Impact of Arable Land Abandonment on Crop Production Losses in Ukraine During the Armed Conflict

Abstract

The outbreak of Russia-Ukraine conflict casted an impact on the global food market, which was believed to be attributed to that Ukraine has suffered significant production losses due to cropland abandonment. Nevertheless, recent outbreaks of farmer protests against Ukraine’s grain exports demonstrated that the production losses might not be as severe as previous estimates. By utilizing the adaptive threshold segmentation method to extract abandoned cropland from the Sentinel-2 high-resolution imagery and calibrating the spatial production allocation model’s gridded crop production data from Ukraine’s statistical data, this study explicitly evaluated Ukraine’s crop-specific production losses and the spatial heterogeneity. The results demonstrated that the estimated area of abandoned cropland in Ukraine ranges from 2.34 to 2.40 million hectares, constituting 7.14% to 7.30% of the total cropland. In Ukrainian-controlled zones, this area spans 1.44 to 1.48 million hectares, whereas in Russian-occupied areas, it varies from 0.90 to 0.92 million hectares. Additionally, the total production losses for wheat, maize, barley, and sunflower amount to 1.92, 1.67, 0.70, and 0.99 million tons, respectively, with corresponding loss ratios of 9.10%, 7.48%, 9.54%, and 8.67%. Furthermore, production losses of wheat, barley, and sunflower emerged in both the eastern and southern states adjacent to the conflict frontlines, while maize losses were concentrated in the western states. The findings imply that Ukraine ought to streamline the food transportation channels and maintain stable agricultural activities in regions with high crop production.

1. Introduction

The outbreak of the Russia-Ukraine conflict on 24 February 2022 disrupted Ukraine’s agricultural production, triggering short-term price fluctuations in the international grain market [1,2]. Several evaluations indicated that the production losses of each crop in Ukraine could go beyond 20% at the beginning of the conflict [3,4]. However, recent outbreaks of farmer protests in Eastern European countries show their dissatisfaction with Ukraine’s significant grain exports [5,6]. Ukraine has sustained its grain exports after the conflict, demonstrating that its actual crop production may deviate from previous estimates.

Even though the conflict has not yet ended, the current grain prices and supply have largely returned to levels before the conflict. The global cereal price index has decreased by 25.3% compared to its peak in 2022 [7,8]. Additionally, Ukraine’s grain exports reached around USD 9.2 billion in 2022, recovering to 77% of the level in 2021 [9]. However, the main destinations for these exports have shifted from Western Asia and North Africa to Eastern Europe [10]. Meanwhile, according to Monitoring Agricultural ResourceS (MARS) of European Commission, Ukraine’s wheat, maize, and barley production in 2022 decreased by 4%, 5%, and 2%, compared to the average level over the past five years [11]. Therefore, the extent of Ukraine’s crop losses may not be as severe as originally estimated [12,13,14].

Remote sensing earth observation data can monitor cropland and crop growth on a broad scale, which enables assessment of Ukraine’s crop production after the conflict. High spatiotemporal resolution remote sensing imagery has been widely used for agricultural monitoring in conflict-affected countries, including Sudan, Syria, and Iraq [15,16,17,18]. Following the outbreak of the Russia-Ukraine conflict, several studies have utilized remote sensing data to assess changes in cropland and crop production in Ukraine [19,20]. At the national scale, Kussul et al., 2022 compared pre- and post-conflict land cover classification maps to estimate a decrease of over 3 million hectares in Ukraine’s national cropland area [21]. Ma et al., 2022 combined remote sensing observations and spatiotemporal statistical methods to identify the Kherson State in eastern Ukraine as a hotspot for abandoned cropland [22]. While these studies provide insights into the spatial distribution characteristics of abandoned cropland, they overlook its impact on crop production. This represents a significant research gap that needs to be addressed to fully understand the implications of cropland abandonment on agricultural production.

The normalized difference vegetation index (NDVI) derived from remote sensing imagery is an index that quantitatively reflects changes in crop yield [15,23]. At the national scale, Lin et al., 2023 used state-level averaged NDVI, average temperature, and other factors to estimate wheat production losses and found that losses exceeded 25% for Ukraine [24]. At the regional level, He et al., 2023 employed machine learning algorithms to directly extract the spatial distribution of abandoned cropland [25]. This work estimated a reduction of 1 million hectares in cropland area across six eastern states of Ukraine, resulting in 8 million tons of production losses. However, these studies are dependent on state-level crop production statistics to assess production losses, overlooking the spatial heterogeneity of crop planting patterns. To reduce the potential errors caused by the high aggregation of administrative statistics data, USDA (United States Department of Agriculture) combined SPAM (spatial production allocation model) crop production gridded data and MODIS (Moderate Resolution Imaging Spectroradiometer) time series NDVI data to assess crop losses in Ukraine. This approach predicted reductions in wheat, maize, barley, and sunflower production in Ukraine at 25%, 20%, 30%, and 32%, respectively [26]. Nevertheless, the SPAM data used include only crop planting information up to the year 2010, which may not comprehensively reflect the changing patterns of crop planting in Ukraine in recent years. To a certain extent, this might overestimate the reduction in crop production during the armed conflict and might fail to precisely represent the spatial heterogeneity of crop losses.

