Journal of Maps ISSN: 1744-5647 (Online) Journal homepage: www.tandfonline.com/journals/tjom20 Rainfed wheat extent (2020/21) across Ethiopia’s complex and highly fragmented agricultural smallholder landscape Gerald Blasch, Louise Lesne, Jolan Wolter, Yoseph Alemayehu, Sophie Bontemps, David P. Hodson & Pierre Defourny To cite this article: Gerald Blasch, Louise Lesne, Jolan Wolter, Yoseph Alemayehu, Sophie Bontemps, David P. Hodson & Pierre Defourny (2025) Rainfed wheat extent (2020/21) across Ethiopia’s complex and highly fragmented agricultural smallholder landscape, Journal of Maps, 21:1, 2602338, DOI: 10.1080/17445647.2025.2602338 To link to this article: https://doi.org/10.1080/17445647.2025.2602338 © 2025 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group on behalf of Journal of Maps View supplementary material Published online: 17 Dec 2025. Submit your article to this journal Article views: 282 View related articles View Crossmark data Citing articles: 1 View citing articles Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=tjom20 https://www.tandfonline.com/journals/tjom20?src=pdf https://www.tandfonline.com/action/showCitFormats?doi=10.1080/17445647.2025.2602338 https://doi.org/10.1080/17445647.2025.2602338 https://www.tandfonline.com/doi/suppl/10.1080/17445647.2025.2602338 https://www.tandfonline.com/doi/suppl/10.1080/17445647.2025.2602338 https://www.tandfonline.com/action/authorSubmission?journalCode=tjom20&show=instructions&src=pdf https://www.tandfonline.com/action/authorSubmission?journalCode=tjom20&show=instructions&src=pdf https://www.tandfonline.com/doi/mlt/10.1080/17445647.2025.2602338?src=pdf https://www.tandfonline.com/doi/mlt/10.1080/17445647.2025.2602338?src=pdf http://crossmark.crossref.org/dialog/?doi=10.1080/17445647.2025.2602338&domain=pdf&date_stamp=17%20Dec%202025 http://crossmark.crossref.org/dialog/?doi=10.1080/17445647.2025.2602338&domain=pdf&date_stamp=17%20Dec%202025 https://www.tandfonline.com/doi/citedby/10.1080/17445647.2025.2602338?src=pdf https://www.tandfonline.com/doi/citedby/10.1080/17445647.2025.2602338?src=pdf https://www.tandfonline.com/action/journalInformation?journalCode=tjom20 Rainfed wheat extent (2020/21) across Ethiopia’s complex and highly fragmented agricultural smallholder landscape Gerald Blascha, Louise Lesneb, Jolan Wolterb, Yoseph Alemayehuc, Sophie Bontempsb, David P. Hodsond and Pierre Defournyb aSustainable Agrifood Systems Program (SAS), CIMMYT, Texcoco, Mexico; bEarth and Life Institute, Université catholique de Louvain, Louvain-la-Neuve, Belgium; cSustainable Agrifood Systems Program (SAS), CIMMYT, Addis Ababa, Ethiopia; dSustainable Agrifood Systems Program (SAS), CIMMYT, Kathmandu, Nepal ABSTRACT Crop-type maps are critical for addressing food insecurity, yet national-scale, high-resolution maps of staple crops like wheat remain unavailable for many African countries due to limited, publicly available ground reference data. Ethiopia’s complex and fragmented agricultural smallholder landscape and frequent cloud cover during rainfed seasons make crop mapping particularly challenging. This study presents the first national-scale, high- resolution wheat area map for Ethiopia, generated through a satellite-based workflow integrating (i) gap-filled, cloud-masked Sentinel-2 time series; (ii) Random Forest classification using temporal-spectral profiles and the EthCT2020 reference dataset; (iii) cropland masking; (iv) multistep validation; and (v) confidence-based consolidation with sub-national statistics. The resulting map estimates approximately 3.19 million hectares of wheat for the 2020/21 rainfed season, with 10 m spatial resolution and temporal coverage from April to December 2020. The dataset, openly accessible via CIMMYT Dataverse, provides a critical resource for operational wheat disease early warning and advisory systems in Ethiopia. ARTICLE HISTORY Received 8 April 2025 Revised 1 December 2025 Accepted 3 December 2025 KEYWORDS Remote sensing; crop type mapping; machine learning; random forest; sentinel-2; satellite 1. Introduction Ethiopian agricultural production and food security are challenged by multiple issues related to armed conflicts, climatic shocks, economic challenges, and recurrent crop disease and pest epidemics (Allen- Sader et al., 2019; Mohamed, 2017; WFP & FAO, 2023). Wheat is an important staple cereal crop having substantial contribution to food security and securing the livelihoods of millions of smallholder farmers in Ethiopia (CSA, 2020; Grote et al., 2021). However, Ethiopia’s wheat production is increasingly threatened by both abiotic and biotic threats, including, trans boundary pathogens such as wheat rusts (Abeyo et al., 2014; Jaleta et al., 2019; Meyer et al., 2021; Olivera et al., 2015). To improve food security in vulnerable countries, systems for crop disease early warning and surveillance as well as agricultural monitoring at multiple scales across cropping systems and geogra phies are key to mitigating regional-global crop pro duction risk and yield loss (Becker-Reshef et al., 2023; Nakalembe, 2020). The Ethiopian Wheat Rust Early Warning and Advisory System (Allen-Sader et al., 2019) provides one example of such a system. The benefit of implementing satellite earth obser vation (EO) in such systems is well documented (Defourny et al., 2019; Fritz et al., 2019; Nakalembe et al., 2021). EO systems can provide in near-real time a detailed picture of cropping systems, cropland area, crop distribution and density at regional to glo bal levels. Crop type maps reveal crucial information © 2025 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group on behalf of Journal of Maps This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrest ricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The terms on which this article has been published allow the posting of the Accepted Manuscript in a repository by the author(s) or with their consent. CONTACT Gerald Blasch g.blasch@cgiar.org CIMMYT global headquarters, Carretera México-Veracruz Km. 45, El Batán, Texcoco, C.P. 56237, México Supplemental data for this article can be accessed online at https://doi.org/10.1080/17445647.2025.2602338. JOURNAL OF MAPS 2025, VOL. 21, NO. 1, 2602338 https://doi.org/10.1080/17445647.2025.2602338 http://crossmark.crossref.org/dialog/?doi=10.1080/17445647.2025.2602338&domain=pdf&date_stamp=2025-12-16 http://creativecommons.org/licenses/by/4.0/ mailto:g.blasch@cgiar.org https://doi.org/10.1080/17445647.2025.2602338 http://www.tandfonline.com/loi/tjom20 http://www.tandfonline.com on which crops are cultivated, to what extent, where, and how. To map cropland and crop types at national and global scales, major investments have been under taken in EO systems, such as Sen2Agri (Defourny et al., 2019), Sen4Stat (European Space Agency, 2022), and WorldCereal (Van Tricht, Degerickx, Gilliams, Zanaga, Battude et al., 2023). However, high-resol ution, up-to-date national-scale EO products on culti vated staple crops such as wheat, maize, rice, and soybeans are clearly missing for almost all of Africa (Becker-Reshef et al., 2023). One reason for this can be found in the scarcity of publicly available georefer enced field data on crop types at high quality (e.g. pre cise location, spatial distribution) and a statistical- sufficient quantity, particularly in the highly dynamic smallholder farming systems of Sub-Saharan Africa. Such data are needed to train and validate