Located in the northwestern part of China, Xinjiang (73°40′–96°18′E, 34°25′–48°10′N) represents the country’s most extensive provincial-level region and experiences a typical temperate continental climate. As China’s primary cotton production base, Xinjiang plays a critical role in national cotton yield and quality, owing to its extensive land resources and favorable thermal and solar conditions (Li et al., 2020b). However, situated within an arid and semi-arid region, Xinjiang experiences scarce natural precipitation, rendering cotton growth highly dependent on artificial irrigation. Therefore, Xinjiang serves as an ideal region for studying the suitability distribution of economic crops in arid and semi-arid environments.

The MaxEnt model is a machine learning technique grounded in the principle of maximum entropy, widely used for species distribution modeling (SDM) (Elith et al., 2011). The fundamental concept is to identify a probability distribution that maximizes entropy, meaning it is the least biased and most uniform given partial environmental data. According to the principle of maximum entropy, when no additional information is available, the most uniform distribution should be chosen (Guiasu and Shenitzer, 1985; Brummer and Newman, 2019). In the MaxEnt framework, a species’ potential habitat distribution is represented by a probability distribution, where various environmental variables are introduced as feature functions (e.g., linear, quadratic, hinge features) to capture the relationship between environmental conditions and species distribution. The objective of the MaxEnt model is to find the optimal predictive model by maximizing the entropy of this distribution. The model’s optimization goal is expressed as follows:

Where p (x) is the probability of the species occurring at location x. f_i(x) is the feature function for the environmental variable at location x, which describes the relationship between environmental factors and species distribution. λ_i is the weight associated with each feature function, representing the impact of various environmental variables on species distribution.

Through adjusting these weights λ_i, the MaxEnt model maximizes the entropy of the species distribution to predict the most suitable habitats for the species under different environmental conditions.

The environmental dataset was processed through bilinear-based interpolation and harmonized to a consistent 250 × 250-meter grid resolution (Xu et al., 2024). To evaluate the combined effects of environmental changes and human activities on the cotton cultivation potential in Xinjiang, three predictive models were developed: Model A: Simulates the potential distribution of cotton under natural conditions using environmental variables representing the current climate, including bioclimatic, soil, topographic, and solar radiation data. Model B: Expands upon Model A by incorporating human footprint data to evaluate the role of human activities in shaping cotton habitat suitability. Model C: Evaluates how climate change may affect cotton distribution by utilizing environmental variables under future climate scenarios, including bioclimatic, soil, topographic, and solar radiation data.

To optimize model parameters and mitigate overfitting, 75% of the data were used for training and 25% for testing. The ENMeval package was employed to optimize the regularization multiplier (RM, ranging from 0.5 to 4 in increments of 0.5) and feature combinations (FC, consisting of six types: L, LQ, H, LQH, LQHP, and LQHPT) in the MaxEnt model. A total of 48 parameter combinations (8 regularization multipliers × 6 feature combinations) were systematically evaluated, and model complexity was assessed (Kass et al., 2021; Huang et al., 2024). The final selection of parameter combinations, based on a balance between model complexity and predictive performance, prioritized lower complexity and stronger ecological interpretability, thus ensuring a balance between predictive accuracy and the risk of overfitting.

The influence of each environmental factor on model predictions was evaluated through the jackknife method, with response curves visualized to demonstrate their impact on cotton suitability. The suitability threshold was determined based on expert knowledge and the Maximum Training Sensitivity plus Specificity (MTSS) criterion within the MaxEnt model. Suitability levels were classified into four categories: unsuitable (<MTSS), low suitability (MTSS–0.3), moderate suitability (0.3–0.4), and high suitability (0.4–0.8) (Shabani et al., 2018; Wang et al., 2019). To ensure comparability across different models and climate scenarios, all predictions were based on the same threshold classification. Comparing Model A and Model B reveals the impact of human activities on cotton suitability, whereas comparing Model A and Model C highlights the potential effects of climate change on cotton distribution and agricultural potential, thus guiding the development of adaptive agricultural management strategies.

