AHAD: African major crops harvested area dataset for the years of 2000, 2010, and 2020

Abstract

Africa faces significant challenges in food security, which are compounded by rapid population growth. The situation is expected to worsen without effective interventions. One of the key obstacles to mitigating food insecurity is the lack of reliable high-resolution data, particularly regarding harvested area datasets, which directly reflect the agricultural situation on the continent. To tackle this challenge, we have developed the African Harvested Area Dataset (AHAD), which covers 22 major crops across the continent at a resolution of 5 arcmin (approximately 10 km) for the years 2000, 2010, and 2020. The dataset is built upon 8 well-used global gridded harvested area datasets, through verifying, merging, calibrating, and confining with available information, including point-specific crop distribution, accurate cropland map data, subnational statistics, and cropping intensity data. In addition to the primary datasets, we also provide data quality assessments for each step of the process. The AHAD could provide geospatial and temporal patterns of harvested area, offering potential for advancing agricultural practices and enhancing food security across Africa.

Background & Summary

Africa is a critical region for global food security research, hosting half of the world’s low-income and food-deficient countries¹. Alarmingly, one in five Africans still faces chronic hunger², a crisis further aggravated by rapid population growth that threatens to escalate food insecurity in the coming decades^3,4. Current initiatives to assess, monitor, and alleviate food insecurity are hindered by severe data gaps, particularly in reliable crop condition monitoring and agricultural production statistics^5,6. Among these challenges, accurate harvested area data is especially vital since it directly reflects two key determinants of food security: crop production trends and disaster-induced losses. Establishing a high-resolution, temporally consistent harvested area dataset for Africa would provide a transformative foundation for improving agricultural management and safeguarding food security across the continent.

Existing research has developed several global gridded harvested area datasets to assess agricultural conditions across the African continent, including notable examples such as the Spatial Production Allocation Model (SPAM) dataset family^7–11, Global Agro-Ecological Zones (GAEZ) datasets^12,13, MIRCA2000¹⁴, M3-Crops Data¹⁵, and CROPGRIDS¹⁶. These datasets typically rely on global cropland maps, yet the accuracy of such global cropland data in Africa remains uncertain. Inaccurate crop masks in particular may lead to significant errors in crop condition assessments, potentially resulting in both over- and under-estimation⁵. Moreover, most global datasets are often constructed and validated using country-level inventory data, which are subsequently downscaled to gridded level resolutions. However, this approach tends to average harvested area values at the national scale, potentially masking local variations, especially in data-scarce regions like Africa¹⁶. Consequently, the application of global datasets in Africa requires verification not only in terms of spatial distribution but also in terms of quantitative accuracy. Meanwhile, due to variations in data sources, validation datasets, and methodologies, different datasets may differ in their accuracy when representing the harvested area of a specific crop in a given region¹⁷. Therefore, integrating the complementary strengths of these datasets may contribute to improving the reliability and precision of harvested area estimates in Africa¹⁶. Additionally, to our knowledge, existing datasets for Africa are generally static^10,18, and even intermittent data remain challenging to obtain despite their critical role in understanding agricultural dynamics. Thus, constructing African harvested area datasets should entail rigorous validation, integrating the complementary strengths of various datasets, and enhanced temporal coverage.

To effectively address the limitations inherent in datasets used across Africa, we developed first specialized harvested area dataset. The dataset is derived from 8 well-used global gridded harvested area datasets, which were spatially verified and merged using African point-specific crop distribution data and high-accuracy cropland maps. The dataset values were calibrated and confined by subnational statistical ranges and dynamic cropping intensity data. The resulting African Harvested Area Dataset (AHAD) spans 22 crops for the years 2000, 2010, and 2020, and is accompanied by a detailed report on data quality at each stage.