machine learning classification algorithms. For Ethiopia, the recently published multi-source EthCT2020 ground reference dataset (Blasch et al., 2024b, 2024a) is a first step to address this gap. Mapping crop types in Ethiopia at the national scale remains highly challenging due to a complex and highly fragmented agricultural landscape and very fre quent cloud coverage during the rainfed growing sea sons. The fragmentation originates from the interrelation of an extremely heterogenous topogra phy influencing temperature and rainfall distribution and a cereal-focused smallholder crop production (Asefa et al., 2020; CSA, 2020; Livelihood Integration Unit, 2010; Tadesse et al., 2006; Teshome, 2014). Moreover, crop calendars are very diverse across Ethiopia’s rainfed and irrigated cropland, as shown for wheat (Bradshaw et al., 2022). These factors result in patterns of spectrally similar crop types (e.g. wheat- barley-teff and maize-sorghum) and their phenologi cal stages. Several case studies related to regional crop type mapping in Ethiopia have been conducted (Eggen et al., 2016; Eisfelder et al., 2024; Kibret et al., 2020; Marshall et al., 2019; Neigh et al., 2018); how ever, to the best of our knowledge, a national-scale mapping attempt for a staple cereal has not been undertaken yet. For improving crop disease early warning systems, the value of EO-derived dynamic maps, targeting crop distribution and growth, as well as rainfed and irri gated cropping areas, was recently highlighted (Gilligan, 2024). Addressing the demand of precise wheat distribution information for the Ethiopian Wheat Rust Early Warning and Advisory System (Allen-Sader et al., 2019), the objective of this research was to develop a satellite-based environmental moni toring tool able to identify wheat production areas across Ethiopia by discriminating wheat from other crop types. To overcome mapping challenges, the method consisted of (i) computing a gap-filled and cloud-masked Sentinel-2 time series, (ii) running a Random Forest classification using temporal-spectral features and the high-quality EthCT2020 ground reference dataset (Blasch et al., 2024b, 2024a), (iii) masking out non-cropland areas, (iv) performing mul tiple validation checks, and (v) consolidating the classification output based on confidence information and sub-national agricultural statistics. The final target was to generate a high-resolution wheat map for the Ethiopian rainfed cropping seasons 2020/21. 2. Materials and methods 2.1. Study area The study area (Figure 1) is located in Ethiopia (lati tudes: 3° 23’ to 14° 53’ N; longitudes: 32° 59’ to 47° 58’ E), a land-locked country in in the Horn of Africa, East Africa. The study area covers the main extent of Ethiopia’s rainfed cropland as derived from the ESA WorldCereal’s temporary crop mask (Van Tricht, Degerickx, Gilliams, Zanaga, Savinaud et al., 2023), a land area of 497,586 km2. It is a representative area for the Ethiopian agricultural landscapes as it contains several major topographic features (e.g. Western and Eastern Highlands, Upper Rift Valley,) with an elevation ranging from 1,000 m to 4,000 m asl. It includes both rainfed single and double wheat cropping systems due to unimodal (Meher only, long rainy sea son June-September) and bimodal (Belg, short rains March-May and Meher) rainfall patterns as well as irri gated wheat production areas during the dry season (October–April). The highlands are composed of elev ated plateaus and table lands, steep escarpments, and high mountains, decreasing abruptly to deep valleys. The upper, southwestern section of the Rift Valley is a higher-elevated narrow depression containing an internal drainage basin of many small rivers draining into several large lakes. Thus, agroecological zones highlight the complexity of spatial temperature and moisture patterns, mostly driven by topography, as well as the high variability of crop production environ ments from arid to per-humid (wet), lowland and high land areas (Asefa et al., 2020; Livelihood Integration Unit, 2010; Tadesse et al., 2006) (Figure 1). Ethiopia’s agriculture is characterized by subsis tence farming with ∼44 million smallholders, culti vating crops at ∼14.5 million hectares (ha) during the main cropping season (Meher) (CSA, 2020). Average farm sizes vary between regions, i.e. 0.54 ha in Tigray, 0.75 ha in Amhara, 0.89 ha in SNNPR, and 1.15 ha in Oromia (national average: 1.02 ha). Farms have 2.3 plots at national average (Teshome, 2014). In terms of cropland area, grain crops (cereals: 72%; pulses: 11%; oilseeds: 6%) are the predominant crop group (Meher) with major cereals, such as teff (24%), maize (18%), sorghum (14%), wheat (14%), and barley (7%) (CSA, 2020). For Meher 2020/21, 2 G. BLASCH ET AL. sub-national agricultural statistics reported an esti mated wheat area of ∼1.9 million ha (CSA, 2021). Wheat is the focus of the current study, as a key objec tive was to provide enhanced information for the operational wheat rust early warning and advisory system. The main growing altitude for wheat, rainfed and irrigated, is between 1,500 m and 3,300 m (Gelagay et al., 2025; Mekuriaw et al., 2023). In this range, other cereals such as barley, teff, and maize are also found. Within the same agroecological zones, smallholders focus on cropping similar cereals (e.g. wheat-barley-teff and maize-sorghum) on small, irregular-shaped fields. For each single crop, this leads to varying crop calendars across relevant agroecologi cal zones and complex patterns of spectrally similar crop types and growth stages. 2.2. Sentinel-2 satellite data To address the spatial and temporal resolution needed for this study, the state-of-the-art Sentinel-2 constella tions (European Space Agency (ESA), Frascati, Italy) was selected. Based on the twin-satellite concept, the Sentinel-2 constellation includes two identical high- resolution optical satellites in the same orbit but phased at 180° for optimal coverage, high revisit frequency (5 days), and data delivery. The satellites are capable of cloud-free and wide-swath (290 km) recordings of multispectral imagery (13 bands), capturing from the visible to the short-wave infrared (SWIR) range of the electromagnetic spectrum, at high spatial resolutions of 10 m (Blue; Green; Red; and near-infrared (NIR)), 20 m (red-edge (RE) bands: RE1, RE2, and RE3; Narrow NIR; SWIR1; and SWIR2), and 60 m (Coastal aerosol; water vapor; and SWIR cirrus). Due to the sen sor specifications, Sentinel-2 enabled the acquisition of a high-resolution imagery time series representative from the crop growing season despite significant cloud cover. To match the entire study area extent, 75 Sentinel-2 tiles (1 tile = 100 km x 100 km) were needed from three different UTM zones (36N, 37N, and 38N). The high cloud coverage over Ethiopia for several months during the 2020/21 agricultural seasons lim ited the number of available images, which was a chal lenge for an optical image-based classification approach. The image availability for each month from June to September 2020 is detailed in Figure 2. 