- 2020 Base Land Cover Model : This deep learning model uses diverse imagery sources at different resolutions to classify land types in local contexts, such as forest, cocoa, coffee, palm oil, rubber, and water. The model is designed to perform reliably across various sensors and resolu tions, and to overcome challenges related to cloud cover or missing data. - Annual Change Detection Model : This takes as input the land cover outputs before and after a given year, along with a series of images in between. It then estimates whether deforestation occurred during the year of interest. Together , these models work to identify deforestation activity and establish the nature of any subsequent agricultural land use (cocoa farming, or another crop). • Manual review : The initial modelled results are subject to rigorous manual review by a team of trained Enveritas analysts with deep regional expertise in coffee, cocoa, and palm -producing origins. These reviews confirm that deforestation occurred and validate the subsequ ent land use. This process allows Enveritas to identify locations where forest was specifically converted to EUDR -restricted commodities, such as coffee or cocoa, after 2020. Enveritas also verifies that the polygons do not intersect or over lap with
Protected Areas delimited maps provided by CFI 2019. If doubt persists, Enveritas is able to rapidly deploy field teams locally to conduct ground
-
truthing.
• Probabilistic risk of deforestation : Enveritas also conducts an analysis to answer the question: “What is the probability of future deforestation for each farm?”. This analysis supports proactive risk mitigation by identifying farms that may require early outreach or agronomic support. Using this mode l, deforestation can be predicted by applying a spatial risk model that considers a variety of indicators around each plot. We assess the probability of future deforestation based on proximity to key risk factors. These include the presence of forest within a defined buffer , which may indicate potential encroachment zones, and historical deforestation within a wider radius, which can reveal patterns of nearby forest loss that signal ongoing or expanding threats. These spatial indicators are layered into a scoring system or mo del that outputs an annual
risk estimate at the plot level.
8
Made with FlippingBook flipbook maker