การวเคราะหเครอขายการเกดรวมกนของปจจยลดผลผลตขาวในระบบนเวศ
นาชลประทานเขตพนทราบลมภาคกลางของประเทศไทย
Co-occurrence Network Analysis of Yield-limiting Factors of Irrigated Lowland Rice
Ecosystems in Central Plain of Thailand
สทธ ใจสงฆ 1) เทดพงษ มหาวงศ 2) Adam H Sparks 3) Ireneo B Pangga 4)
Sith Jaisong 1) Terdphong Mahawong 2) Adam H Sparks 3) Ireneo B Pangga 4)
Abstract
Occurrences of rice injuries caused by weeds, animal pests and pathogens differ in
various seasons and rice-growing areas. To improve pest management, there is a need to
characterize these injuries at the field scale. Detailed on-farm surveys are useful sources of data
to help understand relationships of rice injuries in farmer’s fields. One hundred and two surveys
of this study were specifically conducted in Central Plain of Thailand during 2013-2015. In this
study, co-occurrence network analysis was used for identifying important rice injuries for pest
management. Networks constructed based on the relationships of rice injuries in different
seasons were examined. The node centrality measures were applied to determine the role of
rice injuries embedded in the networks. The syndromes of rice injuries were detected by means
of network analysis. The syndromes of rice injuries (combinations of coexistence of rice injuries)
were detected based on maximal modularity score, meaning that they may share rice field
conditions. According to rice injury occurrence network of dry season, there were four
syndromes. And false smut, weed below infestation and whorl maggot injuries had high value of
centrality, indicating they were more likely to occur and co-occur with same and other
syndromes in dry season. So, they could be used as good indicators to monitor the injury
Keywords: rice, irrigated lowland rice, yield-limiting factors, co-occurrence network analysis,
Central Plain of Thailand
1) Division of Rice Research and Development, Rice Department., Chatuchak, Bangkok, Thailand 10900. 2) Syngenta 25th Fl., Tower A, Cyber World Tower 90 Ratchadapisek Road, HuaiKhwang, Bangkok 10310. 3) Centre of Crop Health, University of Southern Queensland: Toowoomba, QLD, Australia. 4) College of Agriculture, University of the Philippines Los Baños, Laguna, Philippines.
2
occurrence in dry season. While in wet season, there were four syndromes. Brown spot, whorl
maggot injuries, and weed below infestation could be indicators for monitoring pest and
disease incidence. These findings could provide the understanding, analyzing and managing
multiple rice injuries.
บทคดยอ
รปแบบการเกดรวมกนของอาการของตนขาวทมสาเหตจากวชพช แมลงและสตวศตรขาวและ
เชอจลนทรยกอโรคขาว มความแตกตางกนไปตามพนทและฤดกาล เพอเปนการเพมประสทธภาพการ
จดการศตรขาว จาเปนตองทราบลกษณะเฉพาะของรปแบบการเกดรวมกนของอาการของตนขาวตาง ๆ
เหลานทปรากฏในแปลงนา พนทราบลมภาคกลางของประเทศไทยระหวางป พ.ศ. 2556 - 2558 จานวน
102 แปลง นามาวเคราะหโครงสรางรปแบบการเกดรวมกน (co-occurrence network analysis) ของ
อาการตาง ๆ ทปรากฏบนตนขาวในแปลงนา ในฤดนาปและนาปรง คาศนยกลาง (node centrality) ของ
อาการของตนขาวจากครอขายรปแบบการเกดรวมกนบงบอกถงบทบาทและความสาคญของอาการของ
ตนขาวนน ๆ ในการเกดรวมกน โครงสรางดงกลาวสามารถใชจดกลมอาการ (injury syndrome) ทมกเกด
รวมกนไดโดย สงเกตจากคา maximal modularity ซงอาการของตนขาวทอยในกลมเดยวกน อาจจะม
สภาพแวดลอมทเหมาะสมรวมกน เชน สภาพอากาศ หรอ การเขตกรรมแบบตาง ๆ เปนตน การวเคราะห
เครอขายรปแบบการเกดรวมกนของอาการของตนขาวทพบในฤดนาปรง พบวาสามารถจดกลมอาการได 4
กลม และพบวา โรคดอกกระถน การแขงขนของตนหญาทสงตากวาตนขาว และอาการทเกดจากหนอน
แมลงวนเจาะยอดขาวมคาศนยกลางสง บงชไดวา อาการเหลานมแนวโนมทจะเกดบอยในฤดนาปรงและ
เมออาการเหลานเกด จะมแนวโนมทอาการอน ๆทอยในกลมอาการเดยวกนเกดรวมดวย สามารถใชอาการ
ทงสามนเพอเตอนหรอเฝาระวงการเกดการระบาดของโรค แมลงและสตวศตรขาวอน ๆ ในฤดนาปรง สวน
