disruptsc 1.0.2

A spatial agent-based model to simulate the dynamics supply chains subject to disruptions

DisruptSupplyChain Model

Installation

Git clone the project

You need python. The required libraries are in the "dsc-environment.yml" file. To set the right environment, we recommend using an "Anaconda" or "Miniconda" distribution. Then you can create the adequate environment by doing:

 conda env create -f dsc-environment.yml

Note that I observed that, in some situations, this command gets stuck in the "solving environment" step. I think we need to define the package versions in the yml. To be updated...

Calling the script

Activate the right environment:

conda activate dsc

The model is launched by the following command:

python code/main.py <region> <optional argument>

Argument region corresponds to the region that is studied, for instance, Tanzania. This string should be the same as:

• the input subfolder containing the data in the input folder, e.g., "input/Tanzania". This folder should be created manually by the user.
• the suffix of the user-defined parameters yaml file in the parameter folder, e.g., "parameter/user_defined_Tanzania.yaml". This file should be created manually by the user.

The optional argument is made to allow for some part of the initial state to be resued in a subsequent call:

• same_transport_network_new_agents: the transport network is reused, but new agents, supplier-buyer links, logistic routes are generated
• same_agents_new_sc_network: the transport network and the agents are reused, but new supplier-buyer links, logistic routes are generated
• same_sc_network_new_logistic_routes: the transport network, agents, supplier-buyer links, are reused, but new logistic routes are generated
• same_logistic_routes: everything is reused The files are cached in the tmp folder as pickle files. Note that if you change region, you will first need to call the script without optional argument.

The default parameters are defined in the "parameter/default.yaml" file. This file also defines the filepath to the input files. To change a parameter of filepath value, write it into the "parameter/user_defined_.yaml" file.

Concepts

Geographic structure

The model focuses on a country or group of countries. A country is divided into some administrative units (e.g., regions, districts, communes, cantones, etc.) We need to pick a relevant administrative level for which the economic data are compiled. We call it "regions" in the remainder of the text.

Regions are labelled using the country-relevant classification, for instance, the string XXXXXX where X are digits. We call this id the region code. It should always be a string, even if it consists of digits only, e.g., 123456 as string, not as integer.

The transport network is composed of edges and nodes. Edges are LineString, Nodes are Points. Edges are identified by an integer ID, so are nodes.

Each region is associated with one node in the transport network.

Sector structure

We use one sector classification, such as ISIC Rev. 4, which depends on the data available on firms and on the input-output table available, and on their granularity. Each sector is identified by a trigram, such as MTE for Manufacture of Textiles. Note that imports are labeled by IMP.

Objects

The main objects are the economic agents. There are three classes of agents:

• firms
• households
• countries

Firms, households, and countries are associated with nodes in the transport network. There is at most one firm per sector per region, one household per sector per region. Countries are associated to nodes which are located outside the country.

Implementation choices

• For geographical data file, we use GeoJSON and not shapefiles. They should not be projected and use the epsg:4326 coordinate reference system.

Different versions of the model with different inputs needs

The model can be used in different version, depending on how supply-chain linkages are inputed. Such versions are controlled by the firm_data_type parameter, which can takes three values:

• disaggregating IO uses a global input-table which gets disaggregated using regional economic data
• mrio use a multi-regional input-output table
• supplier buyer network uses supplier buyer network data [not yet implemented]

Inputs

Transport inputs are needed whatever the versions

Note that the filepath of each data files are defined in parameter/filepaths_default.py. These default filepaths can be overriden using the parameter/filepaths.py path.

Within a subfolder Transport, there is a series of GeoJSON files and one YAML file.

Transport network files (GeoJSON)

There should be two GeoJSON files per transport mode, one for nodes and one for edges. Acceptable transport modes are : 'roads', 'airways', 'maritime', 'waterways', 'multimodal'. For instance, for roads:

• roads_nodes.geojson
• roads_edges.geojson

There is only one file for the multimodal layer, which describe the edges. There is no multimodal node.

