Development and Validation of the Wireless Sensor Network Dataset of Leaf Area Index in Shandong Yucheng of China (2020)

Li, R. X.^1,2  Zhou, X.^1*  Lyu, T. T.¹  Tao, Z.¹  Wang, J.¹  Xie, F. T.^1,2

1. Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100101, China;

2. School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China

Abstract: With the development of communication technology, ground measurement based on Wireless Sensor Network (WSN) technology has become an important method for obtaining ground surface parameters. With its advantages of long-term and multi-point simultaneous observation, WSN provides reliable data for the validation of remote sensing satellite LAI products. However, unexpected situations such as dead batteries, communication failures or weather influence will cause the measurement data to be unstable, so it is necessary to process a large amount of raw data into relatively true value representing the ground measurement. In this paper, three nodes (0803, 0804 and 0805) are arranged in Yucheng Station of Shandong province of China from May to November 2020, and the LAI WSN system SBLX-034 is used for observation. First, we select the observation between 10 a.m. and 3 p.m., and filter out invalid data for each node. Then, according to the temporal and spatial correlation between the notes, the abnormal time data with NARX model prediction error more than 1 is eliminated, and the LSTM neural network is used to test the processed data pattern. Finally, the daily data are averaged into the measured relative true value of long-term LAI, which is used to provide data support for related research. This dataset is the processed WSN data of 0803, 0804 and 0805 notes in Shandong Yucheng Station, including: (1) geographical location of three WSN nodes in Yucheng Station; (2) daily LAI of three nodes from May to November in 2020. The storage format is .xlsx, .shp and .kml data formats, and consists of 10 files with a data volume of 49.1 KB (compressed into one file, 42.5 KB).

Keywords: Shandong Yucheng; LAI; observation nodes; daily average; ground observation

DOI: https://doi.org/10.3974/geodp.2021.02.04

CSTR: https://cstr.escience.org.cn/CSTR:20146.14.2021.02.04

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.2021.03.01.V1 or https://cstr.escience.org.cn/CSTR:20146.11.2021.03.01.V1.

1 Introduction

Leaf Area Index (LAI) is usually defined as half of the total green leaf area per unit surface area^[1], which is an important parameter to describe the structure and function of vegetation canopy^[2]. Since leaf is the main channel for energy and material exchange between the land surface and atmospheric boundary layer (such as water and carbon dioxide), LAI is also a key input factor for most land surface ecological models^[3–6]. With the rapid development of satellite technology and sensor performance, various global LAI products have been generated from satellite data using various inversion models. However, due to the accuracy limitation of data acquired by sensors and instability of inversion model, LAI products itself contains a certain degree of error^[7,8]. In order to evaluate the data quality better and expand the application filed, it is necessary to validate LAI products. The validation of remote sensing products requires to obtain the relative truth value which can represent the ground target. Due to the limitation of ground observation cost, the traditional LAI field measurement cannot meet the requirements of continuous long-term measurement. With the development of communication technology, WSN has been used in the observation experiments of sites. The ground measurement method based on WSN technology can ensure the continuous observation of parameters, realize long-term and stable multi-point synchronous observation, and facilitate the comparative analysis of multi-point data, so as to provide more reliable ground observation technique for validation.

Vegetation growth is a complex biological process, which is affected by many environmental factors. In the field experiment, LAI is measured every 5 minutes. The performance of plant leaves is different under different natural conditions. Even between 10 a.m. and 3 p.m., wind speed, wind direction and sunlight conditions will affect the tilt degree and opening angle of leaves, which will lead to unstable measurement results of LAI by WSN. At present, the research using vegetation WSN data at home and abroad usually adopts the method of setting a few days of aggregation sliding window when the time resolution requirement is not high^[9]. Most of the other researches with days or hours as the cycle adopt the method of averaging interpolation values over a period of time^[10–13], but there is a certain subjectivity in the process of removing outliers. In order to solve this problem, this paper uses the WSN data of Yucheng Comprehensive Experimental Station in Shandong province from May to November in 2020 to filter the effective data, eliminate outliers, and better retain original data under the premise of ensuring the original trend.