ฤดนาปเครอขายรปแบบการเกดรวมกนของอาการของตนขาว สามารถจดกลมอาการได 4 กลม และ โรค
ใบจดสนาตาล อาการทเกดจากแมลงวนเจาะยอดขาว และ วชพชทสงตากวาตนขาว มความเปน
ศนยกลางสง
คาสาคญ: ขาว นาชลประทาน ปจจยลดผลผลต การวเคราะหโครงสรางรปแบบการเกดรวมกน
ภาคกลางประเทศไทย
3
Introduction
Agricultural crops are frequently damaged by more than one species of pests and
pathogens at the same time, which consequently, affect yields. To develop effective pest
management, there is a need to shift from single pest control to a holistic one. Studies on
complex plant injuries of rice crops have led to suggestions on how to improve pest
management based on the changes in agricultural ecosystem. The characterization of complex
rice injuries, referred to as “injury profiles”, revealed that similar patterns of injury profiles shared
similar patterns of production situations. (Savary, 2000b). Thus, understanding injury profiles is
integral in developing pest management. This study aims to explore patterns of injury
occurrences based on survey data, which can be important in defining new targets and
strategies for pest management.
Co-occurrence relationships are commonly found in nature. They are important patterns
in ecosystem, which are related to niche processes that lead to coexistence and diversity
maintenance within biological communities (Willams et al., 2014). The co-occurrence patterns in
the communities may reveal groups in the co-occurring species that share similar ecological
conditions or niche. Co-occurrence analysis and network theory have recently been used to
reveal the patterns of co-occurrence and identify key entities in a system (Faust and Raes,
2012; Willams et al., 2014). For instance, in the studies of microbial ecology, these two
approaches were applied to explore the co-occurrence between microorganisms in complex
environments, ranging from the human gut to the ocean and soils (Faust and Sathirapongsasuti,
2012; Ma et al., 2016), while network topology has been proven effective in studying the
characteristics of co-occurrence patterns of geological communities and identifying the
keystone species in microbial networks (Williams et al., 2014; Barberan et al., 2012).
To date, network analysis has not been applied to explore co-occurrence patterns
between rice injuries in farmers’ fields based on crop health survey data, and to untangle the
structure of complex data among the various parameters. With the analysis of network,
correlations of rice injury co-occurrences can be better understood. Moreover, results of the co-
occurrences presented in this study showed the associations between injuries proposed by
network analysis, which would help characterize injury syndromes (the combinations of co-
occurring injuries) in these survey data.
4
Materials and methods
Crop health survey data
Crop health survey data were collected at farmers’ fields over two production seasons,
and three consecutive years (2013 to 2015) in Central Plain of Thailand: Central Plain (14o 23’-
14o 53’N, 100o 1’ - 100o 12’E), Thailand (Fig.1).The total number of surveyed fields were 102
fields (20, 20, 14, 21, 15 and 12 fields in dry season and wet season in 2013 to 2015,
respectively)
Data collection and data preparation
The surveyed fields should best represent agricultural production environment of the
region. Villages were carefully selected as representative of villages in Central Plain, Thailand.
In each village, 10 to 15 fields were then chosen as representative of the diversity of cropping
practices and environments that prevail in each village. All fields selected were farmers’ fields
with local cropping practices (including pest control measures and fertilizer application).
