We then tracked moisture from extratropical and tropical forests, and their combination, to agricultural areas and estimated the spatially distributed moisture flows within and between countries. We assessed the moisture flows against crop water requirements (CWR) on agricultural areas that are dominantly fed by green water (GW) and blue water (BW). Finally, we combined the moisture flows with traded crop flows to assess the circular and cascading connectivity between countries (Methods).

Fig. 1: Types of moisture-dependent crop production and export.
Fig. 1: Types of moisture-dependent crop production and export.
Full size image

The flow types from top to bottom are: nationally recycled moisture flow within a crop producer (M_nat,prod), transboundary moisture flow to a crop producer (M_transb,prod), circular transboundary flow between a crop-importing forest country and a crop producer and exporter (MC_transb,circ), transboundary moisture flow to a crop producer and exporter (M_transb,exp), transboundary moisture flow to a crop producer and exporter cascading to a crop importer (MC_transb,casc), nationally recycled moisture flow within a crop producer and exporter (M_nat,exp), nationally recycled moisture flow within a crop producer and exporter cascading to a crop importer (MC_nat,casc). Crop flows are by definition transboundary. Only crop exporters that produce crops domestically are considered in the study.

Results

Global moisture flows from forests to agriculture

Our moisture tracking analyses identify hotspots of agricultural areas with the highest dependence on forest evaporation (high percentile of relative recycled moisture from forests μ in Fig. 2a upper panel) and the forests with the highest evaporation (high percentile of forest weighted evaporation ETw in Fig. 2a, top). Agricultural areas that receive > 6% of their annual precipitation from forests are spatially distributed globally across central Canada, northwestern and eastern South America, south of the Sahel, eastern China, eastern Europe, southern Russia and northern Kazakhstan (high percentile μ in Fig. 2a upper panel). In contrast, values of ETw exceeding 356 mm yr−1 are primarily located in tropical regions, northeastern North America, central-latitudinal and eastern Russia and Japan (high percentile ETw in Fig. 2a, top). Meanwhile, both extratropical and tropical forests supply moisture to distant regions covering all agricultural areas, with a gradient of moisture dependence levels that appear to follow the distance from forests (μ in Supplementary Fig. 17 left panels). Countries with extensive areas over which dependence on precipitation from forests is relatively high are also high crop producers and exporters, such as Brazil, Argentina, Canada, Russia, China and Ukraine (high percentile μ in Fig. 2a, top, and crop production Cprod > 90th percentile P90, crop export Cexp > P90 in Fig. 2b). The results show that agricultural areas worldwide are generally well connected to upwind forests through the atmospheric moisture transport.

Fig. 2: Hotspots of moisture flows, crop production and export.
Fig. 2: Hotspots of moisture flows, crop production and export.
Full size image

a, Evaporation weighted by forest-cover ratio in 0.5° × 0.5° grid cell (ETw in mm yr−1) that partly falls as precipitation on agricultural areas (μ in % annual precipitation) at low (μ < 33rd percentile P33), medium (P33 ≤ μ < 67th percentile P67) and high (μ ≥ P67) dependence levels (top). Percentage of total number of cells fulfilling GW and BW crop water requirements (ρ in % annual CWR) for each moisture dependence level (bottom). b, Precipitation on agricultural areas supports crop production (Cprod in Mtonnes yr−1) and crop export (Cexp in Mtonnes yr−1) at the national scale. P50, P90 and max refer to the 50th percentile, 90th percentile and maximum of the corresponding variable. Values inside legend quadrants represent the number of countries in each superimposed category.

Approximately 46% and 10% of the highly dependent agricultural areas have their respective green and blue water requirements for all crops fulfilled, despite some receiving only 6% of their annual precipitation from forests (high percentile μ in Fig. 2a). In addition, whereas the values of moisture flows expressed in percentage of annual or monthly precipitation can be low, they can still be significant for specific crops with low tolerance to reduced precipitation. Among the five major crops, extratropical forests fulfil the CWR over around 40% of GW and BW agricultural cells for wheat in most months and maize in some months, whereas tropical forests do so for all crops in some specific months except wheat (>100% CWR ρ in Supplementary Fig. 16). Interestingly, only extratropical forests have a concurrent timing between high level of moisture supply and peak growing areas for boreal winter wheat and boreal summer maize (Supplementary Fig. 16).

