Sugarcane Water Requirement and Yield Projections in Major Producing Regions of China Under Future Climate Scenarios

7 Relative soil moisture is of great significance to the growth and yield of sugarcane. 8 In this study, we use the relative soil moisture from the China Meteorological 9 Administration Land Data Assimilation System (CLDAS) to dynamically evaluate the 10 water requirement of sugarcane and its growth adaptability at different growth stages. 11 Based on the data of relative soil moisture, air temperature, precipitation and soil 12 temperature, a sugarcane yield model is established to analyze the projected change 13 trends of sugarcane yield in China from 2020 to 2100 under three future scenarios. 14 Analysis results show that sugarcane requires more water during the elongation stage 15 but less water at the ripening stage. The relative soil moisture from the CLDAS can be 16 used to calculate the proportion of the daily suitable area to the total planting area. The 17 combining of relative soil moisture data and water requirement indicators can better 18 characterize the water requirement during sugarcane growth. Suitable relative soil 19 moisture during the tillering and elongation stages is the most critical factor that directly 20 affects the sugarcane yield. From 2020 to 2100, sugarcane yield will increase first and 21 then decrease sharply. The increase in emissions can lead to an apparent downward 22 trend in sugarcane yield. Based on the CLDAS data and water requirement indicators, 23 a new method for monitoring the sugarcane growth throughout the growth period is 24 proposed in this study. In the SSP370 and SSP460 scenarios, the sugarcane yield 25 showed a downward trend, and there were mutations in 2064 and 2052, respectively. 26 After the mutation, the yield decline trend was more obvious. Under the SSP585 27 scenario model, the sugarcane production showed an upward trend from 2022 to 2033, 28 and a downward trend after 2033, and a mutation occurred in 2051. After the mutation,


