A Multi-factor based Grading Evaluations for Pasture: A Fuzzy Data Fusion Approach
(Rui Zhang)
1
(Dongkai Liu)
1
(Jingsha Zheng)
1
(Yan Wang)
2*
(Gang Li)
3
(Jinchuan Huang)
1
(Xulingyun An)
1
(Gengliang Li)
1
-
(College of Electronic Information and Automation, Tianjin University of Science and
Technology, Tianjin 300222, China)
-
(School of Computer and Information Engineering, Tianjin Chengjian University, Tianjin
300384, China)
-
(School of Computer Science, Inner Mongolia University, Hohhot 010021, China)
Copyright © The Institute of Electronics and Information Engineers(IEIE)
Keywords
Pasture grade evaluation, Fuzzy comprehensive evaluation, Data fusion, Analytic hierarchy process (AHP)
1. Introduction
As social productivity and technology have developed recently, the ability of humans
to understand nature is getting stronger. In order to better exploit and utilize the
nature [1], the concept of sustainable development for the environment has garnered significant
attention [2], especially in the animal husbandry [3]. A potential way to develop sustainable animal husbandry is to rationally use the
grassland resources [4]. As such, a precise evaluation of grassland quality is becoming necessary to realize
rational utilization of resources. Moreover, previous work studied a variety of methods
to evaluate the grassland and understand the degrees and reasons for grassland degradation.
However, this work has overlooked key modeling factors, such as meteorological conditions,
making it difficult to accurately reflect the grassland's actual features. Therefore,
a comprehensive evaluation method for the grassland with the consideration of various
environment factors is meaningful and mandatory. To this end, a promising but challenging
way is to apply advanced Internet of Things (IoT) technology in grassland evaluations.
Note that, as indicated in Sui et al. [5], the result by a single type of sensor is not accurate, and lacks reliability and
validity in evaluations. In this paper, with the consideration of multiple factors,
such as climate, vegetation coverage, et al., a multi-factor based grading evaluation
for the grassland is proposed. However, this type of evaluation is challenging for
several reasons. (1) Because there are no uniform criteria for grassland evaluation,
it would be difficult to design a reasonable method to quantify the evaluation result
of the grassland; (2) In practice, since the grade of the grassland is affected by
many factors, such as wind, temperature, et al., these affected factors must be considered
in the evaluation process. These factors have different units and physical properties.
Thus, it would be challenging to jointly consider these factors, and further develop
a scheme to obtain evaluation results.
In light of the above difficulties, this paper, inspired by the fuzzy data fusion
method, aims to study the grassland grade evaluation. Specifically, the uncertainties
in the evaluation are first analyzed quantitatively by determining both the affecting
factor and the evaluation set of the grassland. Afterwards, the analytic hierarchy
process (AHP) is used to determine the weight of each considered factor from both
subjective and objective perspectives. Finally, both the environment data and the
Normalized Different Vegetation Index (NDVI) data are processed by the fuzzy data
fusion. The proposed grade evaluation method can not only avoid the limitations of
using a single factor, but also overcome the inaccuracy caused by using single factor
only. Besides, this method can monitor and evaluate the grade of grasslands in a real-time
manner, and further provide a theoretical guidance for the development of smart pastures
in practice. The main contributions of this paper can be summarized as follows:
-
To overcome the drawbacks by only modeling single factor, we establish a grassland
evaluation platform by considering multiple affecting factors;
-
With the consideration of multi-factor, we propose a new evaluation method for grasslands.
To the best of our knowledge, this is the first time to propose a fuzzy data fusion
based algorithm for the grassland evaluation;
-
Experimental results are obtained using real data to demonstrate the effectiveness
of our proposed method. In comparison to the traditional methods, the proposed grade
evaluation method has superiorities in terms of real time and accuracy.
2. Related Work
To overcome the shortcomings of a single sensor, the multi-sensor data fusion algorithm
[6] has become a hot research topic today. In literature, there are many studies on grassland
environmental monitoring or evaluation [7], but they mainly focused on environmental conditions or vegetation coverage. Currently,
foreign countries, such as the United States and Britain, have shifted their focus
on investigating remote sensing technology, geographic information system technology,
and sensor information methods to analyze grassland resources [8]. In 2021, Mingdong Chen proposed the grassland monitoring and management system based
on the IoT, in which the collected information was transmitted to the cloud platform
for data computations [9]. For example, in Sun [10], Danish researchers established a comprehensive grassland environment monitoring
system for wireless sensor networks, where the factors affecting the grassland animal
husbandry were studied. To overcome the lower data fusion precision, which was caused
by the restriction of noise measurement to dissimilar-sensor, in Zhuqing et al. [11] conducted a novel data fusion method based on the weighted least square. Compared
to the errors from single-sensor and mean estimations, the measurement accuracy was
higher for dissimilar-sensor data fusion based on the weighted least square. To address
the issues of low quality and particle degeneration in particle filtering, Zhang Chuang
et al. adopted a particle filtering algorithm based on the two-stage adaptive weight
[12]. Towards resolving the problem of low precision in multi-sensor monitoring data fusion,
Guo Rong et al. employed the adaptive weighted fusion technology to monitor the soil
environment [13]. The results proved that the data collected by the system was accurate and reliable,
and the improved algorithm was easy to achieve. To meet the requirements of data compression
and data fusion in structural health monitoring in wireless sensor networks, Ji S
et al. used Bayesian methods to realize data fusion and reconstruction for sparse
signals [14], which can save the network bandwidth and energy according to a good data fusion
performance, anti-noise property and a better data compression effect. For the problem
of low accuracy and reliability of single sensor data in environment monitoring, based
on the Wireless Sensing Network (WSN) monitoring system, Liu Jing et al. used an improved
self-adaptive weighted fusion algorithm and fuzzy neural network to increase the reliability
of environment monitoring [15]. Siyami Karaca et al. developed a soil quality index (SQI) for pasture systems in
the semi-arid high plateau of eastern Van Lake using PCA and AHP-fuzzy methods integrated
with GIS and RS techniques. Their study demonstrated that these advanced methods effectively
identified crucial soil quality parameters and provided reliable data for sustainable
pasture management, aiding farm managers, researchers, and local authorities in evaluating
and planning land management practices [16]. Cai, K et al. examine a dynamic information fusion method in multi-source incomplete
interval-valued information systems, addressing the challenges arising from variations
in information sources and attributes over time [17].
