Skip to content
Snippets Groups Projects

Compare revisions

Changes are shown as if the source revision was being merged into the target revision. Learn more about comparing revisions.

Source

Select target project
No results found

Target

Select target project
  • umr-astre/asf-challenge
  • umr-astre/asf-speed
2 results
Show changes
Commits on Source (6)
Expand all Hide whitespace changes
Inline Side-by-side
reports/Cirad_submission_1_updated.pdf
View file @ 10884b70
No preview for this file type
reports/Cirad_submission_2.pdf
View file @ 10884b70
No preview for this file type
reports/descriptive_analysis.html
View file @ 10884b70
File suppressed by a .gitattributes entry or the file's encoding is unsupported.
reports/model_farms_network.html
View file @ 10884b70
......@@ -1559,17 +1559,10 @@ background-color: #f5f5f5;
}
.lightable-classic {
border-top: 0.16em solid #111111;
border-bottom: 0.16em solid #111111;
width: 100%;
margin-bottom: 10px;
margin: 10px 5px;
}
.lightable-classic tfoot tr td {
border: 0;
}
.lightable-classic tfoot tr:first-child td {
border-top: 0.14em solid #111111;
}
.lightable-classic caption {
color: #222222;
}
......@@ -1587,6 +1580,9 @@ color: #222222;
.lightable-classic thead tr:last-child th {
border-bottom: 0.10em solid #111111;
}
.lightable-classic tbody tr:last-child td {
border-bottom: 0.14em solid #111111;
}
.lightable-classic.lightable-hover tbody tr:hover {
background-color: #F9EEC1;
}
......@@ -1595,16 +1591,9 @@ background-color: #f5f5f5;
}
.lightable-classic-2 {
border-top: 3px double #111111;
border-bottom: 3px double #111111;
width: 100%;
margin-bottom: 10px;
}
.lightable-classic-2 tfoot tr td {
border: 0;
}
.lightable-classic-2 tfoot tr:first-child td {
border-top: 3px double #111111;
}
.lightable-classic-2 caption {
color: #222222;
}
......@@ -1640,12 +1629,6 @@ border: 1px solid #EEE;
border-collapse: collapse;
margin-bottom: 10px;
}
.lightable-material tfoot tr td {
border: 0;
}
.lightable-material tfoot tr:first-child td {
border-top: 1px solid #EEE;
}
.lightable-material th {
height: 56px;
padding-left: 16px;
......@@ -1679,12 +1662,6 @@ border-collapse: collapse;
margin-bottom: 10px;
background-color: #363640;
}
.lightable-material-dark tfoot tr td {
border: 0;
}
.lightable-material-dark tfoot tr:first-child td {
border-top: 1px solid #FFFFFF12;
}
.lightable-material-dark th {
height: 56px;
padding-left: 16px;
......@@ -1715,12 +1692,6 @@ width: 100%;
margin-bottom: 10px;
color: #444;
}
.lightable-paper tfoot tr td {
border: 0;
}
.lightable-paper tfoot tr:first-child td {
border-top: 1px solid #00000020;
}
.lightable-paper thead tr:last-child th {
color: #666;
vertical-align: bottom;
reports/simulationModel.html
View file @ 10884b70
This diff is collapsed.
src/submission_2.Rmd
View file @ 10884b70
......@@ -39,34 +39,34 @@ theme_set(theme_bw())
loadd(
admin,
animal_movements,
data_structure,
distMatPigs,
# data_structure,
# distMatPigs,
# farm_exposition_D80,
farm_pred_inf_ship_CSIF_D50,
farm_pred_inf_ship_CSIF_D80,
fence,
fm_farm_inf_D80, # model prob farm infection
herd_cullings_CSIF_D80,
herd_cullings_CSPZ_D80,
herd_cullings_CSWB_D80,
herd_cullings_CSTR_D80,
hexAll,
# herd_cullings_CSIF_D80,
# herd_cullings_CSPZ_D80,
# herd_cullings_CSWB_D80,
# herd_cullings_CSTR_D80,
# hexAll,
huntZone,
huntedWBisland,
# huntedWBisland,
huntedWBzoom,
IWB_scenarios,
landcover,
lcMap,
lc_lite,
lc_lite_r,
# landcover,
# lcMap,
# lc_lite,
# lc_lite_r,
mcmcFileModel10,
movementAnimation,
movementAnimationHack,
movementFileName,
negative_hunt_samples,
obs_days_D80, ## Number of observation days for the exposure process
outbreaks,
outbreaks_at_D50,
# movementAnimation,
# movementAnimationHack,
# movementFileName,
# negative_hunt_samples,
# obs_days_D80, ## Number of observation days for the exposure process
# outbreaks,
# outbreaks_at_D50,
outbreaks_at_D80,
pig_sites,
pig_sites_pred_CSIF_D50,
......@@ -84,14 +84,14 @@ loadd(
map_expectedI_scenario3_140,
map_expectedI_scenario4_140,
tab_probAS,
tmAdminPigTypesAndOutbreaks,
tmProportionAgricultureMap,
tmProportionForestCoverMap,
tblMoves,
# tmAdminPigTypesAndOutbreaks,
# tmProportionAgricultureMap,
# tmProportionForestCoverMap,
# tblMoves,
tmAdmin,
tmRowAdmin,
# tmRowAdmin,
wb_case_density_D80,
wildBoarObsAll,
# wildBoarObsAll,
)
```
......@@ -332,22 +332,26 @@ We finish the report with a brief section outlining our main conclusions.