Introduction
Sugarcane is an important raw material for sugar production, and the bagasse can also be used to produce energy such as alcohol (Christofoletti et al. 2013;Jaiswal et al. 2017).China is the third-largest sugar-producing country after Brazil and India, where the sugar production reached 2.2319 million tons in 2019, which acts as an essential part of the agricultural trade (Zu et al. 2018).In China, sugarcane is mainly cultivated in Guangxi Zhuang Autonomous Region, Guangdong Province, Yunnan Province and Hainan Province.The gross product of the sugarcane sugar industry is 6.86 billion dollar, with the farmer income being 5.08 billion dollar, which is an essential source of income for farmers (Li and Yang 2015).Therefore, the forecasting of sugarcane yield and its change trend plays a vital role in the formulation of policies by relevant departments (Verma et al. 2021;Wang et al. 2017).
The China Meteorological Administration (CMA) Land Data Assimilation System (CLDAS) can provide a land surface dataset (available online at http://data.cma.cn/search/uSearch.html?keywords=CLDAS) with high spatio-temporal resolutions (Xie et al. 2017).Another datasets of the same type are the Global Land Data Assimilation System (GLDAS) dataset and North American Land Data Assimilation project (NLDAS), which are also widely used in agricultural land drought studies and crop yield studies: Fang(2021) studied the Soil Water Deficit Index (SWDI) and Soil Moisture Deficit Index (SMDI) in spring and summer out of Australia using Soil Moisture Active Passive (SMAP) soil moisture (SM), GLDAS long-term SM and soil attribute products; Mokhtari (2018) input GLDAS data set and leaf area index data as driving factors into the Soil Water Atmosphere Plant (SWAP) model to predict wheat yield, the experimental results show that the accuracy of SWAP model is improved after combining GLDAS dataset.Xia (2014) used the NLDAS dataset to calculate drought indices for each region of the U.S. and to reconstruct typical drought events in U.S. history.CLDAS, NLDAS and GLDAS are all data sets generated by terrestrial assimilation systems.The CLDAS dataset, NLDAS dataset, and GLDAS dataset cover China, North America, and the world, respectively.For the study of the Chinese region, the CLDAS dataset has higher accuracy than the GLDAS dataset (Han et al. 2020;Sun et al. 2020), and this paper will be based on the CLDAS data.
Sugarcane-related researches mainly focus on remote sensing-based planting area extraction and growth monitoring and yield prediction, among which the research on sugarcane planting area extraction is relatively mature at present (Aguiar et al. 2011;Wang et al. 2019).There are also related studies on sugarcane yield prediction, but most of these studies are based on satellite remote sensing supplemented by crop models to experiment with sugarcane yield prediction. Rampazo and Núria (2021) combined the Moderate-Resolution Imaging Spectroradiometer (MODIS) images and the Simple Algorithm for Retrieving Evapotranspiration model to analyze sugarcane growth situation in southern Brazil.However, timely monitoring of soil water deficits cannot be realized by MODIS because of its long production cycle (8 or 16 days).Based on Landsat images, Almeida (2006) studied the spectral characteristics of sugarcane at different growth stages to estimate sugarcane yield.However, satellites like Landsat are susceptible to cloud cover (Dong and Menzel 2016;Foga et al. 2017).With the rapid development of artificial intelligence technology in recent years, some scientists have successively introduced machine learning technology to sugarcane yield forecasting.(Fernandes et al. 2017) obtained the NDVI index through the MODIS sensor, and used the NDVI index combined with the ANN neural network to evaluate the sugarcane yield status. (Xu et al. 2020) used UAV-LIDAR data to simulate sugarcane yield in Chongzuo City, Guangxi Province based on the random forest algorithm.The results show that the random forest algorithm is more effective than the traditional linear regression, and the fitting accuracy is higher.Neither the analysis on sugarcane yield at LIDAR or stations nor the research on different growth stages of sugarcane by satellite remote sensing can meet the requirements of large scale, high spatio-temporal resolutions and strong interference resistance.
The CLDAS can overcome the influence of cloud cover on the monitoring of surface meteorological elements (Chen and Yuan 2020), which has good adaptability to soil moisture monitoring (Long et al. 2019).The dataset has the advantage of short update period and high accuracy for short-term weather condition monitoring, and has achieved well research results in soil moisture monitoring (Suon et al. 2019;Wang and Yu 2021;Yu et al. 2019) and regionalization of crop growth adaptability (Rongsheng et al. 2020;Rongsheng et al. 2021).
The purpose of this study is to set the adaptable indicators for soil water requirement at different sugarcane growth stages according to the relative soil moisture, realize daily growth dynamic monitoring based on high-resolution data from CLDAS, and use CMIP6-related data combined with random forest algorithm to determine sugarcane yield under different scenarios and analyze the trend of the future sugarcane yield.

Data description
The data used in this study include the basic geographic information data, CLDAS version 2.0 data, information from statistical yearbooks and outputs of the Coupled Model Intercomparison Project Phase 6 (CMIP6) models under future scenarios.
Specifically, the basic geographic information data include three administrative boundaries at provincial, municipal and county levels and 1-km digital elevation model data.
The dataset is developed by combining satellite observation data and soil observation data, and is developed using techniques such as multi-grid variational assimilation, optimal interpolation, probability density function matching, physical inversion, and terrain correction.It has extremely high accuracy in China.Based on the integration of multiple land surface models, the CLDAS version 2.0 dataset used in this study include relative soil moisture, maximum temperature, average temperature, average wind speed, soil temperature and precipitation.
Through provincial and municipal statistical yearbooks, the sugarcane yields in main producing regions of China from 2017 to 2019 are obtained at provincial, municipal and county levels.Considering the regional applicability of the model (Zhu et al. 2020), the data selected for yield model construction and prediction include the soil moisture and maximum field capacity from the Canadian Earth System Model version 5 (CanESM5) (Sospedra-Alfonso et al. 2021) of CMIP6 during 2020-2100, the soil temperature and air temperature from the low resolution of climate model 6A of Institut Pierre-Simon Laplace (IPSL-CM6A-LR) model (Boucher et al. 2020), and the precipitation flux and 10-m wind speed from the version 2.1 of Goddard Institute for Space Studies (GISS-E2.1-G)model (Kelley et al. 2020;Nazarenko et al. 2022).
The SSP370 scenario represents the medium to high end of the range of future forcing pathways, the radiative forcing is 7.0 W/m 2 and the temperature increase is about 2.8°C by 2100 (Zhao et al. 2020).The SSP460 scenario represents the medium range of future forcing pathways, the radiative forcing is 6.0 W/m 2 and the temperature increase is about 1.8°C by 2100 (Pu et al. 2020).The SSP585 scenario represents the high range of future forcing pathways, the radiative forcing is 8.5 W/m2 and the temperature increase is about 3.2°C by 2100 (O Neill et al. 2017;O'Neill et al. 2016).