In recent years, there have been many methods developed for evaluating grasslands.
do Valle Júnior, R. F. et al. used multi-temporal NDVI data to reduce the impact of
short-term climate variations on vegetation indices, thereby obtaining more accurate
long-term vegetation change trends. These trends are then validated and calibrated
using ground-based field survey data [18]. Their study concluded that there is significant pasture degradation within the environmental
protection area of the Ubeiba River Basin, particularly in terms of soil fertility
and organic matter content. Samaei, F. et al. assessed soil quality in pastures and
agricultural land using a multi-indicator weighted method [19]. They employed a randomized complete block design, collecting 120 surface soil samples.
These samples were then analyzed in the laboratory for biological, physical, chemical,
and soil nutrient indicators. Michez, Adrien et al. utilized drones to describe pasture
quality and other agricultural parameters [20]. They used a variety of drones and sensors, including the DJI Phantom 4 Pro, Mavic
2 Pro Platinum, and a Parrot Sequoia equipped with a multispectral sensor. The captured
images were processed using Structure from Motion (SFM) techniques to generate grassland
height models and biomass estimates. Marușca, T. et al. proposed an indirect method
to evaluate pasture productivity based on plant community surveys [21]. They described the species present in the turf and their participation using plant
surveys and employed the Pasture Value Index to assess pasture quality.
Although the aforementioned studies have made progress in environmental monitoring
and evaluation, most remain limited to single sensors or focus solely on individual
environmental variables. Few studies have systematically examined the impact of multi-source
data fusion on the evaluation of grassland resources. This paper proposes a novel
grassland evaluation method based on fuzzy data fusion, utilizing a self-developed
platform and multi-factor analysis to provide a more comprehensive assessment of the
overall condition of grasslands.
3. Multi-Sensor Fuzzy Data Fusion Algorithm
In the process of evaluating grassland grades, evaluation results can be unclear in
distinguishing between suitable and unsuitable conditions. As a result, in this paper,
considering multiple factors of grasslands, a fuzzy data fusion based method is proposed
to evaluate the grassland grade, and the Analytic Hierarchy Process (AHP) is used
to determine the weights of both each sensor and the Normalized Different Vegetation
Index (NDVI) [22].
The key idea of the proposed grade evaluation method is summarized in Fig. 1, where the factor set for evaluating environmental information is first determined.
Then, we calculate the weight of each factor and its corresponding vector to form
a fuzzy evaluation matrix. Finally, the formed evaluation matrix and weight vector
are used to obtain the final evaluation results. In the following, we will elaborate
on the detailed procedures of our proposed evaluation method.
Fig. 1. An illustration of the proposed evaluation flowchart.
3.1. Construction of Evaluation Indicators
The construction of the evaluation indicators is crucial to our evaluation system,
which mainly comprises the factors and the evaluation set.
3.1.1 Determination of the factor set for the grassland
The factor set is composed of several elements that can model and characterize the
evaluated object as close as possible. We assume that the factor set $V = \{v_i\},
i = 1,\cdots,n$ , where $n$ is the number of evaluated factors, which means that $n$
factors are adopted to characterize the considered object. In this paper, according
to the characters and features of the grassland, we apply ten different factors [23], where each factor $v_i$, represents the temperature, air humidity, light, pressure,
rainfall, wind speed, wind direction, NDVI, soil moisture, and soil pH, respectively
[24,
25]. The factor set can be rewritten as follows.
3.1.2 Establishment of the evaluation set for the grassland
The evaluation set consists of many elements of the final assessed result. A special
evaluation result set is generally formulated according to the needs of users. The
more the accuracy by result, the richer the result set. If the evaluation results
are $m$, $u_m$ represents an evaluation level. According to the Environmental Quality
Standard For Soils in China [?], the Natural Grassland Grading Technical Specifications
in China, and the Ambient Air Quality Standards in China [?], the evaluation results
can be divided into four grades, where $u_1$ is bad, $u_2$ indicates suitable, $u_3$
indicates more suitable, and $u_4$ is very suitable. The evaluation set can be denoted
as.
3.2. Establishment of Membership Matrix
Let $R$ be the membership matrix, where each entry is usually determined by the membership
function, i.e., $r_{ij}$, $i = 1,2,\dots,n$; $j = 1,2,\dots,m$. Since the traditional
membership functions don’t consider the randomness of subjective judgment, there are
several disadvantages, such as the difficulty in determining the function parameters,
and the ambiguity of fuzzy concept conversion after the determination. Instead, we
utilize the step function to overcome the above drawbacks. It would be much more precise
to describe the membership function by using the step function [26]. Hence, with the consideration of the practical factors in our considered scenario,
we select the trapezoidal function to obtain the membership degree. The mathematical
formulation of membership function can be expressed as
Fig. 2. The membership function.
where $x$ is the value of the environmental influence factors of the grassland; $a$,
$b$, $c$, and $d$ are the predefined parameters, respectively, which can be determined
based on the physical characteristics of the factors. Fig. 2 depicts an example of the membership function $\mu_r(x)$.
The normalization of each row is carried out to get the final evaluation matrix $R$.
