SOD Spread tutorial

From GRASS-Wiki
Revision as of 17:37, 31 March 2020 by Annakrat (talk | contribs) (update)
Jump to navigation Jump to search

r.pops.spread is a model for stochastic landscape spread of the pest and pathogens. It uses PoPS (Pest or Pathogen Spread) library. In this tutorial we will use it specifically to model the spread of Sudden Oak Death tree disease. This tutorial shows how to run the model. You can download the sample dataset and simulate sudden oak death spread in the Rouge River-Siskiyou National Forest region of western Oregon.

Software

Required software includes:

Input data used in this tutorial

Download the sample dataset containing:

  • digital elevation model
  • orthophoto of study area
  • host layer
  • layer of all trees
  • roads and rivers
  • mapset containing weather coefficients

The sample dataset is a GRASS GIS Location, so it goes into your GRASS GIS Database which is usually a directory called grassdata in your home directory or your Documents directory.

Workflow

Download sample data and unzip it into your grassdata. Launch GRASS GIS and select Location PoPS_SOD_tutorial and Mapset tutorial.

Change working directory by typing cd (stands for change directory) into the GUI Console and hit Enter. You can create a directory for this exercise (outside of grassdata) anywhere where you can write and read. Navigate to that directory.

Install addon:

g.extension r.pops.spread

First, we will set computational region of our analyses to an extent covering our study area:

g.region region=small_study_area -p

Create files with lists of input maps

Here we used already prepared weather coefficients, these can be created using this workflow. Mapset weather includes raster layers per each week of a simulation. With that, we need to prepare a text file. List and write the maps in a file using g.list:

g.list -m type=raster pattern="average_weather_*" mapset=weather output=weather.txt

The file should be written to your current working directory.

Display infected data

Download color ramp into your current working directory to be used for better visualization of infected cells. Display the infected data. When displaying the infection, do not display zeros.

r.colors map=eu_infection_2019 rules=color_infected.txt
d.rast map=ortho
d.rast map=eu_infection_2019 values=0 -i
d.vect map=NHDFlowline where="FCODE >= 46006" color=30:144:255
d.vect map=roads where="FULLNAME is not NULL" color=165:159:159 width=2

Compute the spread of SOD for default values

Example output of SOD spread. Overlaid onto the orthophoto image.

Run the model using the text files created and setting only the required parameters. For this analysis we used a wind direction of NE and are looking at the first 5 years of spread from the initial 2019 infection discovery.

We export a result from a single stochastic simulation (with specified random seed).

r.pops.spread host=host total_plants=max_host infected=eu_infection_2019 start_date=2019-01-01 end_date=2023-12-31 step_unit=week reproductive_rate=1.6 weather_coefficient_file=weather.txt natural_direction=NE natural_dispersal_kernel=exponential natural_distance=242 natural_direction_strength=3 single_series=spread_sod output_frequency=yearly runs=1 random_seed=1

We list newly created output layers representing infected trees in each year of the simulation and we set a custom color ramp.

g.list type=raster pattern="spread_sod*" output=series.txt
r.colors rules=color_infected.txt file=series.txt

You can visualize the spread series in Animation Tool, do not display 0 values for infected maps.

We can run multiple stochastic runs and aggregate the results into a probability layer (0.1 if cell was infected once in 10 runs), average layer (average number of infected trees per cell) and standard deviation layer.

Here we run the process 10x and we use 2 cores for parallel processing (providing module was compiled with OpenMP support and cores are available).

r.pops.spread host=host total_plants=max_host infected=eu_infection_2019 start_date=2019-01-01 end_date=2023-12-31 step_unit=week reproductive_rate=1.6 weather_coefficient_file=weather.txt natural_direction=NE natural_dispersal_kernel=exponential natural_distance=242 natural_direction_strength=3 average_series=average probability_series=probability stddev_series=stddev output_frequency=yearly runs=10 nprocs=2 random_seed=1

Set color ramp for probability:

g.list type=raster pattern="probability*" output=series.txt --o
r.colors color=magma file=series.txt

Explore results in Map Display or using Animation Tool, omit displaying 0 values of probability.


Example output of SOD spread for kappa=0. Overlaid onto the orthophoto image.
Example output of SOD spread for kappa=4. Overlaid onto the orthophoto image.

Effect of dispersal kernel

The choice and parametrization of dispersal kernel significantly influences the spread and should be informed by calibration. The natural dispersal kernel (required) typically represents wind dispersal, additionally, we can optionally add the anthropogenic kernel which represents more human affected spread possibly over longer distances.

Each kernel is defined by type (cauchy, exponential), direction (none, N, NE, E, ...), direction strength (concentration around the direction using von mises distribution) and scale (distance).

r.pops.spread host=host total_plants=max_host infected=eu_infection_2019 start_date=2019-01-01 end_date=2023-12-31 step_unit=week reproductive_rate=1.6 weather_coefficient_file=weather.txt natural_direction=NE natural_dispersal_kernel=exponential natural_distance=400 natural_direction_strength=3 single_series=spread_distance output_frequency=yearly runs=1 random_seed=1
g.list type=raster pattern="spread_distance*" output=series.txt --o
r.colors rules=color_infected.txt file=series.txt

Compare our initial run with run with increased kernel scale.