Study area
The study area covers four provinces, namely Guangxi Zhuang Autonomous Region, Guangdong Province, Yunnan Province and Hainan Province, where the annual yield of sugarcane accounts for more than 90% of the total sugarcane yield in China.
The elevation distribution map (Fig. 1) shows that the western part of the study area is relatively high, and the terrain gradually tends to flatten out from the west to the east.
The western part is Yunnan Province, located to the southeast of the Hengduan Mountains, which is an essential part of the Yunnan-Guizhou Plateau.The central part is Guangxi Zhuang Autonomous Region, which is mostly hilly.The eastern part is Guangdong Province, located in the Pearl River Delta region, with numerous alluvial plains.While the southern part is Hainan Province, whose terrain is low around and high in the middle.It can be seen that the major sugarcane producing areas in China belong to the subtropical monsoon climate zone, where the rainy and high-temperature seasons coincide, with the annual sunshine hours being 1000-3000 hours and the annual precipitation being 900-2600 mm (Guga et al. 2021).

Sugarcane water requirement
Water requirement during the crop growth period is one of the critical factors that determine the crop yield, and thus a reasonable evaluation of the soil moisture content throughout the growth period plays a vital role in estimating sugarcane yield.Based on previous studies on sugarcane water requirements (Guozhang 1993;Zhaomin 2019), this study summarizes the previous studies on water requirements of sugarcane to classify the sugarcane growth adaptability.The values of relative soil moisture corresponding to different sugarcane growth stages are shown in Table 1, which can be divided into three grades of most adaptable, adaptableand unadaptable.The sugarcane growth period is divided into four stages, namely germination-seedling, tillering, elongation and ripening stages.The soil depth suitable for sugarcane growth varies at different growth stages, which is relatively shallow in the germination-seedling stage and relatively deep in the middle and late stages due to the more extended root system.The periods and days corresponding to different growth stages are shown in Table 2.The entire growth period of sugarcane lasts 313 days, with 79 days for the germination-seedling stage, 21 days for the tillering stage, 182 days for the elongation stage (the longest) and 31 days for the ripening stage (Zhaomin 2019).Note that the growth periods and days are obtained based on the regional average, while the actural growth periods and days vary due to different producing regions, years and sugarcane varieties.

Data quality verification
In this study, three indicators of absolute error (Gao 2021), relative error (Mohammadi et al. 2015) and root-mean-square error (Wessel et al. 2018) are adopted to perform quality verification analysis.The absolute error measures the difference between the fitted and actual values, whose expression is given below (Eq.2).
( ) where MAE indicates the absolute error;  , the fitted sugarcane yield from pixel to pixel under different scenarios;  , the pixel-by-pixel value of the actual yield data; i and j the row and column of the current pixel; m and n denote the maximum numbers of rows and columns.
The mean relative errors (MRE) measures the confidence level of the fitted value, which can be expressed as follows (Eq.3).
The root-mean-square error (RMSE) is adopted to measure the deviation between fitted value and actual value, and it can be expressed as follows (Eq.4).

