Introduction

Irrigated agriculture accounts for over 70% of global freshwater withdrawals and more than 90% of consumptive water use1,2. Increasing temperatures due to climate change elevate evapotranspiration, which raises crop water consumption and reduces water availability for irrigation3,4. Globally, the top 10% of most water-stressed river basins encompass 35% of the global irrigated calorie production5. Amidst accelerating climate change, monitoring the dynamics of crop water consumption and disentangling its drivers is key to securing livelihoods and food security, especially in water-scarce drylands where agriculture relies on irrigation.

In arid and semiarid basins, actual evapotranspiration (ETa) is the dominant outgoing flux of the water budget and therefore proxies crop water consumption in areas dominated by crop production6,7,8. Due to its close association with soil moisture, vegetative health, and atmospheric demand, ETa is a valuable indicator of landscape response to environmental and climatic drivers9. Quantifying ETa over space and time permits assessing crop water consumption and crop health and outlining the pathways for adaptation of crop production to climate change.

The endorheic river basins of Central Asia are particularly susceptible to climate change due to their highly continental and semi-arid climate, where evaporation dominates the water cycle10,11. Human interventions have further intensified pressure on water resources. Large-scale irrigation projects initiated during the Soviet period (1950s–1991) to achieve self-sufficiency in water-intensive cotton production led to the desiccation and salinization of the Aral Sea, once the world’s fourth-largest inland water body12,13. Since the collapse of the Soviet Union, crop production in Central Asia has gradually shifted away from cotton to less water-intensive winter wheat; concurrently, double-cropping (two harvests per year) with higher water demand became more widespread14. However, the impacts of climate change and changes in land management on the changes in crop water consumption in the irrigated areas of Central Asia remain underexplored.

We assess the spatiotemporal dynamics of crop water consumption for every year from 1987 to 2019 across the Amu Darya Basin (535,000 km2), the main tributary of the Aral Sea. We model ETa of the growing season (April through October) for all irrigated areas and the three dominant cropping practices (dry season cropping, wet season cropping, and double cropping). We use the Landsat-based Operational Simplified Surface Energy Balance model (SSEBop)15, that has been successfully applied in other arid and semiarid regions for ETa estimates16,17,18. The model is driven by land surface temperature (LST) from Landsat imagery19 and reference evapotranspiration (PET)20. The model is parameterized by the Normalized Difference Vegetation Index (NDVI) from Landsat19, daily maximum air temperature, net radiation from ERA5-Land, the land component of the fifth generation of the European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis (ERA5)21, and elevation from the Shuttle Radar Topographic Mission (STRM)22. Our annual ETa estimates provide a long-term, high-resolution analysis of crop water consumption in the Amu Darya Basin. These estimates further allow us to disentangle the contribution of land use change and climate change to the changes in crop water consumption.

Results

Actual evapotranspiration in the Amu Darya Basin

We observed the highest crop water consumption in midstream and downstream areas of the Amu Darya Basin where intensive irrigated agriculture dominates (Fig. 1a, d). Stratifying average ETa by cropping practice, we find the highest average crop water consumption for double cropping (511 mm/year), followed by dry season cropping (415 mm/year) and wet season cropping (370 mm/year) (Fig. 1c). Our SSEBop simulations agree with crop coefficient-based ETa estimates from weather stations located in irrigated areas (r ranges from 0.79 to 0.95, RMSE is 11–54 mm/month, with higher differences at lower altitudes and in drier conditions) (Fig. 1b).

Fig. 1: Actual evapotranspiration in the Amu Darya Basin.
Fig. 1: Actual evapotranspiration in the Amu Darya Basin.
Full size image

Annual median actual evapotranspiration (ETa) estimates from the SSEBop model during the growing season (April to October) from 1987 to 2019 across all irrigated areas of the Amu Darya Basin. The map has a 30-mspatial resolution (ETa values over 1250 mm/year correspond to open water areas) (a). Scatterplots of monthly crop coefficient-based ETa estimates using the Penman-Monteith method based on meteorological data (temperature, wind speed, and relative humidity) and Landsat-derived normalized difference vegetation index (NDVI) from five meteorological stations (Table S3) compared against the SSEBop ETa. The dashed line represents the 1:1 relationship. Root mean square error (RMSE) is expressed in mm/month (b). Median annual ETa estimates from SSEBop aggregated by cropping practice (c). The topographical profile of the study area is the mean elevation across each degree of longitude at 30-m resolution22. The elevation profile was smoothed using a 1000-pixel moving window average across the longitude axis (d). The locator map was generated using Cartopy71 with Natural Earth base layers72.