Dependence of crop production and export on national and transboundary forests

Combining the moisture flow analyses with the crop production and export data reveal the countries that are most dependent on national and transboundary forests. There are 155 countries that depend on transboundary forests (transboundary recycled moisture εtransb), which are 47% more than the 105 countries dependent on national forests (nationally recycled moisture εnat) (Supplementary Data 1). Tropical forests supply much higher percentage of precipitation to downwind agriculture than extratropical forests at the country and monthly scales (Supplementary Fig. 17 right panels). Forests in Brazil, Indonesia, the Democratic Republic of Congo (DRC), the Republic of Congo, Gabon, Central African Republic, Venezuela, Colombia, Peru, Ecuador and Bolivia safeguard a high percentile of precipitation to agricultural areas within their national borders that cumulatively account for 10% of global crop production (εnat > P90 in Fig. 3a representing crop production dependence on nationally recycled moisture as M_nat,prod in Fig. 1 and Supplementary Data 1). On the contrary, 16 countries that depend on high percentile of transboundary moisture contribute to only 7% of global crop production (εtransb > P90 in Fig. 3b representing crop production dependence on transboundary moisture as M_transb,prod in Fig. 1 and Supplementary Data 1). Each group of countries that rely heavily on nationally recycled or transboundary moisture contributes to around 11% of global crop export (εnat > P90 in Fig. 3c, εtransb > P90 in Fig. 3d, representing crop export dependence on nationally recycled moisture as M_nat,exp and transboundary moisture as M_transb,exp, respectively, in Fig. 1 and Supplementary Data 1).

Fig. 3: Dependence of crop producers and exporters on recycled moisture.
Fig. 3: Dependence of crop producers and exporters on recycled moisture.
Full size image

a,c, Crop production (a, Cprod in Mtonnes yr−1) and crop export (c, Cexp in Mtonnes yr−1) in countries that are dependent on nationally recycled moisture (εnat in mm yr−1). b,d, Crop production (b) and export (d) in countries dependent on transboundary moisture (εtransb in mm yr−1). P50, P90 and max, refer to the 50th percentile, 90th percentile and maximum of the corresponding variable, respectively. Values inside the legend quadrants represent the number of countries that fall into each superimposed category. The values of nationally recycled moisture, transboundary moisture, crop production and crop export for individual countries are given in Supplementary Data 1.

Among the large crop producers and exporters, Brazil and Argentina stand out as hotspot countries with high percentile of nationally recycled and transboundary moisture, respectively ([εnat, Cprod, Cexp] > P90 in Fig. 3a,c and [εtransb, Cprod, Cexp] > P90 in Fig. 3b,d, respectively). Forests in Brazil recycle around 9% of the national annual precipitation on its agricultural areas while constituting 6% of global crop production and 9% of global crop export (Supplementary Data 1). Meanwhile, Argentina receives approximately 14% of its annual precipitation from transboundary sources, mainly from Brazil, to support its contribution to merely 3% of global crop production, but 6% of global crop export (Supplementary Data 1). In addition, a country may be moderately or highly dependent on national or transboundary forests while producing a large volume of crops but exporting minimally (εnat > P50 in Fig. 3a or εtransb > P50 in Fig. 3b that go down in row in Fig. 3c,d, respectively), such as Malaysia, Venezuela, DRC and Cameroon. In these cases, moisture supply from forests might not be of a vital importance to the global crop supply but to their national food consumption instead.

Some countries show strong differences between the national and transboundary sources of moisture for their agricultural areas. The DRC, Brazil and Indonesia rely around three times as much on nationally recycled moisture as on transboundary moisture (Supplementary Data 1). Conversely, countries such as Togo, Benin and Kyrgyzstan have no dependence on nationally recycled moisture but receive 47 to 66 mm yr−1 of precipitation from transboundary forest sources (zero εnat, εtransb > 0 in Fig. 3 and Supplementary Data 1). Finally, Ukraine, Romania, Paraguay and Uruguay are important exporters, together accounting for 9% of global crop export, that receive medium to high percentile of transboundary moisture ([εtransb, Cexp] > P50 in Fig. 3d and Supplementary Data 1).

Circular and cascading connectivity between countries

Pairing moisture and crop flows further reveals the indirect dependence of crop importers on forests that are located upwind of their crop exporters. Over 90% of moisture supplied to most of the crop producers and exporters is dedicated for the green water-fed agricultural areas, except in Uruguay, Ecuador, Peru and Kazakhstan, where blue water CWR takes up 14%, 16%, 23% and 21% of total CWR, respectively (middle nodes in Fig. 4). The DRC, Bolivia, Brazil, Peru, Canada and Russia are important both as suppliers of moisture and of crops (cascading nationally recycled moisture within crop producers exporting to crop importers as MC_nat,casc in Figs. 1,4). Together, they are responsible for 13% of global crop production and 20% of global crop export (Supplementary Data 2).