Therefore, the objective of this study is to establish an assessment framework that utilizes remote sensing observations and calibrated SPAM crop production gridded data to estimate the spatial heterogeneity of Ukraine’s crop production losses during the armed conflict. Firstly, the Sentinel-2 10 m resolution imagery data and adaptive threshold segmentation algorithm are employed to extract the distribution of abandoned cropland. Secondly, the SPAM data are calibrated using the most recent statistical data for four crops: wheat, maize, barley, and sunflower, which may help accurately reflect recent changes in crop planting patterns. Lastly, the production losses and spatial distribution characteristics are analyzed based on the abandoned cropland and calibrated SPAM gridded data with the policy implications proposed. This study holds considerable significance for spatially explicit evaluations of production losses in Ukraine, furnishing an integrated framework encompassing remote sensing data and gridded statistical information for crop monitoring throughout the armed conflict.

2. Materials and Method

2.1. Study Area

Ukraine is located in Eastern Europe, bordering Russia to the east, Poland to the west, Belarus to the north, and the Black Sea to the south. It is an important geopolitical crossroad between the European Union and Russia. Ukraine comprises 27 state-level administrations covering approximately 600,000 square kilometers. (Figure 1a). The country has about 420,000 square kilometers of cropland (Figure 1b), of which about 340,000 square kilometers are fertile black soil, accounting for 27% of the world’s total black soil area. As a result, Ukraine is one of the world’s leading grain producers and is referred to as the “breadbasket of Europe”. Ukraine’s major crops are wheat, maize, barley, and sunflower, which account for 10%, 15%, 13%, and 50% of the world grain market, respectively.

The Russia-Ukraine conflict erupted on 24 February 2022, primarily affecting southeastern provinces of Ukraine: Luhansk, Donetsk, Zaporizhia, Kherson, Crimean, Kharkiv, and the Dnipropetrovsk (Figure 1c). The changes in Russian-occupied and Ukrainian-recaptured areas from the outbreak of the conflict until early 2023 are shown in Figure 1d, which shows that the areas controlled by both sides have slightly changed since May 2022.

2.2. Data Sources

This study uses the 10 m resolution Sentinel-2A remote sensing data to calculate NDVI. In addition, the ESA WorldCover v100 land cover data provided by the European Space Agency are used to obtain the distribution of cropland in Ukraine. Furthermore, SPAM crop gridded data with a spatial resolution of 10 km are utilized. SPAM includes crop yield, harvested area, and production for each grid cell. Moreover, the administrative boundary vector data of Ukraine are sourced from GADM (Database of Global Administrative Areas). The distribution of controlled areas by Russia and Ukraine during the conflict is obtained from the LIVEUAMAP website.

Additionally, the study acquires state-level statistical inventory on crop harvested area, yields, and production from 2010 to 2021, as provided by the State Statistics Service of Ukraine (https://www.ukrstat.gov.ua, accessed on 13 July 2023). The statistics data are utilized for calibrating the SPAM gridded crop data. Furthermore, the study obtains crop phenology information from Ukraine from the NASA (National Aeronautics and Space Administration) Harvest project and the USDA-IPAD (United States Department of Agriculture-International Production Assessment Division) project. The data are utilized to determine the extraction periods for abandoned cropland. The full details of these data sources are described in

Table 1. Data sources.

Name	Type	Resolution	Period	Source
Sentinel-2A	Raster	10 m	2019–2022	[27]
ESA WorldCover 2020 v100	Raster	10 m	2020	[28]
SPAM 2010 v2.0	Raster	10 km	2010	[29]
Crop statistics data	Plain text	-	2017–2021	https://www.ukrstat.gov.ua (accessed on 13 July 2023)
GEOGLAM data	Plain text	-	2017–2022	https://www.nasaharvest.org (accessed on 1 July 2023)
IPAD data	Plain text	-	2017–2023	https://ipad.fas.usda.gov (accessed on 1 July 2023)
Ukraine administration boundary	Vector	-	2022	https://gadm.org (accessed on 20 June 2023)
Russia-Ukraine controlled area	Vector	-	2022	https://liveuamap.com (accessed on 25 June 2023)

.