2.3. Ground reference data The open-access Ethiopian Crop Type 2020 (EthCT2020) ground reference dataset (Blasch et al., 2024a) was used to build a wheat mapping model. For the 2020/21 Meher season, this wheat-focused multi-source dataset contains 2,793 in-situ samples on annual single crops at the smallholder field level across key agro-ecologies of Ethiopia’s rainfed cereal production system, overlapping both unimodal and bimodal rainfall seasons. Most of the samples belong to cereals (93%), whereby wheat (74%) is the main crop followed by teff (9%), barley (4%), and maize (3.4%). It should be noted that the dataset was col lected in the context of wheat rust monitoring, there fore it is skewed towards wheat fields, leading to a potential wheat overestimation effect that needs to be addressed in postclassification. Some crop types are not represented in the dataset. These ‘omitted’ crop types can be found in both rainfall patterns or under one specific pattern only. All samples are circu lar polygons of 10-m radius belonging to a specific field with homogenous crop cover. The samples are extensively processed in terms of data harmonization, mixed pixels identification, and spatial accuracy assessment. A detailed description of EthCT2020 data set and related processing methodology can be found in Blasch et al. (2024b). To ensure the compatibility of the EthCT2020 data set with Sentinel-2 satellite imagery, an additional data quality check based on spectral-temporal NDVI profiles (Defourny et al., 2019; Pasternak & Pawlus zek-Filipiak, 2022) was carried out. The Normalized Figure 1. Location of the study area (blue dashed line) in relation to the – (a) Ethiopian rainfed cropland area (orange) (Van Tricht, Degerickx, Gilliams, Zanaga, Savinaud et al., 2023); (b) agro-ecological zones (Tadesse et al., 2006); and (c) topography (NASA JPL, 2013) (basemap: World Terrain (ESRI, 2024)). JOURNAL OF MAPS 3 Figure 2. Monthly Sentinel-2 availability – (a) June, (b) July, (c) August, and (d) September 2020. 4 G. BLASCH ET AL. Difference Vegetation Index (NDVI) profile of each sample location (20-m buffer around the sample cen troid) was extracted from the Sentinel-2 time series. The median NDVI values of each date between June and December 2020 are plotted separately for each location as a function of time, with their variance shown as error bars (Figure 3). The availability of Sen tinel-2 images over the area is displayed at the bottom of the graphs, along the horizontal axis. Missing data are expected because of cloud coverage. The profiles were smoothed using a Whittaker filter. Each NDVI profile was visually checked to decide if it should be dis carded or not. Figure 3(a) is an example of a good qual ity profile, with low variance around the median NDVI values and a NDVI curve matching crop growth. This sample of wheat, harvested during the main season, was therefore retained in the dataset. Conversely, Figure 3(b) is an example of a low-quality profile, with median NDVI values that vary within the polygon and with a NDVI curve that does not look like that expected for a crop. Consequently, this wheat sample was discarded. Through this NDVI-based quality assessment, 86 samples were removed, resulting in a cleaned and har monized in-situ dataset of 2,707 samples to be used for wheat mapping. The resulting samples are distributed over different crop types (Table 1). Finally, the samples were randomly split into two fixed datasets: 75% of the fields of each crop type were set aside for training, and the remaining 25% were used for inde pendent validation of the classifications. 2.4. Wheat mapping remote sensing workflow The wheat mapping remote sensing workflow is com posed of five main processing blocks: preprocessing of Sentinel-2 and ground data, classification, postproces sing, validation, and consolidation. All processing steps are summarized in Figure 4 (ground data prep aration explained in section 2.3). 2.4.1. Preprocessing of Sentinel-2 data For the 2020/21 wheat growing seasons, Sentinel-2 satellite data were obtained from the ESA Open Access Figure 3. Median NDVI profile for a wheat field (a) in the unimodal rain pattern area and (b) of low quality (central thumbnail: undated Google satellite basemap; right thumbnail: monthly Planet composite in (a) December 2020 and (b) August 2020). Table 1. Crop distribution of the final cleaned and harmonized dataset. Crop type Samples (n) Wheat 2038 Teff 245 Barley 97 Maize 96 Faba beans 49 Triticale 44 Oilseed crops (other) 34 Millets 20 Potatoes 17 Spice crops 16 Leguminous crops (other) 12 Vegetables 11 Peas 6 Sweet potatoes 6 Sorghum 5 Chickpeas 4 Oats 3 Sugar cane 2 Groundnuts 1 Lentils 1 Total 2707 JOURNAL OF MAPS 5 Hub (https://scihub.copernicus.eu/) and processed for the period from 01/04/2020 to 31/12/2020. This period was set based on expert knowledge, as it covers both the short (Belg) and long (Meher) rainfed cropping seasons in Ethiopia. All ortho-rectified and georefer enced Sentinel-2 Level 1C (L1C; Top-Of-Atmosphere) and Level 2A (L2A; Bottom-Of-Atmosphere) products with a cloud coverage below 90% were downloaded for this period. The preprocessing was carried out adapt ing the Sen2Agri standard practice (Defourny et al., 2019) to the Ethiopian context, and the main stages are presented in Figure 5. Using the Fmask algorithm (Zhu et al., 2015), an additional cloud mask was generated and applied to the L1C images, because the Sen2Cor algorithm provided in the L2A products was not of sufficient quality. The cloud masked L2A images were used to generate three spectral indices: NDVI, Normalized Difference Water Index (NDWI), and Brightness Index (BI). Thus, cloud-free time series of Sentinel-2 surface reflectance and spectral indices were obtained from L1C and L2A products, respectively, and sub sequently computed to 10-days composites by aver aging the reflectance or spectral index values available for each pixel over each given 10-day period. A Whittaker filter was then applied to ensure a robust extraction of these temporal metrics (Defourny et al., 2019; Liang et al., 2023), resulting in a gap-filled and cloud masked 10-days time series of 27 smoothed reflectance and spectral index values for each pixel. Figure 4. Schematic illustrating the wheat mapping workflow and processing steps (S2: Sentinel-2). Figure 5. Schematic illustrating the main steps of the preprocessing block applied to S2 data. 6 G. BLASCH ET AL. https://scihub.copernicus.eu/ 2.4.2. Classification From the time series, a first crop type map at 10 m spatial resolution was generated using a Random For est classifier, a machine learning algorithm known for its high classification accuracy, especially in complex and heterogeneous agricultural landscapes (Belgiu & Drăgu, 2016). This product was provided with a confi dence layer obtained from the Random Forest algor ithm and corresponding to the probabilities that a pixel belongs to the different classes. As mentioned in Section 2.3, the algorithm was trained on 75% of the cleaned and harmonized in-situ dataset created from the cleaned EthCT2020 dataset, while the remaining 25% were retained for validation purposes. The Random Forest used all Sentinel-2 bands and the three spectral indices as features for each date in the time series, i.e. every 10 days between 01/04/2020 and 31/12/2020. The reduced amount of ground data limited the ability to use multiple classification models to address the high topographic and climatic variabil ity. Consequently, a single classification model was the best approach for mapping the whole study area. 2.4.3. Postprocessing A posteriori, a binary crop mask (10 m pixels) derived from the ESA WorldCereal’s temporary crop mask product (Van Tricht, Degerickx, Gilliams, Zanaga, Savinaud, et al., 2023) was applied to both classifi cation outputs to remove non-cropland areas. Sub sequently, pixels classified as crop types other than wheat and non-crop pixels are grouped together under the term non-wheat to obtain a preliminary binary wheat / non-wheat map at 10 m resolution. The postprocessed classification outputs were a pre liminary wheat map as well as a classification confi dence information layer at the pixel level. 