In South America, Brazil’s role in supplying moisture to crop producers and exporters is prominent as it appears in all three circular and cascade typologies (Fig. 4). Apart from recycling moisture within its borders, Brazil supplies moisture to agricultural areas in Peru, Ecuador, Bolivia, Paraguay, Uruguay and Argentina, which stand for 10% of global crop export, including to countries in Europe, Asia, Africa and Oceania (cascading transboundary moisture to crop producers exporting to crop importers as MC_transb,casc in Fig. 1; Figs. 4 and 5a and Supplementary Data 2). Thus, countries outside South America are indirectly dependent on moisture from forests in Brazil. Bolivia depends on forests in Brazil and Peru for 12% of their annual precipitation, while providing 9% and 5% of the annual precipitation in Paraguay and Uruguay, respectively (Figs. 4 and 5a). The combined contribution of Brazilian, Bolivian and Peruvian forests is pivotal to the regional crop production and the overall export within and from South America.

Fig. 4: Connectivity between countries through moisture and crop flows.
Fig. 4: Connectivity between countries through moisture and crop flows.
Full size image

Moisture flows (μ) and crop flows (Cexp,rel) are expressed relative to the receiving country, which corresponds to % of annual precipitation in crop exporters (middle nodes) and % of annual import to crop importers (right nodes). In each crop exporter (middle node), moisture flows are divided into crop water requirements for GW (unshaded) and BW (shaded) agriculture, expressed in ratio. All μ and Cexp,rel are above a 5% threshold. Continent names are used as crop importers only when the crop exporters are located in a different continent. The magnitude of Cexp,rel is scaled to the total μ received by the corresponding crop exporter. The values of the individual circular and cascading flows are provided in Supplementary Data 2.

Fig. 5: Illustration of the regional moisture and crop connectivity.
Fig. 5: Illustration of the regional moisture and crop connectivity.
Full size image

a–c, Regional forest and agriculture connectivity through nested flows in South America (a), nested flows in sub-Saharan Africa (b) and transboundary cascading flows in Eurasia (c). Green (orange) lines represent the moisture (crop) flows in % of annual precipitation (% of annual import). Green (purple) dots in the background show the forests (agricultural areas) classified in this study. Not all flows in Fig. 4 are shown here to ease visualization. Arrows are approximately scaled to the corresponding values of moisture or crop flows.

Brazil’s moisture supply to Paraguay, Uruguay and Argentina is offset by their combined importance in providing over 77% of Brazil’s annual crop import through circular flows (circular transboundary moisture to crop producers exporting back to forested nations as MC_transb,circ in Figs. 1, 4 and 5a), Supplementary Data 2. The moisture and crop flows between Brazil and Paraguay are balanced at around 17% (Fig. 4 and 5a). This is contrary to the crop flow from Argentina to Brazil that is disproportionately more than seven times larger than the moisture flow from Brazil to Argentina (Fig. 5a). Similarly, in a cascading flow, Kazakhstan significantly supplies more than 75% of the crops imported by Tajikistan, Uzbekistan and Kyrgyzstan, which is disproportionate to the 6% of their annual precipitation supplied by Russia (MC_transb,casc in Fig. 1, Figs. 4, 5c, Supplementary Data 2). Furthermore, various circular and cascading flows can be nested within a region, adding complexity to the secondary dependence between countries (Fig. 5a, b). For example, Chad and Equatorial Guinea are indirectly dependent on the DRC through a series of crop and moisture flows (Fig. 5b).

Discussion and conclusion

In this study, forests are found to have an important role in the global crop supply by providing precipitation to downwind agricultural areas that contribute to global crop production and export. These results suggest that ensuring the provision of precipitation for agriculture could be leveraged by the governance of atmospheric moisture, which aligns well with the ambition to maintain forest cover worldwide. Specifically, the narrative of forests as a leverage for securing global crop supply could gain impetus from existing restoration and conservation agendas with the imperative to sequester carbon as part of the climate mitigation efforts37,38,39, to curtail biodiversity loss40 and to safeguard the livelihoods of Indigenous People and local communities41. Moreover, the concept of governing atmospheric moisture for enhancing precipitation has been proposed by scholars42,43, but its implementation lacks an entry point to existing or new governance mechanisms34. Atmospheric moisture connects places differently from other freshwater resources, such as rivers and groundwater44, which necessitates a novel approach in rethinking the scales to govern it. Institutional fitness, as opposed to institutional fit, may prove useful in approaching the atmospheric moisture governance with flexibility as a way to augment the crisis-responding, anticipatory or adaptive capacity of the global crop supply45. Accordingly, aspects beyond the quantity of moisture from forests to agriculture, such as the geopolitical, economic and institutional factors, shape the contexts within which the opportunity to govern atmospheric moisture lies.