2.3. Methodological Framework

This study establishes an assessment framework for quantitatively analyzing crop production losses resulting from the Russia-Ukraine conflict. The framework uses remote sensing observations and spatially explicit gridded crop production data. It comprises three main components: abandoned cropland extraction, SPAM gridded crop production calibration, and crop production loss estimation, as depicted in Figure 2. In the abandoned cropland extraction module, this research employs Sentinel-2A imagery data to calculate Ukraine’s NDVI changes and employs the OTSU threshold segmentation method to obtain abandoned cropland. Within the SPAM gridded crop production calibration module, the study adjusts SPAM data using Ukraine’s state-level statistical data to match with recent crop planting patterns. Finally, by overlaying abandoned cropland distribution with the calibrated gridded crop production, the study calculates production losses and analyzes the distribution characteristics.

2.3.1. Abandoned Cropland

The vegetation index-based detection method and machine learning-based classification method are utilized to identify abandoned cropland using remote sensing data. It is challenging to apply machine learning-based classification methods to war zones due to the complexity of land cover changes during wartime and the difficulty of obtaining validation samples [30]. Therefore, detection methods based on vegetation indices are employed to avoid error propagation in the classification results of machine learning algorithms [31]. By far, vegetation index-based methods have proven successful for abandoned cropland extraction in conflict areas [32].

This study leverages Sentinel-2A high-resolution remote sensing data and the Google Earth Engine cloud computing platform to extract abandoned cropland based on NDVI changes in the beforemath and aftermath of the conflict. Then, the OTSU threshold segmentation algorithm is employed to categorize images into target and background classes based on their levels [33], identifying an optimal threshold through iteration to maximize inter-class variance or minimize intra-class variance [34]. The detailed algorithmic steps are as follows:

$$MaxNDV{I}_{i,y}=Max(NDV{I}_{i,y,T})$$

(1)

$$y\in \left[2019, 2022\right]$$

$$T\in \left[May, October\right]$$

According to crop phenology data, the major growing seasons for the primary four crops in Ukraine are between May and October each year. Therefore, following Equation (1), the maximum NDVI synthetic image $MaxNDV{I}_{i,y}$ is calculated for each year from 2019 to 2022 within this time frame. $i$ represents the pixel, $y$ represents the year, and $T$ denotes the period from May to October.

$$NDVIChang{e}_i=MaxNDV{I}_{i,2022}-Avg\left(MaxNDV{I}_{i, y^{\prime }}\right)$$

(2)

$$y^{\prime }\in \left[2019, 2021\right]$$

Using the average of the synthesized maximum NDVI images for the years 2019, 2020, and 2021 as the pre-conflict maximum NDVI value $Avg\left(MaxNDV{I}_{i, y^{\prime }}\right)$, the NDVI difference image is obtained by subtracting the maximum NDVI value for the year 2022 $MaxNDV{I}_{i,2022}$, as shown in Equation (2). To improve the accuracy of the abandoned cropland extraction, after multiple tests, the pixels with $NDVIChange$ less than zero were retained for threshold segmentation. Subsequently, applying the OTSU adaptive threshold segmentation algorithm, the NDVI difference image is partitioned into two categories: abandoned cropland and non-abandoned cropland land (0/1).

$${\mu }_s={\displaystyle \frac{Su{m}_s}{N_s}}$$

(3)

The algorithm assumes that threshold $t$ divides the $NDVIChang{e}_i$ image into two segments: abandoned cropland $\left(s=1\right)$ and non-abandoned cropland $\left(s=0\right)$. It calculates the mean values, ${\mu }_0$ and ${\mu }_1$ for pixels in these two segments. In Equation (3), $Su{m}_0$ and $Su{m}_1$ represent the total sum of pixel values in the two segments, while $N_0$ and $N_1$ correspond to the number of pixels in each segment.

$${\omega }_s={\displaystyle \frac{N_s}{N}}$$

(4)

Afterwards, the proportions of the two classes’ pixel count to the total pixel count, respectively, denoted as ${\omega }_0$ and ${\omega }_1$, are calculated using the algorithm, as shown in Equation (4), with $N$ representing the total pixel count.

$${\sigma }_B^2={\omega }_0{\omega }_1{\left({\mu }_0-{\mu }_1\right)}^2$$

(5)

Based on the above results, the between-class variance ${\sigma }_B^2$ between the two categories is defined as Equation (5).

$${\sigma }_B^2\left(T\right)=max\left({\sigma }_B^2\left(t\right)\right)$$

(6)

Then, by iterating through all pixel values as potential threshold values $t$, the segmentation threshold $T$ that maximizes the between-class variance ${\sigma }_B^2\left(T\right)$ is identified as shown in Equation (6).