The survey procedure and data were based on a standardized protocol, “A survey
portfolio to characterize yield-reducing factors in rice” (Savary and Castilla, 2009). Thirty-one
rice injuries were collected including the injuries caused by weeds, animal pests and
pathogens, which are harmful to rice plants, and importantly considered to reduce yield
productivity. The injuries were evaluated at booting and ripening stages according to the survey
procedure. They were found on different organs of rice plants, depending on their natures.
Except for weed infestation and systemic injuries, information pertaining to injuries was
collected in the form of number of injured organs (leaves, tillers and panicles), which later was
made relative to the corresponding total number of organs present in the sampling units (12 hills
per field for transplanted rice crops or 12 quadrats (10 x10 cm) for direct seeded rice crop).
Injuries on leaves such as bacterial leaf blight (BLB), bacterial leaf streak (BLS), brown spot
(BS), leaf blast (LB), leaffolder injury (LF), leaf miner injuries (LM), leaf scald (LS), neck blast
(NB), narrow brown spot (NBS), rice hispa (RH), red stripe (RS), rice thrip injury (RTH), and
whorl maggot injury (WM) were determined as a proportion of injured leaves. Injuries on tillers or
hills such as dead heart (DH), dirty panicle (DP), false smut (FS), neck blast (NB), panicle mite
injury (PM), rice bug injury (RB), rat injury (RT), stem rot (SR), silver shoot (SS), sheath blight
(SHB), sheath rot (SHR), and whitehead (WH) were determined as a proportion of injured tillers
or panicles.
5
Systemic injuries such as bugburn (BB), grassy stunt (GS), hopperburn (HB), ragged
stunt (RGS), and tungro (RTG) were determined as the proportion of area affected. As for weed
infestation, the proportions of soil area covered by any weed species at two levels of the rice
canopy (above the canopy (WA) and below the canopy (WB)) were assessed in three areas of
1 m2 each. The rice injury lists and their descriptions are shown in Table 1.
Before analysis, data were summarized over time during crop growth. Two types of data
were computed, which were based on the natures of injuries as defined by Savary and Castilla
(2009). One is an area under injury progress curve (AUIPC) used for injury variables (which
present on the leaves), systemic injuries, and weed infestation. Another is the maximum level at
any of the two observations used for injury variables that can be observed on tillers, panicles,
hills, and area. The area under injury progress curve (AUIPC) (Campbell et al., 1990) was
calculated by the mid-point method using the following equation:
𝐴𝑈𝐼𝑃𝐶 = �(𝑋𝑖 + 𝑋𝑖+1)
2
𝑛−1
𝑖=1
(𝑇𝑖+1 − 𝑇𝑖)
where Xi is %age (%) of leaves, tillers, panicles, or field areas injured due to rice pests (e.g.,
brown spot and leaffolder) at the ith observation, Ti is time in rice development stage units (dsu)
on a 0 to 100 scale (10: seedling, 20: tillering, 30: stem elongation, 40: booting, 50: heading,
60: flowering, 70: milking, 80: dough, 90: ripening, and 100: fully mature) at the ith observation
and n is total number of observations.
Network construction
A statistical approach written in the R environment, version 3.3.0 (R Core Team 2016),
was designed. All scripts used in this analysis are included in the appendix and in a Github
repository (https://github.com/sithjaisong/SJ_dissertation_appendices). The methodology presented
in this chapter was adopted from Williams et al., (2014) for constructing network models of co-
occurrence patterns of rice injuries across cropping seasons (dry and wet seasons) within
Central Plain of Thailand.
As illustrated in Fig. 2, network construction involved three steps. In step 1, an incidence
matrix was obtained using the data set of rice injury occurrences. An incidence matrix lists each
injury in the data set by row (farmers’ fields) with the columns corresponding to the injuries.
6
In other words, the incidence matrix showed the co-occurrence of injuries by rice fields. In step
2, an adjacency matrix was computed from the incidence matrix using Spearman’s correlation
method. The square adjacency matrix gave the co-occurrence matrix, which contained
Spearman’ correlation coefficient between a pair of injuries. And finally, in step 3, a network
graph was drawn by connecting injuries that had a non-zero entry in the co-occurrence matrix.