The circular and cascade typologies in this study can be used to understand the complex dependence between countries and the contexts guiding the atmospheric moisture governance. Circular flows between moisture and crop (MC_transb,circ) provides an ideal example of how bilateral cooperation can be mutually beneficial for the countries involved, such as between Brazil and either Paraguay, Argentina or Uruguay (Figs. 4 and 5a). In such scenarios, one country derives benefits from stable crop import by conserving their forest ecosystem, whereas the other country gains from the favourable climatic conditions that support their agricultural production and export. Similarly, targeting countries that are both dominant national moisture recyclers and large crop producers and exporters (MC_nat,casc), such as Brazil, can provide a win–win scenario for national interest and the global crop supply at large. Brazil’s adaptability to significant fluctuations in the global market through crop substitutions30 can be undermined by continuing deforestation that might impose climate-related constraints on their own agricultural areas. A 0.25% decrease in annual precipitation is estimated for every percentage point of forest loss in the Amazon18 and the potential economic consequences of reduced precipitation are approximated at annual agricultural productivity losses of up to US$1 billion (ref. 22). The deforestation rate in Brazil has gone through periods of improvements and reversals46, partly due to the intricate influence of political changes and land-use policies47.

A win–win scenario through the nationally recycled cascading flow is also applicable to Russia (MC_nat,casc). The trajectory of forest-cover dynamics in Russia is characterized by net forest gains48, although their forests are particularly prone to fire risks, which are likely to be intensified by climate change49. Changes to Russian forests could potentially affect the crop export from Russia but could also create a cascading effect on other notable crop exporters, such as Ukraine and Kazakhstan (MC_transb,casc). The Russian invasion of Ukraine in 2022 demonstrated that a disruption in crop production and export from Ukraine can lead to propagating shocks of global crop distribution along the supply chain50. Consequently, food access in the Middle Eastern, Asian and African countries relying on staple cereal export from both Russia and Ukraine were affected51. Nevertheless, in contrast to the singular power of Brazil or Russia to manage their forest domestically, the power to sustain precipitation for securing crop production in Ukraine or Kazakhstan lies more with the upwind countries, which complicates the motivation to conserve or restore forests. In the case of Ukraine, geopolitical tensions with Russia restrict the chance for collaboration. Therefore, appropriate contexts that are represented in the different typologies should be considered to determine the opportunity and scales to govern atmospheric moisture.

This study comes with three caveats. First, forests are important for agriculture in this context primarily due to their role in moisture recycling as the primary process. Additional processes and feedback also take place upon forest-cover change but are not considered in this study, such as through the changes in energy balance, cloud cover, atmospheric composition and surface roughness, among others52. Second, the role of forests through moisture recycling is a component of a larger set of hydrological processes that take effect upon forest-cover change. Hence, moisture recycling should be assessed against other more local processes affecting the spatially relevant hydrological balance53. Third, whereas the analysis is aggregated at the annual scale, the sensitivity analysis conducted at the monthly scale reveals that the dependence can be considerably stronger in certain months within the growing season (Supplementary Data 1). Future studies that employ finer-grained scales are essential to refine our understanding based on pertinent land-use change projection, agricultural practices, climate change prediction, the determination of growing season and other socio-economic factors.

In this study, we find that global crop supply relies on forests to provide moisture to agricultural areas. Conserving forests could potentially be used as a leverage for mitigating the escalating climate-induced risks threatening agriculture. The circular and cascading typologies offer context and scale versatility to approaching atmospheric moisture governance that could potentially be synergized with the safeguarding of global crop supply and forest conservation.

Methods

Overview

We map out the connectivity between forests and crop supply across countries in four steps. In the first step, we harmonized a land-cover classification dataset that includes forests and croplands with a more detailed cropland dataset that can be paired with crop production data. In the second step, we applied an algorithm to the harmonized dataset to categorize global land areas into forest-dominated or agriculture-dominated areas. We then further categorized these areas based on de facto national boundaries. In the third step, we calculated the moisture flows between forest-dominated areas and agriculture-dominated areas and established the country-to-country moisture flows. Lastly, we paired the moisture flows with crop production and traded crop flows to study the circular and cascading flows across countries.