2 Metadata of the Dataset

The metadata of the Leaf area index daily dataset from observation nodes in Yucheng of Shandong province, China (2020)^[14] 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

3.1 Operating Principle of the LAI Sensor Network System

The LAI sensor network system SBLX-034 relies on advanced fish-eye photography technology, which uses image processing technology to quickly analyze the vegetation canopy, obtains canopy structure information in real time, and uses self-developed technology to accurately segment the image. It greatly eliminates the influence of flares under strong light conditions, improves the analysis accuracy, and obtains a variety of vegetation parameters including LAI.

Table 1  Metadata Summary of the Leaf area index daily dataset from observation nodes in Yucheng of Shandong province, China (2020)

The sensor network system consists of processing center, node manager and sensor nodes. A large number of sensor nodes are distributed with high density in the monitoring area, and the network is automatically formed. When the sensor nodes get commands from the node manager, the data collectors will perform preliminary processing, and then transmits it to the node managers, level by level among the nodes. The node manager collects and organizes the monitoring data of all sensor nodes in the network, and finally reaches the data processing center through the external network, where the administrator performs batch processing. All sensor nodes adopt automatic observation mode, and obtain observation data every 5 minutes.

This system mainly uses the single-angle method to estimate LAI. LAI is related to canopy porosity and plant leaf tilt angle, so the expression of LAI can be indirectly derived. The leaf tilt angles of common plants can generally be divided into the following five situations: ?? The leaves are distributed horizontally, and the leaf tilt angles are all 0??; ??The leaves are distributed vertically, and the leaf tilt angles are all 90??; ?? The leaves are distributed in a cone shape, and the leaf tilt angles are between 0?? and 90??; ?? The leaves are distributed in a spherical shape, which is a very uniform distribution method. In this distribution method, all leaves have no specific angle, which is a continuous random distribution; ?? The leaves are distributed in an ellipsoid shape which are all distributed on a continuous ellipsoidal surface. In these five cases, the azimuth is random. The fifth case is general, and the others are special. Campbell used the ellipsoidal distribution function to simulate the leaf tilt angle distribution^[16], and expanded the projection function as shown in the following equation (1):

                                                                                     (1)

where G(q) represented the projected acreage which was calculated by the direction q and projection in this direction ; x was the ratio of the horizontal semi-axis and the vertical semi-axis of the ellipsoid. The larger x indicated that the canopy leaf tilt angle tend to be horizontally distributed, and the smaller x indicated that tends to be vertically distributed. Wang deduced the relationship between x and the average leaf tiltaccording to the leaf tilt density function and other relation^[17], as shown in the following equation (2):

                                                                                                            (2)

The value of projection function always equaled to 0.5 when  reached 57.5?? which meant that the value of projection function can be approximately regarded as independent of the plant leaf tilt angle. The single angle method was to derive the calculation equation of LAI based on the characteristic of the projection function, as shown in the following equation (3):

                                                                                          (3)

where T represented the transmittance, and the plant LAI can be calculated according to equation (3).

3.2 Data Collection

Yucheng Comprehensive Experimental Station is located in Shandong province in warm temperate semi-humid monsoon climate zone. Winter wheat and summer maize are the main plants in the typical and representative area of the Huanghuaihai Plain. The WSN layout of Yucheng Station is shown in Figure 1. This dataset includes more than 300,000 valid raw data of the LAI WSN from May 1 to November 8, 2020. The measurement data includes: nodes latitude and longitude, data collection time, air temperature and LAI. Due to unexpected battery failure, communication failure or weather influence, some data was missing or invalid. Therefore, three nodes (0803, 0804, and 0805) with relatively better data quality were selected for processing, and there were more than 90,000 valid raw data.

Figure 1  WSN layout of Yucheng Station

3.3 Data Processing Method

The processing of LAI WSN dataset of Shandong Yucheng Station (2020) is shown in Figure 2.