This study utilized ROC curves and the Area Under the Curve (AUC) metric to evaluate model performance, where AUC values range from 0 to 1. AUC values closer to 1 indicate higher predictive accuracy. According to AUC grading standards, model performance ranges from poor (0 < AUC ≤ 0.6) to excellent (0.9 < AUC ≤ 1) (Zhao et al., 2021). The True Skill Statistic (TSS) was introduced as a supplementary evaluation metric. TSS is calculated by summing sensitivity and specificity and subtracting 1, with values ranging from −1 to 1 (Allouche et al., 2006). Based on evaluation standards, TSS scores are categorized as poor (−1 to −0.4), fair (0.4–0.7), very good (0.7–0.85), and perfect (0.9–1) (Préau et al., 2018). By integrating AUC and TSS, the models’ performance in predicting suitable cotton habitats in Xinjiang was assessed comprehensively. AUC emphasizes overall predictive capacity, whereas TSS refines the accuracy in distinguishing species presence and absence. Together, these two metrics offer a more comprehensive framework for model evaluation (Kabir et al., 2017).

In this study, simulation prediction results were evaluated using a combination of AUC and TSS metrics. Under the MaxEnt model, the average training AUC was 0.947, the average testing AUC was 0.941, and the average TSS was 0.840, indicating a high level of predictive accuracy and model reliability (Table 3). When the regularization multiplier (RM) was set to 1 and the feature combination (FC) to LQHPT, the MaxEnt model achieved optimal performance. In this configuration, the training AUC reached 0.946, the testing AUC reached 0.943, and the TSS was 0.837 (Table 3). Under environmental conditions (Model A), the variables influencing cotton distribution were: slope (36.2%), DEM (15.5%), bio2 (13.5%), bio11 (7.4%), solar radiation (6.2%), t_CaCO3 (5.1%), and t_sand (4.5%), with a cumulative contribution rate of 88.4%. Under human activity influences (Model B), the variables influencing cotton distribution were: HF (43.7%), slope (27.5%), DEM (9.8%), bio2 (3.7%), and bio19 (3.3%), with a cumulative contribution rate of 88%. As the human footprint increased, environmental variables such as slope, DEM, and bio2, each initially contributing more than 6%, exhibited decreases compared to Model A (with reductions of 8.7, 5.7, 9.8, and 4.7%, respectively). Conversely, in Model B, the contribution rate of bio19 increased by 0.7% (Figure 3).

The MaxEnt model was used to simulate cotton suitability habitats under scenarios both with and without human activity interference, followed by classification analysis and area calculations for each suitability level (Table 4). Under the current climate model, when only environmental variables were considered, the total suitable habitat area for cotton in Xinjiang was 7.97 × 10⁴ km². Of this, 4.56 × 10⁴ km² was classified as highly suitable, 1.84 × 10⁴ km² as moderately suitable, and 1.57 × 10⁴ km² as poorly suitable. The suitable habitats were primarily distributed in regions such as Aksu, Kashgar, Tacheng, Bayingolin Mongol Autonomous Prefecture, and Changji (Figure 4A). When human activity interference was considered, the total suitable habitat area expanded to 9.50 × 10⁴ km², representing a 19% increase compared to the scenario based solely on environmental variables. The area corresponding to each suitability category also increased (Figure 4B). Specifically, the highly suitable habitat area expanded by 0.26 × 10⁴ km², with new regions primarily located in northern Kashgar, northern Bayingolin Mongol Autonomous Prefecture, southern Tacheng, and eastern and western Aksu. The moderately suitable habitat area increased by 0.64 × 10⁴ km², with new regions mainly in southern Tacheng, eastern and western Aksu, northern Kashgar, and northern Bayingolin Mongol Autonomous Prefecture. The poorly suitable habitat area increased by 0.63 × 10⁴ km², primarily concentrated in northern Bayingolin Mongol Autonomous Prefecture, with scattered patches in Changji.