Dynamics of crop water consumption

Crop water consumption of wet season cropping exhibited a sharp rise of more than 200% (Fig. 2a) over the study period (calculated as the difference between the average of 1987–1990 and 2016–2019), driven by the expansion of wet season cropping. Still, dry season cropping consumed the most water in terms of volume (7.8 km3), compared with wet season cropping (3.1 km3) and double cropping (3.8 km3) in 2019. The total crop water consumption has risen by 10% over the study period (Fig. 2b). At the 30-meter pixel level (Fig. 2c), average crop water consumption increased for all cropping practices. Dry season crops showed the highest rise in average ETa at 22%, followed by wet season crops at 17% and double cropping at 16% (difference between the average of 1987–1990 and 2016–2019). The basin-wide average ETa increased by 18% (Fig. 2d).

Fig. 2: Dynamics of crop water consumption in the Amu Darya Basin.
Fig. 2: Dynamics of crop water consumption in the Amu Darya Basin.
Full size image

Crop water consumption estimated as actual evapotranspiration (ETa) in the Amu Darya Basin for the period 1987-2019. Annual values of crop water consumption for each cropping practice (dry season cropping, double cropping, wet season cropping) are expressed in km3 (a) and stacked to obtain the total volume of the three cropping practices for the entire region, showing an increasing trend (black dashed line, p = 0.023) (b). Average crop water consumption per pixel for the three main cropping practices, expressed in mm (c) and stacked for all cropping practices, shows an increasing trend (black dashed line, p < 0.001) (d). Dashed lines show significant trends (p < 0.05), estimated with the non-parametric Mann–Kendall test and the Theil-Sen estimator, represent the change over time. We excluded the years 2002 and 2003 from the ETa estimates due to insufficient availability of Landsat images, which resulted in high uncertainties of actual evapotranspiration (Supplementary Fig. S114).

Climate drivers and anomalies in crop water consumption

We examined the temporal trends of precipitation, atmospheric water demand (ET reference), and maximum air temperature for downstream, midstream, and upstream regions of the Amu Darya Basin (Figs. 3a, 1–3). ET reference (2.45–4.55 mm/year, p < 0.001) and maximum air temperatures (0.04–0.06 °C/year, p < 0.05) exhibit a steady increase throughout the study period across all regions. Higher rates of change in maximum temperature and atmospheric water demand occurred in the downstream region compared to the midstream and upstream sections. These changes indicate severe water stress conditions in the downstream areas, particularly during dry years (Fig. 3b).

Fig. 3: Climatic drivers and anomalies of actual evapotranspiration in the Amu Darya Basin.
Fig. 3: Climatic drivers and anomalies of actual evapotranspiration in the Amu Darya Basin.
Full size image

Annual precipitation and maximum air temperature (Max Temperature) and reference evapotranspiration (Reference ET) during the growing season (April to October) for downstream, midstream, and upstream regions (note the different scaling on the y-axis for maximum temperature in the upstream region). The black dashed line represents the 25th percentile of precipitation for each area and is used to classify dry years, defined as those falling below this threshold. Colored dashed lines show significant trends (p <0.05) estimated with the non-parametric Mann–Kendall test and the Theil-Sen estimator to represent their change over time (a, 1–3). Anomalies of actual evapotranspiration are expressed as the percent deviation from the median for dry (2001) and normal years (2018) (b).

At the pixel level, downstream regions experienced greater annual ETa anomalies. Values can drop 50% below the median, especially during extremely dry years, such as in Karakalpakstan province (e.g., 2001, Fig. 3b). We also observed higher spatial heterogeneity of anomalies in normal and dry years, where neighboring agricultural fields can experience contrasting ETa anomalies that might be related to different water management practices and water availability. The basin-wide and regional ETa (Figs. S2.1–4) exhibited positive anomalies from 2004 onwards, indicating increased crop water consumption.

Fig. 4: Climate and land use contributions to changes in crop water consumption.
Fig. 4: Climate and land use contributions to changes in crop water consumption.
Full size image

Net change (green) in actual evapotranspiration (ETa) (comparing the mean of 1987–1990 and 2016–2019) for the Amu Darya Basin expressed in percentage (%), the contribution of land use change (purple), and the potential contribution of climate change (dashed orange) (a). Net change for downstream, midstream, and upstream regions (b). Stream locations are shown in Fig. 3, and sensitivity analysis using different start and end dates is provided in Fig. S3.