The co-occurrence network was inferred based on adjacency matrix, which was
Spearman correlation matrix constructed with R function ‘cor.test’ with parameter method
‘Spearman’ (package stats) (R Core Team, 2016) used to calculate Spearman’s correlation
coefficient (ρ).
The adjacency matrix A of this network expressed co-occurrence matrix of pair of rice
injury i and j, and was written in A = [Cij], which is
where C is rank correlation coefficient (ρ from the Spearman’s correlation at p-value < 0.05)
between pairs of injures, and
where A is the adjacency (correlation) matrix, in which the rows and columns are injuries. If rows
and columns were ordered first by injury (1...n), and second by grid cells (j + 1...n + j), a
square matrix with i + j rows and i + j columns was produced.
From adjacency matrix, the networks were visualized with ‘igraph’ package (Csardi and
Nepusz, 2006) in the R environment using indirect network and the Fruchterman–Reingold
layout (Fruchterman and Reingold, 1991). Nodes in this network represented rice injuries, while
edges that connected these nodes represented correlations between injuries.
7
Topological feature analysis
Topological features of each network were measured using ‘igraph’ package. To
describe the topology of the resulted networks, a set of measures (node degree, local clustering
coefficient and betweenness) was calculated (Newman, 2006). Node degree is measured by
the number of the edges (connections) a node had. Betweenness of a node is defined by the
number of shortest paths going through a node, while the local clustering coefficients of a node
is the ratio of existing edges connecting a node’s neighbors to each other to the maximum
possible number of such edges. The network clustering coefficient measures the degree to
which nodes of the network tend to cluster together. It is also a measure of the connectedness
of the network and is indicative of the degree of relationships within the network.
Detection and characterization of modular structure in rice injury co-occurrence could
help identify groups of injuries that were closely related and often occurred together. The
networks constructed from survey data detected community structures by maximizing the
modularity measure over all possible partitions by using the ‘cluster optimal’ function of igraph
package. Nodes in the same group were called “syndrome”, which was a combination of
injuries that were closely related, and were most likely to co-occur.
The importance and the role of a node are evaluated by multiple indicators including
node degree, betweenness and local clustering coefficient. The injuries with high node degree
would indicate that the injury has relationships with many other injuries. The betweenness of an
injury in a network reflects the importance of control that the injury exerts over the relationships
of other injuries in the network. The injuries with high clustering coefficient are likely to have a
pronounced effect on injury syndrome because they can rapidly affect other injuries in a
syndrome. The importance of a node is equal to the normalized sum of its three indicators. A
candidate for core or bridge is selected from the great degree nodes or the nodes with great
betweenness respectively. Then, the role of a candidate is determined according to the
difference between its indicator’s relationships with the statistical correlation of the overall
network.
8
Result and discussion
The co-occurrence networks for crop health survey data collected in Central Plain,
Thailand are shown in Fig. 2 and 3. A connection stands for a positive (Spearman’s ρ > 0) and
significant (p < 0.05) correlation. To analyze co-occurrence network of rice injuries, node
properties, namely, node degree, local clustering coefficient and betweenness were highlighted
in the networks. Node degree is a measure of the number of connections a node has as
weighted by Spearman’s correlation coefficient. While the local clustering coefficient is a
measure of the degree to which nodes tend to cluster together. It is defined by the frequency or
the number of triangles formed by a node with its direct neighbors that are proportional to the
number of potential triangles the relevant node can form with its direct neighbors. Betweenness
measures how frequently a node lies on the shortest path between every combination of two
other nodes, indicating the importance of the node in the flow of information through the
network. (Toubiana et al., 2013).
The dry season network (Fig. 3a) was comprised of 20 associated injuries and 54
associations (edges). The network showed four groups of injury syndromes (the combination of
injuries) based on the optimal clustering algorithm. The syndrome, WH, SHR, SHB, DP, BS, RH,
NB, DH, FS, HB and RS had high clustering coefficient, which indicated that these injuries
developed complex co-occurrence relationships. Network properties (Fig. 3b) revealed FS and
WB, WM, BLS, RH, and DH were high-betweenness nodes. Compared to other injuries, BLB
and BLS had low scores on at least two centrality measures. Apparently, they were less likely to
occur as evident with its low betweenness and even to occur with other injuries, as suggested
by its low degree and clustering coefficient. Because of high value of centrality, FS, WB and
WM could be used as indicators to monitor the injury occurrence in dry season.