Mann-Kendall test
The Mann-Kendall test has been widely applied in the analysis of abrupt climate changes in fields including meteorology, climatology, hydrology (Gocic and Trajkovic 2013;Wang 2020).In this study, the Mann-Kendall test is applied to analyze the abrupt changes in the long-term time series of sugarcane yield based on the simulations from 2020 to 2100.The most distinguishing feature of this method during a non-parametric test is that the test samples do not have to follow a specific distribution, and this method is independent of a few outliers.The UFK curve greater than 0 indicates an upward trend for the time series, while the curve less than 0 indicates a downward trend.When the curve exceeds the threshold (α=0.05), it indicates that the upward or downward trend is significant, and the part that exceeds α is the time range for abrupt change.
UFK is a standard normal distribution, which is a series of statistics calculated in the order of time series x, UBK=-UFK.The intersection point between UFK and UBK curves is the abrupt change time (Wei 2007).In addition, the separate MK trend analysis does not take into account the seasonal cycle changes, and cannot take into account the impact of the previous time period on the current time period.Therefore, the Correlated Seasonal MK Test is carried out on the basis of the MK trend analysis to compare whether the results of the MK trend analysis and the Correlated Seasonal MK Test are consistent (Yue and Wang 2004;Yue and Wang 2002).

Technical process
In this study, the relative soil moisture data in the CLDAS dataset were used to delineate the different adaptability of sugarcane in conjunction with the optimum relative soil moisture indicators for sugarcane at different fertility stages in Table 1.

Spatial distributions of soil moisture based on the CLDAS data
Requirements of soil water vary at different sugarcane growth stages, i.e., more water at the elongation stage and less water at the ripening stage.Based on the CLDAS data, the daily relative soil moisture at different soil depths that vary at different growth stages are obtained, of which the spatial distributions are shown in Fig. 4. Figure 4a shows the spatial distribution of relative soil moisture on April 1, 2019, when the sugarcane is at the germination-seedling stage.The result indicates that the relative soil moisture is relatively low in the Hengduan Mountains in northwestern Yunnan Province, which is not conducive to sugarcane growth.It is between 60% and 80% in the eastern part of Yunnan Province bordering Guangxi, which is suitable for sugarcane growth.In east Guangxi and Guangdong, the relative soil moisture is generally high (above 80%), which is prone to cause root rot of sugarcane seedlings, thus leading to yield reduction.
Generally, the relative soil moisture gradually increases from the west to the east.
The spatial distribution on May 20, 2019 during the sugarcane tillering stage (Fig. 4b) shows that the relative soil moisture is relatively low in Yunnan, unfavorable to sugarcane growth.Guangxi and southern Guangdong have relative soil moisture of 60%-90%, which is adaptable for sugarcane growth.Guangdong Province has low relative soil moisture that is not conducive to sugarcane growth.In general, the relative soil moisture is low in the west and high in the east at the tillering stage.The spatial distribution of relative soil moisture on July 1, 2019 during the elongation stage (Fig. 4c) demonstrates that the soil moisture is relatively high in the southeast of Guangxi and the southern and northern regions of Guangdong.In northwestern Yunnan, the relative soil moisture is relatively low.While the junction of Yunnan and Guangxi as well as the northeastern part of Guangxi have relative soil moisture of 60%-90%, which is adaptable for sugarcane growth.Figure 4d illustrates the spatial distribution of relative soil moisture on December 1, 2019 during the ripening stage.
The result shows that the relative soil moisture in Hainan Province is relatively high, which is not conducive to sugar accumulation; while in eastern Yunnan, central Guangxi and southern Guangdong, the relative soil moisture is between 40% and 60%, adaptable for sugar accumulation.

