Remote Sensing-Based Monitoring of Cotton Growth and Its Response to Meteorological Factors

Abstract

Cotton is an important economic crop and strategic resource. Monitoring its growth and analysing its response to meteorological factors are crucial for field management and yield estimation. This study selects the primary cotton-producing regions in northern Xinjiang as the study area. Firstly, using the Google Earth Engine cloud platform, the Cotton Mapping Index (CMI) was utilised to extract cotton planting areas from 2019 to 2023. Secondly, Sentinel-2A data were used to calculate the NDVI of cotton during the growing season and analyse its variation characteristics. Finally, correlation, lag, and partial correlation analyses were conducted between cotton NDVI and meteorological factors, including effective accumulated temperature, wind speed, precipitation, and solar shortwave radiation, to explore the response relationship. The results indicate the following: (1) The optimal classification threshold of CMI in the study area was determined to be 0.74, which was applied to extract cotton planting areas over the years. The overall classification accuracy achieved was 84.85%. The R² value for the cotton area extracted by CMI compared to the cotton planting area in the statistical yearbook data is 0.98, with an average relative error of 16.84%. CMI’s classification use effectively distinguishes cotton from other major crops, such as wheat and corn, in the study area. Compared with different classification methods, CMI is more convenient and efficient for extracting cotton planting areas, contributing significantly to yield estimation and management. (2) We found that from 2019 to 2023, some fields were planted with cotton yearly. In order to prevent land degradation, a crop rotation system should be implemented, in which cotton rotates with other crops to reduce the rate of soil nutrient loss and achieve sustainable agricultural development. (3) NDVI can effectively monitor the spatiotemporal changes and regional variations in cotton growth. Sentinel-2 multi-spectral imagery possesses high spatial and temporal resolution, enabling effective monitoring of cotton growth, provision of cotton growth data for field managers, and application in cotton production management. Additionally, cotton yield estimation can be achieved by comparing the overall growth of cotton across different years. (4) Cotton NDVI exhibits a strong correlation with effective accumulated temperature and solar radiation, with the majority passing the significance test, suggesting a significant promotion effect on cotton growth by accumulated temperature and solar radiation. In cotton cultivation management, attention should be directed toward monitoring changes in accumulated temperature and solar radiation. Moreover, NDVI changes in response to solar radiation exhibit a certain lag. The correlation between NDVI and precipitation is low, likely attributed to local cotton cultivation primarily relying on drip irrigation. Cotton NDVI is negatively correlated with wind speed. Cotton planting should consider weather changes and take corresponding preventive management measures. The research results have significant reference value for monitoring cotton growth, disaster prevention, and sustainable agricultural development.

1. Introduction

Global climate change presents one of the most intricate challenges confronting humanity in the 21st century [1]. Agriculture is one of the industries most directly impacted by climate change [2,3,4,5]. Meteorological factors such as temperature, precipitation, and solar radiation significantly affect agricultural production [6,7,8,9]. Plants’ physiological and biochemical changes occur within optimal temperature ranges; deviation from the optimal temperature range can significantly impact plant growth [10,11]. Accumulated temperature is an indicator that investigates the relationship between temperature and cotton’s growth and development rate. Different crops have varying requirements for accumulated temperature, and the suitability of accumulated temperature directly impacts crops’ growth status and yield [12,13]. In most cases, precipitation supplies the necessary moisture for vegetation growth [14]. Wind affects plants in various ways, directly or indirectly influencing their growth and development [15]. Solar radiation’s intensity and spectral composition influence plant growth and development [16,17].

Cotton, the second-largest crop after grains, is a vital economic commodity and strategic resource, holding a prominent position in the national economy [18,19,20,21,22]. The cotton industry encompasses various stages, including cultivation, harvesting, processing, textile production, and exportation, generating employment opportunities directly or indirectly [23]. It has played a pivotal role in driving economic development [24,25]. As a weaving material, cotton can also have its seeds pressed for oil extraction, thus finding applications in both industrial and agricultural sectors. The unique natural conditions of Xinjiang are conducive to cotton cultivation [26]. In recent years, influenced by mechanisation and solar thermal resources, the focus of cotton cultivation in China has steadily shifted towards the Xinjiang cotton region [27]. Cotton production in Xinjiang is high-quality and abundant, accounting for up to 80% of China’s capacity and one-fifth of the world’s total [28].

