Spring Phenological Grid Dataset of Typical
Deciduous Broad-leaved Woody Plants in the Warm Temperate Zone of China

Dai, J. H.^1,2,*  Zhu, M. Y.^1,2  Wang, H. J.¹  Tao, Z. X.^1,2  Hu, Z.^1,2  Dong, X. Y.³

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

2. University of Chinese Academy of Sciences, Beijing 100049, China;

3. School of Earth Science and Resources, Chang’an University, Xi’an 710054, China

Abstract: Plant phenology is important in assessing the effects of global change and in studying ecosystem structure and function, as well as in developing climate and land surface models. The leaf unfolding date (LUD) and the first flowering date (FFD) of eight woody plants (Fraxinus chinensis Roxb. (Oleaceae), Salix babylonica L. (Salicaceae), Robinia pseudoacacia L. (Leguminosae), Albizia julibrissin Durazz. (Leguminosae), Morus alba L. (Moraceae), Ulmus pumila L. (Ulmaceae Mirb.), Armeniaca vulgaris Lam. (Rosaceae), Cercis chinensis Bunge. (Leguminosae)) from 1963 to 2015 were obtained from the Chinese Phenological Observation Network. The 0.5° × 0.5° grid data of the LUD and FFD were developed for each species in the past 50 years by building and testing three phenological models: Unified, UniChill, and Temporal–Spatial Coupling (TSC) model. The verification results showed that the average simulation error of the LUD was approximately 5–9 days and that of the FFD was approximately 5–12 days. The data set includes the following: (1) header file, containing grid data header information of the species phenophase and spatial distribution; (2) phenophase, containing the LUD and FFD of each species from 1963 to 2015; and (3) spatial distribution, containing the actual spatial distribution range grid of each species. The data set consists of three folders and 1,713 files. The data files are stored in TXT, GEOTIFF, and ARCGIS ASCII formats. The data volume is 74.5 MB before compression (9 MB after compression).

Keywords: Spring phenology; Woody plants; Leaf unfolding date; First flowering date; China

1 Introduction

Plant phenology is the seasonal behavior of plants during an annual cycle that is influenced by diverse environmental factors (including climate, hydrology, and soil)^[1] Because of the influence of the environment, phenology is highly sensitive to climate change and has become an important indicator for recording and evaluating the effects of global changes on ecosystems^[2] In recent decades, the increase in spring temperature has led to a substantial advance in plant spring phenology in the Northern Hemisphere, thereby substantially influencing carbon and water cycles and the nutrient balance in terrestrial ecosystems.^[3] Thus, plant phenological data are also important in indicating the impact of global change, in understanding the structure and function of ecosystems, as well as in developing climate and land models. The authors of this paper developed and published grid data of the spring phenology of individual species based on the Chinese Phenological Observation Network (CPON)^[4] Then, using primarily the manual observation data of CPON, they greatly improved understanding of the phenological temporal–spatial patterns and their causes in China^[5,6] According to the experience of other researchers, widely distributed and well observed plant species were selected to produce multispecies grid data sets, which displayed the changes in the phenological temporal–spatial patterns of Chinese woody plants in spring. The grid data sets also improve the use of Chinese phenological data in understanding global change. In addition, the team is currently participating in the special project of “earth big data science and engineering”, in which Chinese phenological data products are an essential part of gathering and sharing ground monitoring data. Therefore, the purpose of this paper is to publish the methods and basic results of the spring phenological temporal–spatial data set of Chinese woody plants. The data set contains the grid data of the spring phenophase (the leaf unfolding date (LUD) and the first flowering date (FFD)) of eight typical deciduous broad-leaved woody plants in China from 1963 to 2015 and provides the scale transformation from station phenology to regional phenology in China.

2 Metadata of Data set

The metadata of the spring phenological grid data set of typical deciduous broad-leaved woody plants in the warm temperate zone of China^[7] are summarized in Table 1. It includes the data set full name, short name, authors, year of the data set, temporal resolution, spatial resolution, data format, data size, data files, data publisher, and data sharing policy.

