Time Series of Land Surface Phenology Dataset in Central Asia (1982–2015)

Ma, Y. G.  Liu, S. H.^*

College of Resource and Environment Sciences, Xinjiang University, Urumqi 830046, China

Abstract: Central Asia (30°N–60°N, 50°E–100°E) is one of the arid and semi-arid regions of the world. It is extremely important to understand the response of surface vegetation to climate change under water stress. Based on a Global Inventory Monitoring and Modeling System (GIMMS) ndvi3g.v1 data, the threshold-based and inflexion-based methods were used to extract surface phenology data in Central Asia from 1982 to 2015. The spatial reference system was a geographic coordinate system at a spatial resolution of 0.083,3° (about 8 km). The dataset consisted of two groups of data files: Group V1 and Group V2. The processing of the Group V1 data file was as follows. First, the NDVI data were filtered and reconstructed by a double logistic function, and then the start of season (SOS), end of season (EOS), and length of season (LOS) were calculated by a 20% dynamic threshold method. The Group V2 data file was developed by reconstructing the NDVI time series data using a series of piece wise logistic functions, and then the SOS, EOS, and LOS were calculated by the inflexion-based method. The dataset was archived in .hdr, .img and .tif formats in 408 data files, with a data size of 126 MB (compressed to 75.6 MB in one file).

Keywords: land surface phenology; remote sensing; Central Asia; 1982–2015

1 Introduction

Vegetation phenology is widely collected, summarized, and analyzed to determine the characteristics and trends of historical climate change ^[1]. The vegetation information can be used as an intermediate parameter in scientific simulations and the calculation of regional and even Earth system energy and material transfers^[2–3].

Land surface phenology data is relatively limited. Vegetation phenological data sources can be divided into in-situ observations, remote sensing phenology, and phenology observations based on digital camera or unmanned aerial vehicle (UAV) data. The three observation methods have their own advantages and disadvantages. For the remote sensing phenology method, vegetation phenological information is determined by extracting characteristic values from the growth curve. This method can obtain phenological information data at the landscape or regional scale, but there are scale differences with human observation data, which need to be verified carefully^[4]. Digital cameras and UAVs use modern technology to measure phenology. With finer scale, but it is difficult to generate long-time series data. It is therefore mainly used for the cross validation and multi-scale analysis of alternative human observation phenology data and remote sensing phenology^[5‒7].

Central Asia is one of the world’s typical arid and semi-arid regions. Vegetation phenology studies in this region could further our understanding of the response mechanism of plant phenology to climate change under water stress. the GIMMS ndvi3g.v1 data is the longest time span vegetation data currently available. The threshold and inflection point methods were applied to this dataset to extract a phenological dataset for the start of season (SOS), end of season (EOS), and length of season (LOS) in Central Asia from 1982 to 2015.

2 Metadata of the Dataset

The metadata of the “Time series of land surface phenology dataset in Central Asia (1982–2015)”^[8] are shown in Table 1.

Table 1  Metadata summary of “Time series of land surface phenology dataset in Central Asia (1982–2015)”

3 Methods

3.1 Algorithm Principle

The GIMMS ndvi3g.v1 data was first fitted with logical functions and reconstructed:

                                                       (1)

where NDVI_max is the maximum NDVI, NDVI_min is the minimum NDVI, t is the Julian day (d), VI(t) is the NDVI value at day t after fitting, and A and B are fitting parameters. The difference between NDVI_max and NDVI_min is the amplitude of the vegetation growth curve.

(1) Dynamic threshold method^[10]

When the NDVI value of Julian day t is greater than or equal to an amplitude of 20%, t is regarded as the SOS, and when the NDVI value of Julian day t is less than or equal to the amplitude of 20%, t is regarded as the EOS.

                                       (2)

(2) Inflection point method^[11]

The curvature is calculated as follows:

                                                                   (3)

                                                                                          (4)

The rate of change of curvature is:

                         (5)

By calculating the local extremum of the rate of change of the curvature, the position of the inflection point can be determined, thereby determining the SOS and EOS.