In this study, the threshold segmentation is conducted for each state. To mitigate the influence of extreme values on threshold determination, the study employed the bootstrap method for pixel sampling. By conducting OTSU segmentation across 1000 resampling iterations, the maximum, minimum, and median thresholds were derived as presented in Table S1 of the Supplementary Materials, which delineate the uncertainty interval in the estimated area of abandoned cropland. Then, the resulting abandoned cropland raster data are transformed into vector data. The noise pixel is removed, and an “eliminate polygon part” operation is performed to enhance the completeness of the cropland parcel polygons.

$$AbandonRati{o}^{grid}={\displaystyle \frac{AbandonAati{o}^{grid}}{CroplandAre{a}^{grid}}}$$

(7)

Finally, the abandoned cropland distribution data are spatially overlaid with the SPAM grids. The abandoned cropland area proportion $\left(AbandonAati{o}^{grid}\right)$ within each grid is calculated as follows. The $AbandonAati{o}^{grid}$ represents the abandoned cropland area within the grid, and $CroplandAre{a}^{grid}$ represents the total cropland area of the grid.

2.3.2. Crop Production Losses Calculation

This research investigates four key crops in Ukraine: wheat, maize, barley, and sunflower. The harvested area of these four crops constitutes 70% of the total area of crops in Ukraine, and they are also the major export products. The SPAM2010 gridded data provide information on harvested areas, yields, and productions for 42 crops worldwide in 2010, at a resolution of 10 km [29]. Since these data capture crop information for 2010, which differs from the actual crop planting patterns in recent years in Ukraine, it is necessary to calibrate the gridded data with the statistical data at the regional scale [35]. For instance, from 2010 to 2021, barley’s harvested area decreased from 4.31 million hectares to 2.47 million hectares, while maize’s harvested area increased from 2.65 million hectares to 5.48 million hectares [36].

Consequently, this study calibrates the SPAM2010 grid data to align with the crop production patterns in Ukraine for the year 2020. The calibration process is carried out in two steps: first, the original grid data are adjusted based on state-level statistical data from 2010; second, the grid data are projected to 2020 by applying the rate of change derived from statistical data spanning 2010 to 2021. The adjustments encompass both harvested area and yield, with production estimates generated from the calibrated results. The detailed steps involved in this calibration process are outlined as follows:

$$Rati{o}_{y,s,c}^{HA}={\displaystyle \frac{H{A}_{y,s,c}^{statistic}}{H{A}_{y,s,c}^{aggregated }}}$$

(8)

$$H{A}_{y,s,c}^{aggregated }={\displaystyle \sum }_{grid}H{A}_{y,s,c}^{grid },grid ϵ s$$

(9)

The calibration of crop harvested area at the grid scale $(H{A}_{y,s,c}^{grid })$ is based on the ratio between state-level statistical data $(H{A}_{y,s,c}^{statistic})$ and the aggregated area $(H{A}_{y,s,c}^{aggregated })$ of the original grid data.

$$H{A}_{y,s,c}^{grid,calibrated }=H{A}_{y,s,c}^{grid }\ast Rati{o}_{y,s,c}^{HA}$$

(10)

where $y$ represents the year; $s$ denotes the state; and $c$ represents the crop type. The aggregated grid-scale harvested area $\left(H{A}_{y,s,c}^{grid, calibrated }\right)$ is aligned with the statistical data by multiplying the grid data by the corresponding $(Rati{o}_{y,s,c}^{HA})$.

$$Rati{o}_{y,s,c}^{YI}={\displaystyle \frac{Y{I}_{y,s,c}^{statistic}}{Y{I}_{y,s,c}^{average}}}$$

(11)

The calibration of yield per unit area for the grid $\left(Y{I}_{y,s,c}^{grid }\right)$ is similarly based on the ratio $\left(Rati{o}_{y,s,c}^{YI}\right)$ of state-level statistical yield $\left(Y{I}_{y,s,c}^{statistic}\right)$ to the average yield $\left(Y{I}_{y,s,c}^{average}\right)$ of the grids within the same state.

$$Y{I}_{y,s,c}^{average}={\displaystyle \frac{H{A}_{y,s,c}^{grid,calibrated}\ast Y{I}_{y,s,c}^{grid}}{H{A}_{y,s,c}^{grid,calibrated}}},grid ϵ s$$

(12)

Meanwhile, the formula for calculating the state-level average yield of the grid data is as follows:

$$Y{I}_{y,s,c}^{grid,calibrated }=Y{I}_{y,s,c}^{grid }\ast Rati{o}_{y,s,c}^{YI}$$

(13)

Then, the calibrated grid yield is calculated as:

$$P{R}_{y,s,c}^{grid,calibrated }=H{A}_{y,s,c}^{grid,calibrated }\ast Y{I}_{y,s,c}^{grid,calibrated }$$

(14)

The calibrated crop production $\left(P{R}_{y,s,c}^{grid,calibrated }\right)$ is obtained by multiplying the grid-scale harvested area by the yield per unit area.