Adaptability for sugarcane growth based on the CLDAS data
Based on the relative soil moisture from the CLDAS version 2.0 data and by referrring to the adaptable indicators of soil water requirement at different sugarcane growth stages (Table 1), the sugarcane growth adaptabilities on the above four representative dates are obtained, of which the spatial distributions are shown in Fig. 5.
Figure 5a shows the distribution of sugarcane growth adaptability on April 1, 2019.
Since the sugarcane water requirement is not high at the germination-seedling stage, eastern Guangxi and Guangdong with high soil moisture are unadaptable for sugarcane growth, and even more, excessive soil moisture can inhibit sugarcane growth.Figure 5b presents the spatial distribution of sugarcane growth adaptability on May 20, 2019.
Since the sugarcane water requirement gradually increases at the tillering stage, the regions with low relative soil moisture are no longer adaptable for sugarcane growth, and the unsuitable areas are mainly located in the Hengduan Mountains of western Yunnan and northern Guangdong.Figure 5c is the same as Fig. 5b, but for July 1, 2019 at the elongation stage, which is a critical stage for sugarcane growth, and the water requirement reaches the highest at this stage.Excessively low relative soil moisture can inhibit sugarcane growth.Therefore, the unsuitable regions are mainly concentrated in west Yunnan and southeast Guangxi.

Analysis of relative soil moisture in typical regions
In the main sugarcane producing areas of China, four typical regions of Fusui, Lincang, Xingbin and Danzhou are selected to analyze the changes of relative soil moisture over time and the corresponding sugarcane growth adaptability (Fig. 6).
As shown in Fig. 6a, the relative soil moisture can meet sugarcane growth needs at the germination-seedling stage in Fusui, and it is generally adaptable for sugarcane growth at the tillering and elongation stages.While at the ripening stage, the relative soil moisture is relatively high in Fusui, which is not conducive to sugar accumulation.
Overall, the relative soil moisture in Fusui County basically meets sugarcane water requirement during the entire growth period.The sugarcane yield per unit area in Fusui was relatively high in 2019, reaching 87.1 t•ha −1 .
In Lincang (Fig. 6b), the relative soil moisture generally meets the optimum demand of sugarcane growth at the seedling-germination stage.It is relatively low at the early and late elongation stage, while high at the ripening stage, which is not conducive to sugar accumulation.The sugarcane yield per unit area in Lincang was 63.2 t•ha −1 in 2019, basically the same as that in 2018 (63.0 t•ha −1 ).
In Danzhou (Fig. 6c), the relative soil moisture is abnormally high at the germination-seedling stage.It does not meet the optimum sugarcane growth conditions in the tillering stage except a few days.At the elongation stage, the relative soil moisture fluctuates at the upper limit of the optimum relative soil moisture, mostly higher than the optimum.Abnormally high relative soil moisture at the ripening stage can easily lead to sugarcane root rotting and death, thereby reducing the yield of sugarcane.In 2019, the relative soil moisture in Danzhou did not meet the water requirement of sugarcane growth, resulting in the sugarcane yield per unit area being only 55.8 t•ha −1 .
For Xingbin (Fig. 6d), the relative soil moisture is abnormally high at the germination-seedling stage, which can easily cause seedling death and leading to yield reduction.At the early elongation stage, the relative soil moisture is greater than the optimum soil moisture; while in the middle and late stages, however, the relative soil moisture is adaptable for sugarcane growth.Losses caused by unexpected deaths at the sugarcane seedling stage can be saved by measures such as timely replanting.Therefore,

Validation of sugarcane yield model and importance analysis of factors
As mentioned in section 2.2.2, 10% of the sample data are left to verify the performance of sugarcane yield model.Based on the three evaluation indicators, the quality evaluation results of the fitting yields are shown in Table 3.It can be seen that the RMSE and MRE of the fitted sugarcane yields under the SSP370 scenario are smaller than those under the other two scenarios, while the MAE of the yield is smaller under the SSP585 scenario than those under the other two scenarios.The overall performance reaches the best under the SSP370 scenario.The comprehensive error of the SSP370 scenario mode is the smallest, so taking the scenario as an example, the spatial distributions of relative errors for sugarcane

Projected sugarcane yields under future scenarios
Figure 8 shows the results from the Mann-Kendall test for sugarcane yields under different scenarios.Under the SSP370 scenario (Fig. 8a), the UFK curve is firstly in the positive-value zone, indicating an upward trend of sugarcane yield.Then, the UFK curve is in the negative-value zone, suggesting that the sugarcane yield has a downward trend.From 2020 to 2100, the sugarcane yield shows a trend of first increasing and then decreasing, where an abrupt change appears in 2064.Figure 8b shows that the UFK curve is all in the negative-value zone under the SSP460 scenario, indicating that sugarcane yield has been in a decreasing trend, where an abrupt change appears in 2052.

