Dataset of Spatial Pattern in the Forest Water Retention in China Based on Meta-analysis

Wu, X.^1,2  Shi, W. J.^1,3*  Tao, F. L.^1,3*

1. Key Laboratory of Land Surface Pattern and Simulation, State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China;

2. Jiangsu Province Surveying & Mapping Engineering Institute, Nanjing 210013, China;

3. College of Resources and Environment, University of Chinese Academy of Sciences, Beijing 100049, China

Abstract: Water retention of forest ecosystem plays important roles in interception, storage and redistribution of precipitation. The spatial pattern dataset of forest water retention is based on the 1,045 observation sites with the parameters and its influencing factors across China. We used empirical model to estimate canopy interception capacity, litter maximum water-holding capacity, and soil water storage capacity. Then, we applied the random forest model to predict the spatial pattern of forest water retention. The results show that the random forest model based on observation sites has credible results in predicting the spatial pattern of forest water retention of China. Our results revealed that the forest water retention capacity in China increased from north to south. The total forest water retention amount in Sichuan, Tibet and Yunnan are relatively high. The dataset includes forest water retention capacity divided by forest types based on a 10 km x 10 km grid, and data of 1,045 observation sites of forest water retention. The dataset is archived in .xlsx and .shp data formats, and consists of 9 data files with data size of 118 MB.

Keywords: China; forest water retention; national scale; spatial pattern

DOI: https://doi.org/10.3974/geodp.2023.01.02

CSTR: https://cstr.escience.org.cn/CSTR:20146.14.2023.01.02

Dataset Availability Statement:

The dataset supporting this paper was published and is accessible through the Digital Journal of Global Change Data Repository at: https://doi.org/10.3974/geodb.2022.03.06.V1 or https://cstr.escience.org.cn/CSTR:20146.11.2022.03.06.V1.

1 Introduction

Under a series of conditions of global climate change, request of restoring the ecological environment and terrestrial water storage depletion, a series of ecological problems such as water pollution, land desertification, and soil erosion in China are needed to be urgently solved^[1,2]. These problems are closely related to water retention volumes. Forest ecosystems play a decisive role in global ecosystems and are one of the most important terrestrial ecosystems^[3,4]. The forest water retention service is an important process of regulating climate and water resources. It promotes rainfall redistribution, moderates surface runoff, and increases soil runoff and underground runoff through the interception, retention and accumulation of rainfall^[5]. China has a vast territory and diverse climatic characteristics. The capacity and volumes of forest water retention in have large variety in different regions. It is urgent to explore patterns of forest water retention in different regions across China.

The methods for water retention estimation are mainly based on water-balance theory or empirical models. The water-balance model has been widely applied to simulate the forest water retention capacity at a large scale, but there is a lack of observation sites to validate model results in various regions^[6]. Empirical models are usually applied at small spatial scales for estimating water retention. Due to the difficulty in obtaining observational data, empirical models are difficult to practically measure at a national scale. Methods of forest water retention spatial pattern based on observation sites in previous studies include assignment method, regression method, machine learning and geostatistical method, among which machine learning and geostatistical method are more suitable for large scale research^[6].

The empirical model is used to estimate canopy interception capacity (CIC), litter maximum water-holding capacity (LWHC), soil water storage capacity (SSC) and forest water retention capacity (WRC) of 1045 observation sites. Then, the random forest model is used to predict the spatial distribution characteristics and further analyze the spatial pattern of forest water retention.

2 Metadata of the Dataset

The metadata of the Dataset of spatial pattern in the forest water retention in China based on meta-analysis^[9] is summarized in Table 1. It includes the dataset full name, short name, authors, year of the dataset, temporal resolution, spatial resolution, data format, data size, data files, data publisher, and data sharing policy, etc.

3 Methods

Based on the 1,045 water retention observation sites across China (Figure 1), the maximum capacity of precipitation intercepted and stored by the forest can be simulated by the empirical model of WRC as the sum of CIC, LWHC and SSC. The 1,045 observation sites selected in this paper are based on meta-analysis to collect articles reporting on the parameters related to water retention, including canopy interception rate, litter storage, litter maximum water-holding rate, soil depth and soil non-capillary porosity or three components related to forest water retention function, including CIC, LWHC and SSC.

Then, the CIC, LWHC, and SSC values of 1,045 observation sites are calculated by the empirical model, which are used as the training samples and verification dataset of the random forest model to construct spatial pattern models of the CIC, LWHC and SSC.