Disentangling the contribution of climate and land use to changes in crop water consumption

We used a decomposition approach to attribute the contributions of climate and land use changes to the observed changes in crop water consumption, based on area-weighted ETa23,24. We compared the average ETa of a baseline period (1987–1990) with that of the recent period (2016–2019). After 2004, climate change increasingly dominated the rising crop water consumption (Fig. S4). Overall, we found that climate change contributed 21% while land use change offset crop water consumption by 3%, resulting in a net increase of 18% (Fig. 4a).

The offset effect varied across the basin (Fig. 4b). The upstream and midstream regions exhibited the most substantial compensatory effects from land use changes. In contrast, land use changes did not compensate for rising water demand downstream, where the increase in ETa was the highest (21%). Although dry season cropping decreased by 25% downstream, it still occupies 60% of the total cropping area, resulting in high water stress conditions amidst accelerating climate change. The 5% offset effect of land use change in the midstream region results from the pronounced changes in the cropping practices away from water-intensive cultivation practices (Fig. S5). Dry season cropping in this region decreased by 40%, while the share of wet season cropping expanded from 13% baseline to 36% in the recent period (2016–2019).

Discussion

Our analysis reveals that, despite the shift toward less water-intensive crops, crop water consumption in the Amu Darya Basin, proxied with ETa, has increased by 10%. Rising atmospheric water demand and increasing temperatures caused the overall and per unit area increase in crop water consumption for all cropping classes. These findings are consistent with global trends of increasing evapotranspiration observed from 1982 to 2010 caused by rising temperatures25 and higher atmospheric evaporative demand both globally and regionally26,27. Similar climate-induced increases in global crop water consumption have been reported for staple crops between 1981 and 2013, with wheat-growing areas experiencing the largest increase28.

Our decomposition analysis revealed that land use changes mitigated 3% of the potential 21% increase in crop water consumption due to climate change, resulting in a net observed increase of 18%. Dry season crops remain the highest water consumers by total volume, but their overall water consumption has remained constant over the last 20 years. Yet, the increasing crop water consumption due to the expansion of wet season crops and double cropping, and the higher use per unit area contributed to the overall increase of crop water consumption. Importantly, these changes in crop water consumption occurred without expansion of total cropland area (Fig. S6)14.

The spatial heterogeneity of the changes in crop water consumption mirrors the pivotal role of environmental conditions, irrigation infrastructure, and water governance in shaping regional patterns of crop water use. The midstream region exhibited the highest reduction of water use due to land use change, thanks to the largest shift from dry season (mainly cotton) to wet season (mainly winter wheat) cropping. Besides, midstream areas benefit from more reliable water availability and better soil quality, allowing for greater agricultural diversification29. Conversely, downstream areas where water-intensive dry season crops continue to dominate witnessed the highest increase in ETa. These areas, which include intensively irrigated regions, such as Khorezm, Karakalpakstan, and Dashoguz (Fig. 1), face higher salinity, shallow groundwater levels, and less favorable irrigation infrastructure, limiting their capacity for transitioning to less water-intensive crops30,31,32.

Governments in Uzbekistan and Turkmenistan retained aspects of the central planning system from the Soviet period, such as controlling crop choices and cultivated area, and imposing cotton production quotas because cotton remained a vital component of their economies33. The centralization disproportionately affected downstream regions, which continued to be burdened with cotton production quotas despite escalating land degradation, rising temperatures along with more frequent heat extremes, and growing water scarcity34,35. However, successful transitions in regions like Australia demonstrate that, even under challenging conditions, targeted land use changes can help mitigate climate impacts36. While our findings underscore that climate change is the dominant driver of increasing ETa, additional factors, such as an increase in applied irrigation water, may also contribute to observed trends in crop water consumption. This study provides a robust framework for quantifying climate-driven changes; future research could advance this understanding by incorporating water balance assessments alongside remote sensing and in-situ observations to disentangle the relative influence of additional water inputs on ETa trajectories.