We can then also change the direction to E and increase the strength of direction:

r.pops.spread host=host total_plants=max_host infected=eu_infection_2019 start_date=2019-01-01 end_date=2023-12-31 step_unit=week reproductive_rate=1.6 weather_coefficient_file=weather.txt natural_direction=E natural_dispersal_kernel=exponential natural_distance=242 natural_direction_strength=10 single_series=spread_direction output_frequency=yearly runs=1 random_seed=1
g.list type=raster pattern="spread_direction*" output=series.txt --o
r.colors rules=color_infected.txt file=series.txt

Finally, we can select a different kernel type.

r.pops.spread host=host total_plants=max_host infected=eu_infection_2019 start_date=2019-01-01 end_date=2023-12-31 step_unit=week reproductive_rate=1.6  weather_coefficient_file=weather.txt natural_direction=NE natural_dispersal_kernel=cauchy natural_distance=242 natural_direction_strength=3 single_series=spread_type output_frequency=yearly runs=1 random_seed=1
g.list type=raster pattern="spread_type*" output=series.txt --o
r.colors rules=color_infected.txt file=series.txt

Effect of reproductive rate

Similarly to kernel, reproductive rate should be informed by calibration. In this example we double it:

r.pops.spread host=host total_plants=max_host infected=eu_infection_2019 start_date=2019-01-01 end_date=2023-12-31 step_unit=week reproductive_rate=3 weather_coefficient_file=weather.txt natural_direction=NE natural_dispersal_kernel=exponential natural_distance=242 natural_direction_strength=3 single_series=spread_rate output_frequency=yearly runs=1 random_seed=1

Wind dispersal

Direction

Kappa influences the wind dispersal direction. Wind dispersal uses the Von Mises circular normal distribution. Large kappa makes the distribution very concentrated about the angle, specifically the wind direction specified. If kappa is set to 0 the wind dispersal direction will have no effect on distribution.

r.pops.spread host=host total_plants=all_trees infected=inf_2016 output=spread_sod_k0 wind=N moisture_coefficient_file=precipitation.txt temperature_coefficient_file=temperature.txt start_time=2016 end_time=2021 step=week kappa=0 random_seed=4

If kappa is larger than 2 the distribution becomes very concentrated around the specified wind direction.

r.pops.spread host=host total_plants=all_trees infected=inf_2016 output=spread_sod_k4 wind=N moisture_coefficient_file=precipitation.txt temperature_coefficient_file=temperature.txt start_time=2016 end_time=2021 step=week kappa=4 random_seed=4

Distance

Parameter short_distance_scale decides how far the spores can travel and it is the input to Cauchy distribution.

r.pops.spread host=host total_plants=all_trees infected=inf_2016 output=spread_sod_sc5 wind=N moisture_coefficient_file=precipitation.txt temperature_coefficient_file=temperature.txt start_time=2016 end_time=2021 step=week short_distance_scale=5 random_seed=4
r.pops.spread host=host total_plants=all_trees infected=inf_2016 output=spread_sod_sc30 wind=N moisture_coefficient_file=precipitation.txt temperature_coefficient_file=temperature.txt start_time=2016 end_time=2021 step=week short_distance_scale=30 random_seed=4


Spore rate

The spore production rate per week for each infected tree. This model uses the Poisson distribution to generate spores.

Decreasing the rate down to 1 scales down the spread a great deal. However there are still new areas of spread.

r.pops.spread host=host total_plants=all_trees infected=inf_2016 output=spread_sod_sr1 wind=N moisture_coefficient_file=precipitation.txt temperature_coefficient_file=temperature.txt start_time=2016 end_time=2021 step=week reproductive_rate=1 random_seed=4
r.pops.spread host=host total_plants=all_trees infected=inf_2016 output=spread_sod_sr6 wind=N moisture_coefficient_file=precipitation.txt temperature_coefficient_file=temperature.txt start_time=2016 end_time=2021 step=week reproductive_rate=6 random_seed=4


Treatments

We will treat the area by removing the host. We will develop several scenarios. First, no treatment is applied:

r.pops.spread host=host total_plants=all_trees infected=inf_2016 output=spread_sod_notreatment wind=N moisture_coefficient_file=precipitation.txt temperature_coefficient_file=temperature.txt start_time=2016 end_time=2021 step=week  random_seed=4

We treat the initial infection and a buffer around it. The treatments are applied at the end of the year.

r.buffer -z input=inf_2016 output=buffer_A distances=200
r.mapcalc "treatment_A = if (isnull(buffer_A), 0, 1)"
r.pops.spread host=host total_plants=all_trees infected=inf_2016 output=spread_sod_treatA wind=N moisture_coefficient_file=precipitation.txt temperature_coefficient_file=temperature.txt start_time=2016 end_time=2021 step=week  treatments=treatment_A treatment_year=2016 random_seed=4

Here we increase the buffer size:

r.buffer -z input=inf_2016 output=buffer_B distances=500
r.mapcalc "treatment_B = if (isnull(buffer_B), 0, 1)"
r.pops.spread host=host total_plants=all_trees infected=inf_2016 output=spread_sod_treatB wind=N moisture_coefficient_file=precipitation.txt temperature_coefficient_file=temperature.txt start_time=2016 end_time=2021 step=week  treatments=treatment_B treatment_year=2016 random_seed=4

Finally, we use the buffer from last example and create a large 1km wide barrier in an attempt to stop the spread. Inspect the annual results to see how effective the solution is:

r.mapcalc "barrier = if (y() > 4684000 && y() <4685000, 1, 0 )"
r.pops.spread host=host total_plants=all_trees infected=inf_2016 output=spread_sod_treatB_barrier output_series=spread_sod_treatB_barrier wind=N moisture_coefficient_file=precipitation.txt temperature_coefficient_file=temperature.txt start_time=2016 end_time=2021 step=week  treatments=treatment_B,barrier treatment_year=2016,2018 random_seed=4