Abstract
Peste des petits ruminants (PPR) is a viral transboundary disease of
small ruminants that causes significant damage to agriculture. The
disease has not been previously registered in the Republic of Kazakhstan
(RK). This paper presents an assessment of the susceptibility of the RK
territory to the spread of this disease in case of its importation from
infected countries. Ordinary Least Squares (OLS) and Geographically
Weighted Regression (GWR) models trained on the PPR outbreaks in China
were used to rank municipal districts of the RK in terms of the risk of
PPR spread. Spatial density of outbreaks was used as a risk indicator
while a number of socio-economic, landscape and climatic indicators were
considered as explanatory variables. The Exploratory Regression tool was
used to reveal a best combination of independent variables based on
specified thresholds of R-squared, variables’ multicollinearity and
residuals’ normality and autocorrelation. The small ruminants’ density,
the maximum green vegetation fraction, the annual mean temperature, the
road length and density as well as the cattle density were the most
significant factors. Both OLS and GWR demonstrated nearly similar model
performance providing a global adjusted R-squared of 0.61. Applied to
the RK, the models show the greatest risk of PPR spread in the
south-eastern and northern regions of the country, especially within
Almaty, Zhambyl, Turkistan, West Kazakhstan and East Kazakhstan regions.
As part of the study, a country-wise survey was carried out to collect
data on the distribution of livestock population the RK, which resulted
in compiling a complete geo-database of small ruminants’ holdings in the
country. The research results can be used to form a national strategy
for the prevention of the importation and spread of PPR in Kazakhstan
through targeted monitoring in high-risk areas.
Keywords : Peste des Petits Ruminants, Republic of Kazakhstan,
Ordinary Least Squares regression, Geographically Weighted Regression,
Explanatory Regression, ArcGIS.
Introduction
The preservation of sustainable epizootic welfare of the country’s
livestock in relation to threats caused by especially dangerous
diseases, such as Peste des petits ruminants (PPR), is the most
important task of veterinary science and practice, which is of paramount
importance in protecting the health and life of people, providing the
population with high-grade and safe food products, and providing
industry with quality raw materials.
PPR is a highly contagious viral disease that affects small ruminants
and causes 30% to 70% mortality among the infected animals (Mahapatra
et al., 2006;http://www.fao.org/ppr/en/).
Due to the significant socio-economic damage and negative impact on food
security in many countries of the world, PPR is included in the list of
priority diseases of the Five-Year Plan of Action of the FAO / OIE World
Framework Program for the progressive control of transboundary animal
diseases aimed at PPR elimination by 2030 (Global Strategy for the
Control and Eradication of PPR, 2015). The high degree of this disease’s
presence among countries close to Kazakhstan provides a need to analyze
the threat of importation and subsequent spread of PPR in the country
(Ahaduzzaman, 2020).
PPR is a typical transboundary disease: first reported in West Africa,
1942, the disease has steadily expanded its range over the years. So, in
the period from 2001 to 2011, the disease spread in 56 countries: 35 in
Africa and 21 in Asia (Munir, 2015), and by 2016 it was registered in
more than 70 countries and became endemic in the regions of Northern and
Eastern and West Africa, the Near and Middle East, South, Central Asia
and Western Eurasia (Balamurugan et al., 2014; Bouchemla et al., 2018;
Zhuravlyova et al., 2020). These regions are home to more than 80% of
the world’s sheep and goats; products such as goat’s milk, lamb and wool
play a huge role in the welfare of many families. FAO estimates that
about 300 million small farming families worldwide depend on small
ruminants because sheep and goats are critical assets for poor rural
households, providing them with protein, milk, fertilizer and wool, and
often representing substantial social capital and access to financial
loans (Global Strategy for the Control and Eradication of PPR, 2015).
According to official information provided by the OIE, the epizootic
situation with PPR in the world remains rather tense (OIE WAHIS, 2020).
Despite intense international, regional and national efforts to combat
the disease, most developing countries around the world are non-free
from PPR, constituting a constraint to free, liberal global trade in
animals and livestock products
(http://www.fao.org/ppr/en/).
The PPR epizootic situation in the Central Asia countries neighboring
Kazakhstan is ambiguous. Thus, in Armenia, Azerbaijan and Turkmenistan,
outbreaks of PPR have not been previously registered, however monitoring
studies and preventive vaccination of 33-70% of animals susceptible to
PPR in risk zones are being carried out (Koshemetov et al., 2014;
Amirbekov et al., 2020). In Uzbekistan and Kyrgyzstan, isolated
outbreaks of the disease were previously recorded, and active monitoring
and preventive vaccination are currently being carried out (Yapici et
al., 2014; Fine et al., 2020).