Table 1  Metadata summary of the Dataset of spatial pattern in the forest water retention in China based on meta-analysis

Figure 1  Map of forest and observation sites distribution in China

Finally, the empirical model is used to calculate the spatial WRC, canopy interception amount (CIA), litter maximum water-holding amount (LWHA), soil water storage amount (SSA) and forest water retention amount (WRA).

3.1 Data Processing

3.1.1 Random Forest Model

China covers an area of approximately 9.6 million km², so it is difficult to extract observation sites for measuring water retention, which are uneven distributed. The relationships between parameters and driving factors of forest water retention are complex and mostly nonlinear. A random forest model is a nonparametric decision tree classifier that can be used to process complex and nonlinear variables^{[11, 12]}, and it is powerful for spatial prediction over complex terrain, which is suitable for spatial prediction of WRC and WRA at a large scale^[13]. In this study, the spatial prediction of water retention is based on the "Random Forest" module in the R software. The water retention value of 1,045 observation sites are the input data, and the significant factors affecting the forest water retention are used as auxiliary information to predict the CIC, LWHC and SSC^[13].

In random forest prediction, the predicted value is the average of the output results of all regression trees, and the expression is:

                                                                                                        (1)

whereis predictor;  are independent identically distributed random vectors; X is an input vector;  is the output result of the ith regression tree; k is the number of regression tree.

3.1.2 Empirical model

An empirical model of the forest water retention capacity, representing the sum of CIC, LWHC and SSC and considering the canopy, litter and soil layers based on observation data, is a relatively comprehensive method for estimating WRC and WRA^[6]. Then, the WRC and WRA are calculated according to Equations (2) – (6), respectively:

                                                                                                (2)

                                                                                                    (3)

                                                                                                             (4)

                                                                                                   (5)

                                                                                                            (6)

where WRC_i, CIC_i, LWHC_i and SSC_i are the WRC, CIC, LWHC, and SSC in the ith forest sublot (mm), respectively; WRA_i, CIA_i , LWHA_i, and SSA_i are the WRA, CIA, LWHA, and SSA (m³) in the ith forest sublot (m³), respectively; A_i is the area of the ith forest sublot (m²).

3.2 Technical Route

Based on forest water retention observation sites, the random forest model is used to predict the spatial pattern of forest water retention in China. The steps are as follows (Figure 2).

Firstly, the China National Knowledge Infrastructure (CNKI) and Web of Science Core Collection Databases were used to search the observation data in the forest water retention articles based on a meta-analysis method. A total of 1045 observation sites were collected for obtaining water retention parameters, basic forest information and site conditions.

Then, the empirical model was used to calculate the CIC, LWHC and SSC of observation sites^[14]. Taking the observation data calculated by the empirical model as the input data and the significant influencing factors as the auxiliary data, the spatial pattern models of the CIC, LWHC and SSC were constructed based on the random forest model.

Finally, according to the CIC, LWHC and SSC predicted by the random forest model, the empirical model was used to calculate the spatial WRC, CIA, LWHA, SSA and WRA. Due to the significant spatial heterogeneity of forest water retention in China, spatial pattern characteristics of CIC, LWHC, SSC, WRC, CIA, LWHA, SSA and WRA were analyzed. However, because Shanghai, Hainan, Macao and Taiwan did not collect water retention observation sites, the spatial pattern characteristics of their forest water retention were not analyzed.

Figure 2  The framework of forest water retention spatial dataset in China based on the meta-analysis

4 Data Results and Validation

4.1 Data Composition

The Dataset of spatial pattern in the forest water retention in China based on meta-analysis is composed of forest water retention data and statistical tables. The statistical tables include the statistical table of forest water retention in various provinces based on the random forest model prediction, and the data of 1,045 observation sites of forest water retention. The relevant fields of forest water retention vector data are shown in Table 2.

4.2 Data Products

4.2.1 Spatial Patterns of the CIC and CIA

The CIC values were in the range of 0–37 mm, which were lower in northern China than in southern China (Figure 3). The CIC values less than 14 mm were mainly distributed in Heilongjiang, northern Inner Mongolia and Shanxi, and the CIC values larger than 26 mm were mostly distributed in Guangdong, Guangxi, Yunnan. The average CIC value varied widely among provinces, ranging from 13.35 to 26.67 mm (Figure 4).