Data

We harmonized the Moderate Resolution Imaging Spectroradiometer (MODIS) MCD12C1 version 6.1 land-cover classification that includes 17 land-cover types54 with the Monthly irrigated and rain-fed crop areas around the year 2000 (MIRCA2000) dataset55. We also used the Spatial Production Allocation Model (SPAM) version 2.0 cropland dataset56 to substitute MIRCA dataset in sensitivity analyses. MODIS is one of the few land-cover datasets that were used as a basis to develop the SPAM dataset57. After harmonization, we categorized the dominant forests or dominant agricultural areas into different countries based on national boundaries defined by the Database of Global Administrative Areas (GADM) version 4.158. We computed the downwind region of forest-dominated areas using the total precipitation and evaporation from the Earth Retrospective Analysis 5 (ERA5) monthly averaged data on single levels59 and the forward trajectory dataset of UTrack moisture flow60 for the period between 2008 and 2017. The moisture that was transported from forest-dominated areas to agriculture-dominated areas is referred to as ‘moisture flow’ hereafter. The moisture flows were assessed against ERA5’s precipitation and the green water (GW) and blue water (BW) crop water requirements (CWR) from a dataset produced by the WATNEEDS model61. Data of CWR at the annual scale for all 26 MIRCA crops and at the monthly scale for five major crops comprising wheat, maize, rice, soybean and sugarcane were used. Furthermore, the moisture flows were paired with national crop production and traded crop flows across countries. Whereas the SPAM dataset consists of croplands and crop production data, MIRCA only includes croplands and hence was complemented with Earthstat’s M3 crop production dataset for the year 200062. The MIRCA dataset was produced based on the M3 dataset55. Finally, the traded crop flows used in this study were the basis for the Coping with water scarcity in a globalized world (CWASI) database63, which was aligned with the moisture flows to include the period between 2008 and 2017. We refer to them as ‘crop flows.’ All data except the UTrack moisture trajectory were rescaled to 0.5° × 0.5° grid cells using the Climate Data Operators (CDO) package64. All analyses were done in Jupyter Notebook version 6.4.12 and Python version 3. Visualizations were created using the matplotlib65 and plotly66 libraries in Python.

Harmonization of land-cover datasets

We started the harmonization by replacing the land areas that are categorized as ‘croplands’ (type 12) and ‘cropland/natural vegetation mosaic’ (type 14) in MODIS54 by the land areas that are categorized as ‘maximum cropped area’ of croplands in MIRCA55. For the sensitivity analysis, MIRCA was substituted by the ‘physical area’ of croplands in SPAM56. By harmonizing MODIS and MIRCA or SPAM, we could attribute the amount of crop production in M3/MIRCA or SPAM to the croplands selected as agricultural areas in this study.

For defining forests, we followed the International Geosphere-Biosphere Programme classification in MODIS that includes only areas within a cell exceeding 60% of tree cover and 2 metres of tree height54. Meanwhile, for substituting MODIS’ cropland with M3/MIRCA or SPAM, we used the approach done in Lu et al., 2020. The approach considers 100% land areas categorized as ‘croplands’ (hereafter referred as ‘old croplands’), added by 60% land areas categorized as ‘cropland / natural vegetation mosaic’ as the total croplands in MODIS (hereafter referred as ‘new croplands’)57. The new croplands were then replaced by MIRCA’s ‘maximum cropped area’ or SPAM’s ‘physical area’ in the harmonized dataset. The remaining 40% of ‘cropland/natural vegetation mosaic’ will still be considered as ‘cropland/natural vegetation mosaic’ in the harmonized dataset. To compensate with the discrepancies between ‘old croplands’ and ‘new croplands’, we adjusted the remaining land-cover types proportionally using the following equation:

$$\begin{array}{l}\mathrm{if}\,{\mathrm{LC}}_{{\mathrm{MODIS}}_{n,i,\,j}} > 0\to {\mathrm{LC}}_{{{\rm{H}}}_{n,i,\,j}}={\mathrm{LC}}_{{\mathrm{MODIS}}_{n,i,\,j}}+({C}_{{\mathrm{MODIS}}_{i,\,j}}-{C}_{{\mathrm{MIRCA}/\mathrm{SPAM}}_{i,\,j}})\\ \,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\times \frac{{\mathrm{LC}}_{{\mathrm{MODIS}}_{n,i,\,j}}}{\mathop{\sum }\limits_{n=0}^{N}{\mathrm{LC}}_{{\mathrm{MODIS}}_{n,i,\,j}}}\end{array}\,$$

where LCMODIS is the land-cover areas in MODIS after 60% of type 14 ‘cropland/natural vegetation mosaic’ has been assigned to ‘old croplands’, LCH is the harmonized dataset, CMODIS is the cropland areas in MODIS (type 12 ‘croplands’ + 60% of type 14 ‘cropland/natural vegetation mosaic’), CMIRCA/SPAM is the physical areas of croplands in MIRCA or SPAM, n is the type number of land-cover category (between 0 and 16), i is the cell latitude, j is the cell longitude. The reassignment of new land areas is done for each land-cover type, except for croplands’ type number 12. Following the reassignment of new land areas, land areas that have negative values were corrected as 0, which were found in negligible amounts. The total adjusted areas of all land-cover types must not exceed the total land areas. The change of each land-cover type through the harmonization process, expressed as the harmonized dataset subtracted by the original MODIS dataset, are provided in Supplementary Fig. 3. Data harmonization between MODIS and MIRCA and MODIS and SPAM both result in forests covering 21 million km2 of land (12% of the land surface) and agricultural areas spanning over 10 million km2 (6% of the land surface) (Supplementary Fig. 4).