Key strategies recognized by Central Asian governments focus on increasing crop diversification by introducing alternative crops in planned rotations, alongside adopting conservation agriculture, to mitigate environmental degradation and sustain production37. Balanced approaches are needed that integrate efficient irrigation schemes, cropping schedules adapted to water availability, and crops with higher crop water productivity. More efficient irrigation management is essential to support these practices, yet challenges such as aging infrastructure and low water productivity hinder more sustainable crop production37. Precision agriculture, deficit irrigation, and drought-tolerant crops offer solutions38,39, but their success depends on resilient institutions and effective coordination to ensure sustainable water allocation and policy implementation. The collapse of Soviet-era irrigation systems has fueled conflicts over transboundary water resources, exacerbating salinization and desertification while limiting land rehabilitation efforts40,41. Recent governance reforms in Central Asia have sought to improve institutional capacity and irrigation efficiency; however, ongoing political tensions, infrastructure deficiencies, and limited integration of scientific research into policy decisions continue to hinder progress toward sustainable water management42.

Remote sensing has become a valuable tool for water management and agricultural decision-making. By providing spatially detailed and temporally consistent data on ETa, models like the SSEBop reveal water consumption, water use efficiency, and irrigation performance. These insights can inform strategic planning and monitoring in water-stressed areas, such as the Aral Sea Basin43. However, challenges remain in estimating irrigation losses and integrating remote sensing data with in-situ observations to better comprehend changes in water availability trends41,44. In Central Asia, aging irrigation systems cause significant water conveyance losses, which are rarely integrated into long-term water availability studies37. Moreover, the reliance of the region on glacial and snowmelt runoff, which is increasingly affected by climate change, underscores the urgency of improving water management strategies13,45. Combining the technological advances of remote sensing applications with robust governance frameworks and transboundary cooperation is essential for sustainable water use40.

Improved cropping practices, more efficient irrigation systems, better monitoring, and governance reforms are critical for achieving more sustainable water and food systems in Central Asia. However, these efforts alone will not be sufficient without a global commitment to reducing emissions. In 2024, the average global temperature surpassed 1.5 °C above pre-industrial levels, with each of the past 10 years (2015–2024) ranking among the warmest on record46. Despite the Paris Agreement signed by 200 countries to limit global warming to 1.5 °C, carbon emissions from fossil fuels and other sources have continued to rise47. This highlights the urgency of preparing for critical warming thresholds and higher evapotranspiration rates, requiring stronger mitigation and adaptation strategies48.

Conclusions

Our study highlights the importance of integrated strategies to ensure the long-term sustainability of agriculture and water management in Central Asia. Despite shifts toward less water-intensive crops, the rising atmospheric water demand, increasing temperatures, and cropping intensification continue to drive higher water consumption. Downstream regions are particularly affected by environmental constraints and aging irrigation infrastructure, which exacerbate the challenges posed by climate change. Land use changes can only provide modest mitigation of water scarcity, amidst accelerating climate change and persistent inefficiencies in irrigation systems.

To achieve sustainable water and food systems, it is imperative to address critical research gaps, including assessing the relative influence of additional water inputs on ETa trends, quantifying irrigation losses, and integrating water availability trends across time and space. Enhancing governance frameworks, fostering transboundary cooperation, and modernizing irrigation infrastructure will be essential to balance agricultural productivity with resource conservation. However, local measures alone cannot fully counter the intensifying impacts of a warming climate. Without a meaningful reduction in global emissions, even well-designed regional interventions risk being overshadowed by broader global climatic shifts. For Central Asia, ambitious global action is essential for building resilience in water and agricultural systems.

Methods

Data and methods

Study area

The present study was conducted in the Amu Darya Basin, covering 535,000 km2 from the Tien Shan and Pamir Mountains to the downstream drylands of Tajikistan, Turkmenistan, and Uzbekistan (Fig. 1a). The Amu Darya is the largest river in Central Asia and the primary source of freshwater. The river depends on the water storage in snow and ice from the headwater catchments of the mountains49,50,51. The basin has a highly continental climate with hot, dry summers and cold, wet winters. Average precipitation in the region ranges from 50 mm/year in the desert regions to 1000 mm/year in the mountains. Temperature has a strong diurnal and intra-annual variation from −40 °C (minimum air temperature in winter) to above 40 °C (maximum air temperature in summer). Over 90% of the water withdrawals in the region are used for irrigation52,53. Irrigated crop production focuses on cereals (mainly wheat) cultivated during the wet season (winter) and cotton with a growing period during the hot and dry summer season. The share of fodder and permanent crop cultivation tends to increase with elevation54,55.