Each entry $r_{ij}$ can be any real number in the range of [0, 1], i.e., $0 \le r_{ij}
\le 1$. By considering ten factors and four grading results in this paper, we can
get a $10 \times 4$ matrix, which can be expressed as
where $r_{ij}$ denotes the membership degree between the $i$-th factor in the set
$V$, and the $j$-th result in the set $U$. Note that the greater the value of $r_{ij}$,
the higher the degree of membership between $i$ and $j$.
3.3. Determination of the Weight for Each Factor by AHP
The weight refers to the degree of importance of temperature, humidity, light, and
NDVI, respectively. The weight has a great influence on the final evaluation result.
Therefore, devising a reasonable way to determine the weight of each factor is of
great importance in the grassland grade evaluation. In this paper, by considering
the following two facts: (1) the workload is non-trivial because a large scale of
data is usually required in the grassland grade evaluation; (2) the system indicator
types are numerous. So, we adopt the AHP to determine the weight of each considered
factors. AHP decomposes complex problems into multiple layers and then forms a progressive
hierarchy, which can significantly simplify the analysis of the original problem.
This method can not only obtain multiple evaluation results so as to increase the
credibility of the evaluation, but also effectively overcome the ambiguity problems
in the evaluation process.
3.3.1 Construction of the weight vector
Before the determination of the weights of all considered factors, we should construct
a judgment matrix. Followed by Hajar et al. [27], each entry, i.e., $a_{ij}$, in the judgment matrix has nine different scales, and
the definition of each scale is given in Table 1.
Table 1. Definition and note of proportional scale.
|
$a_{ij}$
|
Definition
|
|
1
|
The two factors are of the same importance when compared to each other.
|
|
3
|
When two factors are compared, one factor is slightly more important than the other.
|
|
5
|
When two factors are compared, one factor is more important than the other.
|
|
7
|
When two factors are compared, one factor is strongly more important than the other.
|
|
9
|
When two factors are compared, one factor is absolutely more important than the other.
|
|
2,4,6,8
|
Represents the important degree of the two factors is between the adjacent median
values of the above judgment.
|
Then, the judgment matrix can be formally represented as Table 2.
Table 2. Judgment matrix.
|
$A$
|
$B_1$
|
$B_2$
|
$\cdots$
|
$B_n$
|
|
$B_1$
|
$a_{11}$
|
$a_{12}$
|
$\cdots$
|
$a_{1n}$
|
|
$B_2$
|
$a_{21}$
|
$a_{22}$
|
$\cdots$
|
$a_{2n}$
|
|
$\cdots$
|
$\cdots$
|
$\cdots$
|
$\cdots$
|
$\cdots$
|
|
$B_n$
|
$a_{n1}$
|
$a_{n2}$
|
$\cdots$
|
$a_{nn}$
|
The weight of any factor, i.e., $w_i$, can be determined by using both the summation
and averaging operations in row $i$. There is
Let $w_A = (w_1,w_2,w_3,\cdots,w_{10})^T$. By implementing normalization on the vector
$w_A$, we have
where $w'_i > 0$, $i = 1,2...10$.
3.3.2 Consistency test
In order to ensure the effectiveness of hierarchical ordering, it is necessary to
check the consistency of the established judgment matrix. According to the matrix
eigenvalue decomposition, there must be a vector $X$ for matrix $A$, so that
where $\lambda$ is the eigenvalue of $A$, $X$ is the eigenvector corresponding to
the eigenvalue. We denote the largest eigenvalue of the judgment matrix as $\lambda_{max}$.
The following three steps are adopted to check the consistency of the established
judgment matrix.
a. Determination of consistency index
Consistency index ($CI$) is used to test the consistency of the judgment matrix. The
calculation of the consistency index is given by
where $n$ is the number of considered factors. The larger the value of $CI$, the greater
the degree of matrix deviation. The smaller the $CI$, the better the matrix consistency.
b. Determination of the ratio index
Similar to Wang et al. [28], we also apply the same ratio index ($RI$) values, which are listed in Table 3.
Table 3. RI values.
|
n
|
1
|
2
|
3
|
4
|
5
|
6
|
7
|
8
|
9
|
10
|
|
RI
|
0
|
0
|
0.52
|
0.89
|
1.12
|
1.26
|
1.36
|
1.41
|
1.46
|
1.49
|
c. Calculation of the consistency ratio
The consistency ratio ($CR$) is a fraction, which is determined by the ($CI$) over
the $RI$. We have
In general, when $CR \le 0.1$ for a matrix larger than $4 \times 4$, the consistency
of the judgment matrix is acceptable; when $CR > 0.1$, it is considered to fail the
consistency test. As a result, further adjustment actions should be done to the judgment
matrix in order to demand the requirement of the consistency test.
Once the weight vector and the membership matrix $R$ are obtained, the final evaluation
result can be calculated by multiplying $W$ and $R$. We have
where $B$ ($j = 1,2,3,4$) is the final evaluation set. In set $B$, each $b_j$ represents
the probability to choose the potential evaluation result $u_j$ in set $U$. Note that
our proposed evaluation method is different from others. Our proposed is not a specific
value, but a fuzzy vector so that it can bring much more information than other methods.
Finally, according to the principle of maximum membership degree, the final result
of the evaluation is determined by selecting the element in set $U$ whose corresponding
probability in $B$ is maximal.
4. Experiments and Evaluation
In this section, we will use our self-developed platform to conduct experiments to
verify the effectiveness of our proposed evaluation method. We will sense and monitor
ten different factors for the grassland: temperature, air humidity, light, air pressure,
rainfall, wind speed, wind direction, NDVI, soil humidity, and soil pH. The proposed
method is implemented through continuous monitoring and data collection related to
grasslands and plant growth.