2.4.4. Validation The product validation was carried out using the remaining 25% of the in-situ dataset that was not used during the model training, and confusion matrix and accuracy metrics (e.g. overall accuracy) were cal culated. In addition, (i) a visual analysis of the results was carried out by country agriculture experts (from CIMMYT and EIAR) with field experience in Ethiopia and (ii) wheat area mapping results were compared against the wheat area from the national and sub- national agricultural statistics (CSA, 2021). 2.4.5. Consolidation Firstly, the confidence layer provided by the Random Forest algorithm was used to consolidate the prelimi nary wheat map. Only the wheat pixels classified with at least a minimum level of confidence were con sidered as ‘true’ wheat in terms of a high certainty per centage. The certainty threshold was defined (i) to minimize the wheat area gap between the sub-national agricultural statistics (CSA, 2021) and those of the map, while (ii) maintaining a high level of overall accuracy. For the consolidation, the agricultural stat istics at the regional level were selected due to the het erogeneity of the map results, as well as missing or uncertain values at the zonal level. To obtain an opti mal certainty threshold for each region, a sensitivity analysis was conducted, which (i) compared for each region the wheat area derived from the agricultural statistics with the mapped wheat area across all levels of certainty (Figure 6(a)) and (ii) calculated the overall accuracy from the in-situ data present in the specific region across all levels of certainty (Figure 6(b)). Sub sequently, only wheat pixels with a certainty level greater or equal to the threshold were retained in the map. This assessment identified that all pixels with a level of certainty ≥42% should be used to minimize the discrepancy between the sub-national agricultural statistics data (CSA, 2021) and the map result, because using any certainty level above this threshold yields in less than or equal to the estimated wheat area from the sub-national statistics. Also, the overall accuracy declines below 80% if pixels with a level of certainty ≥44% are taken for the region (Figure 6). Secondly, three additional conditions based on local expert knowledge and literature – to ensure that a pixel is wheat – were added: a pixel was confirmed as ‘wheat’ if: (1) the pixel is in the elevation range between 1500 and 3300 meters – the altitude where rainfed wheat is mainly being cultivated in Ethiopia (Gelagay et al., 2025; Mekuriaw et al., 2023) – derived from the ESA Copernicus Digital Elevation Model (European Space Agency, 2019), (2) the pixel had a NDVI time series value greater than 0.3 (NDVImin) to avoid misclassification with bare soil, and (3) the pixels had an average NDVI time series value less than 0.8 (NDVImax) to avoid misclassification with tree cover. In general, NDVImin and NDVImax thresholds for identifying the start and peak of the growing season, respectively, are crop specific and influenced by environmental conditions (Huang et al., 2019; John son et al., 2021). NDVI thresholds used in this study were set in a conservative manner derived from regional studies (Bojago et al., 2025; Mohammed et al., 2023), considering relevant crop types cultivated in Ethiopia and the diverse agroecology. 3. Results and discussion The preliminary (non-consolidated) and final (conso lidated) products are maps with wheat and non-wheat classes at a spatial resolution of 10 m, produced for the JOURNAL OF MAPS 7 2020/21 main cropping seasons (Belg 2020 and Meher 2020/21) (Figures 7 and 8). All maps show only pixels that were classified as wheat by the Random Forest model during the period ranging from April to December 2020. The confusion matrices are shown in Table 2. For the preliminary wheat map, the Random Forest model yielded an overall accuracy of 86%. A produ cer’s accuracy of 87% for wheat indicates that most actual wheat areas were correctly identified. The user’s accuracy of 94% for wheat confirms that most predicted wheat areas were indeed wheat. Wheat is slightly overestimated (commission error: 6%), while 13% of actual wheat areas were missed (omission error). The visual analysis by experts confirmed that the main wheat areas appear to be correctly identified, and the map looks good overall. However, the wheat area seemed to be strongly overestimated, revealing a significant commission error in areas where the possi bility of growing wheat is very low due to unfavorable climatic conditions such as eastern Tigray (Figures 8(a) and (b)) and the Rift Valley (Figures 8(c)–(f)) that were identified as wheat. A quantitative compari son of the wheat area and proportion – in relation to the total crop area and proportion of annual and per ennial crops – from the preliminary map against the sub-national agricultural statistics (CSA, 2021), confirmed this observation (area / proportion: pre liminary map: 7,175,939 ha / 37.8%; statistics: 1,897,405 ha / 12.9%). Moreover, the analysis revealed a high spatial variability at the zonal level (Figure 9). For example, the map appears to be less accurate in terms of wheat area (preliminary map: 278,975 ha; statistics: 51,771 ha) and proportion (preliminary map: 47.9%; statistics: 18.7%) for the North Wello Zone than for the Arsi Zone (area / proportion: Figure 6. (a) Wheat area according to different levels of certainty; (b) overall accuracy according to the different levels of certainty. Figure 7. Wheat area maps (pixel size: 10 m) – (a) preliminary (non-consolidated) product and (b) final (consolidated) product (dashed rectangles: zoom-in-areas related to Figure 8: North Wello zone (I), Arsi zone (II), and West Arsi zone (III); basemap: World Terrain (ESRI, 2024)). 8 G. BLASCH ET AL. Figure 8. Preliminary (non-consolidated) and final (consolidated) wheat area maps zoom-in-views for the (a) and (b) North Wello zone, (c) and (d) Arsi zone, and (e) and (f) West Arsi zone (basemap: Bing Aerial Imagery). JOURNAL OF MAPS 9 preliminary map: 434,220 ha / 39.3%; statistics: 209,433 ha / 33.4%). The overestimation and variation can be explained by comparing the location of used in- situ samples (Figure 10(a)) with the generated confi dence information layer (Figure 10(b)). Higher accu racy (less overestimation) was achieved in areas with denser sampling locations, resulting in higher confi dence levels (e.g. Arsi, Bale, East Bale, and West Arsi zones). In such areas, field sizes tend to be larger (aver age field size in Oromia: 1.15 ha) than the national average of 1.02 ha, and neighboring fields are often clustered to a homogenous field area, growing the same crop, while the topography is relatively flat and less complex. This enables a more accurate detection of farmer fields by the Sentinel-2 satellite. In contrast, more fragmented areas with less sampling locations (e.g. East and West Gojam, North Wello, and South Gondar) have usually smaller fields (average field size in Amhara: 0.75 ha) and consequently showed lower confidence