In this study, we calibrated the SPAM 2010 grid data, utilizing the statistical data from 2010. Subsequently, we established linear regression relationships for harvested area and yield across various crops in each state, as illustrated in Figures S1 and S2 of the Supplementary Materials. It should be noted that due to limitations in statistical data, the yield data for barley and sunflower seeds in 2010 are missing. Furthermore, since Crimea was occupied by Russia in 2014, statistical data for this state are unavailable. Thus, this study does not adjust its 2020 crop gridded data. Based on these regression equations, we calculated the rate of change for both variables from 2010 to 2021 to provide a foundation for calibrating the grid data for 2020.

A comparison with state-level statistical data reveals that the median difference between the calibrated grid data for crop production in 2010 and 2020 and the corresponding statistical figures is less than 5.5% (Figure S3), indicating that the calibration successfully achieved an accurate spatial distribution grid.

$$P{R}_{y,s,c}^{grid,loss}=P{R}_{y,s,c}^{grid,calibrated }\ast AbandonRati{o}^{grid}$$

(15)

Finally, the total crop production within the grid cell $\left(P{R}_{y,s,c}^{grid,calibrated }\right)$ is multiplied by the abandonment ratio $AbandonRati{o}^{grid}$, which helps generate the losses caused by abandoned cropland $\left(P{R}_{y,s,c}^{grid,loss}\right)$.

2.3.3. Spatial Analysis of Ratios of Abandoned Cropland and Production Losses

The bivariable local Moran’s I method can quantitatively calculate the local indicators of spatial association (LISA) between two spatial variables [37]. In this study, this coefficient was used at the grid scale to analyze the spatial relationship between the cropland abandonment rate and crop production, which could reveal the impact of cropland abandonment on the production of major crops and their spatial distribution. This coefficient is calculated as below.

$$I_{i,c}=Z_{productio{n}_{c,i}}{\displaystyle \sum }_{j=1,j\ne i}^N{W}_{i,j}{Z}_{abandoned,j}$$

(16)

In Equation (16), $productio{n}_{c,i}$ represents the total production of crop $c$ in grid $i$, and $abandone{d}_j$ represents the abandonment rate of grid $j$ in the neighborhood of grid $i$. $Z_{productio{n}_{c,i}}$ and $Z_{abandoned,j}$ represent the values that have been standardized using z-scores. $W_{i,j}$ represents the spatial adjacency relationship between grid $i$ and $j$. In this study, a first-order Queen weight matrix was used, where $W_{i,j}=1$ when grid $i$ and $j$ share an edge, and 0 otherwise. When $I_{i,c}$ is a positive value, it indicates that the value in a grid is similar to the high or low values in its neighboring grids. Conversely, when it is a negative value, it reflects the difference in values [38]. In this research, the GeoDa software (version 1.20.0) [39] was employed as the analytical tool to interpret the computational results at a significance level of $p=0.05$. The gridded polygon vector data encompassing information of cropland abandonment rates and crop production were employed as input, and the analyzed attributes of spatial clustering were appended to the same vector file.

According to the spatial distribution of the calculated LISA coefficients, the spatial grids can be classified into four types:

(1)

High-High (H-H) denotes regions characterized by both high crop production and high abandonment rates, suggesting that areas with substantial production volumes have experienced significant reductions.

(2)

Low-Low (L-L) signifies both low crop production and low abandonment rates, indicating that crop production in these regions is relatively modest, and the issue of cropland abandonment is comparatively mild.

(3)

Low-High (L-H) stands for low production and high abandonment rate. It suggests that although these regions have witnessed more abandonment of cropland, their original crop production was relatively low.

(4)

High-Low (H-L) represents high crop production but low abandonment rate, indicating that the crop production is relatively high and the abandonment of cultivated land is not severe.

3. Results

3.1. Abandoned Cropland Distribution

Figure 3 illustrates the spatial distribution of abandoned cropland extracted using median thresholds at grid scales of 10 m and 10 km, respectively. The abandoned cropland is concentrated in the eastern and southern regions, particularly on the Kharkiv-Luhansk border in the east, the Zaporizhia-Kherson border in the southeast, and the Kherson-Mykolaiv border. These regions are close to the frontlines of the conflict and have been major battlegrounds. Additionally, a high proportion of abandoned cropland is observed in western regions such as Vinnytsya and Khmelnytskyi.