Methods

Data resources

Table 1 presents detailed information on all the input datasets employed in this study. The global gridded harvested area datasets are sourced from widely recognized resources, such as the SPAM dataset family^7–9,11, GAEZ v4¹³, MIRCA2000¹⁴, M3-Crops Data¹⁵, and CROPGRIDS¹⁶. The crops and corresponding years covered by these datasets are shown in Table 2. In the SPAM 2010 and SPAM 2020 datasets, millet is represented as the sum of pearl millet and small millet, while coffee is depicted as a combination of arabica and robusta varieties. Furthermore, SPAM and CROPGRIDS have integrated and enhanced previous datasets^7,16. For example, CROPGRIDS combines data from SPAM 2017 SSA¹⁰ and GAEZ+ 2015¹² to generate the dataset for the year 2020¹⁶.

Table 1.

Summary of datasets and statistic used in the study.

Datasets	Category	Resolution	Year
SPAM 20008	Gridded harvested area	5 arcmin	2000
GAEZ v4_200013	Gridded harvested area	5 arcmin	2000
MIRCA200014	Gridded harvested area	5 arcmin	2000
M3-Crops Data15	Gridded harvested area	5 arcmin	2000
SPAM 20109	Gridded harvested area	5 arcmin	2010
GAEZ v4_201013	Gridded harvested area	5 arcmin	2010
SPAM 202011	Gridded harvested area	5 arcmin	2020
CROPGRIDS16	Gridded harvested area	0.05°	2020
Global Yield Gap Atlas19	Crop distribution	Point-level	1980–2022
African Cropland Layer52	Gridded cropland	30 m	2016
HarvestStat Africa20,21	Statistic harvested area	Subnational
RCeA data22	Statistic harvested area	Subnational
Agro-MAPS23	Statistic harvested area	Subnational
Statistical yearbooks24–50	Statistic harvested area	Subnational
FAOSTAT51	Statistic harvested area	National
Global Cropping Intensity53	Gridded cropping intensity	250 m	2001–2019

Table 2.

The crop list of AHAD and other gridded harvested area datasets.

AHAD	SPAM 2000	GAEZ v4 2000	MIRCA 2000	M3-Crops Data	SPAM 2010	GAEZ v4 2010	SPAM 2020	CROPGRIDS
banana		√		√	√	√	√	√
barley	√	√	√	√	√	√	√	√
bean	√			√	√		√	√
groundnut	√	√		√	√	√	√	√
maize	√	√	√	√	√	√	√	√
potato	√		√	√	√		√	√
rice	√	√	√	√	√	√	√	√
cotton	√	√	√	√	√	√	√	√
sorghum	√	√	√	√	√	√	√	√
sunflower		√	√	√	√	√	√	√
wheat	√	√	√	√	√	√	√	√
cassava	√	√	√	√	√	√	√	√
cocoa		√	√	√	√	√	√	√
coffee	√	√	√	√	√	√	√	√
millet	√	√	√	√	√	√	√	√
sesame				√	√	√	√	√
soybean	√	√	√	√	√	√	√	√
sugarcane	√	√	√	√	√	√	√	√
sweetpotato				√	√		√	√
yam				√	√		√	√
cowpea				√	√		√	√
pigeonpea				√	√		√	√

The point-specific crop distribution data come from the Global Yield Gap Atlas (GYGA), which includes agronomic data collected from 1980 to 2022¹⁹. Agronomists from each participating country contribute detailed information on cropping systems, crops, and other site-specific factors. This bottom-up approach ensures the data reflects real-world conditions. Subnational statistical data come from sources such as HarvestStat Africa^20,21, the ReSAKSS Country Atlases (RCeA)²², Agro-MAPS²³, and national statistical yearbooks^24–50. These datasets can complement each other in terms of regions, crops, and years. A small portion of the data comes from subnational level 2 sources, which are at a finer geographic resolution than subnational level 1 data. To reduce inconsistencies caused by naming variations and remove statistical anomalies, we aggregated the subnational level 2 data into subnational level 1 units, and for each combination of country and crop, records were excluded if the total harvested area across all subnational level 1 units exceeded the national harvested area by more than tenfold. Harvested area statistics at the national level were acquired through FAOSTAT⁵¹. Figure 1 illustrates the spatial distribution of subnational data alongside the point-specific crop distribution.