## Methods
The methods used for farm risk assessments are described in the previous submission report with some modifications and additions described next.
The methods used for farm risk assessments were described in the previously submitted report with some modifications and additions described here.
Apart from the infection pathways considered in the last submission, we included fomites as a possible infection pathway, for instance due to veterniarians or visitors that may introduce the virus in their equipment or vehicles, after visiting an infected farm.
In addition to the infection pathways considered in the last submission,
we included fomites as a possible infection pathway,
for instance due to veterniarians or visitors that may introduce the virus,
via their equipment or vehicles, after visiting an infected farm.
We consider the risk higher when infected farms, or farms with high probability of being infected, are in close proximity to the focal site.
We thus model this risk as a exponentially decreasing function of the distance to other farms.
We thus model this risk as a exponentially decreasing function of the distance between farms.
Since infected farms in the vicinity could have not yet been detected, we consider all neighbouring farms weighted by their exposition to infectious wild boar.
Since infected farms in the vicinity may not yet have been detected,
we consider all neighbouring farms weighted by their exposition to infectious wild boar.
<!-- (TODO: probability of infection due to WB?) -->
Another change from the previous submission is a correction of the estimation of the probability of infection via animal shipment.
Previously we considered all the recorded shipments as potential transmission events. Now, we consider only shipments that have taken place in the last 15 days. Earlier shipments can't be transmission events, otherwise any subsequent infection at a destination farm would have been detected by now.
Previously we considered all the recorded shipments as potential transmission events. Now, we consider only shipments that have taken place in the last 15 days. Earlier shipments can't be transmission events, otherwise any subsequent infection at a destination farm would have been detected.
We now also account for the culling strategy of infected herds, which was disregarded last time since there were very few.
From now, herds that were culled can only be infected again after repopulation,
We now also account for the culling strategy of infected herds, which was disregarded in our first report since there were very few.
Now we consider that culled herds can only be infected again after repopulation,
which is 50 days after infection.
This lowers the risk of infection due to wild boar exposure for some farms,
which also impacts the risk of transmission via fomites.
......@@ -602,6 +606,8 @@ tm_shape(admin, bbox = plot_area) +
) +
tm_shape(fence) +
tm_borders() +
tm_shape(huntZone) +
tm_borders() +
tm_legend(
bg.color = "white",
bg.alpha = .7,
......@@ -767,8 +773,8 @@ tm_shape(admin, bbox = plot_area) +
### Impact of additional culling measures
Table \@ref(tab:exp-pigs-cs) shows our preliminary calculations of the impact we expect from each of the proposed four control strategies. The current strategy is culling all infected farms (IF). The alternative strategies\footnote{PZ: Culling all pig herds in protection zones;
WB: Culling all pig herds located at < 3 km from positive wild boar carcasses; TR: Culling all pig herds that have traded pigs with an infected farm less than 3 weeks before detection} are in __addition__ to IF and are therefore at least as effective.
Table \@ref(tab:exp-pigs-cs) shows our preliminary calculations of the impact we expect from each of the proposed four control strategies. The current strategy is culling all infected farms (IF). The alternative strategies\footnote{PZ: culling all pig herds in protection zones;
WB: culling all pig herds located at < 3 km from positive wild boar carcasses; TR: culling all pig herds that have traded pigs with an infected farm less than 3 weeks before detection} are in __addition__ to IF and are therefore at least as effective.