Dominant forests and agricultural areas

After data harmonization, the dominant forests and agricultural areas were determined for each land cell. For a cell to be dominated by forests, it had to have or be surrounded by larger forested areas than agricultural areas. Similarly, for a cell to be dominated by agricultural areas, it had to have or be surrounded by larger agricultural areas than forests.

In the first step, we aggregated all forest types in the harmonized dataset that meet the criteria of over 60% tree cover, which includes the ‘evergreen needleleaf forests’, ‘evergreen broadleaf forests’, ‘deciduous needleleaf forests’, ‘deciduous broadleaf forests’ and ‘mixed forests’ land-cover classifications in MODIS. Secondly, we computed the top three land-cover types for each cell in the harmonized dataset (Supplementary Fig. 5a). We assigned all cells with only aggregated forests in the top three land covers as ‘forest dominated’ and all cells with only agricultural areas in the top three land covers as ‘agriculture dominated’. The cells with neither forests nor agricultural areas in the top three land-cover types were removed from the determination of dominant land cover, as they have negligible areas of forests or agriculture. The remaining cells with both forests and agricultural areas in the top three land-cover types were referred to as ‘overlap’, which cover nearly 18% of all cells with forests and/or agriculture (Supplementary Fig. 6). The overlap cells were processed further to determine the dominance between forests and agriculture (Supplementary Fig. 5b).

In the third step, we calculated the difference of areas between forests and croplands in the overlap cells. If the difference between forests and agricultural areas were more than 5% of the cell area, we proceeded to the next step. If not, we skipped to the fifth step. We used 5% as a minimum difference between two land covers to cater for uncertainty in land-cover classification. In the fourth step, cells with forested areas over twice the size of agricultural areas were categorized as forest dominated, whereas cells with agricultural areas over twice the size of forested areas were categorized as agriculture dominated (Supplementary Fig. 5c). We assume that if one land-cover type is double the size of the other with the difference between the two land covers exceeding 5% of cell area, the dominance would be significant. In the fifth step, the remaining cells from the third and fourth steps were categorized as either forest or agriculture dominated based on the sum of forested or agricultural areas in their surrounding eight cells (Supplementary Fig. 5d). Cells with more extensive forested neighbours were categorized as forest dominated, whereas cells with more extensive agricultural neighbours were categorized as agriculture dominated. The masks of forest- and agriculture-dominated areas before and after the overlapping cells are classified are given in Supplementary Fig. 6.

Forest- and agriculture-dominated areas are referred to as ‘forests’ and ‘agricultural areas’ hereafter. When comparing the results of data harmonization and dominant area selection to the original datasets, forests and agricultural areas categorized in this study represent 94% of total forest areas in MODIS and 84% of total physically cropped areas in MIRCA (Supplementary Fig. 4b). The agricultural areas represent areas where 81% of global crops are produced according to the M3 dataset. The methods are thus considered suitable for distinguishing land areas into forest and agriculture-dominated areas at the global scale, as they result in high representation of both land covers and crop production (over 80%) in comparison to the original datasets.

Downwind region of forests

Using the ERA5 evaporation dataset and UTrack dataset that contains the ratio of evaporation from a source cell that falls on each of the sink cells60, we mapped out the downwind region of evaporation from all forest cells. We used the evaporation of forest cells that was weighted by the ratio of aggregated forest cover in each 0.5° × 0.5° cell in MODIS, expressed as ETw in Fig. 2a. The weighting of evaporation is a conservative approach as previous studies show that forests generally evaporate more moisture, disproportionately to their areas, as compared to other land-cover types67. This means that this study would tend to underestimate, rather than overestimate, the amount of moisture supplied by forests to agriculture, which we assume to be better than overclaiming forests’ importance. Only agricultural areas within the downwind region of forests were included in the following analysis to establish the forest to agriculture moisture flows. We also conducted analysis of downwind regions of extratropical and tropical forests separately. We used the threshold 30° N and 30° S to distinguish extratropical from tropical forests and then we categorized each country as being extratropical or tropical based on the dominant area of the two forest types.

The forest to agriculture moisture flows were then assigned to different countries based on the national boundary vectors58 and are expressed as ε and μ:

$$\varepsilon =\frac{{\mathrm{ET}}_{{\rm{w}},\mathrm{agr}}}{{A}_{\mathrm{agr}}}\times 1,000$$

where ε is the moisture flow from forest to agricultural areas expressed in mm yr−1 or mm per month, ETw,agr is the evaporation weighted by forest-cover ratio from forests that falls on downwind agricultural areas (m3 yr−1 or m3 per month), and Aagr is the extent of downwind agricultural areas (m2). Calculations of ε were done either at the grid cell level (ε) or at national level (εnat or εtransb). ‘National’ and ‘transboundary’ terms refer to whether the upwind forests are located within the same national border with the downwind agricultural areas.

$$\mu =\frac{\varepsilon }{{P}_{\mathrm{agr}}}\times 100 \%$$

where μ is the moisture flow from forest to agricultural areas expressed in % annual or % monthly precipitation (P), Pagr is the total precipitation over the downwind agricultural areas (mm yr−1 or mm per month). Calculations of μ were also done either at the grid cell level (μ) or at the national level (μnat or μtransb). Hotspot regions were identified using a set of percentiles (33rd, 67th, 100th) that rank the values of the variable of interest into ‘low’, ‘medium’ and ‘high’ levels, such as ETw or ε or μ (Figs. 2 and 3 and Supplementary Figs 4, 10, 11 and 17). The hotspot regions are reflected in the high percentile category and should not be interpreted as having a statistically significant value.