Data

For every year of our study period, we isolated the agricultural lands of the Amu Darya using the Landsat-based cropland maps at 30-m spatial resolution of the Amu Darya Basin14. These maps were used to extract the spatial extent and the actual evapotranspiration of the three cropping practices (Wet season cropping, dry season cropping, and double cropping) (Supplementary Table S1). To analyze the temporal trends of the climate drivers, we used daily precipitation from the Climate Hazard Group InfraRed Precipitation with Station Data (CHIRPS version 2) with a spatial resolution of 0.05° (about 5 km at the equator)56, which was considered most accurate for the region57. We also obtained daily maximum air temperature from the global reanalysis ERA5-Land with 0.1° (about 9 km at the equator) of spatial resolution21, and for reference evapotranspiration we used the Potential Evapotranspiration (PET) dataset at a daily scale with 0.1° of spatial resolution20. For trend detection, we employed the non-parametric Mann–Kendall test and the Theil–Sen estimator, both implemented in the pyMannKendall Python package58.

Modeling of actual evapotranspiration

We used the Operational Simplified Surface Energy Balance (SSEBop) model in the Google Earth Engine (GEE)59 cloud computing platform43, along with Python and Quantum GIS (QGIS). SSEBop computes daily total actual evapotranspiration (ETa) using a combination of LST from Landsat and reference ET. We used Landsat collection 2 Tier 1 surface reflectance products19 covering the irrigated agriculture of the Amu Darya Basin during the growing season (April to October) between 1987 and 2019. We only used images with less than 60% cloud cover. In total, 9511 Landsat images were collected, with 42% of the images coming from Landsat 7 (3961), 39% from Landsat 5 (3779), 18% from Landsat 8 (1695), and 1% from Landsat 4 (76). The median number of clear-day images per year was 12 (see Fig. S7). For grass reference evapotranspiration (ETo), we used PET, which was calculated via the FAO’s Penman-Monteith method60. The model also uses as ancillary datasets daily maximum temperature from ERA5-Land and digital elevation model (DEM, 30-m) from the Shuttle Radar Topography Mission (SRTM)22 (See Supplementary Table S2).

SSEBop is a parametric energy balance model that solves the latent heat flux at a daily time scale using a satellite psychrometric approach61. Daily ETa is calculated as the product of ET fraction (ETf) and reference ET15. The difference between the dry-bulb (observed LST) and wet-bulb reference LST determines the ETf in combination with a surface psychrometric constant. The SSEBop model integrates input datasets at their native resolution, with the spatial resolution of the final ETa output determined by the thermal input (LST). Coarser-resolution inputs such as reference ET and air temperature are treated as regional variables, which vary slowly in space and are suitable for integration without resampling, following common practice in thermal-based ET modeling. We used the open-access SSEBop FANO (v.0.2.8) Python implementation62. The model implementation includes an improved parameterization using the Forcing and Normalizing Operation (FANO) algorithm44. This algorithm increases the spatiotemporal coverage of ETa estimations in all landscapes and seasons, regardless of vegetation cover density in comparison with previous versions. ET fractions (ETf) from satellite overpass dates are linearly interpolated to create daily ETf that are multiplied with the corresponding ETo to create daily total ETa which in turn are aggregated to monthly and summed for our period of interest: 1 April to 31 October for all years from 1987 to 2019. All resulting maps are openly available for further use at https://doi.org/10.5281/zenodo.17720311.

Performance evaluation

We evaluated the SSEBop model by comparing its monthly estimates with the commonly used crop coefficient-based (Kc) ET approach63, which combines NDVI and reference evapotranspiration derived from meteorological observations (net radiation, temperature, wind speed, pressure, and relative humidity) to estimate crop water use under optimal, well-watered conditions. Kc-based comparison was conducted at five weather stations, located in irrigated areas of Kashkadarya, Navoi, and Karakalpakstan provinces in Uzbekistan (Fig. 1, Supplementary Table S3). The station data were collected from local research and governmental institutions. While the number and spatial distribution of the stations are limited due to data availability in this region, the performance evaluation prioritizes the consistency of ETa estimates across space and time, rather than absolute accuracy. The Kc-based method serve as the upper boundary for crop ET. Thus, SSEBop ETa is expected to be lower (under water stress) than or similar (under optimal conditions) to the Kc-based estimate for each station. We first estimated grass reference evapotranspiration (ETo) using the standardized Penman-Monteith equation (Eq. 1) for each station:

$${ETo}=\frac{0.408\left({R}_{N}-{G}_{0}\right)+\gamma \frac{900}{{T}_{a+273}}+{u}_{2}({e}_{s}-{e}_{a})}{\Delta +\gamma (1+0.34{u}_{2})}$$
(1)

where u2 is wind speed at 2 m height [m s−1], \({e}_{s}\) is the saturation vapor pressure (kPa), \({e}_{a}\) to the actual vapor pressure (kPa), \(\triangle\) is the slope of the vapor pressure [kPa °C−1], RN is the net radiation at the crop surface [MJ m−2 day−1], G0 is the soil heat flux density [MJ m−2 day−1], and Ta is the mean daily air temperature at 2 m height (°C). We estimated the Penman-Monteith reference evapotranspiration using the PyET package64,65. We then used the Landsat-derived monthly NDVI to estimate the Kc-based ETa (Eq. 2)63.