```{r exp-pigs-cs}
(exp_pigs_cs <-
......@@ -818,8 +824,8 @@ The table of parameters and associated priors of the modified model are as follo
| $x_\text{Intro}$ | ASF introduction (easting) | Unif(extent of island) |
| $y_\text{Intro}$ | ASF introduction (northing) | Unif(extent of island) |
| $\beta$ | Transmission rate | Exp($\lambda=0.1$) |
| $p_\text{Home}$ | Connectivity (prob. in home pixel) | Beta(10,10) |
| $p_\text{AttractI}$ | Attractivity of infectious relative to carcass | Beta(2,2) |
| $p_\text{Home}$ | Connectivity (prob. WB in home pixel) | Beta(10,10) |
| $p_\text{AttractI}$ | Attractivity of living IWB relative to carcass | Beta(2,2) |
| $\tau_\text{P}$ | Passive search detection rate | LogNorm(lmean=0, lsd=3) |
| $\tau_\text{A}$ | Baseline active search detection rate | Exp($\lambda=10^{-7}$) |
| $\tau_\text{Phz}$ | Augmentation of $\tau_\text{P}$ in hunting zone | Exp($\lambda=10^{-7}$) |
......@@ -839,7 +845,7 @@ In our modified model an active search affects only:
1. the pixel within which an infected wild boar was detected -- we call this the _central_ pixel;
2. the pixels immediately adjacent to the central pixel -- i.e. neighbouring pixels.
Recall that our model uses a hexagonal grid with 5km distance between pixel centroids.
Recall that our model uses a hexagonal grid with a 5km distance between pixel centroids.
Thus, we utilise approximations at this resolution to investigate the difference between
1km and 2km active search radii.
......@@ -847,8 +853,8 @@ Note that $\tau_A$ is the initial value given to the active search detection rat
We further assume:
1. a 7 day delay between a case confirmation date and active search initialisation -- this delay is consistently observed in the data;
2. once initialised, active search detection rates decrease each day by a multiplicative factor of $1-\frac{1}{7}$;
3. for a 1km search radius, the daily detection probability throughout a central pixel is initialised at $p_{C1} = 1 - exp(-\tau_A)$;
2. once initialised, active search detection rates decrease each day by a multiplicative factor of $(1-\frac{1}{7})$;
3. for a 1km search radius, the daily detection probability throughout a central pixel is initialised at $p_{C1} = 1 - \exp(-\tau_A)$;
4. for a 1km search radius, the daily detection probability throughout a neighbouring pixel is initialised as $p_{N1} = \frac{\omega_{N1}}{\omega_{C1}} p_{C1}$;
5. for a 2km search radius, the daily detection probability throughout central pixels and neighbouring pixels are
$p_{C2} = \frac{\omega_{C2}}{\omega_{C1}} p_{C1}$
......@@ -858,11 +864,11 @@ respectively;
6. the weights $\omega_{Cx}$ and $\omega_{Nx}$ indicate the expected proportions of a circle of radius $x$ to fall within the central pixel,
or a given neighbouring pixel, respectively;
7. when cases are found in multiple neighbours, detection rates accumulate in a pixel up to a maximum value of
$\tau_{Ax}=-log(1-p_{Cx})$.
$\tau_{Ax}=-\log(1-p_{Cx})$.
Note,
the above procedure is identical inside the zone of increased hunting pressure,
except that the initial detection rate is augmented to $\tau_A + \tau_{Ahz}$,
except that the initial detection rate is augmented to $\tau_A + \tau_{Ahz}$
for a central pixel under the 1km search radius scenario.
......@@ -870,15 +876,14 @@ for a central pixel under the 1km search radius scenario.
### MCMC and simulation
The analysis was peformed in two steps:
The analysis was peformed in two steps.
1. first, we perfrom estimation using Markov chain Monte Carlo (MCMC);
2. secondly, we perfrom a prediction step to explore the relative efficacy of eight alternative control scenarios.
1. First, we perfrom a parameter estimation step, fitting the model to the newly available data via Markov chain Monte Carlo (MCMC).
2. Secondly, we perfrom a prediction step to explore the relative efficacy of eight alternative control scenarios.
In the estimation step, we employed the same Monte-Carlo likelihoods detailed in report 1.
In the parameter estimation step, we employed the same Monte-Carlo likelihoods detailed in report 1.
i.e. for each likelihood calculation, 500 simulations were used to estimate the expected number of
cases for each observation type (hunt -ve, hunt +ve, active search and passive search)
cases for each of four observation types (hunt -ve, hunt +ve, active search and passive search)
in each pixel, aggregated into ten-day bins.
A series of 'burn-in' runs were used to tune the algorithm.
Thereafter, inference was based on the output of 5 different MCMCs, each providing 4000 samples.
......@@ -941,12 +946,16 @@ eventually carry ASF beyond the fence.