The disease annually leads to large economic losses. For example, a
series of epidemics in Kenya in 2006-2008 caused death of 1.2 million
small ruminants, resulting in losses of more than US $23.5 million, and
milk production declining by 2.1 million liters. In general, the annual
damage from PPR is estimated at US $ 1.4–2.1 billion (Kihu et al.,
2015; Jones et al., 2016; Bardhan et al., 2017).
For the successful prevention of PPR, regional studies of the epizootic
process are important, which will allow to study the features of its
manifestation in a specific territory, in specific natural-geographical
and socio-economic conditions, with subsequent forecasting as a reliable
foundation for managing the epizootic process through the development
and implementation of effective contra-epizootic measures.
According to the official data of the State Veterinary Service of the
RK, PPR has never been registered in the country before, although there
are some publications indicating the isolation of PPR pathogen from sick
sheep and goats in the RK in 2003 and 2014 (Lundervold et al., 2004;
Kock et al., 2015). Socio-economic, organizational, structural and
geopolitical changes in Kazakhstan during post-Soviet time, as well as
the expansion of international trade, economic and cultural ties lead to
additional risks of importation of dangerous infectious diseases’
pathogens into its territory, including through cross-border areas. The
Republic of Kazakhstan is historically characterized by unique natural
conditions for the preservation of the activity of many known and the
formation of new focal areas that can cause a sudden aggravation of the
epizootic situation in the region.
The purpose of this research is the assessment of the susceptibility of
the RK territory to the PPR spread, treated as risk of PPR spread in
case of the pathogen importeation into the country.
- Materials and methods
- Study area
The area of interest for modeling the risk of PPR spread was the entire
territory of the Republic of Kazakhstan (RK, Kazakhstan). RK is a
land-locked state in Central Asia, occupying an area of 2,725,000
km2 with a population of 18.28 million.
Administratively, the RK is divided into 14 units of the first level –
regions (“oblasts”). Each of the regions in turn, is sub-divided into
second-level administrative units – districts, whose area ranges from
283 to 138,663 km2 (mean of 15,780
km2). Totally, there are 173 districts in the RK (Fig.
1).
PPR outbreaks in People’s Republic of China (PRC, China) were used to
train regression models. Total area of China is 9,598,962
km2, while population exceeds 1,404.328 million. The
second (prefectural) level of administrative divisions is presented by
333 units with an area from 490 to 473,671 km2 (mean
of 27,670 km2).
In terms of area, China occupies the third place in the world, while
Kazakhstan – the ninth place. Both countries share a land border of
more than 1,600 km.
<Figure 1 about here>
Modeling method
Since the RK is currently free from PPR, no outbreaks were available to
validate any internally built model. Thus, to rank Kazakhstan districts
as to the risk of PPR spread, a regression model trained on outbreaks in
China was applied. Two types of regression were considered, namely
Ordinary Least Squares (OLS) and Geographically Weighted Regression
(GWR). These models reveal a quantitative relationship between the
indicator under study (dependent variable) and a set of potential
influencing factors expressed as geographic variables. The key
difference between the two models is that GWR looks for local variation
of the study relationships by estimating regressions within a certain
area around each feature thus allowing to account for non-stationarity
of variables (Brunsdont et al., 1998).
Second level administrative units (districts in Kazakhstan, counties or
prefectures in China) were chosen as the analysis units for creating a
regression model. For each unit, the number of PPR outbreaks, number of
infected animals and explanatory factors were extracted (see below).
To build the regression model, the following epidemic indicators per
administrative unit were tested as a dependent variable: 1) the total
number of PPR outbreaks; 2) the density of outbreaks per unit area; 3)
the total number of infected animals; 4) the PPR prevalence. To
normalize distribution of the candidate dependent variables, we used
log-transformed values.
The following geographically distributed landscape, climatic and
socio-economic characteristics for each administrative unit were
selected as potential explanatory factors based on the analysis of
scientific publications on PPR spatial and temporal modeling (Ma et al.,
2017, 2019; Mokhtari et al., 2017; Cao et al., 2018; Gao et al., 2019;
Ruget et al., 2019): 1) total road length; 2) road density; 3) average
small ruminants density; 4) average cattle density; 5) average
population density; 6) average elevation; 7) annual mean temperature; 8)
annual precipitation; 9) maximum green vegetation fraction. Measurement
units and data sources are shown in Table 1.