4.1. Experimental Schematic Diagram and Platform
The considered experimental schematic diagram is shown in Fig. 3, which includes the grassland, a computer, a soil sensor (SYS-WSY), a light sensor
(BH1750FVI), a temperature and humidity sensor (Y-WSY301), a rain sensor (JXBS-3001-YL),
an atmospheric pressure sensor (CSDX-BARO), a wind speed and a direction sensor (JXBS-3001-FSFX),
a STM32F103 master control chip, and the Elastic Cloud Server (ECS). The experiment
site is on the Tianjin University of Science and Technology campus, Hexi District,
Tianjin. The ECS is provided by the Guangzhou Shenzhen Huawei Technology Co., Ltd.
Each sensor collects grassland data, while NDVI data is obtained through image processing
methods using the OpenCV vision library. These data are uploaded to the cloud platform
through the NB-IoT.
Fig. 3. The experimental schematic diagram.
Fig. 4. The considered experiment platform.
Our applied experiment platform is shown in Fig. 4. In the experiment, grassland data are gathered at different time intervals on campus.
Moreover, to validate the robustness and effectiveness of the proposed grading method,
experiments are conducted under various seasonal and weather conditions. Additionally,
to verify the superiority of our proposed method, the experimental results are compared
with those obtained through traditional human observations.
4.2. Data Collection
Fig. 5 shows three different test scenarios at the same campus site: a rainy day in September,
a cloudy day in October, and a sunny day in November 2020. In each scenario, both
environmental information and NDVI data are collected at 9:00 am, 10:00 am, and 11:00
am, yielding 30 data records for each time point. Thus, a total of 90 data records
are collected in each scenario. We randomly select 30 records as the training dataset
to evaluate the grassland grade using the proposed algorithm.
Fig. 5. (a) September. (b) October. (c) November.
4.3. Experimental Results
The experimental data is the sensored grassland information of a certain time in the
selected day from September to November in 2020, involving 10 environmental indicators
such as temperature, air humidity, light and air pressure, etc. The specific results
are shown in Table 4.
Table 4. Collected data.
|
Factor
|
September
|
October
|
November
|
|
Temperature
|
22.6$^\circ$C
|
17.3$^\circ$C
|
13.4$^\circ$C
|
|
Air humidity
|
60.9%RH
|
18.8%RH
|
21.0%RH
|
|
Light
|
31237 Lux
|
19338 Lux
|
13381 Lux
|
|
Air pressure
|
101.9 KPa
|
102.0 KPa
|
102.1 KPa
|
|
Rainfall
|
0.1 mm
|
0.0 mm
|
0.0 mm
|
|
Wind speed
|
0.3 m/s
|
2.5 m/s
|
0.2 m/s
|
|
Wind direction
|
N
|
W
|
S
|
|
NDVI
|
0.89
|
0.75
|
0.41
|
|
Soil moisture
|
38.8% RH
|
20.8% RH
|
8.2% RH
|
|
PH
|
6.07
|
5.94
|
6.01
|
Table 5. Indicators evaluation standard.
|
Factor
|
u1
|
u2
|
u3
|
u4
|
|
Temperature
|
[-10, 0]
|
[0, 10]
|
[10, 18]
|
[18, 25]
|
|
Air humidity
|
[0, 0.1]
|
[0.1, 0.4]
|
[0.4, 0.5]
|
[0.5, 0.7]
|
|
Light
|
[0-5000]
|
[5000-10000]
|
[10000,15000]
|
[15000,20000]
|
|
Air pressure
|
[90, 95]
|
[95-100]
|
[100, 101]
|
[101, 103]
|
|
Rainfall
|
[0, 0.5]
|
[0.5, 2]
|
[2, 5]
|
[5, 17]
|
|
Wind speed
|
[0, 1]
|
[1, 1.6]
|
[1.6, 2.0]
|
[2.0, 3.3]
|
|
Wind direction
|
N
|
N
|
S
|
S
|
|
NDVI
|
[0, 0.3]
|
[0.3, 0.6]
|
[0.6, 0.8]
|
[0.8, 1]
|
|
Soil moisture
|
[0, 0.2]
|
[0.2, 0.4]
|
[0.4, 0.5]
|
[0.6, 0.8]
|
|
PH
|
[0, 4.5]
|
[4.5, 5.5]
|
[5.5, 6.0]
|
[6.0, 7.5]
|
According to our previous discussions, we classify the evaluation results into four
different grades, i.e., $u_1$, $u_2$, $u_3$, $u_4$. Based on the previous research
results, the appropriate indicators system and evaluation classification interval
are sorted out in Table 5, where the values of $a$, $b$, $c$, and $d$ can be determined.
Taking the collected data in September as an example, the 10 factors in the Table 5 are taken into each corresponding membership function i.e., Eq. (3). Then, we can obtain the membership function $R$. There is
According to Wang et al. [22], factors such as temperature, NDVI, and soil pH exert the most significant influence
on grassland information compared to others. Therefore, we establish the judgment
matrix in Table 6.
Table 6. Judgment matrix.