Remote sensing technology enables rapid and comprehensive monitoring of crop information [29,30,31,32,33,34]. The application of remote sensing technology in agriculture has advanced agricultural development [35,36,37]. Scholars have found that vegetation indices effectively reflect the growth status of vegetation and can be utilised for various purposes, including crop growth monitoring, yield estimation, crop classification, and pest and disease control [38,39,40,41]. Moreover, analysing temporal vegetation indices alongside meteorological factors can further elucidate the relationship between vegetation growth and meteorological conditions [42]. Rui Sun et al. [43] analysed the response of NDVI variations in the northern grasslands of China to meteorological factors and found that the dominant meteorological driving factors varied across different regions. Qingzhi Zhao et al. [42] employed Pearson correlation coefficient analysis and Mann–Kendall trend analysis to study the interaction between long-time series vegetation indices and meteorological factors. They discovered varying lag effects of temperature and solar radiation on vegetation indices. Abbasali Vali et al. [44] explored the correlation between topographical and climatic factors with vegetation indices in the Kharestan region of Iran. Kurt Heil et al. [45] studied the influence of multiple meteorological factors on wheat yield. Holzman [46] utilised NDVI, land surface temperature, precipitation, and soil moisture data to establish yield-prediction models for corn and soybeans in Iowa. Monitoring cotton growth and analysing its response to meteorological factors can provide data support for cotton growth monitoring and the advancement of smart agriculture in cotton fields.

The Cotton Mapping Index (CMI) is a threshold-based method that directly extracts cotton planting areas without requiring training samples. It distinguishes cotton from other crops based on the spectral reflectance, spectral angles, and backscattering coefficients during the cotton growing season. Its application is crucial in cotton-producing countries such as China, India, Pakistan, and Brazil. In the research area of the United States, the overall accuracy of cotton planting areas extracted using CMI can reach up to 89.75%. In Shandong, China, the R² between cotton planting areas extracted by CMI and statistical data is 0.89. In India, the R² for cotton planting areas extracted using CMI reaches 0.9, with a mean absolute error (MAE) of 9.5 kha. In the research area of Pakistan, the R² between the extracted cotton area and the statistical area is 0.6, with an MAE of 4.60 kha [47].

The main contents of this study include the following three points: (1) extraction of cotton growing areas from 2019 to 2023 using CMI; (2) the acquisition of cotton growth information via the utilisation of NDVI; and (3) response of cotton growth to meteorological factors in Xinjiang. This study can provide data support for agricultural cotton growth monitoring and significantly promote the advancement of smart agriculture in cotton fields.

2. Materials and Methods

2.1. Study Area

The study area is in the northern part of Xinjiang, China, at the northern foot of the Tianshan Mountains (Figure 1). It encompasses the primary cotton planting regions in the north of Xinjiang, with geographical coordinates ranging from 80.66° to 87.60° E and from 43.06° to 45.71° N. It comprises Changji City, Hutubi County, Jinghe County, and adjacent areas. The northern boundary of the study area borders the Junggar Basin, while the southern boundary is adjacent to the Tianshan Mountains. The climate is a temperate continental dry climate with large evaporation, little precipitation, cold and dry winter, and hot and dry summer [48,49]. Cotton in northern Xinjiang is typically sown in mid to late April, with seedlings emerging in early to mid-May. It generally progresses into budding from late May to mid to late June. The period from flowering to boll formation is known as cotton’s flowering and boll period, typically occurring from late July to late August. The end of cotton boll formation marks the bolling period, generally from early September to mid to late October, lasting about 60 days.

2.2. Data Sources

This study conducted field surveys on cotton growth in the research area during June, July, and August 2023. The surveys involved collecting planting information for farmland and cotton growth details. The crops surveyed on farmland included cotton, corn, wheat, sunflower, grapes, peppers, melons, and others. A total of 3360 verification samples were collected.

The Google Earth Engine platform facilitates the acquisition, processing, visualisation, and analysis of remote sensing data at various spatial and temporal scales [50,51]. Additionally, it encompasses diverse data types, including meteorological data, DEM data, and more.

