Forest status assessment in China with SDG indicators based on high-resolution satellite data

ABSTRACT To assess the status and change trend of forest in China, an indicator framework was developed using SDG sub-indicators. In this paper, we propose an improved methodology and a set of workflows for calculating SDG indicators. The main modification include the use of moderate and high spatial resolution satellite data, as well as state-of-the-art machine learning techniques for forest cover classification and estimation of forest above-ground biomass (AGB). This research employs GF-1 and GF-2 data with enhanced texture information to map forest cover, while time series Landsat data is used to estimate forest AGB across the whole territory of China. The study calculate two SDG sub-indicators: SDG15.1.1 for forest area and SDG15.2.1 for sustainable forest management. The evaluation results showed that the total forest area in China was approximately 219 million hectares at the end of 2021, accounting for about 23.51% of the land area. The average annual forest AGB from 2015 to 2021 was estimated to be 105.01Mg/ha, and the overall trend of forest AGB change in China was positive, albeit with some spatial differences.


Introduction
Forests are essential carbon pools in terrestrial ecosystems and have a significant role in the global carbon cycle (Gao et al. 2022;Xu and Jiang 2015).Forest biomass is a crucial factor in estimationg forest carbon storage, and its quantity and spatial distribution are essential parameters for assessing the carbon sink potential of forest ecosystems (Fang et al. 2001;Rossi and Santos 2020).
The United Nations 2030 Agenda for Sustainable Development, adopted in 2015, covers 17 Sustainable Development Goals (SDGs) established by the United Nations to guide global development work from 2015 to 2030, following the expiration of the Millennium Development Goals (MDGs) between 2000 and 2015 (Guo 2020).Forests are of paramount importance to the earth's ecosystem and human development and are closely linked to SDG15 'Life on Land'.The sub-indicators of SDG 15 including SDG 15.1.1 and SDG 15.2.1 , which respectively refer to forest area and sustainable forest management.Governments worldwide have conducted extensive research on forest monitoring and assessment.The Forest Resource Assessment (FRA) of the United Nations' Food and Agriculture Organization (FAO) has carried out the first systematic estimates of global forest resources assessment since the 1970s (FAO 2010(FAO , 2015(FAO , 2020)).Hansen et al. (2008Hansen et al. ( , 2010Hansen et al. ( , 2013) ) have used forest loss to monitor changes in tropical forests and subsequently globally, with forest loss and gain data available since the twenty-first century.Few researchers, including Kussul et al. (2020), Ishtiaque et al. (2020), and Giuliani et al. (2020) have examined SDG15 on forest change detection at national scales.However, most forest-related research has focused solely on changes in their areal extent, which, while an important contribution, may not be adequate for properly tracking SDG 15.Therefore, more in-depth analysis of various aspects of forest health and ecosystems are required.
The government of China has made significant efforts to improve forest quality and to monitor changes in forest biomass.In 2018, the '13th Five-Year Plan for National Forestry Development' was launched to promote the sustainable development of forests and enhance their ecological, economic, and social benefits (Chinese Forestry and Grassland Administration 2018).To achieve these goals, the Chinese government has implemented various policies and measures, such as the Natural Forest Protection Program and the Ecological Forest Construction Project, which aim to increase the forest coverage, improve forest quality, and enhance forest carbon sinks.In addition, the Chinese government has also actively explored the use of remote sensing technology for forest monitoring and management.The annual report on remote sensing monitoring of global ecological environment, which uses satellite data to monitor the dynamics of global vegetation, has been an important tool for the Chinese government to assess the status and changes of forest resources in China and to guide its forestry development policies (Chinese Academy of Sciences 2019).
However, there is still room for improvement in China's forest management and monitoring.For instance, the current forest inventory mainly focuses on forest area and volume, while information on forest biomass and carbon storage is limited (Wang et al. 2020).Furthermore, the existing forest monitoring system is fragmented and lacks integration, which makes it difficult to obtain accurate and up-to-date information on forest resources.Therefore, it is necessary to strengthen the monitoring of forest biomass and carbon storage, and to improve the integration and coordination of forest monitoring systems in China.
In this study, we present a procedure for assessing the forest status in China using SDG indicators based on forest cover and forest AGB data.Firstly, we obtained the required data for calculating SDG 15.1.1 and SDG 15.2.1 .We proposed an improved methodology for calculating SDG indicators, which involves using moderate and high spatial resolution satellite data and stateof-the-art machine learning methodology for forest cover classification and forest AGB estimation.This methodology is expected to provide more accurate and reliable estimates of the SDG indicators.Additionally, we have utilized trend analysis to calculate SDG 15.2.1 , which enables us to better understand changes in forest management practices over time.Finally, we carried out a quantitative assessment of forest state in China using the proposed assessment method and presented the results.
Overall, this study offers a comprehensive approach for evaluating forest status and change trends in China, which can serve as a basis for formulation effective forest management policies and strategies.