|
A
|
B1
|
B2
|
B3
|
B4
|
B5
|
B6
|
B7
|
B8
|
B9
|
B10
|
Weight
|
|
B1
|
1
|
6
|
4
|
7
|
5
|
6
|
8
|
1/2
|
3
|
2
|
0.198
|
|
B2
|
1/6
|
1
|
1/3
|
2
|
1/2
|
1
|
3
|
1/7
|
1/4
|
1/5
|
0.040
|
|
B3
|
1/4
|
3
|
1
|
4
|
2
|
3
|
5
|
1/5
|
1/2
|
1/3
|
0.090
|
|
B4
|
1/7
|
1/2
|
1/4
|
1
|
1/3
|
1/4
|
2
|
1/8
|
1/4
|
1/6
|
0.024
|
|
B5
|
1/5
|
2
|
1/2
|
3
|
1
|
2
|
4
|
1/6
|
1/3
|
1/4
|
0.063
|
|
B6
|
1/6
|
1
|
1/3
|
4
|
2
|
1
|
3
|
1/7
|
1/4
|
1/5
|
0.050
|
|
B7
|
1/8
|
1/3
|
1/5
|
1/2
|
1/4
|
1/3
|
1
|
1/9
|
1/6
|
1/7
|
0.015
|
|
B8
|
2
|
7
|
5
|
8
|
6
|
7
|
9
|
1
|
4
|
3
|
0.244
|
|
B9
|
1/3
|
4
|
2
|
4
|
3
|
4
|
6
|
1/4
|
1
|
1/2
|
0.118
|
|
B10
|
1/2
|
5
|
3
|
6
|
4
|
5
|
7
|
1/3
|
2
|
1
|
0.590
|
Calculations are carried out to check the consistency of the judgment matrix, we have
It is worth noting that the value of the $CR$ satisfies the consistency test so that
the matrix $A$ is constructed reasonably. Then, the weight vector $W$ can be obtained
by normalizing vector $w_A$, which can be expressed as
The evaluation vector $B$ can be obtained according to Eq. (10), which is
According to $B$, and the set $U$, we have the conclusions that 0% of the grassland
may be “bad”, 1.5% may be “suitable”, 39.1% may be “more suitable”, and 59.4% may
be “very suitable”. Therefore, the highest proportion is 59.4%, which indicates that
the evaluation result of the tested grassland in Tianjin University of Science and
Technology in September is “very suitable”.
Similarly, the data collected in October and November are listed in Table 7.
Table 7. Data collected in October and November.
|
Number
|
October
|
November
|
|
Temperature
|
17.8
|
13.4
|
|
Air humidity
|
18.5%
|
21.0%
|
|
Light
|
16854
|
13381
|
|
Air pressure
|
102.0
|
102.1
|
|
Rainfall
|
0.0
|
0.0
|
|
Wind speed
|
2.8
|
0.2
|
|
Wind direction
|
W
|
S
|
|
NDVI
|
0.75
|
0.41
|
|
Soil moisture
|
21.8%
|
8.2%
|
|
PH
|
6.39
|
6.01
|
|
Result
|
suitable
|
general
|
After implementing the proposed evaluation method, we can obtain the evaluation results
in both October and November, which are shown in Fig. 6.
Fig. 6. Evaluation results.
As shown in Fig. 6, the evaluation results for September, October, and November are “Very Suitable,”
“Relatively Suitable,” and “Suitable,” respectively. In reality, as the weather gradually
cools from September to November, the vegetation in the test field decreases, and
the vegetation cover also progressively diminishes. The results of the traditional
method for scoring grassland indicators similarly reflect evaluations of “very suitable,”
“more suitable,” and “suitable” for these periods.
4.4. Comparison of Pasture Grading Evaluation Methods
In the preceding sections, we provided a detailed overview of the multi-factor grassland
grading evaluation method and its experimental results. To further evaluate the effectiveness
of this method, this paper compares it with several other commonly used experimental
approaches. Table 8 outlines the primary methods, characteristics of each experimental approach, and
Performance Analysis.
Table 8. Common pasture grading evaluation methods.
|
Method name
|
Method summary
|
Features
|
Performance analysis
|
|
1. Traditional methods
|
After observing the pasture and based on past experience and judgment, various indicators
of the pasture are scored to ultimately arrive at a comprehensive evaluation of the
pasture.
|
Lacks standardized criteria; Highly subjective; Requires extensive experience and
specialized knowledge; Demands significant time and human resources; Inefficient for
evaluating large areas of grassland.
|
Relies heavily on human observation, requiring significant manpower and time; subjective
scoring leads to inconsistent results; large pastures demand even more resources for
repeat assessments; compared to modern automated systems, slower, costlier, and less
reliable.
|
|
2. Indirect evaluation method based on plant community survey [21]
|
By converting the dominance ranks of plants into percentage participation and combining
this with forage quality and useful plant biomass indices, the grassland value and
yield of the pasture are calculated. The study first classifies grassland plants into
different forage categories and calculates the forage value and green biomass production
index based on participation and quality indices. Then, these indices are used to
estimate the green biomass yield of the pasture, which in turn determines the grazing
capacity of the pasture.
|
Small margin of error; Applicable to historical data; Requires manual data collection;
Dependent on the observer's subjective judgment.
|
The conversion of plant dominance levels to participation percentage may be influenced
by subjective human factors, leading to inconsistent results; the model simplifies
the complexity of the pasture ecosystem, neglecting the effects of dynamic ecological
factors such as seasonal variations and extreme weather; the method is primarily based
on short-term assessments, lacking support from long-term dynamic monitoring data,
making it difficult to accurately capture temporal changes in pasture yield and quality.
|
|
3. Use MODIS NDVI time series data to evaluate pastures [18]
|
Pasture degradation was divided into four stages through field investigations, ranging
from healthy pasture to fully degraded. Based on NDVI satellite image data from 2013
to 2016, nonlinear regression models were used to establish NDVI fingerprints for
different degradation stages. Meanwhile, soil parameters (such as organic matter content
and penetration resistance) were validated using variance analysis to distinguish
the different stages of pasture degradation. Finally, a map of degraded pastures was
generated by comparing the NDVI values of the pasture with the established NDVI fingerprints.
|
Enables monitoring across extensive regions and long time spans; Atmospheric conditions
can significantly affect data quality; NDVI values are influenced by various environmental
factors.
|
NDVI is greatly influenced by climatic conditions, such as cloud cover, precipitation,
and seasonal changes, which can lead to unstable readings and make it difficult to
distinguish between short-term fluctuations in pasture health and long-term degradation;
the spatial resolution of satellite images may be insufficient to detect small-scale
degradation patches in the pasture, affecting accurate monitoring of localized degradation;
variance analysis of soil parameters may not be broadly applicable across different
types of pastures or regions.