3.2 Technical Route

The dataset was developed by pre-processing the GIMMS ndvi3g.v1 data, obtaining the NDVI time series data for Central Asia from 1982 to 2015, and then obtaining the plant growth fitting curves of each grid through logical function fitting. The threshold method and the inflection point method were used to extract two phenological datasets, which included the three phenological parameters of SOS, EOS, and LOS. Finally, data integration was performed to complete the LSP_CA Central Asia region.

4 Results and Validation

4.1 Data Composition

The “Time series of land surface phenology dataset in Central Asia (1982–2015)” includes SOS, EOS, and LOS data that were derived by two phenological extraction methods, i.e., the

Figure 1  Technical route of the dataset development

threshold and inflection methods. Data description, data format, number of files, and the amount of data are shown in Table 2.

(1) Data header file (.hdr). Contains all header file information of the corresponding raster data, including data type, number of rows and columns, and projection information.

(2) Phenological raster data. The SOS, EOS, and LOS extracted by the threshold method are LOG_20% _SOSyyyy.img, LOG_20% _EOSyyyy.img, and LOG_20% _LOSyyyy.img. The SOS, EOS, and LOS obtained by the inflection point method are LOG_inflexion_ SOSyyyy.tif, LOG_inflexion_EOSyyyy.tif, and LOG_inflexion_LOSyyyy.tif. Here yyyy represents the four-digit year, and each phenological raster data has a corresponding data header file (.hdr), which can be operated in ENVI software. For import and export, the raster data value represents the corresponding Julian day time of the corresponding phenological parameter of the year; the invalid value is 0; and the spatial coordinate system is the latitude and longitude as geographic coordinates.

Table 2  List of files in the “Time series of land surface phenology dataset in Central Asia (1982–2015)”

4.2 Results

Figure 2 shows partial images of the phenological parameters obtained by the threshold and inflection point methods. It was found that the SOS, EOS and LOS extracted by the threshold method reflected the spatial distribution throughout the entire Central Asian region, and the results using the threshold method were smoother than those obtained using the inflection point method This was largely due to the obvious fluctuations of the NDVI data at the beginning of the growth curve, which in turn led to large changes in the rate of curvature change and poor spatial smoothness.

4.3 Data Analysis

The temporal change of phenological data is the focus of phenological research. To analyze the time trend characteristics of the two sets of data in this region, a Mann-Kendall trend test^[12] was applied to analyze the phenological data obtained by the two methods in the 34-year period of the dataset (Table 3).

The results obtained by the two methods showed that changes in the SOS, EOS and LOS for about 73%–85% of Central Asia were not significant over the 34-year period, but there were advances detected in the SOS in 15% of the area investigated. There was a significant advancing trend, but the results of the two methods were not consistent for the analysis of the EOS. The use of the two methods produced the opposite results for the change in the LOS. The results of the threshold method showed that the LOS had shortened in 7.3% of the area but had increased in 18.8% of the area. The inflection point method indicated that the LOS was longer in more than 21% of the area investigated, while it had shortened in only 3.5% of the area.

There is currently only limited phenological data available for Central Asia. The only publicly released data are two datasets of the global MCD12Q2 (versions 005 and 006), which were developed using MODIS data. Version 005 mainly uses the inflection point method for data calculation, while version 006 mainly uses the threshold method calculation. Compared with MCD12Q2 data, the new dataset developed in the present study has a higher time resolution and is more suitable for the long-term change analysis of large-scale mean surface area; however, because this data has a coarse spatial resolution, it is more susceptible to surface heterogeneity than MCD12Q2 data. Due to the influence of surface heterogeneity, large errors may occur in areas with a diverse range of surface cover types or that are strongly affected by human interference.

