How does LandGriffon spatially model supply chain data?

LandGriffon helps companies strategize the sustainable transformation of their supply chains by using technology, data and scientific-based analysis to manage environmental impacts. This blog is part of our methodology series. You can find more blogs on our methodology here, or head to our website and download the complete paper.

Spatial sourcing model.

The spatial sourcing model lies at the core of the LandGriffon methodology. The model identifies likely material sourcing areas. It then attributes impacts in those areas to the sourcing of those materials.

Where the location type is unknown or is only known at the scale of the country or region of production, we assume that the commodity is sourced from all locations producing that material within the relevant spatial extent (Table 1). Where the location type is known as the country of delivery, we assume that the material has been produced globally, in all locations exporting the material to the given country (identified using Multi-Regional Input-Output databases e.g. EXIOBASE 3), and sourced in proportion to the production in any location.

The closest matching gridded production dataset for each material is identified to spatially allocate sourcing within the sourcing region. MapSPAM (International Food Policy Research Institute 2019) is used for crop production (Figure 1), and Gridded Livestock of the World v3 (GLWv3) (Gilbert et al. 2018) is used for livestock production. MapSPAM and GLWv3 are the latest publicly available datasets but are representative of the year 2010. Materials are matched using the extended HS commodity codes but where there is no exact match, the closest parent in the hierarchical system will be used. For example, “Apples, Pears, and Quinces” (HS 0808), will be matched to the MapSPAM dataset for Temperate Fruit crops. Commodities with no close match, such as rubber or acacia, will be analyzed on a case-by-case basis using specific additional datasets. More material will be assumed to be sourced from locations with greater production. So, a higher probability of impact is associated with areas of high production and vice versa.

This implies that LandGriffon could underestimate or overestimate the impact associated with a commodity. If the weighted average impact across the whole sourcing area is lower than in the location or locations where the material was produced there is an underestimation. If the commodity came from a production area with a low impact, there is an overestimation.

Figure 1. Distribution of Cotton production (Tonnes) from MapSPAM data. The commodity production datasets are used in order to distribute the purchased volume across the sourcing location type identified. More material will be assumed to be sourced from locations in which there is greater production.

Sub-national understanding of spatial sourcing is critically important for reducing uncertainties in impact estimation and is the focus of tools such as Trase. Future LandGriffon development will focus on using additional supply chain information to infer the likely sourcing profiles of companies. These profiles can be based on the sourcing profile of a country in which that company is based, or using company specific information on supply chains. For example, palm oil or cacao trader information can be extended with Trase data on supply chains. Ultimately, full knowledge of sourcing location, when this data has been collected, can be incorporated and removes the need for modelling.

Spatial representation.

Each sourcing location is geolocated depending on its associated location type ( Figures 2 and 3). LandGriffon v0.1 uses the H3 format for geospatial data and processing. H3 has the benefit of having efficient hash table matrix math performance, high-speed resampling for visualization or calculation, limited distortion at high latitudes, and appealing aesthetics for visualization. This allows for low-latency calculations and visualizations that update rapidly and are enjoyable to explore.

Figure 2. Geolocation of a country of production, showing the distribution of deforestation risk associated with sourcing palm oil from Indonesia. We model the purchased volume as being produced across all areas of palm oil production in Indonesia.

Figure 3. Geolocation of a supplier’s aggregation point (50 km radius buffer), showing deforestation risk of palm oil around aggregation points in Indonesia. We model the purchased volume as being produced across all areas of palm oil production inside supplier aggregation points.

To learn more about the science and technology of LandGriffon, you can download the full methodology from our website.

Interested in LandGriffon and our services? Contact us now at hello@landgriffon.com

LandGriffon is developed by Satelligence and Vizzuality, and advised by the Stockholm Environment Institute and their Trase Initiative.

Thank you Mike Harfoot, Elena Palao, Francis Gassert and Rens Masselink for preparing the methodology.