Due to the large differences in the distribution area and forest types in various provinces, the CIA varied greatly, ranging from 73.36–488,871×10⁴ m³. The CIAs of Beijing, Tianjin, Jiangsu and Ningxia were less than 10,000×10⁴ m³, and the CIAs of Sichuan, Yunnan and Tibet were higher than other provinces (Figure 4). The CIA of various forest type per 100 km² was in the range of 0–275×10⁴ m³. CIAs of various forest type per 100 km² higher than 150×10⁴ m³ were mostly distributed in the Guangdong, Guangxi and Fujian (Figure 3).

Table 2  Forest water retention dataset parameters and their definitions

Note: The corresponding relationship between forest type codes and forest types is as follows: 1–16 represent cold and temperate mountainous needleleaf forest; 17–22 represent temperate needleleaf forest; 23–31 represent subtropical needleleaf forest; 32–33 represent tropical needleleaf forest; 34–62 represent subtropical and tropical mountainous needleleaf forest; 63 represents temperate needleleaf and broadleaf mixed forest; 64–66 represent subtropical needleleaf and broadleaf mixed forest; 67–90 represent temperate deciduous broadleaf forest; 91–93 represent temperate deciduous lobular forest; 94–105 represent subtropical deciduous broadleaf forest; 106–112 represent subtropical broadleaf mixed forest; 113–135 represent subtropical evergreen broadleaf forest; 136–141 represent subtropical sclerophyllous evergreen broadleaf forest; 142–146 represent tropical monsoon rainforest; 147–161 represent tropical rainforest; 162–175 represent subtropical and tropical bamboo forest; 176–264 represent shrubland.

Figure 3  Maps of spatial patterns of CICs and CIAs in China

Figure 4  Distribution characteristics of CICs and CIAs in various provinces in China

4.2.2 Spatial Patterns of the LWHC and LWHA

Contrary to the spatial trend of CIC, the LWHC values gradually decreased from north to south in China, ranging from 0 mm to 17 mm (Figure 5). The LWHC values in the Heilongjiang and northern Inner Mongolia, which were concentrated from 7–16 mm, were higher in China. The LWHC values in the southern provinces of China (Yunnan, Guizhou, Guangdong, Guangxi, and Fujian) mostly ranged from 0 to 5 mm. The average LWHC value had a small difference among provinces, ranging from 2.42 mm to 6.54 mm (Figure 6).

The LWHAs ranged from 6.5×10⁴ m³ to 139,315.43×10⁴ m³ in each province in China. The LWHAs of Tianjin, Jiangsu and Ningxia were less than 2,000×10⁴ m³, and the LWHAs of Heilongjiang and Inner Mongolia were higher than other provinces, above 100,000×10⁴ m³ (Figure 6). The LWHA of various forest type per 100 km² was in the range of 0–109×10⁴ m³. LWHAs of various forest type per 100 km² higher than 30×10⁴ m³ were mostly distributed in the Heilongjiang and northern Inner Mongolia (Figure 5).

Figure 5  Maps of spatial patterns of LWHCs and LWHAs in China.

Figure 6  Distribution characteristics of LWHCs and LWHAs in various provinces in China

4.2.3 Spatial Patterns of the SSC and SSA

The spatial pattern in the SSC values showed an increasing trend from north to south in China, ranging from 0 mm to 104 mm (Figure 7). The SSC values higher than 65 mm were mostly located in the southeastern Tibet, the junction of Hubei and Chongqing, northern Jiangxi and Zhejiang. The SSC values ranged from 0 mm to 45 mm mostly in the Heilongjiang, Inner Mongolia, Liaoning and Hebei. The average SSC value ranged from 43.95 mm to 68.14 mm in various province. Tianjin and Heilongjiang were lower than other provinces, and the average SSC value of Tibet, Zhejiang and Chongqing were higher (Figure 8).

The SSAs ranged from 182.24×10⁴ m³ to 1,578,417.90×10⁴ m³ in each province in China. The SSAs of Tianjin, Jiangsu and Ningxia were less than 20,000×10⁴ m³, and the SSAs of Sichuan, Tibet and Yunnan were higher than other provinces, above 1,000,000×10⁴ m³ (Figure 8). The SSA of various forest type per 100 km² was in the range of 0–685×10⁴ m³. LWHAs of various forest type per 100 km² higher than 400×10⁴ m³ were mostly distributed in the Guangdong, Fujian, Shanxi and Henan (Figure 7).