Fig. 1.

Point-specific crop distribution data locations and subnational data distribution.

The African cropland layer used in this study was constructed by integrating four distinct cropland products, originally generated from multiple remote sensing sensors. It includes cropland under permanent crops, temporary crops, and temporary fallow. The dataset divides Africa into 41 agro-environmental zones (AEZs), selecting the most accurate cropland layers within each zone for reconstruction, and validating them against both point-level observations and statistical data⁵². These features collectively improve the quality of cropland mapping across the continent, making the dataset more accurate and better suited for the objectives of this study. Global Cropping Intensity (GCI) datasets provide an annual dynamic dataset of global cropping intensity from 2001 to 2019, at a 250 meter resolution, with an overall average accuracy of 89%⁵³. In this study, we use the cropping intensity data from 2001 to approximate conditions in 2000, and data from 2019 to represent those in 2020. Additionally, the country administrative shapefile and first-level administrative boundaries were obtained from GADM⁵⁴.

Data processing steps

Figure 2 illustrates the data processing steps, in which we based on 8 well-used harvested area datasets using comprehensive African reliable data to generate 5 arcmin (approximately 10 km) gridded maps of harvested area for 22 major crops in Africa for the years 2000, 2010, and 2020.

Fig. 2.

Diagram of the African Harvested Area Dataset (AHAD) processing workflow.

Pretreatments

The pretreatment process comprised two main components: aligning the temporal resolution of subnational data with that of the gridded datasets, and standardizing the spatial resolution of the gridded datasets. For temporal alignment, most gridded harvested area datasets represent average values for specific target years, typically covering the two years preceding and following the target year. Accordingly, harvested area data for subnational regions were averaged using the two years nearest to the target year. When multiple data sources were available for a given subnational region, the dataset with broader subnational coverage and a longer temporal span was selected.

To match the desired spatial resolution of 5 arcmin, 30 m African cropland data were aggregated into 5 arcmin by summing the values. For cropping intensity data, the maximum value within each 5 arcmin grid was used. In standardizing the CROPGRIDS 0.05° data to a 5 arcmin resolution, the original resampling techniques were applied, including bilinear interpolation and pycnophylactic methods¹⁶. Furthermore, values of zero or less were treated as missing across all datasets to ensure a consistent interpretation of zero and negative values. For example, in CROPGRIDS, ocean and water areas are marked as –1¹⁶.

Step 1: Verify and merge datasets in Africa for each subnation

In this step, point-specific crop distribution data and cropland data were used to verify the accuracy of crop distributions across multiple datasets for specific years. To assess each dataset’s capability in representing crop distribution within each subnational region, two components were combined. In particular, the ratio of crop type precision (R1) was employed to evaluate the consistency between the spatial distribution of crops in a dataset and the point-level crop distribution. Since point-level data do not cover all crop types, a weight of 0.5 was assigned to R1 for those crop types not represented. R1 primarily reflects the overall representation of crop distribution in the datasets, and its score is calculated as follows:

$${R1}_i=\left\{\begin{array}{c}\frac{{\mathit{Num}(\mathit{HA}}_{i_p,g}\cap {\mathit{Point}}_{i_p})}{\mathit{Num}\left({\mathit{Point}}_{i_p}\right)}i\in i_p\\ \frac{{\sum }_{i\in i_p}{\mathit{Num}(\mathit{HA}}_{i_p,g}\cap {\mathit{Point}}_{i_p})}{{\sum }_{i\in i_p}\mathit{Num}\left({\mathit{Point}}_{i_p}\right)}\times 0.5i\notin i_p\end{array}\right)$$ 1

where $i_p$ represents the crops included in the point-specific crop distribution data. ${\mathit{Num}(\mathit{HA}}_{i_p,g}\cap {\mathit{Point}}_{i_p})$ denotes the number of harvested area datasets for crop $i_p$, in grid $g$ that are confirmed as correct by the point-specific crop distribution data, while $\mathit{Num}\left({\mathit{Point}}_{i_p}\right)$ indicates the total number of point-level data entries for crop $i_p$.