Table 1  Metadata summary of the spring phenological grid data set of eight typical deciduous broad-leaved woody plants in the warm temperate zone of China (1963–2015)

(To be continued on the next page)

(Continued)

3 Methods

3.1 Basic Data Collection

The LUD and the FFD of eight woody plants from 1963–2015 were extracted from the Chinese Phenological Observation Network (CPON). The eight plants were Fraxinus chinensis Roxb. (Oleaceae), Salix babylonica L. (Salicaceae), Robinia pseudoacacia L. (Leguminosae), Albizia julibrissin Durazz. (Leguminosae), Morus alba L. (Moraceae), Ulmus pumila L. (Ulmaceae Mirb.), Armeniaca vulgaris Lam. (Rosaceae), Cercis chinensis Bunge. (Leguminosae), which are typical species of deciduous broad-leaved forests in the warm temperate zone of China^[9] The temperature data were derived from the China Meteorological Data Service Center (CMDSC, data.cma.cn). The daily temperature data of the corresponding stations and time periods in the “China surface daily climate dataset (v.3.0)” were used for the parameterization of phenological models, whereas the grid average daily temperature data obtained from the “China surface 0.5° × 0.5° daily temperature grid dataset (v. 2.0)” were used for phenophase simulation and scale expansion.

3.2 Algorithm Principle

To develop the spring phenological grid data, three steps were included: building phenological models, estimating the corresponding parameter values, and validating the models^[10] To obtain the most accurate simulation of phenophase, three models were tested: the Unichill model^[11], the Unified model^[11], and the Temporal–Spatial Coupling (TSC) model^[12], which was based on the spring warming (SW) model. The minimum root mean square error (RMSE) of the three spring phenological models was used to select the optimal phenological model to simulate the LUD and FFD temporal–spatial patterns of each species in the past 50 years.^[13]

(1) The Unified model divides the process of bud burst into two phases: dormancy and quiescence, with nine species-specific parameters (a, b, c, d, e, w, k, C*, and t_c) fit to phenological observations. Parameters a, b, and c define the response function to temperature (R_c(x_t), also called “chilling units”, eq. (1)). Parameters d and e define the other response function to temperature (R_f (x_t), also called “forcing units”, eq. (2)). Parameter t₀ is the time when R_c(x_t) begins to accumulate, which determines the threshold C* of the cumulative sum of chilling units (eq. (3)). Parameters w, k, and t_c determine the threshold F* of the cumulative sum of forcing units (eq. (4)). t_b is the time of LUD or FFD when  reaches the threshold F*.

                                                                                         (1)

                                                                                          (2)

                                                                                      (3)

                                           (4)

(2) The UniChill model is a simplified version of the Unified model, which contains seven parameters: a, b, c, d, e, C*, and F*. On the basis of Unified model, this model fixes the time t₁ for the start of the forcing units as September 1 of the previous year and assumes that the accumulation of forcing units (F*) required for the start of the spring phenology each year is fixed.

(3) The TSC model is built on the SW model^[14] (eqs. (5, 6)), which uses the winter average temperature to reflect the difference in the accumulated temperature threshold^[15] (eq. (7)). The TSC model includes six parameters: T_b1, F*, t₀, a, b, and f. R_f(x_t) is the forcing unit, x_t is the average daily temperature on day t, and T_b1 is the critical temperature. F* is the threshold value of temperature accumulation. The temperature accumulation threshold F_i of different sites is determined by the winter (December of the previous year to February) temperature  of site i and parameters a, b, and f. t₀ is the time when the forcing unit begins to accumulate, and y is the time of LUD or FFD when  reaches F*.