In the wet season, the co-occurrence network of rice injuries (Fig. 4a) revealed 4
syndromes, 22 injuries and 68 significant relationships (edges). Syndrome of BLS, RS, HB, SHB,
SHR and WM were placed closer to each other than other syndromes based on the structure
and clustering coefficient (Fig. 4b). Based on network structure and betweenness, BS, WM and
WB could be indicators for monitoring pest and disease incidences in this season.
9
Conclusion
To establish priorities and strategies for pest management program, there is a need for
characterization of multiple pests (Mew et al., 2004). Network analysis was used to characterize
co-occurrence patterns of rice injuries from crop health survey data, which were collected from
the farmers’ fields in Central Plain, Thailand for three consecutive years (2013-2015). The
resulting networks, which revealed varying structures, depicted the co-occurrence patterns of
rice injuries in different seasons. From the structures, the networks showed syndromes of rice
injuries (groups of injuries) that are co-occurring injuries in the networks. Moreover, based on
three components of node centrality measures (node degree, clustering coefficient and
betweenness), networks suggest important injuries that in turn could be used for monitoring and
predicting possible trends and occurrence of related injuries under certain conditions. The
networks also revealed the clusters of rice injuries, which were considered as injury syndrome
may share common favorable conditions. The structure of dry season network revealed high
value of centrality of false smut, Weed below infestation and whorl maggot injuries. These
indicated that they have high potential to be associated with other injuries. When these injuries
occurred, other injuries also are likely to occur. So, they could be used as indicators to monitor
the injury occurrence in dry season, whereas in wet season, based on network structure and
betweenness, brown spot, whorl maggot injuries, and weed below. Thus, they could be
indicators for monitoring pest and disease incidences in this season. This information was used
to better understand the variation of rice injury co-occurrence, and to develop more effective
strategies of pest management, specifically those that are based on seasons.
Acknowledgements
Syngenta provided the financial support and collected the on-farm survey data as part
of the SKEP collaboration.
10
References
Barberan, A., T.S Bates, E. O. Casamayor and N. Fierer. 2012. Using network analysis to
explore co-occurrence patterns in soil microbial communities. The ISME Journal.
6: 343–351.
Csardi, G. and T. Nepusz. 2006. The igraph software package for complex network research.
International Journal of Complex Systems 1695: 1-9.
Faust, K., J.F. Sathirapongsasuti, J. Izard, N. Segatta, D. Gevers, J. Raes and C. Huttenhower.
2012. Microbial co-occurrence relationships in the human microbiome. PLoS
Computational Biology 8: 145.
Faust, K. and J. Raes. 2012. Microbial interactions: from networks to models. Nature Reviews
Microbiology 10(8): 538–550.
Fruchterman, T.M. and E.M. Reingold. 1991. Graph drawing by force-directed placement.
Software: Practice and Experience 21: 1129-1164.
Ma, B., H. Wang, M. Dsouza, J. Lou, Y. He, Z. Dai, P.C. Brookes, J. Xu and J. A. Gilbert. 2016.
Geographic patterns of co-occurrence network topological features for soil microbiota at
continental scale in eastern China. The ISME Journal. 10: 1891-1901.
Newman, M. E. J. 2006. Modularity and community structure in networks. Proceedings of the
National Academy of Sciences 103: 8577–8582.
R Core Team 2016. R: A Language and Environment for Statistical Computing [Computer
software manual]. R Foundation for Statistical Computing, Vienna, Austria. Retrieved
from http://www.r-project.org/
Savary, S. and N. Castilla. 2009. A survey portfolio to characterize yield-reducing factors in rice.
IRRI Discussion Paper No 18. 32 p.