The edge's geometry is LineString, the node's geometry is Point.

Nodes should contain at least the following attributes:

• id: int, one unique id per mode

Edges should contain at least the following attributes:

• id: int, one unique id per mode
• end1: int, the id of the node of the same mode where the linestring starts (leave empty for multimodal edges)
• end2: int, the id of the node of the same mode where the linestring ends (leave empty for multimodal edges)
• surface: paved or unpaved. If unknown you can put everything as "paved"
• class: class category, for instance primary, secondary, etc. If unknown you can leave it empty.
• km: length in km
• multimodes (for multimodal edges only): define which type of multimodel link,
• for instance "roads-waterways" or "railways-maritime"
• capacity: maximum handling capacity per time step. If unknown you can leave it empty.
• disruption: flag here the disruption scenarios that affect this edge. Leave empty if unused.

Nodes and Edges should not contain the following attributes:

• index

Based on these input files, the model creates one networkx.Graph object representing the transport network.

Note for multimodal edges:

• in multimodes, it should start with mode1-mode2. The order of the modes does not matter.
• can write something after, e.g., 'roads-maritime-dom', but not before, e.g., 'roads-dom-maritime'.

Transport Parameters (YAML)

A YAML file transport_parameters.yaml with the following structure. It needs to be adjusted to the transport modes modelled.

speeds: #km/hour
  roads
    paved: 31.4
    unpaved: 15
  railways: 23
  waterways: 7
  maritime: 35

loading_time: #hours 
  roads-waterways: 5
  roads-maritime: 12
  roads-railways: 12
  railways-maritime: 24
  waterways-maritime: 24

variability: #as fraction of travel time
  roads:
    paved: 0.01 
    unpaved: 0.075
  railways: 0.02
  waterways: 0.02
  maritime: 0.005
  multimodal:
    roads-waterways: 0.1
    roads-maritime: 0.1
    roads-railways: 0.1
    railways-maritime: 0.1
    waterways-maritime: 0.1

transport_cost_per_tonkm: #USD/(ton*km)
  roads:
    paved: 0.053
    unpaved: 0.1
  railways: 0.04
  waterways: 0.029
  maritime: 0.0017

loading_cost_per_ton: #USD/ton
  roads-waterways: 2.2
  roads-maritime-shv: 2.2
  roads-maritime-vnm: 2.2
  roads-railways: 5
  railways-maritime: 2.2
  waterways-maritime: 2.2

custom_cost: #USD/ton
  roads: 27
  waterways: 27
  maritime: 27

custom_time: #hours
  roads: 1.5
  waterways: 6
  maritime: 2
  multimodal: 2

travel_cost_of_time: 0.49 #USD/hour

variability_coef: 0.44 #USD/hour

Constraints on transport modes (optional)

An additional file transport_modes.csv can be used to prevent specific supply-chains flows from taking specific transport modes. To be further described

Inputs for disaggregating IO version

Three subfolders are needed:

• National: country-wide data derived from input--output tables or other country-wide data
• Subnational: subnational data, typically, for each region, population, location of the main city, sector size
• Trade: import, export, transit flows

National - Sector Table

A CSV file sector_table.csv. One row = one sector. Required columns:

• sector: the sector's trigram, for instance AGR
• type: one of 'agriculture', 'mining', 'manufacturing', utility', 'transport', 'trade', 'service'
• output: the total yearly output, derived from the input-output table. The unit should be the same as defined in the parameter monetary_unit_in_model.
• final_demand: the total yearly final demand, derived from the input-output table. The unit should be the same as defined in the parameter monetary_unit_in_model.
• usd_per_ton: the average monetary value, in USD, of a ton of good. This value can be computed from UN COMTRADE data, in which trade flows are both reported in tons and in USD. Set to 0 for sectors whose type is 'utility', 'transport', 'trade', 'service'
• share_exporting_firms: the percentage of the firms that export per sector. This value can be derived from country-specific data. Without good data, we can simply use for instance the share of export per sector.
• supply_data: the attribute of the 'region_data' file that the model will use to disaggregate the sectoral data
• cutoff: the cutoff value to apply. We will not model any firm for this sector in regions whose supply_data is below the cutoff

Note that, when no network data are used, the model will create firms based on the geospatial economic data. To speed up computation, firms that would be too small are dropped.