Overview of China's geography and forest distribution
China, located in the eastern region of Asia and along the western coast of the Pacific Ocean, covers an area of approximately 9.6 million square kilometers.It extends 62 degrees from east to west, covering about 5,200 km, and nearly 50 degrees from north to south, spanning about 5,500 km.China currently holds the fifth-largest total forested area globally, with forest volume continuously expanding at an average annual rate of 2% (CMF 2014).Nevertheless, China's vast territory and broad geographical span result in complex and diverse terrain, causing the distribution of forest resources to be extensive and highly uneven.Forest density varies considerably, with dense forest in the east and sparse in the west, less in the north and more in the south (see Figure 1).Additionally, the implemention of afforestation policies in China has led to the area of planted forests exceeding 80 million hectares, accounting for over one-third of the total forest area.As a result, China holds the highest planted forests area globally.

Workflows for assessment SDG indicators in China
For a more accurate estimation of forest conditions and change trend, we proposed a workflow that calculate the specific SDGs indicators for forests, as depicted in Figure 2.

Forest field survey data
Forest field survey data are crucial for extrapolating forest classification and retrieving AGB from remotely sensed datasets.In this study, we collected 6667 plot measurements from projects: the Chinese Qinghai-Tibet Plateau Second Scientific Expedition Project and the National key area biomass survey, both of which were conducted in collaboration with the Research center for Eco-Environmental Sciences at the Chinese Academy of Sciences.The plot were rectangular and has a size of 1000 m 2 .Field survey data were collected from May to October in 2011-2012, which corresponds to the growth period of trees.These measurements were obtained from both plantations and natural forests located throughout China.The geolocation of each individual plot record, as well as its corresponding attributes (including stand origin, forest type, AGB value, collection time and location, measurement method, and landform), were included.The 6667 data points were used as training datasets, and their distribution is shown in Figure 1.

Forest cover extraction
In this study, the definition of forests used followed the Global Forest Resources Assessment (2020), which defines forests as 'Land spanning more than 0.5 hectares with trees higher than 5 m and a canopy cover of more than 10 percent, or trees able to reach these thresholds in situ.It does not include land that is predominantly under agricultural or urban land use' (FAO 2018, Terms and Definitions).
The satellite images used in this study were obtained from GF-1 and GF-6 satellites with a resolution of 16 m (data source: China Land Observation Satellite Data Center).The data were processed primarily in two ways.First, radiometric correction was performed, which was completed at the China Remote Sensing Satellite Ground Station (RSGS) using the method described in She et al. ( 2020) paper.Second, a time-series synthetic image of China that is consistent in space and spectrum was generated using a pixel-based time-series data mosaic method.Tao et al. (2018) demonstrated that the decision tree method based on forest region partition is superior to the non-partition decision tree method for extracting forest information from GF1 WFV images.The overall accuracy increased by 3.80-4.65%,and the Kappa coefficient improved by approximately 0.07-0.10.Based on these findings, the territory of China is divided into eight forest subregions with coherent space and consistent forest types, and high-quality forest sample points of subregions were obtained by using field survey forest data, enabling the establishment and implementation of a workflow for the rapid production of forest cover products.This workflow is based on our previous research conducted by Zhang et al. (2020) and Yantao Guo et al. (2022), and is illustrated in Figure 3.
To ensure the accuracy and completeness of forest cover product, a stratified random sampling method was employed to generate check points for validation.The selection of the number of check points was based on the principle that they should be distributed across all provinces of China, with the number of check points in each province determined by the proportion of forest area in the province.In provinces with a relatively small proportion of forest area, 100 verification points were collected uniformly, with 50 forest points and 50 non-forest points each.This approach was adopted to ensure that the validation process was comprehensive and representative.