Savary, S., L. Willocquet, F.A. Elazegui, P.S Teng, P. Van Du,D. Zhu, Q. Tang, S. Huang, X. Lin,
H. Singh, et al. 2000. Rice pest constraints in tropical asia: characterization of injury
profiles in relation to production situations. Plant Disease 84(3): 341–356.
Williams, R.J.,A. Howe and K.S. Hofmockel. 2014. Demonstrating microbial co-occurrence
pattern analyses within and between ecosystems. Frontiers in microbiology 5: 358.
11
Fig. 2 Workflow used for constructing a network that represents the co-occurrence of rice
injuries based on survey data. (A) sub-setting survey by season, and production
environment, (B) constructing correlation matrix using Spearman’s correlation method,
and (C) building network models.
Fig. 1 Map showing the surveyed farmers’ fields in Central Plain, Thailand.
(A) (B) (C)
12
Table 1 List of variables describing individual fields in surveys of rice injuries
Variable Acronym Description Unit1/
Bugburn BB Maximum %age of bugburn in a one-sqm area %
Bacterial leaf blight BLB Area under the progress curve of the mean %age of leaves with bacterial leaf
blight
%dsu
Bacterial leaf streak BLS Area under the progress curve of the mean %age of leaves with bacterial leaf
streak
%dsu
Brown spot BS Area under the progress curve of the mean %age of leaves with brown spot %dsu
Deadheart DH Maximum %age of tillers with deadheart %
Dirty panicle DP Maximum %age of panicles with dirty panicle %
False smut FS Maximum %age of panicles with false smut %
Grassy stunt disease GS Maximum %age of grassy stunt disease in a one-sqm area %
Hopperburn HB Maximum %age of hopperburn in a one-sqm area %
Leaf blast LB Area under the progress curve of the mean %age of leaves with leaf blast %dsu
Leaffolder injury LF Area under the progress curve of the mean %age of leaves with leaffolder injury %dsu
Leaf miner injury LM Area under the progress curve of the mean %age of leaves with leaf miner injury %dsu
Leaf scald LS Area under the progress curve of mean %age of leaves with leaf scald %dsu
Neck blast NB Maximum %age of panicles with neck blast %
Narrow brown spot NBS Area under the progress curve of the mean %age of leaves with narrow brown
spot
%dsu
Panicle mite injury PM Maximum %age of tillers with panicle mite injury %
Rice bug injury RB Maximum %age of panicles with rice bug injury %
Ragged stunt disease RGD Maximum %age of ragged stunt disease in a one-sqm area %
Rice hispa injury RH Area under the progress curve of the mean %age of leaves with rice hispa injury %dsu
Rat injury RT Maximum %age of tillers with rat injury %
Red stripe RS Area under the progress curve of mean %age of leaves with red stripe %dsu
Tungro RTG Maximum %age of tungro in a one-sqm area %
Rice thrip injury RTH Area under the progress curve of the mean %age of leaves with rice thrip injury %dsu
Sheath blight SHB Maximum %age of tillers with sheath blight %
Sheath rot SHR Maximum %age of tillers with sheath rot %
Stem rot SR Maximum %age of tillers with stem rot %
Silver shoot SS Maximum %age of tillers with silvershoot %
Weed above WA Area under the progress curve of the mean %age weed infestation (ground
coverage) above the crop canopy
%dsu
Weed below WB Area under the progress curve of the mean %age weed infestation (ground
coverage) below the crop canopy
%dsu
Whitehead WH Maximum %age of panicles with whitehead %
Whorl maggot injury WM Area under the progress curve of the mean %age of leaves with whorl maggot
injury
%dsu
1/dsu :development stage units on a 0 to 100 scale
13
(a)
(b)
Fig. 3 Co-occurrence network analysis of survey data in dry season at Central Plain. (a) Co-occurrence
network model (b) Three node centrality measures.A :Node degree, B :Clustering coefficient, and
C :Betweenness.
14
(a)
(b)
Fig. 4 Co-occurrence network analysis of survey data in wet season at Central Plain. (a) Co-occurrence
network model (b) Three node centrality measures . A :Node degree, B :Clustering coefficient, and
C :Betweenness.