Figure 2  Spatial distribution of SOS, EOS, and LOS of the dataset

Table 3  Change of the land land surface phenology among 34 years (198–2015) (%)

5 Discussion and Conclusion

The 34-year-old land surface phenology database developed in this dataset will be useful for researchers working on regional climate change and surface plant ecosystem monitoring. However, the phenology parameter set based on GIMMS ndvi3g.v1 data has a coarse spatial resolution of 8 km, which is certain to be affected by the problem of spatial heterogeneity. The accuracy of the data cannot be accurately evaluated from the perspective of manual observations. Through field inspections, areas could be selected in parts of Central Asia with a high surface consistency and long-term unchanged ground cover types. The collection of field observation data in these areas would lead to a greater accuracy of this dataset. In addition, the use of other phenological datasets, such as MCD12Q2 products, or the development of new phenological product datasets based on long-term mid-resolution remote sensing time series data, such as Landsat and sentries, is the next stage of this work. This new dataset can identify the phenological changes at the spatiotemporal scale, but it is susceptible to the influence and interference of changes in surface cover types. Before using this data for further analysis, the use of historical land use cover data, such as MCD12Q1, to remove the impact of different land cover types will improve the reliability of the research outputs.

Author Contributions

Liu, S. H. designed the algorithms of the dataset and evaluated the data. Ma, Y. G. contributed to the design of the research framework, data processing, data analysis, and writing of the data paper. Liu, S. H. reviewed the paper.

References

[1]       Chuine, I., Yiou, P., Viovy, N., et al. Historical phenology: grape ripening as a past climate indicator [J]. Nature, 2004, 432(7015): 289‒290.

[2]       Migliavacca, M., Sonnentag, O., Keenan, T. F., et al. On the uncertainty of phenological responses to climate change, and implications for a terrestrial biosphere model [J]. Biogeosciences, 2012, 9(6): 2063.

[3]       Randerson, J. T., Hoffman, F. M., Thornton, P. E., et al. Systematic assessment of terrestrial biogeochemistry in coupled climate-carbon models [J]. Global Change Biology, 2009, 15(10): 2462‒2484.

[4]       Cleland, E., Chuine, I., Menzel, A., et al. Shifting plant phenology in response to global change [J]. Trends in Ecology & Evolution, 2007, 22(7): 357‒365.

[5]       Nagai, S., Ichie, T., Yoneyama, A., et al. Usability of time-lapse digital camera images to detect characteristics of tree phenology in a tropical rainforest [J]. Ecological Informatics, 2016, 32: 91‒106.

[6]       Vrieling, A., Meroni, M., Darvishzadeh, R., et al. Vegetation phenology from Sentinel-2 and field cameras for a Dutch barrier island [J]. Remote Sensing of Environment, 2018, 215(15): 517–529.

[7]       Zhang, X., Jayavelu, S., Liu, L., et al. Evaluation of land surface phenology from VIIRS data using time series of PhenoCam imagery [J]. Agricultural and Forest Meteorology, 2018, 256‒257: 137‒149.

[8]       MA, Y. G., Liu, S. H. Time series of land surface phenology dataset in Central Asia (1982‒2015) [DB/OL]. Global Change Data Repository, 2020. DOI: 10.3974/ geodb.2020.01.05.V1.

[9]       GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. DOI: 10.3974/dp.policy.2014.05 (Updated 2017).

[10]    Jonsson, P., Eklundh, L. Seasonality extraction by function fitting to time-series of satellite sensor data [J]. IEEE Transactions on Geoscience and Remote Sensing, 2002, 40(8): 1824‒1832.

[11]    Zhang, X., Friedl, M. A., Schaaf, C. B., et al. Monitoring vegetation phenology using MODIS [J]. Remote Sensing of Environment, 2003, 84(3): 471‒475.

[12]    McLeod, A. I. Kendall rank correlation and Mann-Kendall trend test [Z]. R Package “Kendall”, 2005. http://www.stats.uwo.ca/faculty/aim.