Forest aboveground biomass estimation
Forest aboveground biomass (AGB) serves as an important indicator for assessing the amount of carbon stored in a forest above the soil, including stems, stumps, branches, bark, seeds, and foliage of trees (Ravindranath and Ostwald 2007).In this study, the estimation of forest AGB was conducted based on forest ecological zones as a unit, with sample data obtained through field survey data.An algorithm model for forest AGB and its input parameters were established on the Google Earth Engine cloud platform using situ remote sensing satellite data and the features closely related to forest biomass.The technical flowchart for the estimation and validation process is illustrated in Figure 4.
In previous studies, the selection of the best features set for estimating forest biomass was a critical step in model optimization (Belgiu and Drăguţ 2016).To improve the accuracy of the model, optional variables such as band spectral parameters, vegetation index, terrain and climate parameters were used in this study.A correlation analysis was conducted between the optional variables and the measured data of AGB using SPSS 19.0 software.Factors significantly related to biomass were used as independent variables of the prediction model, as shown in Table 1.
Random forest is a popular machine learning model that can be used not only for classification but also for regression analysis.Breiman (2001) proposed that RF has the ability to yield accurate estimation with abundant input features (Belgiu and Drăguţ 2016).Furthermore, the nonlinear characteristics of random forest make it more advantageous than linear algorithm.This is the main reason why we choose RF instead of traditional linear regression model to estimate forest biomass.
To ensure the stable results of RF model with different parameters, several parameters such as the maximum depth of trees (MDT), minimum number of samples to split an internal node (MSP), and minimum number of samples to be a leaf node (MSL) are critical for the overall accuracy of the RF model.The 4-folded cross-validation technique is applied.During the 4-folded crossvalidation process, three-quarters of samples are selected randomly and nonrepeatedly to train the RF model, while the remaining quarter is used to validate the overall accuracy of the model.For each set parameter, 10 times 4-folded cross-validation is implemented to get the average value of Root Mean Square Error (RMSE) and Coefficient of determination R 2 .
Here, AGB mean is the annual average value from 2015 to 2021; AGB i is the forest biomass in the ith year; n is the number of years involved in the calculation, where n = 7.The trend analysis method adopted here is SLOPE, which predicts the change trend through linear regression analysis of variables changing with time.The slope of interannual change of forest AGB can represent the forest change trend.The formula for calculating the inter-annual variation trend is as follows: Here, AGB SLOPE is the slope of the pixel regression equation; AGBi is the forest biomass in year i; n is the number of years involved in the calculation, where n = 7.When Slope > 0, it means that the AGB of the pixel is increasing; when Slope = 0, it means that the AGB of the pixel is unchanged; when Slope < 0, it means that the AGB of the pixel is decreasing.The Slope trend analysis method is more suitable for the processing of large amounts of data such as time series data, and the processing speed is relatively fast; but this method requires the trend to be linear, and it cannot be accurately fitted when the number fluctuates greatly.

Forest data and its validation result
The forest cover data and the forest AGB data are all published, which are respectively available on DOI: 10.12237/casearth.637eed1d819aec05c471007dand DOI: 10.12237/ casearth.637eed1d819aec05c471007e.
To insure the accuracy of forest cover data, 6897 check points were created using a stratified random sampling method and manual interpretation of high resolution satellite images, as described in Section 2.3.2.These check points were used for final accuracy validation, which result in an overall accuracy (OA) of 91%.Additionally, the accuracy of forest cover across 32 province was also validated and presented in Figure 5. Comparing the Kappa coefficient, OA, user accuracy (UA), and production accuracy (PA) of products in the provincial scale, it can be found that, the four accuracy indicators of this product in Taiwan Province and Shandong Province are all relatively high (both 94% and above), while in Hebei Province, Chongqing City, Xinjiang Uygur  Autonomous Region, and Hong Kong, three accuracy coefficients ranked at the bottom.This analysis provides a more detailed picture of the accuracy of the forest cover data across different regions in China.
The accuracy of forest AGB data was assessed according to the method described in section 2.3.3.The R 2 value of forest biomass products in the period from 2015 to 2021 were found to range between 0.6107 and 0.682, while the RMSE values is ranged between 50.226 and 56.278 Mg/ha (refer to Figure 6 for details).These results suggest that the RF regression tree model, which was built based on plot measurements without considering location uncertainties, can explain over 61% of the variance in forest AGB, and the root-mean-squared residual is more than 56.28 Mg/ ha.The estimation accuracy of forest AGB is adequate to meet the analytical requirements for assessing China's forest change trend from 2015 to 2021.The forest coverage rate is generally high in the east and lower in the west, as shown in Figure 7 (left).
In terms of administrative units, the SDG 15.1.1 indicators of various provinces are shown in Figure 7 (right) and Table 2.In the southern coastal provinces such as Fujian, Jiangxi, and Taiwan, the SDG 15.1.1 value is the highest, above 60%, and these areas are rich in forest resources.The value for Beijing, Heilongjiang, Shaanxi in the north, and Hubei, Hunan, Guangdong, Guangxi, Yunnan, Guizhou in the south are next, and the SDG 15.1.1 value is 40%−60%; while the vast Xinjiang, Qinghai, Ningxia, central and western Inner Mongolia and in most areas of Tibet, the distribution of forest resources is relatively small, and the SDG 15.1.1 value is 5% and below.From the statistics of AGB of forests in various vegetation areas in China (Figure 9), the biomass of subtropical evergreen broad-leaved forest increased significantly, with an increase of 5.74%; followed by warm temperate deciduous broad-leaved forest, with an increase of 3.19% and tropical monsoon rainforest increased 2.23%.The forest biomass increased in the southern region.The reason is that the southern region has sufficient sunlight, abundant water, and the water and heat conditions are suitable for the development of forest trees, and the afforestation and forest protection in this region are strong, which is conducive to the accumulation of biomass.The temperate coniferous and deciduous broad-leaved mixed forest did not change significantly, and the biomass was relatively stable.The cold temperate coniferous forests are mainly distributed in the northeast and northwest regions of our country.Except for the decrease in some areas of the Daxing'an Mountains, other areas generally have little change.These areas are mainly primary forests, with relatively stable biomass and less human disturbance.