<Table 1 about here>
The pre-selection of dependent and independent variables was performed
using the Exploratory Regression procedure (ArcGIS, Esri). This
procedure iteratively fits multiple ordinary least squares regression
models (OLS) using various combinations of potential explanatory
variables and provides a recommendation for choosing the optimal
combination based on a set of statistical metrics, which include
Akaike’s Information Criterion (AIC), Adjusted R-squared
(R2) and Variance Inflation Factor (VIF). Besides,
regression residuals are checked for normality using Jarque-Bera test,
and for spatial autocorrelation using Moran’s I test. A combination of
dependent and independent variables was chosen that provided the lowest
AIC, R2 > 0.5, VIF < 2,
Jarque-Bera p-value > 0.1 (that does not allow rejecting
the null-hypothesis about residuals’ normality) and SA p-value
> 0.1 (that does not allow rejecting the null-hypothesis
about residuals’ spatial randomness) (Mitchell, 2005).
Further, the OLS and GWR were both fitted using the selected combination
of dependent and independent variables. For GWR, the adaptive kernel
radius was chosen determined by the input features density based on
cross validation (Fotheringham et al., 2002). The quality of the models’
fit was assessed using the global adjusted R-squared and AIC. Residuals
were tested for spatial autocorrelation by means of Global Moran’s I
test. The best model was chosen that provides a lower AIC, higher
R2 and demonstrate random spatial distribution of
residuals.
The regression coefficients of the resulting model were further used to
build a predictive model for the entire model region including China and
Kazakhstan. Predicted spatial PPR outbreaks’ density values for
Kazakhstan were ranked in four classes by quantiles conventionally named
as ‘very low’, ‘low’, ‘medium’ and ‘high risk’, and mapped using a
choropleth method.
Data sources
Data on PPR outbreaks in China for the period 2007 - 2020 (as of
30.08.2020) were obtained from the FAO EMPRES-I database
(http://empres-i.fao.org/eipws3g/).
During this period, 289 PPR outbreaks were registered in the China (Fig.
2). For 248 (86%) of these outbreaks, the OIE is indicated as a source
of data, while the rest 14% are attributed to “national authorities”.
Of those outbreaks, the vast majority (258; 89%) were recorded in 2014.
Within each outbreak, a number of infected animals ranges from one to
3290 with a mean of 152.
<Figure 2 about here>
Detailed data on the small ruminants’ distribution in Kazakhstan were
obtained during the national wise survey undertaken by the research team
members in 2018 – 2019. The survey included a series of expeditionary
trips coordinated with regional veterinary authorities. During the
survey, complete information was collected about livestock farms in the
RK, including geographic coordinates and the population size that
enables mapping the livestock population at any required level of
spatial resolution. A total of 2,478 small ruminants holdings (farms)
were georeferenced with 18 to 167,918 (mean 8,988) animals. The total
population of small ruminants in the RK thus sums up to 22,271,628 head
providing a district-level density of zero to 277 (mean 9)
head/km2. The density of small ruminants’ population
at the district level overlaid with the farms’ locations is presented in
Figure 1.
Software
Data processing, geospatial analysis and visualization were conducted
using ArcMap Desktop 10.8.1 geographic information system with an
extension Spatial Analyst (Esri, USA).
- Results and discussion
- Selection of dependent and independent variables
Testing of various combinations of dependent and independent variables
using the Exploratory Regression procedure has shown that the best
results are obtained by using the log-transformed PPR outbreaks’ density
per unit area as a dependent variable, as well as: road length and
density, small ruminants density, maximum green vegetation fraction,
annual mean temperature and cattle density as explanatory variables.
This combination of variables provided Jarque-Bera p-value of 0.104 and
Spatial Autocorrelation p-value of 0.868 with VIF of 1.378 and was
further used to build the OLS and GWR models.
Fitting the regression models
Both OLS and GWR models demonstrated nearly similar performance in
explaining the PPR distribution in China with OLS providing slightly
lower AIC (table 2). Testing the residuals using the Moran’s I global
autocorrelation tool returned low z-values with high p-values that
suggests residuals’ spatial distribution close to a normal one and
allows speaking of a fairly good fit of the models. A kernel radius for
the GWR model was set to include 123 nearest features that comprises all
Chinese prefectures with PPR outbreaks thus making the GWR model
virtually identical to the OLS one. For the further modeling, the OLS
was used.