The alignment ratio of cropland to harvested area (R2) quantifies the degree of agreement between the crop spatial distribution in a dataset and the Africa cropland data. It is calculated as follows:

$${R2}_{i,j}=\frac{\mathit{Num}({\mathit{HA}}_{i,j,g}\cap {\mathit{CL}}_{j,g})}{\mathit{Num}\left({\mathit{HA}}_{i,j,g}\right)}$$ 2

where $\mathit{Num}({\mathit{HA}}_{i,j,g}\cap {\mathit{CL}}_{j,g})$ is the number of grid cells for crop $i$, in subnational region $j$, within grid $g$, where both cropland data and the harvested area dataset register non-zero values. $\mathit{Num}\left({\mathit{HA}}_{i,j,g}\right)$ refers to the number of grid cells for crop $i$, in subnational region $j$, grid $g$, as recorded in the datasets.

For each dataset in each subnational region, the R1 and R2 scores are summed to obtain a total score. The dataset with the highest total score is selected to represent the crop distribution in that subnational region. Following these steps, the merged harvested area datasets for Africa are generated.

Step 2: Calibrate harvested area by statistical data

After merging multiple harvested area datasets, the optimal African crop distribution datasets were obtained. However, calibration was required to align the gridded harvested area with subnational statistics for consistency. To maintain consistency with the gridded harvested area datasets, the average of the statistics from the two years before and after the target year was used for calibration^7,13–15. If the total harvested area in a subnational region derived from the merged datasets exceeds the corresponding statistics, the excess area in all grid cells of the initial map is proportionally reduced according to the ratio of each grid cell’s area to the total area of all grid cells, ensuring consistency with the statistical magnitude. Conversely, if the harvested area falls below the inventory data, the deficit is distributed across all grids^7,55,56. In cases where subnational data are unavailable, FAOSTAT datasets are used to calibrate the merged harvested area at the country level. The detailed procedure is described as follows:

$${\mathit{calHA}}_{i,j,g}={\mathit{merHA}}_{i,j,g}+\frac{({\mathit{HA}}_{\mathit{stati},j}-{\sum }_g{\mathit{merHA}}_{i,j,g})\times {\mathit{merHA}}_{i,j,g}}{{\sum }_g{\mathit{merHA}}_{i,j,g}}$$ 3

where ${\mathit{calHA}}_{i,j,g}$ represents the calibrated harvested area for crop $i$, in subnation or country $j$, for grid $g$; ${\mathit{merHA}}_{i,j,g}$ denotes the merged harvested area obtained from the previous steps; and ${\mathit{HA}}_{\mathit{stati},j}$ is the statistical data for crop $i$, in subnation or country $j$.

In this step, the relative error between the $\mathit{merHA}$ and the statistical data, denoted as R3, is used to quantify the magnitude of the adjustments applied to the harvested area. In other words, a higher R3 value indicates a smaller adjustment or a better match with the statistical data. For regions validated using national statistical data, the R3 value is assigned a weight of 0.5, while in the absence of any statistical data, R3 is set to 0. This relationship is expressed as follows:

$${R3}_{i,j}=(1-\min \left\{1,\left(\frac{\left|{\mathit{HA}}_{\mathit{stati},j}-{\sum }_g{\mathit{merHA}}_{i,j,g}\right|}{{\sum }_g{\mathit{merHA}}_{i,j,g}}\right)\right\})\times w_j$$ 4

where $w_j$ is the weight assigned to the level of statistical data. When $j$ denotes a subnational region, $w_j$ is set to 1; when $j$ represents a national region $w_j$ is set to 0.5. Consequently, ${R3}_{i,j}$ ranges from 0 to 1 for subnational regions and from 0 to 0.5 for national regions. If no statistical data is available, ${R3}_{i,j}$ is equal to 0, indicating that there is no statistical calibration for the harvested area.