Discussion
Based on the relative soil moisture data from the CLDAS, this study determines the water requirement indicators for sugarcane growth, realizes dynamic monitoring of the daily growth and conducts sugarcane yield forecasts and trend analysis under future scenarios by CMIP6 models.
In this study, the sugarcane adaptable indicators is mainly based on relative soil moisture.As indicated earlier, the relative soil moisture is essential for sugarcane growth.At the early stage of sugarcane growth, the water requirement is relatively low-excessively more water can cause the root rot and death of seedlings, while excessively less water will inhibit the sugarcane growth.In the middle stage of growth, the soil moisture affects the growth and thickening of sugarcane as the water requirement is high-excessively more water can cause the sugarcane leaves to turn yellow, worse growth, rotten roots and even death, while excessively little water will slow down the sugarcane elongation rate (Guozhang 1993), and even more, the stems and leaves may dry out to death.In the later stage of sugarcane growth, the water requirement is relatively low, and soil moisture affects the sugar accumulation.
Excessively less water can slow down the sugar accumulation rate, while excessively more water can lead to sugarcane re-growth.
However, there are many other factors affecting the sugarcane growth, such as high temperature, low temperature and other extreme weather disasters (Verma et al. 2019).If these refined data can be obtained, a composite indicators influencing the sugarcane growth can be proposed, and thus the monitoring and analysis of sugarcane growth will be more accurate.In addition, the training data acquired at two levels (municipal and county levels) in constructing the sugarcane yield model are not fine enough, where the data from Yunnan is only at the municipal level, and the data for Guangxi, Guangdong and Hainan are at the county level.In the following work, we should collect more refined yield data, which plays a vital role in yield retrieval by remote sensing data.For example, we can collect yield data on specific sugarcaneproducing regions through field surveys.
Combined with weather forecast services, the sugarcane water requirement evaluation based on the CLDAS data conducted in this study can provide daily water requirement of sugarcane planting fields in the next few days for farmers and sugarcane planting companies, aiming to timely supplement soil water for sugarcane growth.
Transition from Regional Competition (SSP3) to Traditional Fossil Fuel Combustion (SSP5), we will find that with the increase of shared socio-economic path ( SSPs ) scenarios, sugarcane yield under SSP3 socio-economic scenario decreased, and sugarcane yield under SSP5 socio-economic scenario increased between 2022 and 2033.
However, after 2050, the decline trend of sugarcane yield under the SSP5 socioeconomic scenario model was larger than that under the SSP3 socio-economic scenario model.Therefor, a country's strategy may have major impact on future sugarcane production.The main reasons are that under three future scenarios, air temperature and soil temperature continue to rise, and thus the relative soil moisture continues to decrease.Such meteorological conditions are not adaptable for sugarcane growth.Some effective measures that can be taken to improve the declining trend of sugarcane yield include cultivating new sugarcane varieties with higher yields, improving the ability to cope with climate change and prevent meteorological disasters, promoting the transformation of sugarcane planting from farmer-based to farm-based, and upgrading the modernization level of farming and field management techniques.

Conclusions
In this study, we use the CLDAS data to dynamically evaluate the sugarcane water requirement at different growth stages and forecast the sugarcane yields from 2020 to 2100 under three future scenarios.The results suggest that the relative soil moisture from the CLDAS dataset can effectively characterize the growth status of sugarcane and directly affect the final yield.
1) Sugarcane requires more water during the tillering and elongation stages, while less water at seedling-germination and ripening stages.
2) Relative soil moisture has a more significant impact on sugarcane yield during the tillering and elongation stages than during the seedling and ripening stages.
3) Under three future scenarios, sugarcane yield shows an overall decreasing trend during 2020-2100.The sugarcane yield decreases more obviously under the higher emission scenario of SSP585 compared with SSP370.
The evaluation of sugarcane growth adaptability in this study is mainly based on relative soil moisture.In the following work, composite meteorological and environmental indicators will be applied to investigate their adaptabilities for sugarcane growth, aiming to evaluate the sugarcane growth at different stages in a more comprehensive and fine way.
FIG. 3. Flow chart of sugarcane yield prediction based on random forest algorithm.