                                                                        (5)

                                                                                       (6)

                                                                                      (7)

3.3 Technical Route

The main steps in the technical roadmap are shown in Figure 1. First, the in situ spring phenological observation data of the eight woody plants from 1963–2015 were obtained from 45 sites in CPON. Second, three models of spring phenology were established from the

spatial–temporal phenophase and corresponding temperature data, and the parameters were estimated using the simulated annealing algorithm.^[16] Next, the goodness of fit (r²) and root mean square error (RMSE) of the three models were compared by both internal and cross- validation to select the optimal model. Finally, the spring phenology 0.5° × 0.5° grid data of the eight woody plants in China were developed by simulation and scale expansion from temperature grid data and spatial distribution grid data.

4 Results and Validation

4.1 Data Composition

The grid data set of spring phenology consists of three groups of files: header file, phenophase, and spatial distribution. The naming format, data description, data format, file number, and data volume of data files are shown in Table 2.

(1) Header file contains grid data header information of the species phenophase and spatial distribution.

(2) Phenophase contains LUD and FFD folders, each of which consists of eight species subfolders. Each species subfolder contains 106 phenophase files from 1963 to 2015. The data files are stored in ARCGIS ASCII and GEOTIFF formats. By copying the grid data header file in front of each ASCII phenophase file, it can be converted into a visual grid format using the “ASCII to Raster” tool in ArcMap.

(3) Spatial distribution contains the actual distribution grid of each species.^[4] The phenophase data outside the distribution range need to be masked by the user. Data files are stored in ARCGIS ASCII and GEOTIFF formats, which can also be transformed into visual GRID format in ArcMap after copying header files.

Table 2  Short form of the spring phenological grid data set of eight typical deciduous broad-leaved woody plants in the warm temperate zone of China (1963–2015)

4.2 Data Products

(1) Products of leaf unfolding date data

The above methods and steps were used to calculate the average LUD of each species from 1963 to 2015, as shown in Figure 2. According to the spatial distribution pattern, the LUD of each species differentiated by latitude and altitude, arriving earlier in the southeast and later in the northwest of the distribution. In terms of temporal distribution, the first species with leaf unfolding was Armeniaca vulgaris (Figure 2g), which was followed by Salix babylonica, Robinia pseudoacacia, Ulmus pumila, Fraxinus chinensis, Morus alba, and Cercis chinensis, with Albizia julibrissin the last species with leaf unfolding (Figure 2d). A linear regress between LUD and year was used to determine the trend of LUD. The results indicated that the LUD of each species advanced from 1963 to 2015 (Table 3), with an average trend of 0.9–3.4 days decade⁻¹. Specifically, in 83.1%–99.7% of the cells the LUD were advanced, whereas in 22.0%–79.8% of the cells the advance were significant (p<0.05).

Figure 2  Average leaf unfolding date (LUD) of eight typical deciduous broad-leaf woody plants in the warm temperate zone of China from 1963 to 2015. Panels (a)–(h) represents mean LUD of each species. DOY represents day of year refer to a Julian calendar.

(2) Products of first flowering data

Similarly, the FFD of the eight plants from 1963 to 2015 was calculated, and the simula­tion results are shown in Figure 3. According to the spatial distribution pattern, for all spe­cies, the latitudinal differentiation of the FFD was less than that of the LUD. Albizia juli­brissin (Figure 3d) and Cercis chinensis (Figure 3h) showed little spatial difference, whereas the other species showed distinct differences. In terms of temporal distribution, the sequence of the FFD for each species in the same area varied. Ulmus pumila (Figure 3f) and Armeni­aca vulgaris (Figure 3g) had the earliest FFD, whereas Robinia pseudoacacia and Albizia julibrissin had the latest. A linear regress between FFD and year was used to determine the trend of FFD. The results of trend analysis indicated that the FFD of the eight woody plants in China was also earlier in the past 50 years, with an average advance of 0.7–3.3 days decade⁻¹. In the grid, 77.1%–97.7% of the cells of the FFD were advanced, whereas 32.5%– 79.7% of the cells were significant (p<0.05).

Table 3  Statistics on the shifting trend in the spring phenophase of eight typical deciduous broad-leaved woody plants in China from 1963 to 2015

Figure 3  Average first flowering date (FFD) of eight typical deciduous broad-leaf woody plants in the warm temperate zone of China from 1963 to 2015. Panels (a)–(h) represents mean FFD of each species. DOY represents day of year refer to a Julian calendar.