The Sentinel-1A and Sentinel-1B satellites carry a dual-polarisation SAR instrument at the C-band, which can capture images with a spatial resolution of 5–40 m and a revisit period of 12 days. Four imaging modes, which include Strip Map (SM), Interferometric Wide-swath (IW), Extra Wide-swath (EW), and Wave (WV), are provided [52].

The Sentinel-2 Multi-Spectral Instrument (MSI), onboard the Sentinel-2A and 2B satellites, is developed by the European Space Agency (ESA) and can capture images with a spatial resolution of 10–60 m and a revisit period of 5 days. It provides 13 spectral bands ranging from visible to shortwave infrared spectrum [53].

ERA5-Land is a reanalysis dataset that provides daily meteorological data from 1950 to the present at a higher resolution (0.1°) than ERA5 [54]. This dataset includes various meteorological variables.

The statistical yearbook data are sourced from the website of the Xinjiang Bureau of Statistics (https://tjj.xinjiang.gov.cn/ (accessed on 10 August 2023)) and the Statistical Bureau of the Xinjiang Production and Construction Corps (http://tjj.xjbt.gov.cn/ (accessed on 10 August 2023)). The data processing flow is shown in Figure 2.

2.3. Cotton Mapping Index

The classification method based on indices has the advantages of simplicity and efficiency and has been used to identify water [55], vegetation [56], plastic greenhouses [57], plastic materials [58], etc. Classifier-based cotton classification methods [59,60] may need to be improved due to the limitations of field-measured samples, and most applications are limited to the identification and classification of small-scale crops in specific areas.

The Cotton Mapping Index (CMI) is a method proposed by Lan Xun et al. for distinguishing between cotton- and non-cotton planting areas, providing a simple way to automatically identify cotton planting areas without the need for training samples. It distinguishes cotton from other crops based on the spectral reflectance, spectral angles, and backscattering coefficients during the cotton growing season. Compared to traditional supervised classifiers, CMI performs similarly without a training dataset. Compared to the supervised classifiers, the proposed CMI is convenient and performs well even without training samples in identifying the cotton-cultivated area. However, the classification results of the experimental area in China have yet to be calibrated with on-site samples. Therefore, this study attempts to use field-measured sample points to calibrate the optimal CMI threshold, further validating the feasibility of CMI, facilitating the extraction of cotton planting areas on a large scale, and benefiting related fields such as cotton management and yield estimation. CMI requires the combination of Sentinel-1 SAR and Sentinel-2 images for computation. This study utilises the Google Earth Engine platform for computation. The calculation formula is as follows:

$$CMI=\frac{{\rho }_{RE1}^{PG}\times {\rho }_{RE2}^{PG}\times \theta }{{\alpha }_R^{PG}\times abs({\displaystyle {\sum }_{\mathrm{t}=PG-2}^{PG}{\beta }_{VV}^t})}$$

(1)

In the formula, $\rho$ represents spectral reflectance and $\alpha$ represents a spectral angle, $\beta$ represents the backscattering coefficient, and $\theta$ set to 1000.t represents the synthesis period of the image. $PG$ represents the peak period of NDVI. ${\rho }_{RE1}^{PG}$ and ${\rho }_{RE2}^{PG}$ representing the spectral reflectance of red-edge 1 and red-edge 2 at the peak of NDVI. Abs represents the absolute value.

2.4. Normalised Difference Vegetation Index

Normalised Difference Vegetation Index ($NDVI$) is an important parameter reflecting crops’ growth status and nutritional information [61]. $NDVI$ is widely used in studies such as crop growth monitoring, yield prediction, and health assessment [62,63]. Its formula is as follows:

$$NDVI=\frac{(B_8-B_4)}{(B_8+B_4)}$$

(2)

In the formula, $NDVI$ represents the Normalised Difference Vegetation Index, where $B_8$ denotes the Sentinel-2 Band 8, the near-infrared band, and $B_4$ denotes the Sentinel-2 Band 4, the red band. NDVI ranges from −1 to 1.