Step 3: Confine harvested area by cropping intensity data

Cropland data were multiplied by the maximum crop intensity for each cropland grid to derive the gridded maximum harvested area. The total harvested area across all crops was then verified to ensure that it did not exceed this maximum. If the total harvested area surpassed the maximum, it was confined accordingly. The harvested area for each crop was determined by multiplying the ratio of the crop’s gridded harvested area to the total gridded harvested area by the maximum harvested area. Any excess harvested area was then evenly allocated among all other grid cells within the same subnational region. The formula is expressed as follows:

$${\mathit{resHA}}_{i,j,g}=\left\{\begin{array}{c}{\mathit{max}\mathit{HA}}_{i,j,g\in g0}\times \frac{\mathit{cal}{\mathit{HA}}_{i,j,g\in g0}}{{\sum }_i{\mathit{calHA}}_{i,j,g\in g0}}g\in g0\\ {\mathit{calHA}}_{i,j,g\in g1}+\frac{{\sum }_{g\in g0}(\mathit{cal}{\mathit{HA}}_{i,j,g\in g0}-{\mathit{resHA}}_{i,j,g\in g0})}{\mathit{Num}\left({\mathit{calHA}}_{i,j,g\in g1}\right)}g\in g1\end{array}\right)$$ 5

where ${\mathit{resHA}}_{i,j,g}$ denotes the resultant harvested area for crop $i$, in subnational region $j$, at grid $g$, and $\max \mathit{HA}$ is the maximum harvested area. The set $g0$ comprises grid cells where the total harvested area across all crops exceeds $\max \mathit{HA}$, while $g1$ includes grid cells in subnational region $j$, where the total harvested area is less than $\max \mathit{HA}$. $\mathit{Num}\left({\mathit{calHA}}_{i,j,g\in g1}\right)$ is the number of grid cells for crop $i$, within the set $g1$, in subnation $j$.

In this step, the ratio of the reallocated area to the calibrated area (R4) was used to quantify the extent of reallocation in the harvested area. A higher R4 value indicates less modification. It is calculated as follows:

$${R4}_{i,j,g}=1-\min \left\{1,\left(\frac{\left|{\mathit{resHA}}_{i,j,g}-{\mathit{calHA}}_{i,j,g}\right|}{\mathit{cal}{\mathit{HA}}_{i,j,g}}\right)\right\}$$ 6

where ${R4}_{i,j,g}$ ranges from 0 to 1. A higher value of R4 corresponds to fewer changes. If ${R4}_{i,j,g}=1$, it means no adjustment has been made.

In the final step, we used R as the overall data quality score, which is calculated by summing R1, R2, R3, and R4. A higher value of R indicates better data quality.

Data Records

The AHAD is publicly available for download from the Zenodo repository⁵⁷. It is stored in GeoTIFF format within the AHAD folder, with each file named according to the format “AHAD_<year>_<crop>.tif”. Each file contains a 5 arcmin gridded map detailing the harvested area in hectares (ha) for one of the 22 African crops during a specific year, and is uniformly georeferenced to the WGS84 geographic coordinate system. The complete list of AHAD crops can be found in Table 2.

Maize occupies the largest harvested area among crops in Africa⁵¹, while cassava plays a vital role in sustaining farmers’ livelihoods and boosting African economies^51,58. In addition, coffee is widely cultivated and remains an economically important crop on the continent⁵⁹. As an example of AHAD, Fig. 3 presents the harvested areas for these three key crops over the years 2000, 2010, and 2020. The dataset demonstrates a steady expansion in the harvested areas for both maize and cassava over time. Moreover, it reveals a shift in coffee cultivation hotspots, with production moving toward lower latitudes and a decline in small-scale coffee farming in southern Africa. These findings provide evidence of a transition from small-scale coffee farming to more intensive production methods, as well as a reduction in suitable land for coffee cultivation likely driven by climate change^60–62.