First, we select the data of the three stations with observation time from 10 a.m. to 3 p.m., a total of 20,610 pieces of raw data. Since each node is affected by the environment to varying degrees, the data-lack moment is different. It is necessary to filter the moments when valid data co-occur at the three sites, this results a total of 4,576 moments with 13,728 pieces of raw data.

Then, according to the temporal and spatial correlation between nodes, the time-series neural network NARX is used to model. After adjusting the model to minimize the error, the final parameter settings are as follows: 80% of the data is used for the training of the neural network, and 10% is used to verify whether the network is generalizing, and stop training before that, 10% will be used for external verification; the hidden layer has 10 nodes; the delay number is set to 2 moments; the kernel function of the model is Bayesian regularization. The neural network modeling accuracy is shown in Figure 3. The model accuracy of the training set is 0.935,39, the model accuracy of the test set is 0.920,53, and the overall modeling accuracy is 0.933,91, indicating that the WSN measured LAI in-situ data has a strong time regularity. This time correlation can be characterized by the NARX model and the simulation effect is good. The data with the prediction error of the neural network model greater than 1 is regarded as abnormal moments. The 0803 node screened out 65 abnormal moments, the 0804 node screened out 57 abnormal moments, and the 0805 node screened out 28 abnormal moments. After the three nodes were merged, a total of 130 moments were eliminated, leaving 4446 moments with 13,338 valid original data.

Finally, the daily raw measurement data after the above processing is averaged as the relative true value of the LAI ground measurement on the day, that is a total of 170 days with 510 site data, so as to better verify the authenticity of the product on the time series.

Figure 3  Modeling accuracy of neural network

4 Data Results and Validation

4.1 Data Composition

This dataset includes: (1) geographic location data of three wireless sensor network nodes; (2) daily LAI of three nodes from May to November in 2020.

4.2 Data Products

The LAI of the note ID 0803 (36??50'2.4174"N, 116??34'56.1"E), note ID 0804 (36??50' 3.048"N, 116??34'56.316"E) and note ID 0805 (36??50'4.7682"N, 116??34'54.84"E) in Shandong Yucheng Station were daily recorded from May to November 2020.

4.3 Data Validation

In order to ensure that the processed data still has a strong spatio-temporal correlation, the LSTM time series neural network is used to model the 13,338 valid data after the abnormal value is eliminated. If the model has a accuracy, low prediction error and high prediction correlation, then it is considered that the processed data has been improved. The parameters are set as follows: the first 80% of the continuous time series is used for training, and the remaining 20% is used for testing; the solver is used for 250 rounds of training; in order to prevent the gradient from exploding, the gradient threshold is set to 1; the initial learning rate is specified as 0.005. After 125 rounds of training, the learning rate is reduced by multiplying by a factor of 0.2; the specified LSTM layer has 200 hidden units. The prediction accuracy of the trained model is 0.83. It can be considered that the processed WSN data retains the changes and relevance of the original measured data.

5 Discussion and Conclusion

This article described the data development of the LAI WSN of Shandong Yucheng Station from May to November 2020, and sets the model error threshold through the time-series neural network NARX. In consideration of retaining as much ground measurement data as possible, this article eliminates 3% of the original data, that is, extreme abnormal points. The LSTM neural network modeling test has a good data law after the elimination. Finally, the data within a day is averaged to obtain a long-term series of continuous measured ground relative truth values as data support for related research.

This dataset can be applied to the point-to-point validation of LAI products of different scales, and has obvious advantages in the test of long-term series, such as studying the changes in the accuracy of satellite products at different vegetation growth stages. In addition, due to the different environments of the three nodes, the missing data is also at different times, so the processed valid data only accounts for 64% of the original data. Among them, the lack of data caused by external factors is force majeure, but with the continuous development of communication technology and deep learning principles, how to use existing data to better interpolate missing values and improve the utilization of data is a problem worthy of research.