2.5. Correlation Analysis

Correlation analysis assesses the degree of association between variables by quantifying it with correlation coefficients [64,65]. Its calculation formula is as follows:

$$r_{xy}=\frac{{\displaystyle \sum _{i=1}^n(x_i-\overline{x})(y_i-\overline{y})}}{\sqrt{{\displaystyle \sum _{i=1}^n{(x_i-\overline{x})}^2{\displaystyle \sum _{i=1}^n{(y_i-\overline{y})}^2}}}}$$

(3)

In the formula, $r_{xy}$ represents the correlation coefficient between variables $y$ and $x$. $n$ is the sample size, and $\overline{x}$ and $\overline{y}$ are the means of the two variables. If $r_{xy}$ is greater than 0, it indicates a positive correlation between the two variables, and if it is less than 0, it indicates a negative correlation. Where $x$ represents cotton NDVI and $y$ represents accumulated temperature, precipitation, solar radiation, and wind speed, respectively. The t-test method is used for significance testing, and its calculation formula is as follows:

$$t=\frac{r\sqrt{n-m-1}}{\sqrt{1-r^2}}$$

(4)

In the formula, $r$ represents the correlation coefficient, $n$ represents the sample size, and m represents the number of independent variables.

2.6. Classification Accuracy Verification

User accuracy (UA) in image classification is the probability that a certain type of pixel is correctly classified into that class, calculated using the rows of the confusion matrix. Producer’s accuracy (PA) refers to the ratio of pixels correctly classified by the classifier as class A (diagonal value) to the sum of class A columns in the real reference total of class A mixed slip array. Overall accuracy (OA) equals the total number of correctly classified pixels divided by the total number of pixels. The number of pixels correctly classified is distributed diagonally along the confusion matrix, and the total number of pixels is equal to the total number of pixels of all real reference sources. The F1 score is an index used in statistics to measure the accuracy of a binary classification model. It considers both the accuracy rate of the classification model and the recall rate. The F1 score can be seen as a harmonic average of model accuracy and recall, with a maximum value of 1 and a minimum value of 0.

2.7. Data Preprocessing

The Cotton Mapping Index was used to extract cotton planting areas over the years using the Google Earth Engine platform. Subsequently, during the cotton growing season (April–October), Sentinel-2 multi-spectral images were used to calculate the vegetation index, and image synthesis was generated every half month (15 days). All data were resampled to 10 m to maintain a uniform data resolution.

This study selected four meteorological factors: temperature, precipitation, incident ground short-wave radiation, and wind speed. Effective accumulated temperature refers to the cumulative temperature difference between the ambient temperature and cotton’s biological lower limit temperature, representing the heat required for cotton growth and development. The study considers a temperature greater than 10 °C the effective accumulated temperature. Precipitation is used to calculate the accumulated precipitation every 15 days. The average daily solar radiation for every 15 days is calculated using the incoming short-wave radiation from the ground. Wind speed is used to calculate the average wind speed every 15 days.

3. Results

3.1. Extraction of Cotton Planting Areas Based on CMI

Firstly, Sentinel-1 SAR and Sentinel-2 images from the 2023 cotton growing season are selected through the GEE cloud platform. Through image selection, filtering, clipping, and other preprocessing steps, remote-sensing images are synthesised into composites at a 15-day interval. Agricultural fields are extracted by masking using land use data. Then, the Cotton Mapping Index (CMI) of farming fields in the study area is calculated using Formula (1) of the Cotton Mapping Index at the peak NDVI period.

Next, field-measured data were used as verification points to calculate the classification accuracy of cotton under different CMI thresholds and correct the optimal CMI classification thresholds, as shown in Figure 3A. Subsequently, the optimal CMI classification threshold was verified using the cotton planting area data from the statistical yearbooks of counties and cities in 2019 and 2020. It is found that when the CMI threshold is 0.74, the overall classification accuracy of cotton can reach 84.85%, and the f1 score is 0.846. In addition, using the threshold value of 0.74, the R² value of the cotton planted area extracted by CMI and the cotton planted area in the statistical yearbooks of different counties and cities in the study area in 2019 and 2020 reached 0.98. The mean absolute error is 8.65 kha, and the mean relative error is 16.84%, as shown in Figure 3B.