levels. The lowest accuracy (high overestimation) was found in areas without sampling locations, resulting in lowest confidence levels. The consolidation step led to a significant reduction of the mapped wheat area and proportion (final map: 3,192,972 ha / 16.8%). Although the overestimation was considerably reduced, the mapped wheat remains larger than the reported wheat (area / proportion: 1,897,405 ha / 12.9%) (CSA, 2021). The final wheat map based on the improved classification model yielded a slightly lower overall accuracy of 81%. The slight reduction in overall accuracy (5%) was metho dologically introduced during the consolidation step to decrease the previously mentioned commission error. Iterative visual assessment of the final map confirmed a considerable decrease of the commission error in such areas of eastern Tigray and the Rift Val ley (Figure 8), representing a more realistic rainfed wheat extent. Compared to the preliminary product, the final wheat map improved on overestimation, yielding a higher user’s accuracy of 96% and a lower commission error of 4%, meaning more precision in terms of labeling wheat. However, this comes at the cost of missing more actual wheat areas (producer’s accuracy: 78%; omission error: 22%). The preliminary map offers a more balanced performance across both classes, with higher overall accuracy and better detec tion of actual wheat areas (higher producer’s accu racy). However, it shows very high overestimation compared to sub-national agricultural statistics. In Table 2. Confusion matrices of the preliminary (non- consolidated) and final (consolidated) wheat map products (P: predicted; A: actual; UA: user accuracy; PA: producer accuracy; OA: overall accuracy; CE: comission error; OE: omission error). Preliminary wheat map Non-Wheat (A) Wheat (A) Total UA CE Non-Wheat (P) 563 273 836 67% 33% Wheat (P) 106 1765 1871 94% 6% Total 669 2038 2707 PA 84% 87% OA: 86% OE 16% 13% Final wheat map Non-Wheat (A) Wheat (A) Total UA CE Non-Wheat (P) 598 449 1047 57% 43% Wheat (P) 71 1589 1660 96% 4% Total 669 2038 2707 PA 89% 78% OA: 81% OE 11% 22% Figure 9. Zonal comparison of sub-national agricultural statistics (CSA, 2021) versus preliminary (non-consolidated) and final (consolidated) products in terms of – (a) wheat area and (b) wheat area proportion (based on cropland area derived from World Cereal product (Van Tricht, Degerickx, Gilliams, Zanaga, Savinaud et al., 2023)). 10 G. BLASCH ET AL. contrast, the final map prioritizes reducing overesti mation, with higher user’s accuracy and fewer false positives, and yields a considerably reduced spatial variability at the zonal level (Figure 9), making the final map more suitable for precise wheat area esti mation. In the context of the Ethiopian Wheat Rust Early Warning and Advisory System, the reduced overestimation decreases the risk of overestimating the disease occurrence, while the weakness of missing actual wheat area might be compensated during spatial upscaling to match the system’s 10 km grid resolution. For mapping the wheat extent across Ethiopia’s complex and highly fragmented agricultural small holder landscape, efficient pre-processing and classifi cation methods have been applied, and a consolidation method was developed where necessary. However, several limitations were encountered during these steps. The low quantity of in-situ data only allowed for using a single classification model, limiting classification finetuning to address the agroecological heterogeneity. Mixing all sampling locations from different agroecological zones might have led to less accurate classification results in agroecological zones that are underrepresented in the in-situ data. In-situ dataset skewed towards wheat led to an imbalance with respect to other crop types, negatively affecting the classification results in terms of overestimating the wheat area. Although the consolidation step con siderably reduced the overestimation effect, additional data balancing and augmentation approaches, such as the synthetic minority over-sampling technique (Blagus & Lusa, 2013) and generative adversarial net works (Lalitha & Latha, 2022), might provide solutions to this issue and further research needs to be con ducted. The high cloud cover limited the availability of Sentinel-2 imagery during the peak of the growing season. Reflectance data for the months of July and August are generally unavailable or very limited, result ing in a lack of information for the period of strong veg etation growth. In that respect, the cloud-penetrating advantage of the Sentinel-1 SAR sensor could improve presented approach by providing additional temporal features for specific crop types, especially during cloudy months, as shown in recent studies (Eisfelder et al., 2024; Faqe Ibrahim et al., 2023; Kpienbaareh et al., 2021; Mercier et al., 2020). To account for the complex ity of the crop calendar in Ethiopia linked to the high spatial variability of rainfall and topography, heteroge neously distributed field data in the study area would be needed for the classification algorithm. An additional difficulty in detecting wheat in Ethiopia is the presence of similar cereals such as teff and barley, which are difficult to distinguish visually or using an NDVI profile. While focused on wheat mapping in Ethiopia for disease early warning, this study addressed broader challenges in national and global crop type mapping and agricultural monitoring, particularly in data- scarce regions (Defourny et al., 2019; European Space Agency, 2022; Van Tricht, Degerickx, Gilliams, Zanaga, Battude et al., 2023). By leveraging satellite imagery and classification techniques, it demonstrates a scalable approach that can be adapted to other crops and geographies, especially across Africa where reliable agricultural data is limited (Becker-Reshef et al., 2023; Sadeh et al., 2024). The methodology can be scaled by processing data within different Figure 10. (a) Spatial distribution of in-situ samples, aggregated per zone; (b) confidence level map, indicating the percentage of decision trees for each pixel that produced the wheat result during the Random Forest classification. JOURNAL OF MAPS 11 agroecological zones or administrative units from municipal to national levels, allowing for regional spe cificities while capable of contributing to global pro ducts related to crop condition assessments and land use statistics. The primary data requirements for scal ing are: (1) consistent, high-resolution, and open- access satellite data and (2) high-quality in-situ data. The methodology aligns with global initiatives such as GEOGLAM, offering a framework for enhancing food security monitoring and supporting evidence- based agricultural policy at regional and international levels (Fritz et al., 2019; Nakalembe et al., 2021). 