Collecting accurate nature data to support climate mitigation and adaptation solutions is often complex and time-consuming. Data can be scattered across different entities and often be incomplete leading to challenges in:
- Obtaining complete data to scale nature-positive impact, support decision-making and reporting to stakeholders, in a cost- and time-efficient manner;
- Efficiently manage and leverage big amounts of data for sustainability decision-making and added value to their services.
To deal with these challenges there are multiple approaches that rely on technical and data solutions:
- More efficient data collection thanks to adapted tools and processes;
- Remote sensing and sensors to support data collection, in addition to on-the-ground data collection processes;
- Data treatment and management to ensure insights are accurate, actionable and support decision-making.
In the past years, we helped a company with multiple Agroforestry Nature-Based Solutions projects improve the estimations of captured carbon by developing more precise science-based models. Due to the long time period between verification, it is key to have a robust forecasting of the carbon sequestration to avoid both overstock and shortage of supply. This difference between the forecast and the reality can be due to multiple factors:
- Over or underestimation of the trees’ survival rate, especially during the first 2 years;
- Usage of linear growth curve (from IPCC) that do not reflect reality, despite being a market standard;
- Error in the data collection and/or selection of a sample that is not representative of the project.
Additionally, often times approaches for carbon estimation are general and standardised across different species, planting models and regions. However, tree survival rates, species composition and growth conditions within the project area can significantly impact carbon capture.
By analysing years of historical data, we reached the conclusion that the empirical data showed lower carbon sequestration in the first years of the tree life, when compared to IPCC data, proving that a linear growth curve was not reflecting the reality on the field. The analysis of over 15+ years of historical data obtained through on-the-ground biomass inventories allowed us to identify robust trends on the evolution of several Nature-based projects. Moving from an area-level to a tree-level estimation with higher granularity, through a census-based estimation of biomass at individual tree level and leveraging historical data we were able to successfully improve carbon estimations.
This transition required a more rigorous and complex data collection process and an update on the foundational scientific assumptions. Scientific literature was used to complement historical data, leveraging recent and peer-reviewed species- and location-specific data on tree growth and survival rates. This would ultimately guarantee that the new model provides more accurate estimations with less uncertainties and be a better fit for complex agroforestry systems.
In another client case, the challenge faced by the company was related to the time-consuming and inefficient process of obtaining data from multiple local public entities, which was often was incomplete and without key parameters that would enable the measurement of impact of urban greening projects. In order to attract investment into nature regeneration and social impact solutions, accurate reporting to all involved stakeholders is key, and that’s where Impact Labs contributed by guaranteeing high-quality impact data to support nature-positive decisions.
In this case, the automated detection of trees and their characteristics, resourcing to a machine learning model developed by Impact Labs would provide significant benefits to the company, among which the acceleration of project implementation timelines, support for negotiations based on scientific and transparent data and allow the scaling of a cost-efficient model to create nature-positive impact.
To achieve these goals we tested different approaches to automate the estimation of tree characteristics, relying on different data layers:
- Google Earth to detect and localize trees at a large scale;
- Google Street View to determine the location of individual trees and their characteristics;
- LiDAR datasets as a verification layer due to their precision, but this data is only available for a limited number of locations.
To develop an accurate and actionable model, first, we focused on location mapping based on ground-truth data to fine-tune the model and obtain a concept prototype with high levels of reliability. Ultimately, the development of this model at a Proof of Concept stage enabled the company to clearly define the next steps to fulfil their business and nature-positive goals.
The goal was to ascertain which approach provided the most reliable, complete and actionable insights, based on the frequency of update, the required computing power and the level of accuracy. Business considerations were also important, such as the cost-efficiency, the scaling capabilities and the efficiency gain. These factors were all taken into consideration in the Proof of Concept stage, to provide the best recommendations on how to scale the automated detection of tree characteristics and the automated computation of tree location.
These cases show how data management, tech, and scientific modelling can support scalable climate solutions by moving from standardized assumptions to more granular data.