Fig. 3.

AHAD examples for maize, cassava, and coffee over the years 2000, 2010, and 2020.

Technical Validation

Data quality

In this study, diverse data sources, including African point-specific crop distribution data, high-accuracy cropland data, statistical records, and cropping intensity data, were used to verify, merge, calibrate, and confine 8 well used gridded harvested area datasets. During the construction of AHAD, the dataset exhibiting the highest spatial accuracy was selected for each region to represent the gridded harvested area, and all available statistical data were employed to calibrate these datasets wherever possible. The quality of the AHAD at each processing step is represented by the values of R1, R2, R3, and R4. Specifically, R1 and R2 evaluate the accuracy of crop distribution within the dataset, with R1 quantifying the agreement between the dataset’s spatial crop distribution and the point-specific crop distribution, while R2 measuring its correspondence with Africa’s cropland data. R3 and R4, on the other hand, reflect the quality of the harvested area estimates. R3 measures the magnitude of adjustments required to align the harvested area with statistical data, and R4 quantifies the extent of reallocation during the process of confining the gridded total harvested area. The AHAD data quality values for individual crops in specific years are available for download from the Zenodo repository⁵⁷ and are stored in GeoTIFF format within the AHAD_DataQuality folder.

As examples of data quality, Fig. 4 presents the quality metrics for maize, cassava, and coffee in the years 2000, 2010, and 2020. In this study, higher R values indicate better data quality. Over time, the quality of the AHAD improved, with distinct crop-specific patterns emerging. Maize harvested area proved to be more verifiable, whereas cassava and coffee data exhibited lower quality, despite their significance as food and economic crops in Africa. Moreover, spatial analysis revealed clear trends, with lower data quality observed in South Sudan, Central Africa, and Angola.

Fig. 4.

Data quality for maize, cassava, and coffee over the years 2000, 2010, and 2020.

Known uncertainties and limitations

The uncertainties in the AHAD arise from several aspects. First, AHAD inherits the uncertainties and errors present in the original input datasets, which may originate from various sources. For instance, datasets within the SPAM family are based on inventory data, and incorporating subnational level 2 data could enhance the accuracy of the results⁷. However, subnational level 2 boundaries and names have changed over the past two decades, leading to inconsistencies among datasets. Despite these limitations, targeted agricultural surveys in Central and Southwestern Africa could help clarify agricultural conditions in these regions (Figs. 1 and 4). Additionally, we selected a cropland dataset specifically developed for African agriculture, which, despite covering only a single year, offers the most suitable characteristics for our research. It provides a comprehensive definition of cropland, including permanent crops, temporary crops, and temporary fallow land, and was designed to support accurate cropland mapping in Africa, making it the most appropriate choice among available options⁵². In contrast, other datasets may underestimate cropland due to narrower definitions. For example, the Global Land Analysis and Discovery (GLAD) dataset, although widely used and available across multiple years, adopts a more conservative definition that excludes perennial woody crops⁶³. The dataset used in this study integrates multiple remote sensing sources from 2015 to 2017 to produce a detailed and statistically consistent representation of cropland distribution across Africa. This spatial accuracy is critical for evaluating the consistency between crop harvested areas and actual cropland distribution patterns. Furthermore, for the cropping intensity data, information was unavailable for the years 2000 and 2020. Consequently, the cropping intensity for these years was assumed to be similar to that of the adjacent years, 2001 and 2019 respectively, based on the assumption that significant changes in cropping intensity rarely occur within a single year^53,64. In this study, we used the maximum cropping intensity at a 5 arcmin resolution to constrain the harvested area. Annual changes in maximum cropping intensity were predominantly stable, with approximately 70 percent of the grids showing no variation and around 30 percent experiencing changes (Fig. 5). Although the 30 percent may reflect some over- or underestimation due to data uncertainty, the subset of grids used to constrain harvested area is less, and within these grids, the proportion of changes is likely lower. Finally, the GYGA dataset has limited spatial coverage in certain regions. To assess the consistency between the spatial distribution of crops in each dataset and the point-level crop distribution recorded by GYGA, the ratio of crop type precision (R1) was employed. The accuracy of R1 assessments improves with the number of available point-level observations. In regions lacking such data, localized discrepancies in crop distribution may be masked in the overall evaluation. Nevertheless, R1 remains a useful metric for evaluating spatial agreement across datasets, particularly when interpreted alongside the alignment ratio of cropland to harvested area. Together, these indicators enable a more comprehensive comparison of spatial patterns. Overall, reducing uncertainties in future versions of AHAD will require improvements across multiple dimensions, including more detailed and consistent subnational agricultural statistics, cropland and cropping intensity data with extended temporal coverage, and broader spatial coverage of point-level observations such as those from GYGA. Incorporating these enhancements would substantially improve the robustness and practical applicability of AHAD in agricultural research.