Finally, the optimal CMI classification threshold was determined to be 0.74, and it was used to extract cotton planting areas in historical years. Using the same method as the first step, the CMI values of farmland in different years were calculated. The CMI threshold was set to 0.74 and classified areas greater than or equal to this threshold as cotton fields, otherwise classified as non-cotton fields. Finally, the annual cotton planting area from 2019 to 2023 was obtained. To intuitively understand the cotton planting situation, the results of five years of cotton planting in the study area were superimposed (Figure 4). In Figure 4, the numbers from 0 to 5 represent the years cotton was planted, with 0 indicating that cotton was never planted and 5 indicating that cotton was planted in all five years.

Studies show that in some areas, the desire to plant cotton is so strong that cotton has been planted every year for five years. These places are concentrated in the south of Jinghe, the south of Wusu, the north of Shawan, and the east of Manas. There are also areas where cotton has not been grown in five years, which may be related to the local climatic conditions. It is mainly concentrated in the west of Bole, the south of Wusu, the south of Shawan, Hutubi, and the south of Changji.

3.2. Extracting Cotton Growth Information

NDVI is a commonly used index in crop growth monitoring. NDVI will show regular changes during crop growth. When crops are affected by water shortage, pests and diseases, and meteorological disasters, NDVI will show different performance. At the same time, the growth of cotton in different regions is different due to various factors. Through the difference in NDVI, the growth difference in different areas can be compared. By comparing the difference in NDVI in different years simultaneously, the cotton growth in different years can be compared, and the reasons can be analysed.

The GEE platform was used to calculate NDVI, and the images were synthesised every 15 days from April to October every year, with 12 images synthesised every year. No mapping was performed since there was no significant change in cotton NDVI on remote sensing images in April. The average NDVI value of the cotton growth period from May to October 2019 to 2023 in the study area is shown in Figure 5.

Statistics of cotton NDVI in each image were conducted to obtain the average NDVI in different periods, which more intuitively reflected the time change in cotton NDVI (as shown in Figure 6). From early April to mid-April, when cotton planting began, NDVI stayed mostly the same, possibly due to the small scale of cotton seedlings, which made it difficult to detect changes in remote sensing images [66]. Only in early May 2020, a significant change in cotton NDVI was observed. Cotton NDVI increased slightly in late May compared with early May. From late May to late June, cotton NDVI increased rapidly, corresponding to the budding stage of cotton [67]. The period from germination to cotton flowering is called the budding period, during which cotton experiences vegetative and reproductive growth, mainly reproductive growth. During this period, cotton roots expanded rapidly, the absorption capacity increased, the leaf area index increased rapidly, and the increase in NDVI reflected the enhancement of photosynthetic productivity and the rapid accumulation of dry matter. From the beginning of July to the end of August, cotton NDVI stabilised at about 0.8, corresponding to cotton’s flowering and bolling period [68]. The period from flowering to the opening of the bell is called the flowering period of cotton. During this period, cotton plants gradually transition from balanced growth of nutrition and reproduction to mainly reproductive growth. The reproductive growth is accelerated, the flowering is increased, and the defoliation rate is generally low, which is the key period to determine the yield and quality of cotton. From early September to late October, cotton NDVI decreased gradually under the influence of farmers’ use of defoliant and the date of flopping.

3.3. The Relationship between Cotton NDVI and Meteorological Factors

3.3.1. Meteorological Factors during the Cotton Growing Season

The meteorological data were resampled to a 10 m resolution and clipped using the cotton planting area. The meteorological data statistics for each 15 days are shown in Figure 7. The spatial distribution of meteorological data for each year, either averaged or accumulated, is illustrated in Figure 8.

Figure 7 shows that the accumulated temperature in 2020 was significantly higher than other years at the beginning of the growing season, while in 2023, they were notably lower compared to other years. This is similar to the observed NDVI changes in cotton for April and May. Regarding precipitation, significantly lower precipitation was observed in April and May 2020 compared to other years, while in the same period, higher precipitation was recorded in 2019 and 2023. The peak average wind speed during the cotton growing season was concentrated from late May to early June. Surface incoming shortwave radiation increased from early April and gradually decreased by the end of July.

The spatial distribution of meteorological factors in Figure 8 reveals that areas closer to mountains had significantly lower accumulated temperatures than plains. The 15-day accumulated precipitation was highest in mountainous areas, with less precipitation in plain areas. Areas with higher wind speeds were mainly concentrated in the plains.