The total area of abandoned cropland is estimated to be approximately 2.34 to 2.40 million hectares, which constitutes about 7.14% to 7.30% of the total cropland area in Ukraine. The extent of abandoned cropland is ample enough to fulfill the food requirements of 10.5 million people (0.22 hectares per person [40]), which is comparable to the total population of Sweden or the Czech Republic. Specifically, the abandoned area within Ukrainian-controlled zones ranges from 1.44 to 1.48 million hectares, accounting for 11.82% to 12.00% of the cropland in those regions. In areas occupied by Russia, the extent of abandoned cropland is between 0.90 and 0.92 million hectares, representing approximately 7.89% to 8.02% of the arable land in those regions (Table S2).

3.2. Calibrated Crop Production

The calibrated gridded harvested areas aggregated at the national level are presented in Figure 4a. It is evident that the calibrated area significantly differs from the original SPAM data, with barley decreasing from 24% to 13%, maize increasing from 15% to 24%, sunflower rising from 25% to 30%, and wheat declining from 36% to 33%. Furthermore, Figure 4b illustrates the production shares of each major crop in terms of quantity. Notably, the share of maize production has shifted from 29% to 40%, while barley has decreased from 19% to 11%, sunflower has increased from 14% to 17%, and wheat has declined from 38% to 32%.

The results indicate that Ukraine’s proportion of maize and sunflower production has gradually increased, while the proportion of barley production has declined, and the proportion of wheat production has remained relatively stable. This trend reflects a gradual replacement of barley cultivation by maize and sunflower in Ukraine since 2010. Therefore, it is both effective and necessary to employ actual state-level statistical data to calibrate the original SPAM 2010 dataset and project it onto the SPAM 2020 dataset, which reflects the changing trends in Ukraine’s crop cultivation patterns in recent years.

3.3. Crop Production Losses

Figure 5 illustrates the spatial distribution of production losses for various crops based on abandoned cropland at median segment thresholds. The regions adjacent to the Zaporizhia-Kherson border and the Kherson-Mykolaiv border are identified as hotspots for production losses in wheat, barley, and sunflower. Additionally, significant losses in wheat and sunflower are observed along the border between Kharkiv and Luhansk states, with comparatively lower losses noted for maize. Consequently, losses in wheat, barley, and sunflower occur in both eastern and southern regions near the conflict frontlines, while maize losses are primarily concentrated in southwestern areas.

Furthermore, the production losses of wheat, maize, barley, and sunflower amount to 1.92, 1.67, 0.70, and 0.99 million tons, with corresponding production loss ratios of 9.10%, 7.48%, 9.54%, and 8.67%, respectively. The loss amounts are approximately equivalent to the quantities of wheat produced in Belgium, maize in Croatia, barley in Morocco, and sunflower in Tanzania during 2022 (Table S4). Within the area controlled by Ukraine, the production of wheat, maize, barley, and sunflower decreased by 1.18, 1.54, 0.46, and 0.64 million tons, respectively. In contrast, the losses in the area occupied by Russia were 0.74, 0.13, 0.25, and 0.35 million tons, respectively (Figure S4). The detailed crop production losses for various segment thresholds are presented in Table S3 of the Supplementary Materials.

3.4. Spatial Heterogeneity of Crop Production Losses

To further elucidate the impact of abandoned cropland on crop production losses, this study examined the spatial clustering relationship between rates of cropland abandonment and crop yields (Figure 6).

The regions exhibiting high production and high abandonment rates (H-H) for wheat, barley, and sunflower are primarily found in the Mykolaiv, Kherson, and Vinnytsya states. For maize, the H-H zones are predominantly located in the Vinnytsya, Kyiv, and hernivtsi states. The abandoned cropland in the H-H regions significantly expands the overall declining crop production in Ukraine. Concurrently, the regions exhibiting high production and low abandonment rates (H-L) for wheat, barley, and sunflower are predominantly located in Kharkiv, Luhansk, and Zaporizhia states. The H-L zones of maize are identified in Chernihiv, Sumy, and Poltava states. The regions have large crop production capacities and have been less affected by the armed conflict. They serve as major contributors to Ukraine’s grain supply and play a critical role in alleviating the pressure of Ukraine’s production losses.

In addition, the low production regions (L-L and L-H) for wheat, barley, and sunflower are primarily situated in northern and western Ukraine. In contrast, the L-L and L-H regions for maize are predominantly found in the southeastern part of the country. With a lower amount of crop production, the impact of the cropland abandonment in these areas has a relatively small effect on the overall national grain reduction. Among them, the L-L regions have significant potential for increasing grain production and are less affected by the armed conflict. Increasing crop yields in these areas is essential to offset losses in other high production regions.