4. Conclusions National-scale crop type mapping in Ethiopia is highly challenging due to a complex and highly fragmented agricultural landscape, cereal-focused smallholder crop production with small fields sizes, high crop diversity and varying crop calendars, and very fre quent cloud coverage during the rainfed growing sea sons. In this research, a satellite-based environmental monitoring tool was developed to identify wheat pro duction areas across Ethiopia by discriminating wheat from other crop types. By applying a Random Forest classification model to a gap-filled Sentinel-2 time series, a final wheat area map was produced, showing the spatial distribution and density of the rainfed wheat extent at a high spatial resolution (10 m) for the 2020/21 cropping seasons. The binary product depicts whether a pixel is classified as wheat or non- wheat during the period ranging from April to December 2020. This study is the first attempt to map a staple cereal at the national scale in Ethiopia, and significant progress was made in developing an improved wheat map by integrating classification confidence information and sub-national agricultural statistics to overcome mapping challenges. However, several constraints were encountered originating from quantity and spatial distribution of ground refer ence data, limited satellite data availability during peak of the growing season, and crop calendar diversity. Although further refinement is needed, proposed workflow is applicable to other years and geographies but depending on in-situ data availability and quality. The 2020/21 product is aimed to complement the Ethiopian Wheat Rust Early Warning and Advisory System’s baseline data for the disease dispersal and environmentally suitability forecast models as well as building the basis for new information layers regard ing the host susceptibility. Software Data pre-processing was performed with Fmask and ESA Sen2Cor processors for the Sentinel-2 data. Additional scripting was undertaken using python as well as GDAL and OTB libraries for the various stages of the workflow. Pixel-based classification was per formed using the Random Forest algorithm. The Geo graphical Information System (GIS) environment adopted for the management, analysis and layout edi tion of all the layers is QGIS 3.34.8 version (http://qgis. osgeo.org). Also, Google Earth images and the Open Street Map layer have been integrated in QGIS using a specific plugin (i.e. QuickMapServices). Acknowledgements We greatly acknowledge the support of partnering insti tutions and financial support for the study through the CGIAR Initiative on Plant Health (https://www.cgiar.org/ initiative/plant-health/), the Improved Disease Monitoring and Management for Wheat and Cassava Through Epide miological Modelling project, led by Cambridge University (Prof. Christopher A. Gilligan) and funded by the Bill & Melinda Gates Foundation (INV 010472) and FCDO, UK (Investment Opportunity OPPGD448). Disclosure statement No potential conflict of interest was reported by the author(s). Data availability statement The dataset and metadata resulting from presented workflow and used for the final map in this paper is openly available from the CIMMYT Dataverse at https://hdl. handle.net/11529/10549175 (Blasch et al., 2025). Main pro cessing blocks of presented data processing pipeline is based on the publicly available ESA Sen4Stat open-source Earth Observation toolbox (European Space Agency, 2022). The consolidation step is available from the corresponding author upon reasonable request. References Abeyo, B., Hodson, D., Hundie, B., Woldeab, G., Girma, B., Badebo, A., Alemayehu, Y., Jobe, T., Tegegn, A., & Denbel, W. (2014). Cultivating Success in Ethiopia: The contrasting stripe rust situations in 2010 and 2013. https://globalrust.org/content/cultivating-success-ethiopi a-contrasting-stripe-rust-situations-2010-and-2013 Allen-Sader, C., Thurston, W., Meyer, M., Nure, E., Bacha, N., Alemayehu, Y., Stutt, R. O. J. H., Safka, D., Craig, A. P., Derso, E., Burgin, L. E., Millington, S. C., Hort, M. C., Hodson, D. P., & Gilligan, C. A. (2019). An early warning system to predict and mitigate wheat rust diseases in Ethiopia. Environmental Research Letters, 14(11), 115004. https://doi.org/10.1088/1748-9326/AB4034 Asefa, M., Cao, M., He, Y., Mekonnen, E., Song, X., & Yang, J. (2020). Ethiopian vegetation types, climate and topo graphy. Plant Diversity, 42(4), 302–311. https://doi.org/ 10.1016/j.pld.2020.04.004 Becker-Reshef, I., Barker, B., Whitcraft, A., Oliva, P., Mobley, K., Justice, C., & Sahajpal, R. (2023). Crop type maps for operational global agricultural monitoring. Scientific Data, 10(1), 172. https://doi.org/10.1038/ s41597-023-02047-9 12 G. BLASCH ET AL. http://qgis.osgeo.org http://qgis.osgeo.org https://www.cgiar.org/initiative/plant-health/ https://www.cgiar.org/initiative/plant-health/ https://hdl.handle.net/11529/10549175 https://hdl.handle.net/11529/10549175 https://globalrust.org/content/cultivating-success-ethiopia-contrasting-stripe-rust-situations-2010-and-2013 https://globalrust.org/content/cultivating-success-ethiopia-contrasting-stripe-rust-situations-2010-and-2013 https://doi.org/10.1088/1748-9326/AB4034 https://doi.org/10.1016/j.pld.2020.04.004 https://doi.org/10.1016/j.pld.2020.04.004 https://doi.org/10.1038/s41597-023-02047-9 https://doi.org/10.1038/s41597-023-02047-9 Belgiu, M., & Drăgu, L. (2016). Random forest in remote sen sing: A review of applications and future directions. ISPRS Journal of Photogrammetry and Remote Sensing, 114, 24– 31. https://doi.org/10.1016/J.ISPRSJPRS.2016.01.011 Blagus, R., & Lusa, L. (2013). SMOTE for high-dimensional class-imbalanced data. BMC Bioinformatics, 14(1), 106. https://doi.org/10.1186/1471-2105-14-106 Blasch, G., Alemayehu, Y., Lesne, L., Wolter, J., Taymans, M., Tesfaye, T., Negash, T., Andulalem, M., Gutu, K., Debela, M., Eshetu, Z., Tesfaye, K., Mottaleb, K., Defourny, P., & Hodson, D. P. (2024a). Ethiopian Crop Type 2020 (EthCT2020) dataset. 1. https://doi.org/10. 17632/MFPVMK8CNM.1 Blasch, G., Alemayehu, Y., Lesne, L., Wolter, J., Taymans, M., Tesfaye, T., Negash, T., Andulalem, M., Gutu, K., Debela, M., Eshetu, Z., Tesfaye, K., Mottaleb, K., Defourny, P., & Hodson, D. P. (2024b). Ethiopian Crop Type 2020 (EthCT2020) dataset: Crop type data for environmental and agricultural remote sensing applications in complex Ethiopian smallholder wheat- based farming systems (Meher season 2020/21). Data in Brief, 54, 110427. https://doi.org/10.1016/J.DIB. 2024.110427 Blasch, G., Lesne, L., Wolter, J., Alemayehu, Y., Bontemps, S., Hodson, D., & Defourny, P. (2025). National-scale wheat area dataset of Ethiopia for the 2020/21 rainfed cropping seasons. https://hdl.handle.net/11529/10549175 Bojago, E., Tadila, G., & Masha, M. (2025). Monitoring spa tio-temporal changes in land use, land cover, and NDVI using MODIS data in Ethiopia’s Gambela region. Discover Applied Sciences, 7(11), 1361. https://doi.org/ 10.1007/S42452-025-07879-1/FIGURES/7 Bradshaw, C. D., Thurston, W., Hodson, D., Mona, T., Smith, J. W., Millington, S. C., Blasch, G., Alemayehu, Y., Gutu, K., Hort, M. C., & Gilligan, C. A. (2022). Irrigation can create new green bridges that promote rapid intercontinental spread of the wheat stem rust pathogen. Environmental Research Letters, 17(11), 114025. https://doi.org/10.1088/1748-9326/AC9AC7 CSA. (2020). Agricultural sample survey - Report on area and production of major crops (2019/2020): Vol. I. CSA. (2021). Agricultural sample survey - Report on area and production of major crops (2020/2021). http://www. statsethiopia.gov.et/agriculture/area-and-production-for -major-crops-private-peasant-holdings-meher-season-20 20-2021-2013-e-c/ Defourny, P., Bontemps, S., Bellemans, N., Cara, C., Dedieu, G., Guzzonato, E., Hagolle, O., Inglada, J., Nicola, L., Rabaute, T., Savinaud, M., Udroiu, C., Valero, S., Bégué, A., Dejoux, J. F., El Harti, A., Ezzahar, J., Kussul, N., Labbassi, K., … Koetz, B. (2019). Near real-time agri culture monitoring at national scale at parcel resolution: Performance assessment of the Sen2-agri automated sys tem in various cropping systems around the world. Remote Sensing of Environment, 221(March 2018), 551– 568. https://doi.org/10.1016/j.rse.2018.11.007 Eggen, M., Ozdogan, M., Zaitchik, B. F., & Simane, B. (2016). Land cover classification in complex and frag mented agricultural landscapes of the Ethiopian high lands. Remote Sensing, 8(12), 1020. https://doi.org/10. 