|
|
4. Evaluation method based on a drone system [20]
|
Various drones and multispectral sensors were employed to capture images from different
perspectives, followed by 3D reconstruction and spectral calibration using Agisoft
Metashape software. The digital surface model of pasture height generated by the drones
was combined with a LiDAR terrain model for correction, enabling accurate estimation
of pasture height. Subsequently, a multiple linear regression (MLR) model was utilized
to compare drone data with field measurements for modeling the dry matter biomass
and quality of the pasture.
|
Cost-effective; Highly flexible; Wide coverage; Complex data processing; Easily affected
by weather conditions.
|
Highly reliant on the precise calibration of LiDAR data; improper calibration can
severely affect height estimation and biomass prediction; drone operations are constrained
by weather conditions, with adverse factors such as wind, rain, or insufficient light
potentially leading to decreased data quality, thus delaying or interrupting data
collection; 3D reconstruction and multispectral image processing require significant
computational resources and technical expertise.
|
|
5. GrassQ web-based decision platform [29]
|
The height of the compressed grass layer was measured using a grass height compression
measurement device, while near-infrared spectroscopy was employed to determine the
dry matter and crude protein content of the grass. Multispectral data from drones
and the European Sentinel-2 satellite were utilized, employing partial least squares
regression and multiple linear regression models to estimate grass biomass.
|
Integration of multi-source data; Use of modeling techniques; High-cost equipment
requirements.
|
High degree of automation, reducing human intervention, enabling rapid processing
of large amounts of data to meet real-time requirements; combines deep learning and
physical models, offering high accuracy and strong robustness, especially performing
well in noisy environments; requires some human involvement in data collection, model
tuning, and result interpretation; high computational cost, complex algorithm, and
substantial resources needed for model design and training; Strong dependence on data
quality and quantity.
|
|
6. Multi-factor based grading evaluations for pasture
|
Analyze ten factors, including temperature, humidity, and light, and use the analytic
hierarchy process to determine the weight of each factor. Calculate the grassland
grade by constructing a fuzzy evaluation matrix and weight vector, and describe the
fuzzy membership degree with a step function.
|
Uses multi-factor comprehensive analysis; High accuracy and reliability; Flexibly
handles vague and uncertain data; Requires the deployment of a certain number of sensors.
|
Strong resistance to interference; AHP assigns appropriate weights to each factor,
ensuring a reasonable balance among factors and effectively responding to sudden changes
in the environment; the use of fuzzy evaluation matrices enhances the system's ability
to handle uncertainty, maintaining high accuracy within a probabilistic range even
when data is difficult to measure precisely; fuzzy membership degrees defined by step
functions ensure smooth transitions between pasture grades, enhancing the system's
stability and adaptability.
|
We compared several commonly used grassland evaluation methods presented in the above
table. While each method has its unique application value in specific scenarios, the
method proposed in this paper demonstrates significant advantages, particularly in
terms of efficiency and adaptability. Compared to the first traditional method and
the second method, our proposed approach automatically collects data through environmental
sensors, significantly saving time and labor while avoiding inconsistencies arising
from subjective assessments. This method considers multiple environmental variables,
such as temperature, humidity, and light, during the evaluation process, enhancing
its resilience to interference and enabling it to maintain stability and accuracy
in the face of environmental changes. In contrast, the third and fourth methods primarily
focus on a single environmental variable, often influenced by climatic conditions
and operational constraints, which can lead to unstable results and limit their comprehensiveness
and reliability. Compared to the fifth method, our proposed approach not only maintains
a high level of evaluation accuracy but also incurs lower computational costs and
equipment investment. Although the fifth method also incorporates multi-sensor technology
to achieve high-precision assessments, its reliance on large datasets and complex
algorithms poses high costs and technical barriers in practical applications.
5. Conclusion
In this study, we introduced a novel grassland evaluation method that surpasses traditional
approaches by integrating multiple factors and employing the Analytic Hierarchy Process
(AHP) combined with a fuzzy data fusion approach. This method determines the weight
of each factor using AHP and constructs a membership matrix, allowing for a comprehensive
grassland grade evaluation. Extensive experiments on real data validate our method's
effectiveness. Results show that our approach provides richer information and significantly
enhances the reliability and accuracy of evaluations.
Our method demonstrates several key advantages over traditional and commonly used
approaches. Traditional and commonly used methods often rely on a limited set of factors
and are frequently time-consuming and labor-intensive. In contrast, our approach incorporates
a broader range of factors, including temperature, humidity, light, pressure, rainfall,
wind speed, wind direction, NDVI, soil moisture, and soil pH. This comprehensive assessment
of grassland conditions is achieved while only requiring the deployment of low-cost
sensors. By utilizing the Analytic Hierarchy Process (AHP) to determine factor weights
and employing a fuzzy data fusion approach, our method effectively addresses the inherent
uncertainties and subjectivities present in traditional evaluations, resulting in
more consistent and accurate outcomes.
However, it is important to note that in Tianjin, there are no natural grasslands
available for grazing. This limitation prevented us from conducting field experiments
on natural pastures. Instead, we selected artificial turf as the subject of our experiments.
This choice allowed us to test our method effectively within the constraints of the
available environment, although it may introduce differences in evaluation compared
to natural settings. Future studies could explore the application of our method in
different types of grasslands to further validate its versatility and robustness.
Acknowledgement
This research is supported by the Inner Mongolia Science and Technology Innovation
Guide Project under Grant 2022CXYD001, and in part by Inner Mongolia Autonomous Region
Key Research and Development and Achievement Transformation Plan Project under Grant
2023YFJM0007, Key Projects of Natural Science Foundation of Inner Mongolia Autonomous
Region under Grant 2024ZD26.