FIG. 4 .
FIG. 4. Spatial distributions of relative soil moisture in the main sugarcane producing regions on (a) April 1, 2019 during the seedling-germination stage, (b) May 20, 2019 during the tillering stage, (c) July 1, 2019 during the elongation stage and (d) December 1, 2019 during the ripening stage.
FIG. 5. Spatial distributions of adaptability in the main sugarcane producing regions on (a) April 1, 2019 during the seedling-germination stage, (b) May 20, 2019 during the tillering stage, (c) July 1, 2019 during the elongation stage and (d) December 1, 2019 during the ripening stage.(Based on the adaptability indicators in TABLE 1 and the relative soil moisture in FIG 4. ) FIG. 6.Comparison of relative soil moisture and water requirement during the sugarcane growth period in 2019 in four typical regions of (a) Fusui, (b) Lincang, (c) Danzhou and (d) Xingbin.
FIG. 7. Relative errors of sugarcane yields under the SSP370 scenario.The relative errors between simulated and actual sugarcane yields of CMIP6 for different years was calculated according to Eq 3.

TABLE 1 .
Adaptable indicators for soil water requirement at different sugarcane

TABLE 2 .
(Saeed et al. 2022)days of sugarcane at different growth stages.In addition to relative soil moisture, the growth of sugarcane is also closely related to meteorological conditions.Air temperature and relative soil moisture are important factors affecting the growth and development of sugarcane, and relative soil moisture is also affected by wind speed, soil temperature and precipitation(Saeed et al. 2022).Yield , represents the sugarcane yield on the grid ( ,  ), GST , the soil temperature, PRE , the precipitation, RSM , the relative soil moisture, TMP , the (Yuan and Hu 2021)Hu 2021).Its main idea is to draw n samples from the original training set with release, and the sample size of each sample is the same as the size of the original training set; then each sample is modeled as a decision tree separately, and n modeling results are obtained, and finally the average of each decision tree prediction result is used as the final prediction result(Rajković et al. 2022).Based on the random forest algorithm, the relative soil moisture data, air temperature data, soil temperature data, precipitation data, and wind speed data in the CLDAS dataset were used to train the random forest model by using the sugarcane yield in each region from 2017 to 2019.The relative soil moisture data, air temperature data, soil temperature data, precipitation data, and wind speed data from CMIP6 were then input into the random forest model to forecast sugarcane yields for 2020-2100 under different scenarios.A total of 277 areas with sugarcane yields have been collected in this study, of which 10% of the yield data are used for verification, while the rest are adopted as training samples.The constructed sugarcane yield model is as follows (Eq.daily average temperature, TMP_MAX , the maximum daily average temperature, WIN , the 10-m wind speed, and  the empirical coefficient.i and j refer to the row and column numbers of raster data, respectively.According to the above random forest model construction process, Fig 2 shows the flow chart of random forest model construction, code available online at: https://github.com/FunnyBiscuit613/random-forest.git.FIG. 2. Flow chart of sugarcane yield prediction based on random forest algorithm.

TABLE 3 .
Verifications of sugarcane yields simulated under different scenarios.

Table 4
is a comparison table of the correlation coefficient and slope between different meteorological elements and sugarcane yield.From the analysis in Table4, it can be seen that the relative soil moisture maintains a high correlation with sugarcane yield both in the correlation coefficient and the slope.Precipitation has the largest negative contribution to sugarcane yield among all variables

TABLE 4 .
Correlation coefficient and slope comparison table between different meteorological elements and sugarcane yield.

TABLE 5
Correlated Seasonal MK Test results in different scenarios.