Figure 7  Maps of spatial patterns of SSCs and SSAs in China.

Figure 8  Distribution characteristics of SSCs and SSAs in various provinces in China

4.2.4 Spatial Patterns of the WRC and WRA

The SSC values and SSAs explained most of the WRC values and WRAs, respectively, and their distribution trends were consistent. The spatial pattern in the WRC values showed an increasing trend from north to south in China, ranging from 0 mm to 130 mm (Figure 9). The WRC values higher than 90 mm were mostly located in the Southeast Tibet, Hubei, Sichuan and the provinces to the south. The WRC values ranged from 0 mm to 65 mm mostly in the Heilongjiang. The average WRC value in various province ranged from 61.70 mm to 95.04 mm. Beijing and Tianjin were lower than other provinces, and the average SSC value of Guangxi, Jiangxi, Zhejiang and Chongqing were higher (Figure 10).

The WRAs ranged from 262.71×10⁴ m³ to 2,163,771.88×10⁴ m³ in each province in China. The WRAs of Tianjin, Jiangsu and Ningxia were less than 20,000×10⁴ m³, and the SSAs of Sichuan and Tibet were higher than other provinces, above 2,000,000×10⁴ m³ (Figure 10). The WRA of various forest type per 100 km² was in the range of 0–966×10⁴ m³. WRAs of various forest type per 100 km² higher than 500×10⁴ m³ were mostly distributed in all provinces (Figure 9).

Figure 9  Maps of spatial patterns of WRCs and WRAs in China

Figure 10  Distribution characteristics of WRCs and WRAs in various provinces in China

4.3 Data Validation

An independent dataset including 30% sampling points was randomly selected from the original canopy, litter and soil layer samples as the validation samples to assess model performance. The root mean square error (RMSE) and mean absolute error (MAE) were calculated for 30 times to assess the accuracy of the predicted CIC, LWHC and SSC values based on the random forest model. In addition, the standard deviations (SD) of the performance indicators RMSE and MAE for the CIC, LWHC, SSC values were also presented.

                                                                                                  (7)

                                                                                                        (8)

whereis the observed CIC, LWHC or SSC value in the ith sample;is the simulated CIC, LWHC or SSC value in the ith sample; m is the total number of samples.

The RMSEs and MAEs of random forest for spatial prediction of CIC, LWHC and SSC in China were shown in Table 3. The RMSE and MAE for CIC were 7.19±0.14 mm and 4.64±0.06 mm, respectively. For the predicted LWHC values, the random forest model produced the lowest RMSE and MAE (RMSE = 3.50±0.07 mm and MAE = 1.97±0.05 mm). For the predicted SSC values, the values of these two indices were 35.05±0.43 mm for RMSE and 22.56±0.18 mm for MAE.

Table 3  The RMSEs and MAEs of random forest for spatial prediction of CIC, LWHC and SSC in China

5 Discussion and Conclusion

In order to estimate forest water retention services at the national scale, the 1,045 observation sites were collected to construct a forest water retention parameter dataset. The random forest model was used to predict the spatial distribution characteristics. Finally, we further analyzed the spatial pattern of forest water retention.

The results showed that the random forest model based on observational sites had good results in predicting the spatial pattern of forest water retention in China. The SSC values and SSAs explained most of the WRC values and WRAs, respectively, accounting for about 54%–97%, followed by CIC values and CIAs, LWHC values and LWHAs. Contrary to the spatial trend of LWHC, the CIC, SSC and WRC gradually increased from north to south in China. The CIAs, SSAs and WRAs of Sichuan, Tibet, and Yunnan were higher than other provinces, and the LWHAs of Heilongjiang and Inner Mongolia were higher.

However, the observation data collected based on meta-analysis had a large age span (1987-2017), and the observation time was also inconsistent. There was a lack of research on the spatial pattern of forest water retention in different time series. In addition, natural factors, such as meteorological factors, terrain factors, soil factors, etc., were taken into account when constructing the random forest model. However, with the continuous expansion of human footprints on land, the original forest had been destroyed, the forest area had decreased, and China has attached great importance to the construction of ecological civilization. A series of major decision-making arrangements have been issued, and a series of ecological protection and restoration projects have been carried out. In the future, the impact of human activities and policy implementation on forest ecosystems should be considered.