Fig. 5.

Annual changes in max cropping intensity from 2002 to 2019. Values are calculated as differences from the preceding year.

Despite its strengths, AHAD has several limitations that users should be aware of. With a spatial resolution of 5 arcmin, AHAD delivers robust patterns at national and regional scales, but it smooths over sub-cell variability that is characteristic of smallholder landscapes. For applications requiring village-level or farm-level analysis, AHAD can be downscaled by integrating high-resolution crop masks and crop growing season datasets to identify intra-cell crop differences and enable more detailed spatial refinement. These downscaled results should be validated using extensive bottom-up observational data. AHAD currently provides decadal snapshots for only three years: 2000, 2010, and 2020. These years were selected based on the availability and consistency of input datasets. The construction process relies on bottom-up statistical data, remote sensing products, and gridded harvested area datasets. The selected years correspond to periods with multiple comprehensive and high-quality datasets, which ensures greater confidence in the construction of AHAD. Although datasets such as SPAM 2005 and GAEZ+ 2015 are available, their information has already been absorbed into later datasets such as SPAM 2010 and CROPGRIDS, which were used in this study. The current framework, which is based on multi-step integration of global gridded harvested area datasets and Africa-specific information, provides a solid foundation for incorporating newly developed products. With the emergence of more high-quality annual datasets, finer-scale crop distribution data, and higher-resolution spatial inputs, developing a continuous annual time series and improving spatial granularity will be key priorities in future updates of AHAD.

Usage Notes

Continuous and systematic surveys of harvested area datasets in Africa are urgently needed for future research. Built upon widely used global gridded harvested area datasets, AHAD enhances the characterization of the spatial and temporal patterns of harvested areas across Africa by verifying, merging, calibrating, and constraining these datasets with available African information. It supports practical needs in agricultural monitoring, food security analysis, and policy development. AHAD enables the assessment of spatial distribution and temporal changes in agricultural land use, provides essential input for evaluating agricultural interventions, and guides the allocation of cropland resources. It also serves as a benchmark for validating national statistics and remote-sensing products, applicable in data-scarce countries. It is a key tool for planning and developing regional agricultural strategies.

Author contributions

Wenmeng Zhang calculated the dataset and wrote the manuscript; Hui Zhang and Hengbin Yu collected the data; Fang Wu, Lijun Zuo and Xuefeng Cui conceived the research idea.

Data availability

The AHAD is publicly available from the Zenodo repository⁵⁷: 10.5281/zenodo.15182820. Further information on the dataset is provided in the Data Records section.

Code availability

The code of the AHAD is archived at the Zenodo repository: 10.5281/zenodo.15182820.

Competing interests

The authors declare no competing interests.