3390/RS8121020 Eisfelder, C., Boemke, B., Gessner, U., Sogno, P., Alemu, G., Hailu, R., Mesmer, C., & Huth, J. (2024). Cropland and crop type classification with sentinel-1 and sentinel-2 time series using google earth engine for agricultural monitoring in Ethiopia. Remote Sensing, 16(5), 866. https://doi.org/10.3390/RS16050866/S1 ESRI. (2024, November 26). World terrain base. https:// www.arcgis.com/home/item.html?id=c61ad8ab017d49e1 a82f580ee1298931 European Space Agency. (2019). Copernicus DEM - Global and European digital elevation model | Copernicus data space ecosystem. https://doi.org/10.5270/ESA-c5d3d65 European Space Agency. (2022). ESA SEN4STAT - Sentinels for agricultural statistics. https://www.esa-sen4stat.org/ Faqe Ibrahim, G. R., Rasul, A., & Abdullah, H. (2023). Improving crop classification accuracy with integrated sentinel-1 and sentinel-2 data: A case study of barley and wheat. Journal of Geovisualization and Spatial Analysis, 7(2), 22. https://doi.org/10.1007/s41651-023- 00152-2 Fritz, S., See, L., Bayas, J. C. L., Waldner, F., Jacques, D., Becker-Reshef, I., Whitcraft, A., Baruth, B., Bonifacio, R., Crutchfield, J., Rembold, F., Rojas, O., Schucknecht, A., Van der Velde, M., Verdin, J., Wu, B., Yan, N., You, L., Gilliams, S., … McCallum, I. (2019). A comparison of global agricultural monitoring systems and current gaps. Agricultural Systems, 168, 258–272. https://doi. org/10.1016/j.agsy.2018.05.010 Gelagay, H. S., Leroux, L., Tamene, L., Chernet, M., Blasch, G., Tibebe, D., Abera, W., Sida, T., Tesfaye, K., Corbeels, M., & Silva, J. V. (2025). A crop-specific and time-variant spatial framework for characterizing rainfed wheat pro duction environments in Ethiopia. Agricultural Systems, 227, 104360. https://doi.org/10.1016/J.AGSY.2025.104360 Gilligan, C. A. (2024). Developing predictive models and early warning systems for invading pathogens: Wheat rusts. Annual Review of Phytopathology, 62(1), 217–241. https:// doi.org/10.1146/ANNUREV-PHYTO-121423-041956 Grote, U., Fasse, A., Nguyen, T. T., & Erenstein, O. (2021). Food security and the dynamics of wheat and maize value chains in Africa and Asia. Frontiers in Sustainable Food Systems, 4, 317. https://doi.org/10.3389/FSUFS.2020. 617009 Huang, X., Liu, J., Zhu, W., Atzberger, C., & Liu, Q. (2019). The optimal threshold and vegetation index time series for retrieving crop phenology based on a modified dynamic threshold method. Remote Sensing, 11(23), 2725. https://doi.org/10.3390/RS11232725 Jaleta, M., Hodson, D., Abeyo, B., Yirga, C., & Erenstein, O. (2019). Smallholders’ coping mechanisms with wheat rust epidemics: Lessons from Ethiopia. PLoS ONE, 14(7) e0219327. https://doi.org/10.1371/journal.pone. 0219327 Johnson, D. M., Rosales, A., Mueller, R., Reynolds, C., Frantz, R., Anyamba, A., Pak, E., & Tucker, C. (2021). USA crop yield estimation with MODIS NDVI: Are remotely sensed models better than simple trend ana lyses? Remote Sensing, 13(21), 4227. https://doi.org/10. 3390/RS13214227 Kibret, K. S., Marohn, C., & Cadisch, G. (2020). Use of MODIS EVI to map crop phenology, identify cropping systems, detect land use change and drought risk in Ethiopia – an application of Google Earth Engine. European Journal of Remote Sensing, 53(1), 176–191. https://doi.org/10.1080/22797254.2020.1786466 Kpienbaareh, D., Sun, X., Wang, J., Luginaah, I., Kerr, R. B., Lupafya, E., & Dakishoni, L. (2021). Crop type and land cover mapping in northern Malawi using the integration of sentinel-1, sentinel-2, and PlanetScope satellite data. Remote Sensing, 13(4), 700. https://doi.org/10.3390/ rs13040700 Lalitha, V., & Latha, B. (2022). A review on remote sensing imagery augmentation using deep learning. Materials JOURNAL OF MAPS 13 https://doi.org/10.1016/J.ISPRSJPRS.2016.01.011 https://doi.org/10.1186/1471-2105-14-106 https://doi.org/10.17632/MFPVMK8CNM.1 https://doi.org/10.17632/MFPVMK8CNM.1 https://doi.org/10.1016/J.DIB.2024.110427 https://doi.org/10.1016/J.DIB.2024.110427 https://hdl.handle.net/11529/10549175 https://doi.org/10.1007/S42452-025-07879-1/FIGURES/7 https://doi.org/10.1007/S42452-025-07879-1/FIGURES/7 https://doi.org/10.1088/1748-9326/AC9AC7 http://www.statsethiopia.gov.et/agriculture/area-and-production-for-major-crops-private-peasant-holdings-meher-season-2020-2021-2013-e-c/ http://www.statsethiopia.gov.et/agriculture/area-and-production-for-major-crops-private-peasant-holdings-meher-season-2020-2021-2013-e-c/ http://www.statsethiopia.gov.et/agriculture/area-and-production-for-major-crops-private-peasant-holdings-meher-season-2020-2021-2013-e-c/ http://www.statsethiopia.gov.et/agriculture/area-and-production-for-major-crops-private-peasant-holdings-meher-season-2020-2021-2013-e-c/ https://doi.org/10.1016/j.rse.2018.11.007 https://doi.org/10.3390/RS8121020 https://doi.org/10.3390/RS8121020 https://doi.org/10.3390/RS16050866/S1 https://www.arcgis.com/home/item.html?id=c61ad8ab017d49e1a82f580ee1298931 https://www.arcgis.com/home/item.html?id=c61ad8ab017d49e1a82f580ee1298931 https://www.arcgis.com/home/item.html?id=c61ad8ab017d49e1a82f580ee1298931 https://doi.org/10.5270/ESA-c5d3d65 https://www.esa-sen4stat.org/ https://doi.org/10.1007/s41651-023-00152-2 https://doi.org/10.1007/s41651-023-00152-2 https://doi.org/10.1016/j.agsy.2018.05.010 https://doi.org/10.1016/j.agsy.2018.05.010 https://doi.org/10.1016/J.AGSY.2025.104360 https://doi.org/10.1146/ANNUREV-PHYTO-121423-041956 https://doi.org/10.1146/ANNUREV-PHYTO-121423-041956 https://doi.org/10.3389/FSUFS.2020.617009 https://doi.org/10.3389/FSUFS.2020.617009 https://doi.org/10.3390/RS11232725 https://doi.org/10.1371/journal.pone.0219327 https://doi.org/10.1371/journal.pone.0219327 https://doi.org/10.3390/RS13214227 https://doi.org/10.3390/RS13214227 https://doi.org/10.1080/22797254.2020.1786466 https://doi.org/10.3390/rs13040700 https://doi.org/10.3390/rs13040700 Today: Proceedings, 62, 4772–4778. https://doi.org/10. 1016/J.MATPR.2022.03.341 Liang, J., Ren, C., Li, Y., Yue, W., Wei, Z., Song, X., Zhang, X., Yin, A., & Lin, X. (2023). Using enhanced gap-filling and whittaker smoothing to reconstruct high spatiotem poral resolution NDVI time series based on landsat 8, sentinel-2, and MODIS imagery. ISPRS International Journal of Geo-Information, 12(6), 214. https://doi.org/ 10.3390/IJGI12060214 Livelihood Integration Unit. (2010). An atlas of Ethiopian livelihoods (T. Boudreau, Ed.). Ministry of Agriculture. https://books.google.com.et/books/about/An_Atlas_of_ Ethiopian_Livelihoods.html?id=ZoHIXwAACAAJ&redir _esc=y Marshall, M., Crommelinck, S., Kohli, D., Perger, C., Yang, M. Y., Ghosh, A., Fritz, S., de Bie, K., & Nelson, A. (2019). Crowd-driven and automated mapping of field bound aries in highly fragmented agricultural landscapes of Ethiopia with very high spatial resolution imagery. Remote Sensing, 11(18), 2082. https://doi.org/10.3390/ RS11182082 Mekuriaw, A., Blasch, G., Ademe, D., Ndour, A., Gardi, M. W., Worku, M., & Fantaye, K. T. (2023). Climate change risks to the wheat value chain in Ethiopia: Findings, impli cations and adaptation options. https://hdl.handle.net/ 10883/22791 Mercier, A., Betbeder, J., Baudry, J., Le Roux, V., Spicher, F., Lacoux, J., Roger, D., & Hubert-Moy, L. (2020). Evaluation of sentinel-1 & 2 time series for predicting wheat and rapeseed phenological stages. ISPRS Journal of Photogrammetry and Remote Sensing, 163, 231–256. https://doi.org/10.1016/j.isprsjprs.2020.03.009 Meyer, M., Bacha, N., Tesfaye, T., Alemayehu, Y., Abera, E., Hundie, B., Woldeab, G., Girma, B., Gemechu, A., Negash, T., Mideksa, T., Smith, J., Jaleta, M., Hodson, D., & Gilligan, C. A. (2021). Wheat rust epidemics damage Ethiopian wheat production: A decade of field disease surveillance reveals national-scale trends in past outbreaks. PLoS ONE, 16(2), e0245697. https://doi.org/ 10.1371/JOURNAL.PONE.0245697 Mohamed, A. A. (2017). Food security situation in Ethiopia: A review study. International Journal of Health Economics and Policy, 2(311), 86–96. https://doi.org/10. 