4.3 Data validation

(1) Leaf unfolding date validation

Internal and cross-validation methods were used to verify the reliability of the LUD (Figure 4). According to internal verification results, the fit of all three phenological models to the data was good. The r² values of the three models of the eight species were greater than 0.81, except for Albizia julibrissin (r² = 0.47, 0.51, 0.45) and Cercis chinensis (r² = 0.51, 0.59, 0.53). For most species, the UniChill model had a relatively better fit to the LUD data, with RMSE values ranging from 5.07 to 9.90 days. The TSC model had the best fit for Salix babylonica and Robinia pseudoacacia, with RMSE val- ues of 5.27–8.36 days. The Unified model was a relatively poor fit to data of all species. According to the cross-validation results, the simulations of the LUD by the UniChill and TSC models were similar for most species. Except for Albizia julibrissin (r² = 0.50 and 0.46, respectively) and Cercis chinensis (r² = 0.60 and 0.64, respectively), the r² values of other species reached 0.84–0.91, with RMSE values of 5.09–11.04 days. The simulation results of the Unified models were relatively poor, with RMSE values ranging from 5.30 to 12.56 days. Overall, the cross-validation results indicated that the

Figure 4  Internal and cross-validation of the simulation results of the leaf unfolding date (LUD) of eight species by three phenological models. Panels (a)–(p) represents internal and cross-validation on LUD of each species. DOY represents day of year refer to a Julian calendar.

Note: The circle, cross, and triangle represent the simulation results of the Unified, UniChill, and TSC models, respectively. Solid lines are linear regression curves fit to observed and predicted values. The number after the symbol and the number in parentheses represent the r² and root mean square error, respectively, of the model. The root mean square minimum value represents the simulation result of the optimal phenological model.

simulation error of the optimal phenological model of the LUD of each species was 5.09–8.97 days.

(2) First flowering date validation

The results of the FFD were similar to those of the LUD. Internal verification showed that the fit of the models to the FFD of the eight woody plants differed (Figure 5). Except for Albizia julibrissin (r² = 0.45), the average r² of the optimal model was 0.68–0.92. The UniChill model had the best fit for the FFD of most species, with RMSE values ranging from 5.05 to 11.34 days. The fit of the TSC model was better than that of the other models for Salix babylonica and Robinia pseudoacacia, with RMSE values ranging from 5.26–13.48 days. According to cross-validation results, the simulations of the different models were obviously different. The simulations of the FFD of the UniChill model were more accurate for half of the study species, and RMSE values were 4.78–8.38 days. The TSC model was the most suitable for Salix babylonica and Robinia pseudoacacia, whereas the Unified model was the best for Fraxinus chinensis and Albizia julibrissin. The cross-validation results indicated that the simulation by the optimal phenological model for the FFD was inferior to the simulation by the optimal phenological model for the LUD, with an error of 4.78–12.35 days.

Figure 5  Internal and cross-validation of the simulation results of the first flowering date (FFD) of eight species by three phenological models. Panels (a)–(p) represents internal and cross-validation on FFD of each species. DOY represents day of year refer to a Julian calendar.

Note: The circle, cross, and triangle represent the simulation results of the Unified, UniChill, and TSC models, respectively. Solid lines are linear regression curves fit to observed and predicted values. The number after the symbol and the number in parentheses represent the r² and root mean square error, respectively, of the model. The root mean square minimum value  represents the simulation result of the optimal phenological model.

5 Discussion and Conclusion

In this paper, in situ phenological data from the China Phenological Observation Network from 1963–2015 were used with three common spring phenological models to develop a spring phenological grid data set of eight typical deciduous broad-leaved woody plants in the warm temperate zone of China. The data set provides grid data of the LUD and the FFD of the eight species. The temporal and spatial resolution are one year and 0.5° × 0.5°, respectively. For each species, the mean simulation error was approximately 5–9 days for the LUD and approximately 5-12 days for the FFD. The data set represents the spatial–temporal patterns of plant phenology in China over the past 50 years, which can provide basic data for many areas of research, such as global change, meteorology, ecology, natural resources, the environment, tourism, and human health.