Predicted suitable habitat ranges for cotton show significant variation across different future climate scenarios (Figure 5). Under the SSP126 scenario, suitable cotton habitats are predominantly concentrated in the southern part of Tacheng, northern Kashgar, central and western Shihezi, Kuitun, and northern Bayingolin Mongol Autonomous Prefecture. Cotton habitats are projected to contract primarily in Bayingolin Mongol Autonomous Prefecture and Aksu (Figure 6). Under the SSP245 scenario, the contraction trend is most pronounced during the 2030s. Suitable habitats are concentrated in the southern part of Tacheng, central and western Shihezi, Kuitun, and northern Bayingolin Mongol Autonomous Prefecture. By the 2050s, the cotton distribution pattern remains largely stable, with suitable habitats primarily located in the northern part of Kashgar, southern Tacheng, northern Bayingolin Mongol Autonomous Prefecture, western Shihezi, and Kuitun, with minor contraction observed. Under the SSP585 scenario, distribution patterns in the 2050s and 2070s remain similar to current patterns, whereas the 2030s exhibit a notably greater contraction trend. Suitable habitats are projected to be predominantly located in northern Kashgar, central and western Shihezi, northern Bayingolin Mongol Autonomous Prefecture, southern Tacheng, and Kuitun.

The predicted suitable habitat area for cotton shows substantial differences compared to the current climate scenario (Figure 7). Under the current climate scenario, the total suitable habitat area for cotton is 7.97 × 10⁴ km², accounting for 5.1% of Xinjiang’s total area. Under the SSP126 scenario, the suitable habitat area is projected to range between 2.37 × 10⁴ km² and 3.37 × 10⁴ km², representing 1.52 to 2.15% of Xinjiang’s area, with the highly suitable habitat area peaking at 0.57 × 10⁴ km² in the 2030s, accounting for 17% of the total suitable area. In the SSP245 scenario, the suitable habitat area is projected to range between 2.06 × 10⁴ km² and 8.09 × 10⁴ km², corresponding to 1.32 to 5.17% of the total area. By the 2050s, the suitable habitat area is expected to exceed the current extent, with highly suitable habitats reaching a maximum of 3.35 × 10⁴ km², representing 41% of the total. Under the SSP585 scenario, suitable habitat areas are expected to range between 2.12 × 10⁴ km² and 6.96 × 10⁴ km², accounting for 1.36 to 4.45% of Xinjiang’s total area, with the highly suitable habitat area peaking at 2.23 × 10⁴ km², or 32% of the total, in the 2070s. Area percentage analysis indicates that under the SSP126 scenario, cotton’s suitable habitat continuously contracts over time (Figure 8), while under the SSP245 scenario, the habitat follows a contraction–expansion–contraction pattern. In contrast, under the SSP585 scenario, an initial contraction is followed by expansion. Overall, changes are most pronounced under the SSP245 scenario, followed by SSP585, with relatively stable conditions observed under SSP126.

The “Centroid Shift (Line)” tool in SDMToolbox was employed to analyze the changes in the centroid of suitable cotton habitats under three Shared Socioeconomic Pathways (SSP126, SSP245, and SSP585) for the 2030s, 2050s, and 2070s, relative to the current centroid. These changes include shifts in latitude, longitude, and elevation (Figure 9). The current centroid of suitable cotton habitats in Xinjiang is situated in the northwest of Bayingolin Mongol Autonomous Prefecture (42.57°N, 83.86°E), at an elevation of 4,082 m. In the SSP126 scenario, the centroid of cotton distribution shifts northeastward along a horizontal gradient. In the SSP245 and SSP585 scenarios, the centroid shifts southwestward along a horizontal gradient, with a vertical shift toward lower altitudes.