References
L. Chai , Q. Zhang , J. Xie , J. Liu , R. Jiang , Dynamic evaluation mode
for ecological environment in irrigation based on theme classify and component technology,
Transactions of the Chinese Society of Agricultural Engineering, Vol. 33, No. 24,
pp. 174-181, 2017
W. Li , Evaluating ecological restoration technology: A new era for ecosystem protection
in vulnerable ecological regions in China, Journal of Resources and Ecology, Vol.
8, No. 4, pp. 313-314, 2017
A. L. Mischenko , O. V. Sukhanova , V. N. Melnikov , P. N. Amosov , Meadow
birds under waning traditional pasture animal husbandry, Biology Bulletin, Vol. 46,
No. 10, pp. 1431-1441, 2019
R. Han , H. Jiang , D. Che , N. Bao , D. Xiang , F. Liu , H. Yang ,
Z. Ban , G. Qin , Effects of environmental temperature and dietary fat content
on the performance and heat production and substrate oxidation in growing pigs, Protein
& Peptide Letters, Vol. 24, No. 5, pp. 425-431, 2017
H. M. Sui , Q. L. Wu , H. Yang , G. F. Yang , Z. Q. Ma , A remote monitoring
system for data-center based on GSM/GPRS network, Advanced Materials Research, Vol.
846-847, pp. 388-391, 2014
Z. Liu , G. Xiao , H. Liu , H. Wei , Multi-sensor measurement and data fusion,
IEEE Instrumentation & Measurement Magazine, Vol. 25, No. 1, pp. 28-36, 2022
X. Zhu , Y. Bi , J. Du , X. Gao , T. Zhang , W. Pi , Y. Zhang , Y.
Wang , H. Zhang , Research on deep learning method recognition and a classification
model of grassland grass species based on unmanned aerial vehicle hyperspectral remote
sensing, Grassland Science, Vol. 69, No. 1, pp. 3-11, 2023
G. Chai , J. Wang , G. Wang , L. Kang , M. Wu , Z. Wang , Estimating
fractional cover of non-photosynthetic vegetation in a typical grassland area of northern
China based on Moderate Resolution Imaging Spectroradiometer (MODIS) image data, International
Journal of Remote Sensing, Vol. 40, No. 23, pp. 8793-8810, 2019
M. Chen , Grassland data acquisition based on Internet of Things and cloud computing,
Proceedings of the 2021 5th International Conference on Trends in Electronics and
Informatics (ICOEI), pp. 485-488, 2021
C. Sun , Application of RFID technology for logistics on Internet of Things, AASRI
Procedia, Vol. 1, pp. 106-111, 2012
Z.-Q. Jiao , W.-L. Xiong , B.-G. Xu , Dissimilar-sensor data fusion based on
weighted least square, Journal of Jilin University (Engineering and Technology Edition),
Vol. 40, No. 3, pp. 816-820, 2010
C. Zhang , C. Guo , Multi-sensor information fusion of a particle filter on the
basis of a two-stage adaptive weight, Journal of Harbin Engineering University, Vol.
37, No. 10, pp. 1347-1352, 2016
R. Guo , P. Yang , J. Wang , Design of monitoring system for soil environment
based on data fusion algorithm, Jiangsu Agricultural Sciences, Vol. 44, No. 6, pp.
419-422, 2016
S. Ji , Z.-Y. Chen , P. Guo , Y.-J. Sun , J. Shen , J. Wang , C.-F. Lai
, Bayesian approach for multi-sensor data fusion based on compressed sensing for
wireless structural damage signal, Journal of Internet Technology, Vol. 17, No. 7,
pp. 1363-1370, 2016
J. Liu , F. Li , R. Jing , Research on multi-sensor data fusion in environment
monitoring, Journal of Shanxi Agricultural University (Natural Science Edition), Vol.
37, No. 5, pp. 340-344, 2017
S. Karaca , O. Dengiz , İ. D. Turan , B. Özkan , M. Dedeoğlu , F. Gülser
, B. Sargin , S. Demirkaya , A. Ay , An assessment of pasture soils quality
based on multi-indicator weighting approaches in semi-arid ecosystem, Ecological Indicators,
Vol. 121, Art. no. 107001, 2021
K. Cai , W. Xu , An efficient multi-source information fusion approach for dynamic
interval-valued data via fuzzy approximate conditional entropy, International Journal
of Machine Learning and Cybernetics, Vol. 15, No. 9, pp. 3619-3645, 2024
R. F. do Valle Júnior , H. E. Siqueira , C. A. Valera , C. F. Oliveira ,
L. F. S. Fernandes , J. P. Moura , F. A. L. Pacheco , Diagnosis of degraded
pastures using an improved NDVI-based remote sensing approach: An application to the
Environmental Protection Area of Uberaba River Basin (Minas Gerais, Brazil), Remote
Sensing Applications: Society and Environment, Vol. 14, pp. 20-33, 2019
F. Samaei , H. Emami , A. Lakzian , Assessing soil quality of pasture and agriculture
land uses in Shandiz County, northwestern Iran, Ecological Indicators, Vol. 139, Art.
no. 108974, 2022
A. Michez , P. Lejeune , D. Knoden , S. Cremer , C. Decamps , J. Bindelle
, Can low-cost unmanned aerial systems describe the forage quality heterogeneity?
Insight from a Timothy pasture case study in southern Belgium, Remote Sensing, Vol.