Conclusions
In conclusion, this study has proposed a workflow based on satellite data to acquire forest cover data and forest AGB data for calculating SDG indicators 15.1.1 and 15.2.1.The proposed workflow provided an precise and efficient way to monitor and evaluate forest status and progress towards sustainable forest management.The findings indicate that the forest area in China accounts for approximately 23.51% of the land area at the end of 2021, and the forest biomass in the southern region increased, while it remained relatively stable in the temperate coniferous and deciduous broad-leaved mixed forests.Moreover, the proposed approach has the potential to extend to monitoring and evaluate other SDGs, thereby providing a comprehensive and integrated way to evaluate sustainable development.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Figure 1 .
Figure 1.Study area and distribution of Forest field survey data, the Land cover data from MODIS (MCD12C1) product.

Figure 3 .
Figure 3. Flowchart of forest cover production.

Figure 5 .
Figure 5. Kappa coefficient, OA, UA and PA of products in provinces in China.

Figure 6 .
Figure 6.Relationships between field measured AGB and the estimated AGB.

Figure 7 .
Figure 7. Forest spatial distribution and SDG 15.1.1 in provinces of China.
3.2.The status of forest cover and sdg15 .1.1 in provinces of China According to statistical data, China's total forest area by the end of 2021 is 219 million hectares, with a forest coverage rate of 23.51%.The distribution of forest resources is characterized by being dense in the east and sparse in the west, and less in the north and more in the south, bounded by the Daxing'an Mountains, Taihang Mountains, Qinling Mountains, and Hengduan Mountains.

Figure 8 .
Figure 8. Annual mean (left) and change trend (right) of AGB from 2015 to 2021 in China.

Figure 9 .
Figure 9. Change trend of forest AGB from 2015 to 2021 in China.

3. 3 .
The quality of forest and change trend from 2015 to 2021 in China3.3.1.The status of forest quality in ChinaThe annual average forest AGB from 2015 to 2021 was 105.01Mg/ha, and the distributions are shown in Figure8(left).The spatial distribution of forest biomass in China is roughly consistent with the spatial distribution of forest cover (see Figure7left), and the biomass is higher in large mountain ranges.3.3.2.Change trend of forest quality from 2015 to 2021 in ChinaFigure8(right) and Figure9illustrate the trend of forest biomass change in China from 2015 to 2021.Overall, the trend is positive, with the southern region showing a relatively favorable trend.However, there are spatial differences with no significant change or a decreasing trend observed in the northern region.

Table 1 .
Input factors and characteristics.Calculation of sdg15 .1.1 indicators SDG 15.1.1 has been defined to be a specific, effective indicator that can be used to quantitatively monitor and evaluate forest conditions.It is defined as the ' Forest area as proportion of total land area'; and can be calculated directly using the following formula (complete list: https:// sustainabledevelopment.un.org/sdg15).Here, both forest area and land area are based on statistical data from forest cover data of reference year.According to the definition of land area by FAO (2018), namely the total area of the land excluding area under Inland water, such as rivers, lakes and dams.Therefore, the land area is the spatial area of the calculated area excluding the area of inland water bodies.The calculation of inland water is based on data from the Global Surface Water project of the Joint Research Centre (JRC).2.4.2.Calculation of sdg15 .2.1 indicators SDG 15.2.1 aims to monitor 'progress towards sustainable forest management', which lacks a standardized quantitative evaluation formula, unlike SDG 15.1.1 .To provide a quantitatively assessment of sustainable forest management progress, this paper proposes a novel method that measures forest quality and change trends using the average value of forest AGB and the slope of inter-annual change.These values for SDG 15.2.1 during the periods 2015-2021 were utilized to determine the state and change of forests in China.The specific methods employed are elaborated below.The annual average of forest AGB can provide an overall representation of forest quality.The formula for calculating the annual average is as follows:

Table 2 .
Forest area and SDG 15.1.1 in provinces of China.