<Table 2 about here>
Table 3 shows the regression coefficients of the OLS model. For each
coefficient, the standard error and the standardized value of the
coefficient are also indicated, which enables clear comparison between
the relative contribution of each variable.
<Table 3 about here>
Analysis of the obtained coefficients allows making conclusions about
the largest contribution of the vegetation index (MGVF) that
demonstrated a positive relation with the dependent variable, so that
more vegetation was found to be more suitable to the PPR spread, which
is obviously explained by higher number of small ruminants in vegetated
areas. The road length showed a negative influence on the outbreaks’
density while road density had a positive effect, which may suggest the
higher suitability of pastoral regions with poorly developed road
network, and potentially be resulted by a bias introduced by prevailing
PPR outbreaks reporting from smaller and densely populated prefectures
of China. Small ruminants density was also among the most contributing
factors demonstrating a positive relation with the PPR, which can be
naturally thought as an indicator of contact rate between herds. Annual
mean temperature was found to be positively associated with the PPR
density that may suggest higher suitability of warmer areas to small
ruminants breeding, particularly that using open pastures. The least
important predictor was a cattle density, which is negatively associated
with the density of PPR outbreaks. This may be explained by the
competitive use of pasture areas by both species: taking into account
the technology of their keeping, feeding and characteristics of
pastures, cattle and small ruminants are mostly bred in different
regions, therefore the presence of cattle may indicate an insignificant
number of small ruminants, and vise versa.
Extrapolation of the model to the entire model territory
Using the obtained coefficients (Table 3), the OLS model was
extrapolated to the entire territory of China and Kazakhstan. We
excluded administrative units of ‘cities’ type, which normally do not
have small ruminants population, but may result in overprediction
because of relatively small area and high road density. The resulting
predicted distribution of PPR outbreaks’ density is shown in Fig. 3. In
general, the model demonstrates good agreement with the distribution of
PPR outbreaks in China, 2007–2020, and suggests overall lower
suitability to the PPR spread in Kazakhstan as compared to China.
<Figure 3 about here>
Creating a risk map for the Republic of Kazakhstan
Allocation of the territory of Kazakhstan from the obtained model allows
constructing a risk map specific for the RK (Fig. 4).
<Figure 4 about here>
This map shows the increased expected density of outbreaks in areas
along the western and especially south-eastern borders of the RK. These
areas are characterized by a higher density of small ruminants (Fig. 1).
In particular, Turkistan, Zhambyl and Almaty regions are historically
leading areas in terms of the small ruminants breeding. In these areas,
there is also a high probability of the importation of the disease from
the border regions of Turkmenistan, Kyrgyzstan and Uzbekistan, which are
characterized by a high density of the small ruminants and the presence
of sporadic outbreaks in the past.
Model limitations
The constructed model demonstrates satisfactory yet not very high
ability to explain variation of the input data, which can be partly
explained by the need of extrapolation of dependencies obtained for
another country to the territory of Kazakhstan that was determined by
the absence of PPR outbreaks in the RK, which could be used for direct
model validation. The geographical and socio-economic risk factors used
in the model are the most general indicators and, perhaps, not
exhaustive for explaining the observed patterns of the epizootic
situation in China. Since PPR does not belong to environmental diseases,
the registration of outbreaks and the spread of the disease depends to a
large extent on the transmission of the virus during transport links,
interstage and interfermal contacts, which can be introduced into the
model only indirectly through the geographical factors used. As another
model limitation, we can mention that the data on PPR outbreaks in China
may be incomplete due to possible underreporting of PPR from less
populated prefectures of central and western parts of the country.
It should also be noted that the information of small ruminants’
population distribution used for modeling is the most accurate and
relevant for the Republic of Kazakhstan, since it was obtained by direct
collection of georeferenced data in 2018 - 2019, while for China we used
modelled data obtained by the dasymetric mapping based on 2010 national
survey results.
In general, it can be noted that the created model demonstrates
reasonable distribution of PPR spread risks across the RK districts that
would be expected based on the information on the density of small
ruminants’ population and intensity of economic links, and thus can be
used by the national veterinary authorities as a scientific support of
the national strategy of PPR prevention. Development of a more accurate
risk assessment study, as well as assessing the pathways of possible
importation of the disease require building a more comprehensive model
and taking into account a larger number of factors, both landscape and
socio-economic (in particular, building a network on animal movements
requires movements data that are not currently collected in the RK on a
regular basis), as well as knowledge of the current epizootic situation
and the results of monitoring studies on PPR in countries bordering the
Republic of Kazakhstan.