12, No. 10, Art. no. 1650, 2020
T. Marușca , Contributions to the evaluation of pasture productivity using the
floristic relevé, Romanian Journal of Grassland and Forage Crops, Vol. 19, pp. 33-47,
2019
C.-F. J. Kuo , C.-H. Lin , M.-W. Hsu , M.-H. Li , Evaluation of intelligent
green building policies in Taiwan–Using fuzzy analytic hierarchical process and fuzzy
transformation matrix, Energy and Buildings, Vol. 139, pp. 146-159, 2017
Y. Bai , J. Wu , Q. Xing , Q. Pan , J. Huang , D. Yang , X. Han , Primary
production and rain use efficiency across a precipitation gradient on the Mongolia
Plateau, Ecology, Vol. 89, No. 8, pp. 2140-2153, 2008
Y. Yang , J. Fang , W. Ma , W. Wang , Relationship between variability in
aboveground net primary production and precipitation in global grasslands, Geophysical
Research Letters, Vol. 35, No. 23, Art. no. L23710, 2008
M. F. Garbulsky , J. Peñuelas , J. Gamon , Y. Inoue , I. Filella , The
photochemical reflectance index (PRI) and the remote sensing of leaf, canopy and ecosystem
radiation use efficiencies: A review and meta-analysis, Remote Sensing of Environment,
Vol. 115, No. 2, pp. 281-297, 2011
S. H. Khairuddin , M. H. Hasan , M. A. Hashmani , M. H. Azam , Generating
clustering-based interval fuzzy type-2 triangular and trapezoidal membership functions:
A structured literature review, Symmetry, Vol. 13, No. 2, Art. no. 239, 2021
H. Xie , J. Ma , X. Fu , Y. Tian , Regional credit environment evaluation
based on analytic hierarchy process (AHP) method and fuzzy comprehensive evaluation
(FCE), Journal of Physics: Conference Series, Vol. 1813, No. 1, Art. no. 012060, 2021
C. Wang , F. Meng , X.-E. Li , L. Jiang , S. Wang , Factors affecting plant
primary productivity of grasslands: A review, Acta Ecologica Sinica, Vol. 34, No.
15, pp. 4125-4132, 2014
B. O'Brien , D. Murphy , M. S. Askari , R. Burke , A. Magee , C. Umstätter
, L. Hart , F. Oudshoorn , M. Murphy , J. Kaivosoja , R. Näsi , E. Honkavaara
, T. McCarthy , Modelling precision grass measurements for a web-based decision
platform to aid grassland management, Proceedings of the 9th European Conference on
Precision Livestock Farming, pp. 858-863, 2019
Rui Zhang received his Ph.D. degree from Tianjin University, Tianjin, China, in 2014.
He is currently an associate professor and a Master Supervisor with the Tianjin University
of Science and Technology, Tianjin, and won the honor of young and middle-aged backbone
innovative talents of universities in Tianjin. He was a Visiting Scholar with Concordia
University, Montreal, QC, Canada. More than 40 academic articles of him were published,
six articles were searched by SCI and 11 articles were searched by EI. His research
interests include wireless sensing, the Internet of Things technology and application,
spectrum detection technology application, and so on. Dr. Zhang won the third prize
of 2017 Tianjin Science and Technology Progress Award. He also presided over or completed
many projects, such as NSFC and NSSP as the main executor. He has applied for six
invention patents, authorized one, applied for more than 20 utility model patents,
authorized 16, and obtained eight software copyrights.
Dongkai Liu graduated from Hebei Agricultural University in 2017 with a bachelor's
degree. Since 2022, he has been pursuing a bachelor's degree in electronic information
from Tianjin University of Science and Technology in China. His current research focuses
on machine learning.
Jingsha Zheng graduated from Hebei Agricultural University with a bachelor's degree
in 2020, and then was admitted to Tianjin University of Science and Technology, and
received a master's degree in engineering in 2023. Her areas of interest are digital
twins and algorithms.
Yan Wang received her Ph.D. degree from Tianjin University, Tianjin, China, in 2018.
She is currently a Teacher with Tianjin Chengjian University and one of “Excellent
Enterprise Science and Technology Specialists,” Tianjin. More than 20 academic articles
of her were published, two articles were searched by SCI and 6 articles were searched
by EI. She has applied for many invention patents and utility model patents. Her research
interests include wireless sensing, mobile communication and wireless technology,
underwater sensor network, and so on. Dr. Wang won the third prize of Tianjin Science
and Technology Progress Award in 2017. She also presided over or completed many projects
as the main executor of National Spark Program and Tianjin University Science and
Technology Program.
Gang Li received the M.S. degree in information and communication systems from the
Guilin University of Electronic Technology, Guilin, China, in 2016, and the Ph.D.
degree from the Concordia University, Montreal, QC, Canada, in 2021. Since December
2021, he has been with the School of Electronic and Electrical Engineering, Wuhan
Textile University, Wuhan, China, where he is currently an Assistant Professor. His
current research interests include mobile edge computing, cooperative communications,
online algorithm, mechanism design, and machine learning. Dr. Li was awarded the International
Graduate Student Scholarship from University of Manitoba in 2018, and was the recipient
of Concordia International Tuition Award of Excellence for 2019–2020. He is a Member
of IEEE.
Jinchuan Huang received the B.E. degree in Electronic Information Engineering from
Tianjin University of Science and Technology in 2017, Tianjin, China.Since 2021,he
has been pursuing the B.E. degree in Electronic Information, Tianjin University of
Science and Technology, Tianjin China. His current research focuses on machine learning
and wireless sensing.
Xulingyun An has been pursuing a Bachelor's degree in Communication Engineering at
Tianjin University of Science and Technology in China since 2022. His research interests
lie in information processing and mobile communication networks.
Gengliang Li obtained a Bachelor's degree in Automation from Hebei University of Water
Resources and Electric Power in 2018 in Hebei. Since 2022, he has been pursuing a
master's degree in Electronic Information at Tianjin University of Science and Technology
in China. His current research focus is on machine learning and millimeter wave radar
signal processing.