Author Contributions

Zhou, X. and Lyu, T. T. made the overall design for the development of the dataset; Li, R. X. processed the LAI WSN data and verified the data; All the authors wrote the data paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]      Chen, J. M., Black, T. A. Defining leaf area index for non-flat leaves [J]. Plant, Cell & Environment, 1992, 15(4): 421‒429.

[2]      Garrigues, S., Lacaze, R., Baret, F., et al. Validation and intercomparison of global leaf area index products derived from remote sensing data [J]. Journal of Geophysical Research, 2008, 113: G02028.

[3]      Liu, J., Chen, J. M., Cihlar, J., et al. A process-based boreal ecosystem productivity simulator using remote sensing inputs [J]. Remote Sensing of Environment, 1997, 62(2): 158‒175.

[4]      Andrew, D., Richardson, R. S., Anderson, M., et al. Terrestrial biosphere models need better representation of vegetation phenology: results from the North American carbon program site synthesis [J]. Global Change Biology, 2012, 18(2): 566‒584.

[5]      Sellers, P. J., Dickinson, R. E., Randall, D. A., et al. Modeling the exchanges of energy, water, and carbon between continents and the atmosphere [J]. Science, 1997, 275(5299): 502‒509.

[6]      Bonan, G. B. Land-Atmosphere interactions for climate system models: coupling biophysical, biogeochemical, and ecosystem dynamical processes [J]. Remote Sensing of Environment, 1995, 51(1): 57‒73.

[7]      Friedl, M. A., Davis, F. W., Michaelsen, J., et al. Scaling and uncertainty in the relationship between the NDVI and land surface biophysical variables: an analysis using a scene simulation model and data from FIFE [J]. Remote Sensing of Environment, 1995, 54(3): 233‒246.

[8]      Ding, Y. L. Remote sensing estimation of vegetation coverage and its authenticity verification [D]. Changchun: Graduate School of Chinese Academy of Sciences (Northeast Institute of geography and agricultural ecology), 2015.

[9]      Shi, Y. C., Wang, J. D., Qin, J., et al. An upscaling algorithm to obtain the representative ground truth of LAI time series in heterogeneous land surface [J]. Remote Sensing, 2015, 7(10): 12887‒12908.

[10]   Zhang, J. L., Liu, Q., Li, X. H., et al. Calibration and data validation of wireless sensor network [P]. Intelligent Earth Observing Systems, 2015.

[11]   Dou, B. C., Wen, J. G., Li, X. H., et al. Wireless sensor network of typical land surface parameters and its preliminary applications for coarse-resolution remote sensing pixel [J]. International Journal of Distributed Sensor Networks, 2016, 12(4): 55‒60.

[12]   Zhou, S. Y. Spatiotemporal variation of soil moisture based on improved thermal inertia model [D]. Kaifeng: Henan University, 2018.

[13]   Qu, Y., Zhu, Y., Han, W., et al. Crop leaf area index observations with a wireless sensor network and its potential for validating remote sensing products [J]. IEEE Journal of Selected Topics in Applied Earth Observations & Remote Sensing, 2014, 7(2): 431‒444.

[14]   Li, R. X., Zhou, X., Lyu, T. T., et al. Leaf area index daily dataset from observation nodes in Yucheng of Shandong province, China (2020) [J/DB/OL]. Digital Journal of Global Change Data Repository, 2021. https://doi.org/10.3974/ geodp.2021.03.01.V1. https://cstr.escience.org.cn/CSTR:20146.11.2021.03.01.V1.

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

[16]   Campbell, G. S. Extinction coefficients for radiation in plant canopies calculated using an ellipsoidal inclination angle distribution [J]. Agricultural and Forest Meteorology, 1986, 36(4): 317‒321.

[17]   Wang, Y. P., Jarvis, P. G. Mean leaf angles for the ellipsoidal inclination angle distribution [J]. Agricultural and Forest Meteorology, 1988, 43(3): 319‒321.