11648/J.HEP.20170203.11 Mohammed, M., Birhanu, B., & Abegaz, F. (2023). Multi- spectral remote sensing for current irrigated area map ping of the Rift Valley Lakes Basin in Ethiopia. AQUA — Water Infrastructure, Ecosystems and Society, 72(12), 2396–2407. https://doi.org/10.2166/AQUA.2023.413 Nakalembe, C. (2020). Urgent and critical need for sub- saharan African countries to invest in earth obser vation-based agricultural early warning and monitoring systems. Environmental Research Letters, 15(12), 121002. https://doi.org/10.1088/1748-9326/ABC0BB Nakalembe, C., Becker-Reshef, I., Bonifacio, R., Hu, G., Humber, M. L., Justice, C. J., Keniston, J., Mwangi, K., Rembold, F., Shukla, S., Urbano, F., Whitcraft, A. K., Li, Y., Zappacosta, M., Jarvis, I., & Sanchez, A. (2021). A review of satellite-based global agricultural monitoring systems available for Africa. Global Food Security, 29, 100543. https://doi.org/10.1016/J.GFS.2021.100543 NASA JPL. (2013). NASA Shuttle Radar Topography Mission Global 1 arc second [Data set]. https://doi.org/ 10.5067/MEaSUREs/SRTM/SRTMGL1.003 Neigh, C. S. R., Carroll, M. L., Wooten, M. R., McCarty, J. L., Powell, B. F., Husak, G. J., Enenkel, M., & Hain, C. R. (2018). Smallholder crop area mapped with wall-to-wall WorldView sub-meter panchromatic image texture: A test case for Tigray, Ethiopia. Remote Sensing of Environment, 212, 8–20. https://doi.org/10.1016/j.rse. 2018.04.025 Olivera, P., Newcomb, M., Szabo, L. J., Rouse, M., Johnson, J., Gale, S., Luster, D. G., Hodson, D., Cox, J. A., Burgin, L., Hort, M., Gilligan, C. A., Patpour, M., Justesen, A. F., Hovmøller, M. S., Woldeab, G., Hailu, E., Hundie, B., Tadesse, K., … Jin, Y. (2015). Phenotypic and genotypic characterization of race TKTTF of Puccinia graminis f. sp. tritici that caused a wheat stem rust epidemic in southern Ethiopia in 2013-14. Phytopathology, 105(7), 917–928. https://doi. org/10.1094/PHYTO-11-14-0302-FI Pasternak, M., & Pawluszek-Filipiak, K. (2022). The evalu ation of spectral vegetation indexes and redundancy reduction on the accuracy of crop type detection. Applied Sciences, 12(10), 5067. https://doi.org/10.3390/ APP12105067 Sadeh, Y., Baber, S., Ma, Y., Oliinyk, O., Leroux, L., Shiferaw, T., Blasch, G., Justice, C. J., Munshell, B., Wronkiewicz, M., Lee, J., Lobell, D. B., Becker-Reshef, I., Sadeh, Y., Baber, S., Ma, Y., Oliinyk, O., Leroux, L., Shiferaw, T., … Becker-Reshef, I. (2024). Advancing field-scale yield prediction for data-limited agricultural regions. AGUFM, 2024, GC22E-02. https://ui.adsabs. harvard.edu/abs/2024AGUFMGC22E..02S/abstract. Tadesse, M., Alemu, B., Bekele, G., Tebikew, T., Chamberlin, J., & Benson, T. (2006). Atlas of the Ethiopian rural economy. In Atlas of the Ethiopian rural economy (p. 94). Washington, DC: International Food Policy Research Institute; Addis Ababa: Central Statistical Agency. https://doi.org/10.2499/0896291545 Teshome, M. (2014). Population growth and cultivated land in rural Ethiopia: Land use dynamics, access, farm size, and fragmentation. Resources and Environment, 4(3), 148–161. https://doi.org/10.5923/j.re.20140403.03 Van Tricht, K., Degerickx, J., Gilliams, S., Zanaga, D., Battude, M., Grosu, A., Brombacher, J., Lesiv, M., Laso Bayas, J. C., Karanam, S., Fritz, S., Becker-Reshef, I., Franch, B., Mollà-Bononad, B., Boogaard, H., Pratihast, A. K., Koetz, B., & Szantoi, Z. (2023). Worldcereal: A dynamic open-source system for global-scale, seasonal, and reproducible crop and irrigation mapping. Earth System Science Data, 15(12), 5491–5515. https://doi.org/ 10.5194/ESSD-15-5491-2023 Van Tricht, K., Degerickx, J., Gilliams, S., Zanaga, D., Savinaud, M., Battude, M., Buguet de Chargère, R., Dubreule, G., Grosu, A., Brombacher, J., Pelgrum, H., Lesiv, M., Bayas, J. C. L., Karanam, S., Fritz, S., Becker- Reshef, I., Franch, B., Bononad, B. M., Cintas, J., … Szantoi, Z. (2023). ESA WorldCereal 10 m 2021 v100. https://doi.org/10.5281/ZENODO.7875105 WFP, & FAO. (2023). Hunger hotspots. FAO–WFP early warnings on acute food insecurity: November 2023 to April 2024 Outlook. https://doi.org/10.4060/cc8419en Zhu, Z., Wang, S., & Woodcock, C. E. (2015). Improvement and expansion of the Fmask algorithm: Cloud, cloud sha dow, and snow detection for Landsats 4–7, 8, and Sentinel 2 images. Remote Sensing of Environment, 159, 269–277. https://doi.org/10.1016/J.RSE.2014.12.014 14 G. BLASCH ET AL. https://doi.org/10.1016/J.MATPR.2022.03.341 https://doi.org/10.1016/J.MATPR.2022.03.341 https://doi.org/10.3390/IJGI12060214 https://doi.org/10.3390/IJGI12060214 https://books.google.com.et/books/about/An_Atlas_of_Ethiopian_Livelihoods.html?id=ZoHIXwAACAAJ%26redir_esc=y https://books.google.com.et/books/about/An_Atlas_of_Ethiopian_Livelihoods.html?id=ZoHIXwAACAAJ%26redir_esc=y https://books.google.com.et/books/about/An_Atlas_of_Ethiopian_Livelihoods.html?id=ZoHIXwAACAAJ%26redir_esc=y https://doi.org/10.3390/RS11182082 https://doi.org/10.3390/RS11182082 https://hdl.handle.net/10883/22791 https://hdl.handle.net/10883/22791 https://doi.org/10.1016/j.isprsjprs.2020.03.009 https://doi.org/10.1371/JOURNAL.PONE.0245697 https://doi.org/10.1371/JOURNAL.PONE.0245697 https://doi.org/10.11648/J.HEP.20170203.11 https://doi.org/10.11648/J.HEP.20170203.11 https://doi.org/10.2166/AQUA.2023.413 https://doi.org/10.1088/1748-9326/ABC0BB https://doi.org/10.1016/J.GFS.2021.100543 https://doi.org/10.5067/MEaSUREs/SRTM/SRTMGL1.003 https://doi.org/10.5067/MEaSUREs/SRTM/SRTMGL1.003 https://doi.org/10.1016/j.rse.2018.04.025 https://doi.org/10.1016/j.rse.2018.04.025 https://doi.org/10.1094/PHYTO-11-14-0302-FI https://doi.org/10.1094/PHYTO-11-14-0302-FI https://doi.org/10.3390/APP12105067 https://doi.org/10.3390/APP12105067 https://ui.adsabs.harvard.edu/abs/2024AGUFMGC22E..02S/abstract https://ui.adsabs.harvard.edu/abs/2024AGUFMGC22E..02S/abstract https://doi.org/10.2499/0896291545 https://doi.org/10.5923/j.re.20140403.03 https://doi.org/10.5194/ESSD-15-5491-2023 https://doi.org/10.5194/ESSD-15-5491-2023 https://doi.org/10.5281/ZENODO.7875105 https://doi.org/10.4060/cc8419en https://doi.org/10.1016/J.RSE.2014.12.014 Abstract 1. Introduction 2. Materials and methods 2.1. Study area 2.2. Sentinel-2 satellite data 2.3. Ground reference data 2.4. Wheat mapping remote sensing workflow 2.4.1. Preprocessing of Sentinel-2 data 2.4.2. Classification 2.4.3. Postprocessing 2.4.4. Validation 2.4.5. Consolidation 3. Results and discussion 4. Conclusions Software Acknowledgements Disclosure statement Data availability statement References