3.3.2. Correlation Analysis between Cotton NDVI and Concurrent Meteorological Factors

Based on previous analysis, the variation in cotton NDVI during the growing season mainly manifests as rapid growth from May to August, followed by a decrease in NDVI due to factors such as spraying defoliants and phenological stages. In this study, when conducting the correlation analysis between cotton NDVI and meteorological factors, only the period from May to August, when cotton is in the seedling, budding, and pre-flowering stages, with NDVI showing growth and no leaf shedding, was selected to avoid errors in data analysis.

Using MATLAB 2020b, the cotton NDVI and raster data of various meteorological factors were subjected to correlation analysis, with the results shown in Figure 9 and Figure 10. Wind speed negatively correlates with cotton NDVI, with most passing the significance test at 0.05. Accumulated temperature is positively correlated with cotton NDVI (p < 0.05). The correlation between cotton NDVI and precipitation is mainly positive in the eastern and northwestern parts of the study area but did not pass the significance test. In the central-eastern part of the study area, it is negatively correlated and passed the significance test. Cotton NDVI is mainly negatively correlated with solar shortwave radiation (p < 0.05), concentrated primarily on the central part of the study area.

Using the average values of cotton NDVI raster images and meteorological factor raster images, correlation analysis was conducted using IBM SPSS Statistics 27, with the results shown in Figure 11, where “Wind” represents wind speed, “ATemp” represents accumulated temperature, “Pre” represents precipitation, and “Radiation” represents solar radiation. The correlation coefficient between NDVI and wind speed is −0.04, and between NDVI and accumulated temperature is 0.72, both passing the significance test at the 0.01 level, consistent with the results of raster correlation analysis. The correlation coefficient between NDVI and precipitation is 0.03, while between NDVI and radiation is 0.21, failing to pass the significance test.

3.3.3. Correlation Analysis between Cotton NDVI and Meteorological Factors Lagged by Half a Month

The correlation analysis between cotton NDVI and meteorological factors with a lag of half a month is shown in Figure 12 and Figure 13. The correlation between cotton NDVI and lagged half-month accumulated temperature is mainly positive, with most correlations (p < 0.05). The correlation between NDVI and precipitation shows negative correlations in the northeast of Shawan, the north of Manas, and the south of Jinghe. In contrast, positive correlations are observed in other areas, with significance only in Wusu and the west of Shawan. The correlation between NDVI and solar radiation is mainly positive across the study area. The correlation between NDVI and lagged half-month wind speed shows negative correlations in regions passing the significance test. In contrast, most areas showing positive correlations do not pass the significance test.

Using the cotton NDVI image and lagged half-month meteorological factor raster images to calculate the average values and conduct correlation analysis, the results are shown in Figure 14. The correlation coefficient between NDVI and accumulated temperature is 0.87 (p < 0.05), consistent with the raster results. The correlation coefficient between NDVI and precipitation is 0.03. The correlation coefficient between NDVI and radiation is 0.38 (p < 0.05). The correlation coefficient between NDVI and wind speed is 0.02, not passing the significance test.

3.3.4. Correlation Analysis between Cotton NDVI and Meteorological Factors Lagged by One Month

The correlation analysis results of cotton NDVI and meteorological factors with a lag of one month are shown in Figure 15 and Figure 16. Cotton NDVI exhibits mainly positive correlation with lagged one-month accumulated temperature, and most of them pass the significance test at the 0.05 level. NDVI shows a negative correlation with precipitation in the southern part of Wusu and the eastern part of Shawan and a positive correlation in the southern part of Changji. All of these correlations pass the significance test. For most areas, the correlation between NDVI and lagged one-month solar radiation is positive and significant at 0.05. NDVI shows a positive correlation with lagged one-month wind speed in the southern part of Jinghe, which passes the significance test, as do the correlations in other regions.

A correlation analysis was conducted using the cotton NDVI image and raster grid images of lagged one-month meteorological factors to compute the mean values. The results are shown in Figure 17. The correlation coefficient between NDVI and accumulated temperature is 0.82 (p < 0.05), consistent with the raster results. The correlation coefficient between NDVI and precipitation is 0.04, not passing the significance test. The correlation coefficient between NDVI and radiation is 0.66 (p < 0.05). The correlation coefficient between NDVI and wind speed is 0.31 (p < 0.05).