The centroid shifts under different climate scenarios are summarized as follows: Under the SSP126 scenario, the centroid in the 2030s is located at 41.69°N latitude, 81.26°E longitude, and 1,518 meters elevation; by the 2050s, it shifts to 42.58°N latitude, 82.75°E longitude, and 3,087 meters elevation; and by the 2070s, it moves to 43.05°N latitude, 83.48°E longitude, and 3,451 meters elevation. Under the SSP245 scenario, the centroid in the 2030s is located at 43.89°N latitude, 84.89°E longitude, and 3,060 meters elevation; by the 2050s, it shifts to 41.84°N latitude, 82.60°E longitude, and 1,520 meters elevation; and by the 2070s, it moves to 42.29°N latitude, 82.40°E longitude, and 2,306 meters elevation. Under the SSP585 scenario, the centroid in the 2030s is located at 41.79°N latitude, 81.98°E longitude, and 1,231 meters elevation; by the 2050s, it moves to 41.51°N latitude, 81.53°E longitude, and 1,379 meters elevation; and by the 2070s, the centroid is located at 42.10°N latitude, 83.13°E longitude, and 1,620 meters elevation.

Optimal MaxEnt model predictions reveal that, under the combined influences of climate change and human activities, the primary factors influencing cotton growth in Xinjiang are the Human Footprint (HF), solar radiation, the mean diurnal range (bio2), the mean temperature of the coldest quarter (bio11), and the precipitation of the coldest quarter (bio19). The response curves for each environmental variable demonstrate the range of cotton’s adaptability to these factors. At the Maximum Training Sensitivity Specificity (MTSS) threshold, the probability response curves reveal the relationship between cotton occurrence probability and the environmental variables. Specifically, the suitable range for human footprint is between 5.82 and 41.80, solar radiation ranges from 14,866.70 to 16,138.84 KJ, the optimal range for bio2 is from 8.68 to 11.99°C and 12.77 to 15.29°C, for bio11, it is from −13.59 to −9.66°C and −5.46 to 0.45°C, and for bio19, it is from 2.41 to 23.98 mm (Figure 10).

Cotton cultivation in Xinjiang is predominantly concentrated in oasis plain regions, characterized by flat terrain and controllable water resources, making them typical artificial agricultural ecosystems. The simulation results of this study indicate that the inclusion of the human footprint (HF) variable significantly increased the area of suitable cotton habitat, with an approximate 19% expansion compared to the no-disturbance scenario. This finding quantitatively highlights the positive influence of human activities on shaping the spatial distribution of cotton, particularly in regions where natural climatic resources are limited. Through the construction of irrigation systems, improvements in farmland infrastructure, soil preparation, and fertilizer management, human interventions have effectively enhanced the ecological suitability for cotton cultivation (Duan et al., 2021; Onuegbu et al., 2024; Zhang et al., 2024). For example, certain areas in the lower reaches of the Tarim River were not suitable for cotton cultivation under natural climatic conditions; however, following the establishment of canal irrigation systems, habitat suitability for cotton significantly improved. These findings align with previous research highlighting the reliance of agricultural expansion in arid regions on artificial irrigation (Abou Zaki et al., 2022; Ahmed et al., 2023).

Previous studies have highlighted the crucial role of solar radiation in the growth and development of cotton (Pinnamaneni et al., 2022). While various experimental methods have explored its relationship with cotton growth, a systematic quantitative analysis of the role of solar radiation in regional suitability assessments remains lacking (Lin et al., 2023). This study utilizes the MaxEnt model to quantitatively characterize the impact of solar radiation on the distribution of suitable areas for cotton. Furthermore, adequate precipitation contributes to increased soil moisture, thereby providing a favorable growth environment for cotton. However, excessive precipitation may impair cotton’s photosynthetic capacity, hinder growth, and ultimately affect its final yield (Jans et al., 2021). In this study, we found that the suitable range for cotton in relation to bio2 is between 8.68°C and 15.29°C, effectively supporting its growth. Additionally, cotton is particularly sensitive to low temperatures, and the suitable range for bio11 is from −13.59°C to −9.66°C and from −5.46°C to 0.45°C, which helps prevent frost damage and supports regenerative growth. Similarly, bio19, with a range of 2.41 mm to 23.98 mm, ensures the accumulation of soil moisture, satisfying cotton’s drought tolerance requirements. These findings are consistent with previous research, where temperature and precipitation are key drivers of cotton growth, with similar trends in suitable ranges and influence mechanisms observed across different regions (Chen et al., 2019; Sun et al., 2024; Khan et al., 2025).