Author Contributions

Shi, W. J. and Tao, F. L. made the overall design for the development of dataset; Wu, X. collected and processed the data; Shi, W. J., Tao, F. L. and Wu, X. designed the model and algorithm of the dataset. Wu, X. did data verification; Wu, X. wrote the data paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]      Lang, Y., Song, W., Zhang, Y. Responses of the water-yield ecosystem service to climate and land use change in Sancha River Basin, China [J]. Physics and Chemistry of the Earth, Parts A/B/C, 2017, 101: 102‒111.

[2]      Moiwo, J. P., Tao, F., Lu, W. Analysis of satellite-based and in situ hydro-climatic data depicts water storage depletion in North China Region [J]. Hydrological Processes, 2013, 27(7): 1011‒1020.

[3]      Zhang, B. A., Li, W. H., Xie, G. D., et al. Water conservation of forest ecosystem in Beijing and its value [J]. Ecological Economics, 2010, 69(7): 1416‒1426.

[4]      Feng, J. G., Ding, L. B., Wang, J. S., et al. Case-based evaluation of forest ecosystem service function in China [J]. Chinese Journal of Applied Ecology, 2016, 27(5): 1375‒1382.

[5]      Julian, J. P., Gardner, R. H. Land cover effects on runoff patterns in eastern Piedmont (USA) watersheds [J]. Hydrological Processes, 2014, 28(3): 1525‒1538.

[6]      Wu, X., Shi, W. J. Spatial simulation methods of regional forest water conversation based on observed data: A review [J]. Journal of Ecology and Rural Environment, 2019, 35: 1505‒1515.

[7]      Tang, Y. Z., Shao, Q. Q. Water conservation capacity of forest ecosystem and its spatial variation in the upper reaches of Wujiang River [J]. Journal of Geo-information Science, 2016, 18, 987‒999.

[8]      Liu, L. L., Shao, Q. Q., Liu, J. Y., et al. Estimation of forest water conservation capacity in Qiongjiang River Watershed [J]. Ecology and Environmental Sciences, 2013, 22, 451‒457.

[9]      Wu, X., Shi, W. J., Tao, F. L. Spatial dataset of forest water retention in China based on meta-analysis [J/DB/OL]. Digital Journal of Global Change Data Repository, 2022. https://doi.org/10.3974/geodb.2022. 03.06.V1. https://cstr.escience.org.cn/CSTR:20146.11.2022.03.06.V1.

[10]              GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. https://doi.org/10.3974/dp.policy.2014.05 (Updated 2017).

[11]   Zhang, L., Wang, L. L., Zhang, X. D., et al. The basic principle of random forest and its applications in ecology: a case study of Pinus yunnanensis [J]. Acta Ecologica Sinica, 2014, 34(3): 650‒659.

[12]   Breiman, L. Random forests [J]. Machine learning, 2001, 45: 5‒32.

[13]   De’ath, G., Fabricius, K. E. Classification and regression trees: a powerful yet simple technique for ecological data analysis [J]. Ecology, 2000, 81(11): 3178‒3192.

[14]   Wu, X., Shi, W. J., Guo, B., et al. Large spatial variations in the distributions of and factors affecting forest water retention capacity in China [J]. Ecological Indicators, 2020, 113: 106152.

[15]   Wu, X. Spatial distributions and factors affecting of forest water conservation in mainland China [D]. Qingdao: Shandong University of Science and Technology, 2019.

[16]   Li, H., Li, Z., Li, Z., et al. Evaluation of ecosystem services: A case study in the middle reach of the Heihe River Basin, Northwest China [J]. Physics and Chemistry of the Earth, Parts A/B/C, 2015, 89: 40‒45.

[17]   Laino-Guanes, R., González-Espinosa, M., Ramírez-Marcial, N., et al. Human pressure on water quality and water yield in the upper Grijalva river basin in the Mexico-Guatemala border [J]. Ecohydrology & Hydrobiology, 2016, 16(3): 149‒159.

[18]   Fan, Y. N., Liu, K., Chen, S. S., et al. Spatial pattern analysis on water conversation functionality of land ecosystem in northern slope of Qinling Mountains [J]. Bulletin of Soil and Water Conservation, 2017, 37(2): 50‒56.

[19]   Ouyang, Z., Zheng, H., Xiao, Y., et al. Improvements in ecosystem services from investments in natural capital [J]. Science, 2016, 352(6292): 1455‒1459.