3.3.5. Cotton NDVI Partial Correlation Analysis with Meteorological Factors

Partial correlation analysis of cotton NDVI with meteorological factors at the same time lagged half a month and one month, the results are shown in

Table 1. The third-order partial correlation coefficients and p-values of cotton NDVI with meteorological factors at different time periods.

		Accumulated Temperature	Precipitation	Wind Speed	Solar Radiation
Concurrent	r	0.93	0.39	−0.49	0.91
	p-value	<0.01	0.01	<0.01	<0.01
Lagged by half a month	r	0.81	0.17	−0.21	−0.06
	p-value	<0.01	0.24	0.15	0.65
Lagged by one month	r	0.88	0.43	−0.43	0.59
	p-value	<0.01	<0.01	<0.01	<0.01

. In controlling for precipitation, wind speed, and solar radiation, the partial correlation coefficient between temperature and NDVI exceeds 0.8 (p < 0.01); in the partial correlation analysis of concurrent meteorological factors, the partial correlation coefficient between temperature and NDVI is 0.93; in the partial correlation analysis of meteorological factors lagged half a month, the partial correlation coefficient between temperature and NDVI is 0.81; in the partial correlation analysis of meteorological factors lagged one month, the partial correlation coefficient between temperature and NDVI is 0.88.

In controlling for temperature, wind speed, and solar radiation, the partial correlation coefficient between precipitation and NDVI is mostly greater than 0.35, and all pass the significance test at the 0.05 level; among them, in the partial correlation analysis of concurrent meteorological factors, the partial correlation coefficient between precipitation and NDVI is 0.39; in the partial correlation analysis of meteorological factors lagged half a month, the partial correlation coefficient between precipitation and NDVI is 0.17, which does not pass the significance test; in the partial correlation analysis of meteorological factors lagged one month, the partial correlation coefficient between precipitation and NDVI is 0.43. In controlling for precipitation, temperature, and solar radiation, the partial correlation coefficient between wind speed and NDVI shows a negative correlation; in the partial correlation analysis of concurrent meteorological factors, the partial correlation coefficient between wind speed and NDVI is −0.49; in the partial correlation analysis of meteorological factors lagged half a month, the partial correlation coefficient between wind speed and NDVI is −0.21; in the partial correlation analysis of meteorological factors lagged one month, the partial correlation coefficient between wind speed and NDVI is −0.43. In controlling for temperature, wind speed, and precipitation, the partial correlation coefficients between solar radiation and NDVI, EVI mostly show positive correlations; among them, in the partial correlation analysis of concurrent meteorological factors, the partial correlation coefficient between solar radiation and NDVI is 0.91; in the partial correlation analysis of meteorological factors lagged one month, the partial correlation coefficient between solar radiation and NDVI is 0.59. In the partial correlation analysis, cotton NDVI shows a mainly positive correlation with temperature and solar radiation, a weak correlation with precipitation, and a negative correlation with wind speed.

4. Discussion

4.1. Cotton Environmental Characteristics

The adverse weather conditions during the cotton sowing and emergence period in 2023 significantly negatively impacted the emergence and growth of cotton in the study area. Field surveys revealed that local cotton farmers had to repeat sowing multiple times, undoubtedly increasing the cost of cotton cultivation and affecting the growth and development of cotton plants.

During field surveys, it was observed that cotton plants around reservoirs exhibited significantly better growth conditions than other areas; insufficient water supply will affect cotton’s normal growth and development [69]. For instance, in June field surveys, cotton plants around reservoirs were found to have a plant height of around 16 cm, notably higher than the plant height of cotton in other areas, which ranged from 8 to 13 cm. In field investigations, it was also observed that water accumulation occurred in uneven, low-lying areas, making cotton more susceptible to poor growth.

Hail often has a significant impact on agriculture [70,71]. When cotton is hit by hail, reduced leaf area and decreased photosynthesis affect the quantity and quality of cotton bolls, ultimately decreasing cotton yield [72]. In late June 2023, Anjihai Town of Shawan County suffered a hail disaster, and some cotton fields were seriously damaged. In hard-hit areas, farmers have no choice but to give up growing cotton and switch to crops such as corn to mitigate their losses.