Crop water requirements and crop production

The volumetric ETw,agr was then assessed against GW and BW CWR from the WATNEEDS model61 (Fig. 2a bottom), as expressed in ρ:

$$\rho =\frac{{\mathrm{ET}}_{{\rm{w}},\mathrm{agr}}\times {r}_{\mathrm{CWR}}}{\mathrm{CWR}}\times 100 \%$$

where ρ is the fulfilment of CWR by moisture provided by upwind forests expressed in % annual CWR or % monthly CWR, rCWR is the ratio of CWR for the corresponding agricultural system (GW or BW) or major crop (either one of wheat, maize, rice, soybean or sugarcane as five major crops) and CWR is the total CWR summing all volumetric water required to grow GW- and BW-fed crops on agricultural areas. The five major crops account for nearly half of global crop production61 and cover around 60% of the global monthly agricultural areas on average. In this study, the agricultural system is grouped into the GW system (including GW-fed rain-fed agriculture and GW-fed irrigated agriculture) and BW system (only BW-fed irrigated agriculture), based on the output dataset of WATNEEDS model61.

The ratio rCWR is adjusted based on the agricultural system or major crop of interest. For either GW or BW agricultural system, rCWR would be calculated as the CWR of the respective system divided by the total GW and BW CWR. The rCWR for a major crop for an agricultural system would be the multiplication of (1) the ratio of either GW or BW agricultural system, (2) the ratio of the specific crop CWR over the CWR of all major crops and (3) the ratio of areas of the five major crops over areas of all crops. For the individual major crops, ρ is thus weighted by the area of major crop within the cell’s agricultural area.

Moreover, both ε and μ at the national level were superimposed with national crop production (Cprod) and crop export (Cexp), both in million tonnes per year (Mtonnes yr−1) (Fig. 3 and Supplementary Fig. 7). National crop production is aligned with the type of crop grown in agricultural areas based on CWR for GW and BW within each country. Accordingly, only crops that are grown within each country were included in the crop export. The crop classes are categorized differently between MIRCA cropland dataset (26 crop classes), M3 crop production dataset (175 crop classes), SPAM cropland and crop production dataset (42 crop classes) and CWASI traded crop flows (125 applicable crop classes). For the purpose of matching the moisture flows, national crop production and traded crop flows, the different crop codes are matched according to the Food and Agriculture Organization (FAO) crop codes or names (Supplementary Data 8).

Circular and cascading flows between moisture and crops

We further paired the forest to agriculture moisture flows with the crop flows. The crop flows represent the average amount of crops (Mtonnes yr−1) that were traded between exporting and importing countries63. The selection of crops from the CWASI database for estimating the crop flows are based on the crops physically grown in the agricultural areas and included in the MIRCA dataset (Supplementary Data 8). As such, the types of crop between those accounted in crop production and crop export align. The amount of crop export should not exceed that of crop production. Only primary crops, and not derived or meat products, are considered in the export (Supplementary Data 8). To match the moisture with crop flows, we selected the moisture flows whose downwind country is also a crop exporter. We then used the relative measures of moisture and crop flows that are expressed in the percentage of moisture transported over total annual precipitation in the downwind country (εnat or εtransb in % annual precipitation) and in the percentage of imported crops over total annual imported crops in the importing country (Cexp,rel in % annual import), respectively. Only moisture and crop flows exceeding 5% of the corresponding units are considered (Fig. 5). We assume that a minimum of 5% would lead to the selection of flows that meaningfully contribute to either precipitation or import in the receiving country. Relative measures were used to compare the importance of moisture and crop flows across countries, considering the various size and the economic development stage of a country.

The distinctive categorization of countries into either crop producer, exporter or importer is challenging as many countries assume most roles through the ever-increasingly complex trade network. We assume that when producers of a primary crop are also exporters, they dedicate a part of their crop production to export to other countries. Similarly, the accounting of crop supply within a country is intricate as the attribution of how national crop production is consumed within one country depends on the production, import, export, stock reserves, consumption and re-export of crops63. Export ban is a common response to shocks in national crop supply, which would indirectly affect crop export and the importing countries31,68. Hence, when national production is disrupted in a country, we assume that their export is also likely to be disrupted. However, there are exceptions where high export variability is not necessarily aligned with high yield variability, such as the production of rice in Uruguay and Italy and of wheat in Canada30.