```{r posteriorOmegaFence}
mcmcOut <- read.table(mcmcFileModel10, header=TRUE)
mySummary <- mean95(ilogit(mcmcOut$logit_omegaFence))
mySummary <- matrix(mySummary, nrow=1)
colnames(mySummary) <- c("2.5%","Mean","97.5%")
rownames(mySummary) <- ""
kable(mySummary,
format = "latex",
caption = "Summary of the posterior distribution of fence efficacy.",
digits = 3,
booktabs = TRUE
)
row.names = FALSE, ## "p_~attractI~",
format = "latex",
caption = "Summary of the posterior distribution of fence efficacy.",
digits = 3,
booktabs = TRUE
)
```
......@@ -1051,12 +1060,16 @@ kable(tab_probAS,
```{r posteriorAttractI}
mcmcOut <- read.table(mcmcFileModel10, header=TRUE)
mySummary <- mean95(ilogit(mcmcOut$logit_attractI))
mySummary <- matrix(mySummary, nrow=1)
colnames(mySummary) <- c("2.5%","Mean","97.5%")
rownames(mySummary) <- ""
kable(mySummary,
format = "latex",
caption = "Summary of the posterior distribution of the relative importance of IWB compared to carcasses in transmission events.",
digits = 3,
booktabs = TRUE
)
row.names = FALSE, ## "p_~attractI~",
format = "latex",
caption = "Summary of the posterior distribution of the relative importance of IWB compared to carcasses in transmission events.",
digits = 3,
booktabs = TRUE
)
```
Finally, we use our simulations to quantify the expected exposure to IWB at each of the outdoor pig farms in the affected area.
......@@ -1085,25 +1098,22 @@ kable(pigSitesExposure %>% st_drop_geometry() %>%
- With the current (D110) policy of disease control on farms, the disease is most likely to continue spreading locally between neighbouring farms within the fenced area and buffer zone. However, the risk of infectious animal shipments out of the area is very low according to our model (and it's assumptions).
- We identified farm id
- We identified farms with site-ids
`r tab_farms_pred_CSIF_D80 %>% filter(!detected, p_pred >= .3) %>% pull(id) %>% paste(collapse = ", ")` as likely ($p \geq .3$) to have been
infected in this period (D80 - D110).
To a lesser extent ($.1 \leq p < .3$), farm sites `r tab_farms_pred_CSIF_D80 %>% filter(!detected, p_pred > .1, p_pred < .3) %>% pull(id) %>% sort %>% paste(collapse = ", ")` also have some risk of being infected.
A few other farms have even smaller risks (see Table \@ref(tab:table-farms-pred)).
- The risk associated with pig trade is very low ($<1\%$).
- The total number of animals affected in these farms are
`r tab_farms_pred_CSIF_D80 %>% filter(!detected, p_pred >= .3) %>% pull(size) %>% sum` in the first case and `r tab_farms_pred_CSIF_D80 %>% filter(!detected, p_pred > .1, p_pred < .3) %>% pull(size) %>% sum` in the second.
- The risk associated with __pig trade__ is very low (below $1\%$ probability of infection for all farms).
- The __expected number of affected animals__, under the current control strategy (in addition to those from already detected farms) is `r (exp_nanimals = exp_pigs_cs %>% filter(CS == "IF") %>% pull("Exp pigs")) %>% round()` out of `r round((total_animals = pig_sites_risks_D80 %>% filter(!detected) %>% pull(size) %>% sum) %>% "/"(1e6))` M heads (`r round(exp_nanimals / total_animals * 100, 2)`%).
- The current control strategy on farms works well. The combination of culling of infected farms and ban trading in protection and surveillance zones are really effective to contain the spread of the disease.
- Among the additional culling measures, the one with largest impact would be culling herds in the protection zones. But the reduction is marginal (see Table \@ref(tab:exp-pigs-cs)). However, these results are preliminary as we didn't fully account for the culling strategy in the assessment of the transmission via fomites. Thus, the impacts are likely a bit larger. In particular for the culling strategy based on the proximity to infectious wild boar carcasses.
<!-- David, you removed this? -->
<!-- - Among the proposed supplementary culling measures concerning domestic pigs, the one with largest impact would be culling herds in the protection zones. But the reduction is marginal (see Table \@ref(tab:exp-pigs-cs)). However, these results are preliminar as we didn't fully account for the culling strategy in the assessment of the transmission via fomites. Thus, the impacts are likely a bit larger. In particular for the culling strategy based on the proximity to infected wild boar carcasses. -->
- Among the additional culling measures, the one with largest impact would be culling herds in the viccinity of wild boar carcasses. But the reduction is marginal (see Table \@ref(tab:exp-pigs-cs)). However, these results are preliminary as we didn't fully account for the culling strategy in the assessment of the transmission via fomites. Thus, the impacts are likely a bit larger.
- The strategy consisting on increasing the size of the active-search area around wild boar carcasses from 1 to 2 km was not assessed, but we think that it would not have a major impact.
......