Example:

sector	type	output	final_demand	usd_per_ton	share_exporting_firms	supply_data	cutoff
AGR	agriculture	415641365	246213152	950	0.16	ag_prod	3.50E+06
...	...	...	...	...	...	...	...

National - Technical coefficients

Technical coefficients are derived from the symmetric industry-by-industry input-output table. The matrix of technical coefficients are sometimes directly available. If only the supply and use tables are available, additional steps are required, which are not described here. Make sure that the sectors in rows and columns are the one to be used in the model, i.e., that there are consistent with the firm-level data.

For the model, a CSV file tech_coef_matrix.csv should be provided with the following structure.

Example:

	AGI	FOR	...
AGI	0.11	0	...
FOR	0.01	0.2	...
...	...	...	...

National - Inventory Duration Target

Firms hold inventory of inputs, quantified by the number of weeks the firm can continue producing without new supply, called "inventory duration target". Those targets are defined nationally for each combination of input*sector.

Example:

input_sector	buying_sector	inventory_duration_target
TRD	AGR	3.5
...	...	...

Subnational - Region Data

In this file are summarized the socioeconomic and spatial data on economic production for each region. A GeoJSON file is expected.

• admin_code: a string with the administrative code of the region
• geometry: a 'Point', which represent where the firms and household modelled for this region will be located. Typically the location of the largest city in the region, or, as a second best option, the region's centroid.
• population: the number of people in the region
• economic supply data: at least one column per sector which captures the size of the sector in this region. For instance, there could be the number of employees in the manufacturing of basic metals, the total sales of construction, and the value of production of agricultural products. The column name should corresponds to the values of the supply_data column in the sector table

A GeoJSON file economic_data.geojson with Points and the following attribute table:

Example:

admin_code	nb_workers_MAN	nb_workers_ELE	crop_production	...	geometry
0101	124	12	465120	...	Point(-45.15631, 0.48654)
...	...	...	...	...	...

Trade - Country Transit Matrix

A CSV file representing a matrix. Country codes are row and column headers. It represents the yearly trade flow from the row country to the column country that goes through the country-under-study's transport network. The unit should be the same as defined in the parameter monetary_unit_in_model.

Example:

...	BDI	COD	...
BDI	4563	4516	...
...	...	...	...

Trade - Import Table

A CSV file representing a matrix. Country codes are row headers. Sector codes are column headers. It represents the yearly imports from the different countries, in row, to the domestic sectors, in column. The unit should be the same as defined in the parameter monetary_unit_in_model.

Example:

...	AGR	FOR	...
BDI	132	0	...
...	...	...	...

Trade - Export Table

A CSV file representing a matrix. Country codes are row headers. Sector codes are column headers. It represents the yearly exports from the domestic sectors, in columns, to the foreigh countries, in row. The unit should be the same as defined in the parameter monetary_unit_in_model.

Example:

...	AGR	FOR	...
BDI	2	456	...
...	...	...	...

Inputs for the mrio version

One subfolder is needed: Network

MRIO (CSV)

A CSV file which contains the multi-regional input-output data.

It is a double-indexed matrix, with the following structure :

...	...	AFG	AFG	FRA	FRA	AFG	FRA	ROW
...	...	Sector1	Sector2	Sector1	Sector2	final_demand	final_demand	exports
AFG	Sector1	...	...	...	...	...	...	...
AFG	Sector2	...	...	...	...	...	...	...
FRA	Sector1	...	...	...	...	...	...	...
FRA	Sector2	...	...	...	...	...	...	...
ROW	imports	...	...	...	...	...	...	...