Sensitivity analysis

We conducted two types of sensitivity analysis. The first sensitivity analysis is focused on testing the data harmonization and the selection of dominant forests and agricultural areas by substituting MIRCA with SPAM dataset. The sensitivity analysis results in the representation of 94% of total forest areas in MODIS and 85% of total physically cropped areas in SPAM, which is similar to the results based on MIRCA (‘Dominant forests and agricultural areas’; Supplementary Fig. 4). The agricultural areas represent areas where 83% of global crops are produced according to the SPAM dataset. The values of moisture flow at the annual scale are generally consistent across the use of SPAM and MIRCA datasets (Supplementary Figs. 9 and 10). The differences between dominant areas determined by SPAM- and MIRCA-based analyses for all applicable countries and the resulting differences in national or transboundary moisture flows are listed in Supplementary Data 14.

Only seven out of the 181 applicable countries have a more than 5-percentage-point difference in either national or transboundary moisture flows (Supplementary Data 14). Gabon stands out as an outlier with a 40-percentage-point difference in the dependence on transboundary moisture flows, due to undetected agricultural cells in SPAM-based analysis as compared to the two agricultural cells detected in MIRCA-based analysis. The difference in detection might rise from the different base year of SPAM and MIRCA croplands, with SPAM representing data from around 2010 while M3/MIRCA around 2000 (ref. 56). Croplands appear to be changing remarkably from 2005 to 2010 in the SPAM dataset56. However, the overall averaged difference of national and transboundary moisture flows are still really low at 0.3 and 1.2 percentage points, respectively (sample size n is 182; Supplementary Data 14). Moreover, SPAM cropland and crop production dataset (42 crop classes) is used for calculating the crop production for the sensitivity analysis. The discrepancy in global crop production between M3/MIRCA and SPAM at the country level is merely 3.5%, despite Brazil having much lower value in M3/MIRCA (Supplementary Data 1). Thus, the data harmonization and dominant area selection based on MIRCA and SPAM are considered robust for the purpose of this study.

The second sensitivity analysis targets the assessment of the yearly moisture flows used in the main analysis against the monthly variability of moisture flows. Only the MIRCA dataset, and not SPAM, includes the monthly spatial distribution of cropped areas globally. We used the monthly differing cropped areas in MIRCA to calculate the moisture flows from forests to agricultural areas every month. The yearly agricultural areas selected in the analysis should cover the cumulative areas represented in all months of the year. The yearly aggregation is used for the main analysis because precipitation on agricultural areas is crucial not only during growing seasons, but throughout the year69. Additionally, the traded crops also include crops that need to be grown all year round, such as fruit trees55. The sensitivity analysis at the monthly scale shows that the moisture flows at the yearly scale are within the range of the monthly scale (Supplementary Data 1 and 2). The monthly moisture flow can get significantly higher in some specific months, which is aligned with previous findings on the variability of moisture recycling within the year, especially during the dry season18.

The calculation of moisture flow at the monthly scale also results in some areas with higher than 100% of monthly precipitation in some months, mainly in boreal summer (cyan shaded cells in Supplementary Fig. 11). These areas are located primarily in dry and mountainous regions, which are identified as areas with large deviations in tracked moisture in the UTrack dataset70. The high relative values of recycled precipitation here might also be due to the uncertainty of evaporation and precipitation, as they apply mostly to cells with precipitation lower than 200 mm per month (Supplementary Fig. 12).

Uncertainties and limitations

The nationally recycled and transboundary moisture flows from forests are in the order of magnitude of national precipitation variability. The mean standard deviation of precipitation over the period of 1979 to 2023 at the national scale across all countries is 16% (Supplementary Fig. 13), whereas the country-to-country moisture flows that are considered in the analysis can reach up to 25% of annual precipitation and up to 53% of monthly precipitation (Supplementary Data 1).

Validation of moisture flows estimated by UTrack is not part of the study. Such a validation can be performed by tracing stable isotope composition but is currently limited due to the lack of global coverage of in situ measurements, data standardization and data availability71. Regardless, since its development in 2020, the UTrack model has been increasingly used and compared with other moisture tracking models, mainly against the Eulerian-based WAM-2layers72. The model intercomparison typically shows that WAM-2layers result in more extensive spatial coverage of regions in which moisture recycling take place44,73. However, both UTrack and WAM-2layers have higher agreement with each other than when each is compared with a mass balance model that estimates moisture recycling74. Finally, the nationally recycled moisture flows that consider the entire areas of countries, rather than only from forests to agricultural areas here, have been compared across earlier studies, resulting in a generally good agreement36.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.