4. Discussion

4.1. Comparison with Previous Estimates

This study reveals that approximately 2.34 to 2.40 million hectares of cropland were abandoned in Ukraine after the conflict, accounting for 7.14% to 7.30% of the total cropland area. This finding is consistent with prior research estimates indicating that the proportion of abandoned cropland ranges from 7.5% [13] to 10% [21]. Moreover, the distribution pattern of abandoned cropland closely aligns with previous research [15,21,25], as shown in Figure S5. On the other hand, the production of wheat, maize, barley, and sunflower decreased by 9.10%, 7.48%, 9.54%, and 8.67%, respectively, which indicated that the crop production losses in Ukraine are less than previously estimated.

For validation of the accuracy, this study compares the results in this study with the estimates of production losses from multiple sources, as presented in

Table 2. Different assessments of changes in Ukraine’s crop production in 2022.

	USDA * 2023	MARS ** 2022/6	MARS 2022/8	This Study	He et al., 2023 [25]	Lin et al., 2023 [24]	Wagner et al., 2023 [14]
Wheat	−25%	−1%	−4%	−9%	Eastern states production losses 7%	Nationwide winter wheat losses 25%	Nationwide winter crop harvested 94%
Maize	−20%	−30%	−5%	−7%
Barly	−30%	−20%	−18%	−10%
Sunflower	−32%	−15%	−2%	−9%

* USDA: United States Department of Agriculture. ** MARS: Monitoring Agricultural ResourceS. The years in the table indicate the release time of the assessment data, where 2022/6 represents June 2022, and 2022/8 represents August 2022. The percentage values in the table are derived by preserving integers from the original data.

. According to the assessment by USDA, the losses of the four major crops exceed 20%. The losses are overestimated due to the lower spatial resolution of MODIS images and the employment of uncalibrated SPAM data [26]. Meanwhile, Lin et al., 2023 estimated a 25% loss for winter wheat based on state-level statistical data [24]. The overestimation is mainly caused by the limited sample data in the evaluation model. In contrast, Wagner et al., 2023 found that 94% of wheat in Ukraine was harvested based on satellite observations [14]. He et al., 2023 also revealed that wheat production in eastern Ukraine has decreased by approximately 7% [25]. Meanwhile, the assessments from MARS have also gradually lowered expectations regarding the reduction in crop production in Ukraine [11,41]. Our results are mostly consistent with this recent evaluation, which is based on multiple sources of information, including remote sensing, field surveys, and crop growth models.

Therefore, the reasons for the lower-than-expected degree of crop reduction in Ukraine may include the following aspects: Firstly, the meteorological conditions in 2022 were favorable for crop growth, resulting in relatively high production across the Black Sea region. For example, Russia’s grain production increased by 24.2% in 2022; however, the statistical data from Ukraine do not account for changes in production in the territories occupied by Russia [42]. Furthermore, Ukraine experienced bumper crop production in 2021 [43,44], which significantly amplified the degree of production reduction when compared year-on-year to 2022. Nevertheless, in comparison to the average production levels from 2017 to 2021, the decrease in grain production is relatively minor [45]. Additionally, significant alterations have occurred in Ukraine’s cropping patterns [46], such as the expansion of corn and sunflower areas, while the sowing areas for wheat and barley have gradually declined (Figure S1), contributing to the overestimation of barley losses.

4.2. Factors Influencing Cropland Abandonment

The abandoned cropland of Ukraine is concentrated in the eastern and southern regions, as shown in Figure 3. The abandonment of cropland in these areas is primarily due to the direct impact of the war, including fires, explosions, and movement of heavy equipment. However, various croplands are abandoned in the western regions, such as Khmelnytskyi and Vinnytsya, despite the lower direct impact of armed conflict in these areas, which is attributed to the indirect effects of war [12].

The labor shortage is a major factor contributing to cropland abandonment and crop production loss [47]. Ukraine has seen an outflow of over 6.33 million refugees since the outbreak of the conflict, representing 14% of its total population [48]. Figure 7 presents the displaced population data for various states of Ukraine, as sourced from the UNHCR (United Nations High Commissioner for Refugees). The eastern states, including Luhans’k, Kharkiv, and Dnipropetrovs’k, along with the northern region of Kiev, have experienced the most significant population outflow. Concurrently, the western regions of Odessa and Vinnytsya have also seen a substantial loss of labor force, further exacerbating the abandonment of cropland in these areas. Moreover, the conscription policy implemented by the government during the war further reduced the workforce available for farming [49].

Additionally, the conflict has led to significant increases in energy and fertilizer prices, raising agricultural costs and jeopardizing farmers’ inclination for crop planting [50,51]. Moreover, domestic grain prices in Ukraine have fallen below export prices due to rising transportation costs and lower domestic demand, further discouraging farmers from growing grain [52,53,54]. In light of these challenges, the government should increase financial support for farmers engaged in grain production, reduce energy and fertilizer prices, and ensure smooth channels for grain sales.