Author Contributions

Dai, J. H. was responsible for the overall design and development of the dataset. Hu, Z. and Dong, X. Y. collected and processed spring phenological data. Wang, H. J. and Tao, Z. X. designed the models and the algorithm. Zhu, M. Y. and Hu, Z. performed data verification. Zhu, M. Y. and Dai, J. H. wrote the dataset papers.

References

[1]       Lieth H. Phenology and Seasonality Modeling [J]. Ecological Studies, 1974, 120(6): 461.

[2]       Richardson A. D., Keenan T. F., Migliavacca M., et al. Climate change, phenology, and phenological control of vegetation feedbacks to the climate system [J]. Agricultural and Forest Meteorology, 2013, 169: 156–173.

[3]       Piao S. L., Liu Q., Chen A. P., et al. Plant phenology and global climate change: current progresses and challenges [J]. Global Change Biology, 2019, 25: 1922–1940.

[4]       Ge Q. S., Dai J. H., Wang H. J. The gridded dataset of spring phenology of Fraxinus chinensis in China from 1952 to 2007 [DB/OL]. Global Change Research Data Publishing & Repository, 2014. DOI: 10.3974/geodb.2014.01.09.V1.

[5]       Ge, Q. S., Wang, H. J., Rutishauser, T., et al. Phenological response to climate change in China: a meta- analysis [J]. Global Change Biology, 2015, 21(1): 265–274.

[6]       Dai, J. H., Wang, H. J., Ge, Q. S. The spatial pattern of leaf phenology and its response to climate change in China [J]. International Journal of Biometeorology, 2014, 58(4): 521–528.

[7]       Dai, J. H., Zhu, M. Y., Wang, H. J., et al. Spring phenological grid data set of typical deciduous broad-leaved woody plants in the warm temperate zone of China [DB/OL]. Global Change Research Data Publishing & Repository, 2019. DOI: 10.3974/geodb.2019.06.04.V1.

[8]       Change Research Data Publishing & Repository. Data sharing policy for scientific research on global change [OL]. DOI: 10.3974/dp.policy.2014.05 (updated 2017).

[9]       Fang, J. Y., Wang, Z. H., Tang, Z. Y. Altas of woody plants in China: distribution and climate [M]. Beijing: Higher Education Press, 2009.

[10]    Wang, H. J., Dai, J. H., Ge, Q. S. The spatiotemporal characteristics of spring phenophase changes of Fraxinus chinensis in China from 1952 to 2007 [J]. Science China-Earth Sciences, 2012, 55(6): 991–1000.

[11]    Chuine, I. A unified model for budburst of trees [J]. Journal of Theoretical Biology, 2000, 207(3): 337–347.

[12]    Ge, Q. S., Wang, H. J., Dai, J. H. Simulating changes in the leaf unfolding time of 20 plant species in China over the twenty-first century [J]. International Journal of Biometeorology, 2014, 58(4): 473–484.

[13]    Tao, Z. X. Simulation of the distributions of five widespread tree species in China under the influence of phenological changes [D]. Beijing: University of Chinese Academy of Sciences, 2018.

[14]    Hunter, A. F., Lechowicz, M. J. Predicting the timing of budburst in temperate trees [J]. Journal of Applied Ecology, 1992, 29(3): 597–604.

[15]    Wang, H. J., Ge, Q. S., Rutishauser, T., et al. Parameterization of temperature sensitivity of spring phenology and its application in explaining diverse phenological responses to temperature change [J]. Scientific Reports, 2015, 5(1): 8833.

[16]    Chuine, I., Cour, P., Rousseau, D. D. Fitting models predicting dates of flowering of temperate-zone trees using simulated annealing [J]. Plant Cell and Environment, 1998, 21(5): 455–466.