4.2. Remote Sensing-Based Cotton Classification

In recent years, the research of cotton classification mainly relies on supervised classification, which needs to select the best feature from various features and classify according to the classifier and training samples. Hao Fei et al. [60] extracted cotton at the county scale using multiple features, such as spectral and vegetation indices, employing the random forest method. Haolu Li et al. [73] used convolutional neural network models combined with measured sample points to identify cotton fields in the Weiku Oasis. These classification methods rely on selecting classifiers and the reliability and quantity of training samples. However, they mainly focus on small areas, and further training or processing is required for large-scale, cotton planting areas.

The classification method based on classifiers and training samples still requires further validation and verification to extract cotton planting areas over large regions and time series. The Cotton Monitoring Index (CMI) is a threshold-based method that does not require complex feature selection and model parameter tuning, making it more convenient and efficient than classifier-based classification methods. CMI can be used to extract cotton planting areas of large areas over many years, which is of great significance for agricultural research and management. This method achieved good results in extracting historical cotton planting areas and can be used for crop studies over time series. Compared to supervised classification methods, the threshold-based CMI method is more efficient and convenient for extracting cotton planting areas.

4.3. The Relationship between Cotton Growth and Meteorological Factors

Studies have shown a significant positive linear correlation between cotton NDVI and accumulated temperature, with a correlation coefficient of 0.76 (p < 0.05). This indicates that the accumulated temperature plays a significant role in cotton growth. As shown in Figure 18, the accumulated temperature in April and early May 2023 was lower than in 2019–2022, also reflected in the cotton growth. The NDVI of cotton in May 2023 was lower than 2019–2022. Accumulated temperature is, therefore, an important factor influencing cotton growth and development.

4.4. Uncertainty Analysis

In determining the CMI optimal classification threshold by field test, it is found that wheat, corn, and other common crops are easily distinguishable from cotton. However, as the variety of crops has diversified, a misclassification has emerged between cotton and other crops such as sunflowers, grapes, and peppers. The authors suggest that in future studies, CMI can be supplemented and refined by combining various spectral characteristics of crops with larger planting areas to improve the precision of cotton extraction further. In terms of accuracy verification, only the cotton planting area of each county and city in the statistical yearbook of 2019 and 2020 was obtained for verification.

Human factors also play a significant role in cotton growth. Farmers’ experience and cultivation practices can also influence cotton growth. The next step of the work should focus on field experiments to further explore the effects of various meteorological factors and human factors on cotton growth and development.

5. Conclusions

This study used the GEE platform and Cotton Mapping Index (CMI) to extract the cotton annual planting area in the study area. NDVI was used to extract cotton growth information and analyse its response to meteorological factors. Specific conclusions are as follows:

When the optimal classification threshold for CMI is set to 0.74 in the study area, it can effectively extract the annual cotton planting areas with an overall accuracy of 84.85%. The correlation coefficient (R²) between the cotton planting areas extracted using CMI and the cotton sowing area in the statistical yearbook reaches 0.98, with an average absolute error of 8.65 kha and an average relative error of 16.84%. A fixed threshold for CMI can be used to extract annual cotton planting areas.

Cotton NDVI is positively correlated with effective accumulated temperature and shortwave radiation, with most correlations significant at the 0.05 significance level. The correlation between cotton NDVI and precipitation is relatively low, possibly related to local cotton planting conditions. Solar radiation has a lag effect on the change of cotton NDVI. In the partial correlation analysis, cotton NDVI negatively correlates with wind speed. In cotton planting management, attention should be paid to the changes in accumulated temperature and solar radiation.

In extracting cotton acreage over the past years, we found that some fields planted cotton every year in 2019–2023. Due to the deep root system of cotton plants, the absorption of nutrients increases, coupled with a large amount of irrigation water; the loss of nutrients in the soil is obvious, resulting in a gradual decline in soil quality. In addition, this monoculture model is prone to diseases and pests, resulting in a heavy reliance on cotton pesticides, resulting in many lands severely polluted by long-term heavy use of pesticides. Therefore, to prevent land degradation, a crop rotation system should be adopted, where cotton is alternated with other crops, thereby reducing the rate of soil nutrient loss and achieving sustainable agricultural development.