$${ETa}=1.25\,{NDVI}* {ETo}$$
(2)

We evaluated the SSEBop ETa estimates by averaging the values of all pixels within a 100 m radius circular buffer around each station and comparing them with the Kc-based ETa at weather station sites using Pearson correlation, slope, and root mean square error (RMSE) using Scikit-learn66,67.

To verify that the observed trends were not an artefact of varying data availability, we assessed the relationship between the maximum time gap in clear Landsat observations and the maximum residuals of estimated ETa at the reference sites. This analysis confirmed the absence of any systematic relationship (Fig. S8).

Uncertainty considerations

Remote sensing-based ET estimations have demonstrated good accuracy for water management applications, with a recent evaluation of six OpenET models, including SSEBop, against 139 eddy covariance flux towers in the United States68. Growing season ET was estimated with an average percent mean bias error (%MBE), percent root mean square (%RMSE) and r2 of −2.1%, 19.8%, and 0.87, respectively, for the Ensemble, with individual models ranging from 0.5 to 7.6 (%MBE), 18.5 to 20.7 (%RMSE), 0.79 to 0.84 (r2)69. As part of the OpenET Ensemble, the SSEBop model was evaluated with −2.1% (%MBE), 19.8 (%RMSE), and 0.82 (r2). Earlier evaluation reported the uncertainties or random errors from input variables and model parameters of the SSEBop led to monthly ET estimates with relative errors less than 20% across multiple flux tower sites distributed across different biomes in the United States70.

Decomposition analysis

Decomposition analysis (DA) allows us to distinguish the effects of climate and land use changes on crop water consumption. We adapted a methodological approach from previous studies23,24 and incorporated land use change through an area-weighted ETa method. DA decomposes the change in ETa into two components: climate change and land use effects. To mitigate interannual variability, we compared average ETa over a baseline period (1987–1990) with the average ETa during a more recent period (2016–2019).

The total change in ETa (ΔETa) is a sum of climate-induced change (\(\Delta {{\mathrm{ETa}}}_{{\mathrm{climate}}}\)) and a change induced by land use \((\Delta {{\mathrm{ETa}}}_{{\mathrm{landuse}}})\):

$$\Delta {{ET}_{a}}={\Delta {ETa}}_{{climate}}+{\Delta {ETa}}_{{landuse}}$$
(3)

The contribution due to climate change is estimated by calculating how much evapotranspiration per pixel has changed for each crop class, weighted by crop class baseline area:

$${\Delta {ETa}}_{climate}=\frac{{\sum }_{i}\,{A}_{i,{baseline}}\,x\,({{ETa}}_{i,{recent}}-{{ETa}}_{i,{baseline}})\,}{{\sum }_{i}{A}_{i,{baseline}}},\,{\mathrm{where}}$$
(4)

Ai – is the area under crop class i so that i ∈ {dry season, wet season, double cropping},

ETa i, baseline – is ETa of crop class i in the baseline period,

ETa i, recent – is ETa of crop class i in the recent period.

The contribution due to land use change is estimated by comparing the average ETa in the recent period to what it would have been if the crop structure had been maintained as in the baseline period:

$${\Delta {ETa}}_{{l}{a}{n}{d}{u}{s}{e}}=\,\frac{{\sum }_{i}\,{A}_{i,{recent}}\,* \,{{ETa}}_{i,{recent}}}{{{\sum }_{i}\,A}_{i,{recent}}}-\,\frac{{\sum }_{i}\,{A}_{i,{baseline}}\,* \,{ETa}_{i,{recent}}}{{\sum }_{i}\,{A}_{i,{baseline}}\,}$$
(5)

The additive framework of this decomposition method assumes linearity and does not capture potential interaction between drivers. While climate change may influence land use decisions in other regions, such interactions are considered negligible in the Amu Darya Basin, where irrigation is the primary determinant of agricultural land use.

We conducted a sensitivity analysis (Fig. S3) to confirm that the results remain consistent regardless of how the baseline and recent periods are defined in terms of the number of years.