Future climate change scenarios suggest substantial uncertainty regarding the trends in suitable habitats for cotton cultivation in Xinjiang. Compared to the current distribution, the total area of suitable habitats is projected to decline under both the SSP126 and SSP585 scenarios. Rising temperatures, changes in precipitation patterns, and an increased frequency of drought events are expected to reduce suitable planting areas under both low-emission (SSP126) and high-emission (SSP585) scenarios, consistent with previous studies on the negative impacts of climate change on crop production in arid regions (Feng et al., 2024; Nazarova et al., 2025). Temperature plays a crucial role in cotton growth, and yields decline significantly when temperatures exceed 32°C. If global temperatures rise by 1.5–2.0°C, cotton yields could decrease by up to 40% by 2,100 (Schlenker and Roberts, 2009). Historical evidence also indicates that cotton yields in the southwestern United States have declined by approximately 26% due to heat stress (Elias et al., 2018). In the low desert region of Arizona, cotton seed yields are projected to decrease by 40% by mid-century (2036–2065) and by 51% by the end of the century (2066–2095), compared to the baseline period of 1980–2005 (Ayankojo et al., 2020). These findings are broadly consistent with the trends of habitat suitability changes for cotton cultivation identified in this study.

In contrast, under the SSP245 scenario, predictions for the 2050s indicate an expansion trend, whereas the 2030s and 2070s are projected to experience continued contraction. Under the moderate emission scenario (SSP245), moderate warming, combined with improved precipitation patterns and optimized irrigation systems, is likely to provide relatively favorable growing conditions for cotton. This result further emphasizes the key role of human activity management in shaping cotton suitability. Additionally, advancements in cotton adaptive breeding technologies, such as the promotion of drought-resistant varieties, could further facilitate the expansion of suitable habitats (Wang et al., 2023). Relevant studies suggest that with appropriate climate adaptation strategies, crops like cotton can adjust to environmental changes induced by climate change, expanding their suitable habitats even under rising temperatures and shifting precipitation patterns (Li et al., 2021).

This study shows that under the SSP126, SSP245, and SSP585 climate scenarios, the center of suitable cotton habitats in Xinjiang generally migrates from higher to lower altitudes. This shift is mainly driven by a combination of rising temperatures, changing precipitation patterns, and human water resource management (Wang et al., 2021; Zhou et al., 2022). As temperatures rise and precipitation patterns shift, droughts intensify, and water shortages become more severe in high-altitude regions, thus restricting cotton growth. In contrast, low-altitude areas, with richer water resources and well-developed irrigation infrastructure, are becoming increasingly suitable for cotton cultivation (Li et al., 2020a). Recently, Xinjiang has strengthened water security for agriculture in low-altitude regions by promoting efficient water-saving irrigation technologies and implementing integrated regional water resource management strategies (Liang et al., 2019). Therefore, future efforts to address the spatial reorganization of cotton cultivation areas under climate change should integrate water resource management with agricultural adaptation strategies.

This study utilized the MaxEnt model to assess the impacts of climate change and human activities on the suitability of cotton habitats in Xinjiang. However, several limitations should be acknowledged. First, the temporal and spatial scales considered in this study were relatively limited; future research should consider incorporating longer temporal spans and a broader range of climate scenarios to enhance the robustness and comprehensiveness of projections. Second, although the Human Footprint (HF) index was used to represent human activities, it primarily reflects pressures such as population density and infrastructure development and does not directly account for agricultural irrigation systems or water resource availability. Given the significant reliance of cotton cultivation in Xinjiang on irrigation, future studies should consider integrating irrigation-related datasets or conducting sensitivity analyses to more accurately assess potential impacts. Furthermore, this study did not fully consider factors such as pest and disease outbreaks, genetic modification, and policy interventions. Future research should aim to include a broader set of driving forces, thus enabling the development of more targeted adaptation and management strategies to support the sustainable development of the cotton industry in Xinjiang (Khan et al., 2023; Farooq et al., 2024; Zafar et al., 2024).