The first row/column indicates the "admin_code" of the regions. The second row/column indicates the "sector_code" of the sectors.

region_table (GeoJSON)

A GeoJSON file with Point geometries indicating the origin-destination point associated to each region. It should contain the following attributes:

• admin_code: should correspond to the admin_code of the mrio.csv
• geometry: Point

sector_table (CSV)

A CSV file sector_table.csv. One row = one region*sector. Required columns:

• sector: the combination of <region_code>_<sector_code>
• type: one of 'agriculture', 'mining', 'manufacturing', utility', 'transport', 'trade', 'service'
• output: the total yearly output, derived from the input-output table. The unit should be the same as defined in the parameter monetary_unit_in_model.
• final_demand: the total yearly final demand, derived from the input-output table. The unit should be the same as defined in the parameter monetary_unit_in_model.
• usd_per_ton: the average monetary value, in USD, of a ton of good. This value can be computed from UN COMTRADE data, in which trade flows are both reported in tons and in USD. Set to 0 for sectors whose type is 'utility', 'transport', 'trade', 'service'
• share_exporting_firms: the percentage of the firms that export per sector. This value can be derived from country-specific data. Without good data, we can simply use for instance the share of export per sector.
• supply_data: the attribute of the 'region_data' file that the model will use to disaggregate the sectoral data
• cutoff: the cutoff value to apply. We will not model any firm for this sector in regions whose supply_data is below the cutoff

Note that, when no network data are used, the model will create firms based on the geospatial economic data. To speed up computation, firms that would be too small are dropped.

Example:

sector	type	output	final_demand	usd_per_ton	share_exporting_firms	supply_data	cutoff
AFG_agriculture	agriculture	415641365	246213152	950	0.16	ag_prod	3.50E+06
...	...	...	...	...	...	...	...

Inventory Target (CSV)

Firms hold inventory of inputs, quantified by the number of weeks the firm can continue producing without new supply, called "inventory duration target". Those targets are defined nationally for each combination of input * country*sector.

Example:

input_sector	buying_sector	inventory_duration_target
ESP_Electrical and Machinery	ESP_Recycling	13
ESP_Electrical and Machinery	SYR_Others	14
ESP_Electrical and Machinery	TUR_Public Administration	10
ESP_Electrical and Machinery	USR_Agriculture	8
...	...	...

Indicatives steps for data prep for disaggregating IO version

Requisite:

• Say you have sales data per sector per region, sales_csv, preprated from business census data. It should be coded with a relevant sector classification.
• You should also have a list of countries, or group of countries, to represent the rest of the world. Typically, isolate the neighbooring countries, or those that very relevant in terms of trade, and then group the rest based on continents. Code the countries with a trigram. For instance, VNM for Vietnam, THA for Thailand, ASI for Asia, OCE for Oceania, AFR for Africa, EUR for Europe, AME for America.

Steps:

1. Prepare an initial sector table

• Prepare a sector_csv, with all the sector and their corresponding identifier (recommended identifier: a trigram, e.g. CER for cereal production or MTE for manufacturing of textile)
• The column with the identifier should be called sector

2. Prepare the region data

• Prepare a region_geojson, with all region, their adequate identifier (typically the administrative code, beware whether you want it to me string or integer), the point of their main city as geometry
• Make sure the coordinates system is EPSG 4326 (unprojected, degrees)
• The column with the admin code should be called admin_code
• Make sure that in your sales_csv, the regions and sectors are adequately coded
• Create the region_data_geojson (see above) from region_geojson and sales_csv, which includes population per region

3. Add the sector cutoff to the sector table

• Add a column supply_data in the sector table: it should correspond to the name of the attribute of the regional data (region_geojson) contains the economic data of that sector
• Add a column cutoff in the sector table: only region with data above this cutoff will be modeled as production place for that sector

4. Prepare the technical coefficient matrix

• Get the symmetric industry-by-industry input--output (IO) table with the adequate sector classification
  • Note that you may need the adjust the sector classification of the IO table yourself, and/or the one of the sales data. What is important is that the same sector classification is used throughout the data
• Open it in excel (or equivalent) and control that you get the GDP using the 3 approaches (income, production, expenditure)
• Create the technical coefficient matrix, add the row for imports, coded IMP