4.3. Policy Implications

Ukraine’s crop production capacity remains relatively stable in 2022 based on this study’s assessment. Consequently, the public panic induced by information asymmetry is the primary factor contributing to fluctuations in food prices, especially during wartime when field surveys and government statistics are challenging to publish promptly [55]. The seamless integration of remote sensing observations and statistical data provides a robust and reliable tool for objectively and transparently evaluating changes in agricultural production [56,57]. For countries in conflict, assessing agricultural yields can significantly support adjustments to food production policies. Importantly, nations that rely on food imports face an urgent need for agricultural assessment tools to aid in adapting their import strategies, thereby alleviating domestic food security crises and inflation [58].

As the ongoing conflict continues, the Ukrainian government needs to implement region-specific policies to ensure stable crop production. For the eastern and southern states most directly affected by armed conflict, such as Mykolaiv and Kherson, revitalizing agricultural production should be a priority. The government must enhance security measures to ensure safety in agricultural areas while providing financial assistance to farmers to restore damaged cropland [59]. In contrast, crop production is low in the western regions. To mitigate these production losses, the government should actively encourage farmers to return to safe areas for agricultural activities. Additionally, the northern states have been less impacted by the conflict, resulting in sustained crop production. It is crucial to ensure adequate agricultural resources and energy supplies in regions like Chernihiv and Sumy. Targeted transportation policies must be implemented to facilitate the efficient movement of products in these states [60].

On the other hand, trade restrictions and transportation blockades are also leading to an increase in global food prices [61,62]. To address the issue of grain prices caused by the conflict, international cooperation is crucial. For instance, the “Black Sea Grain Initiative” aims to restore Ukraine’s grain maritime export channels, bringing its grain exports back to pre-conflict levels [63]. However, Russia has suspended its commitment to this agreement, which could potentially lead to further volatility and disruption in the international grain market. To maintain Ukraine’s role in global food security, it is crucial to promote adherence to the agreement and reestablish maritime export channels among all parties involved. Furthermore, alternative avenues for grain exports should be explored. Initiatives like the “solidarity lanes” could support exporting grains through overland routes [64].

4.4. Limitations and Future Prospects

While high-resolution Sentinel-2 images can capture fine NDVI differences, this study focused only on changes in crop growth within arable land, lacking information on crop type distribution. Moreover, the coarse statistical units used for calibrating the crop production grid (state level) increase uncertainty in spatial distribution. The armed conflict further undermines the reliability of statistical data, as government statistics often miss parts of Ukraine, and crop information in Russian-occupied areas lacks transparency [56]. Furthermore, the study neglected the effects of climate change and soil quality on crop yield, such as increased yields from rising temperatures and land degradation due to soil erosion [65,66]. Additionally, the method employed in this study for identifying abandoned cropland has limitations in differentiating between various drivers of abandonment, such as government-led ecological programs and farmer-initiated abandonment.

To enhance the accuracy of crop production evaluations, future research should integrate multiple remote sensing data sources to identify abandoned cropland [67]. Moreover, high-resolution data on crop production distribution should be developed to avoid errors from mismatches between the scales of abandoned land and crop production grid cells. Furthermore, incorporating data from various sources—such as remote sensing, ground observations, crop growth models, and social media—could improve the assessment [68]. This could help accurately predict the heterogeneity of crop reduction and the disturbance caused to the global food market.

5. Conclusions

This study first identifies Ukraine’s abandoned cropland distribution during the armed conflict based on remote sensing imagery. The original SPAM crop production gridded data are then calibrated according to the authoritative statistical data. Finally, the abandoned cropland is coupled with the crop production grids to obtain the crop production loss distributions. Based on the assessment framework, this study draws the following conclusions:

(1)

The Russia-Ukraine conflict has led to the abandonment of over 2.34 million hectares of cropland in Ukraine, accounting for more than 7% of the total cropland area. In addition, the abandoned cropland area in Ukrainian-controlled regions is approximately 1.5 times larger than that in Russian-occupied territories.

(2)

Crop production losses due to cropland abandonment in Ukraine amount to 1.92, 1.67, 0.70, and 0.99 million tons for wheat, maize, barley, and sunflower, respectively. Moreover, crop production is worse in Ukrainian-controlled areas than in Russian-occupied zones.

(3)

The crop production losses of Ukraine are concentrated in the war frontline areas of the eastern and southern and the western regions affected by population outflow.

Therefore, this study concludes that the impact of conflict on Ukraine’s food production is not as severe as previous estimates. To maintain stable food production, the Ukraine government should urge the reclamation of abandoned cropland in the eastern and southern regions. Additionally, it is recommended that the Ukraine government should preserve agricultural production in the major grain-producing areas located in the north and enhance production in the low-production western regions. In addition, the international community should assist Russia and Ukraine in reaching a ceasefire agreement to safeguard global food security.