5. Add output, final demand, share_exporting_firms to the sector table

• Use the input--output table of step 2 to retrieve total output, final demand, and % of export (exports / total output) per sector
• Add those three columns to the sector table with the following names: output, final_demand, share_exporting_firms
  • Note that we use here % of export as a proxy for the proportion of firms that exports. If better data are available on this, use it instead of % of export

6. Prepare the import and export tables

• Get trade data for the relevant year, choose the country as "Partner", get the csv file, such as comtrade_2016_EcuadorAsPartner_AG2.csv
• Prepare a country mapping table, which maps the 3-digit ISO code of all countries listed in the ComTrade data to your country classification
• Prepare a sector mapping table, which maps the ComTrade commodity code (see https://unstats.un.org/unsd/classifications/Econ) into your sector classification
• Map the trade data into your country and sector classification
• Create the import table and the export table

7. Prepare the transit matrix

• Identify the pair of neighboring countries, whose physical trade flows could go through the country under study (e.g., Peru - Colombia for Ecuador)
• For each pair, get the total import and export of good between the two countries of the pair on ComTrade for the relevant year (not splitted by sector)
• With any data available, estimate how much of these trade flows goes through the country
• Produce the resulting transit matrix

8. Add usd_per_ton to the sector table

• Get comTrade data, with the country under study as reporter, splitted by commodity, for the relevant year (not split by partners)
• Check that the trade flows are both in qty and in USD
• Map it onto the sector classification
• Calculate the ratio USD per ton for each sector
• Add this column to the sector table

9. Calculate the inventory target table

• Get the inventory data and map ii into the sector classification
• Calculate the inventory duration targets for each pair of supplying sector to buying sector
• Create the inventory target table

Running a disruption scenario

In the user_defined parameters file, set simulation_type to disruption. In the events parameter, write a list of disruption events. Each disruption event is a dictionary.

There are two types of events: transport disruptions and capital destructions.

A transport disruption is defined by:

• type [str]: transport_disruption
• description_type [str]: edge_attributes (more are possible, to be described)
• attribute [str]: the attribute of the edge.geojson file that the model will use to identify the edge to disrupt, for isntance
• value [list[str]]: the value of the attribute that indicates wether that an edge is disrupted or not. This is a list, so multiple values can be given
• start_time [int]: the time step at which the disruption starts
• duration [int]: the amount of time steps the disruption lasts. After that, the edge is back to normal

A capital destruction is defined by:

• type [str]: - type: capital_destruction`
• description_type [str]: region_sector_file
• region_sector_filepath [str]: filepath of the CSV file describing the amount of destroyed capital in each region sector, e.g., "Disruption/earthquake_capital_destruction.csv" (other formats to be described)
• unit [str]: monetary units used in the region_sector_filepath, possible values are USD, kUSD, mUSD
• reconstruction_market [bool]: if True, a reconstruction market is created, and capital is rebuilt. The parameters of this market is, until now, hard-coded in the model. If False, the capital is never rebuilt.
• start_time [int]: the time step at which the disruption starts

Model run

The model is launched by the following command:

python code/main.py <region> <optional argument>

See the section "Calling the script" for more details.

This is the general workflow of the model:

graph TD;
    A[1. Setup transport network]-->B[2. Setup agents];
    B-->C[3. Setup supply chain network];
    C-->D[4. Setup logistic routes];
    D-->E[5. Initialize economic variables];
    E-->F[6. Run time step 0 to set up flows];
    F-->G[7. Run disruption scenario];

Effect of the optional argument in the model run:

• When using no optional argument, the model starts at step 1.
• When using same_transport_network_new_agents, the model starts at step 2.
• When using same_agents_new_sc_network, the model starts at step 3.
• When using same_sc_network_new_logistic_routes, the model starts at step 4.
• When using same_logistic_routes, the model starts at step 5.

Note that only step 3, setup supply chain network, is stochastic. To capture the stochasticity, the model can be run multiple times restarting at step 3.