As a growing number of companies set targets to zero out their emissions of planet-warming greenhouse gases, many are turning to carbon offsets to lower their climate impact. For example, an airline might buy an offset based on a project that plants trees to absorb carbon dioxide, and count it against its own footprint. But the market is unregulated and many offsets don’t deliver on their climate promises.

London-based startup Sylvera Ltd is trying to boost transparency by assigning ratings to projects, similar to what S&P Global Ratings and Moody’s Investors Service do for debt issuances. So far the company has analyzed about 35 projects and found that nearly half of them don’t deliver what they claim. At least three projects received the equivalent of a junk rating, according to founder Samuel Gill.

The company uses machine learning to analyze data such as satellite images to look at historical land use and assess how a project is performing against the benchmark set by its developers. “Nobody can actually tell which projects are real and which aren't,” said Gill. “Our data and services should push more money to the best projects.”

One of the projects that Sylvera concluded had failed set out to prevent deforestation in a certain area. That was true, but there was significant deforestation in areas nearby, meaning that the specific site’s contribution was negligible.

With about 40% of the projects Sylvera has worked on, there was some carbon reduction, just not as much as promised by the developer. In one instance, a fire wiped out part of the project. That wasn’t the fault of the developers, but it did reduce the potential abatement.

“Having an independent auditor looking at the accuracy and conservativeness of the baseline is one of the key challenges for forestry projects,” said Gilles Dufrasne, policy officer at nonprofit Carbon Market Watch. “If they’re monitoring things like leakage and risk of fires maybe they can detect reversals that wouldn’t be detected otherwise and bring some independent review of the quality of the project.”

Still, Dufrasne said the amount of failing projects would likely be much higher than 3 out of 35. Not many systematic studies have been done on the credibility of offset projects, but almost all the ones that do exist have found severe limitations. Most recently, nonprofit Carbon Plan found that California’s offset projects were overclaiming 30% of their carbon savings. That means about $400 million was paid for 30 million metric tons worth of emissions reduction that didn’t actually occur.

Because companies buy offsets to negate their own carbon, an ineffective offset that fails to reduce emissions could give the buyer an excuse to pollute more—and cause an increase in overall emissions.

Sylvera said Thursday it raised $5.8 million in a seed funding round led by Index Ventures, which has backed companies including Roblox Corp. and Deliveroo Plc. It will get another $2 million from a research contract with Innovate UK.

It will use the money from Innovate UK to work with academics at NASA’s Jet Propulsion Laboratory, the University of California, Los Angeles and University College London to create a new method of acquiring data on the carbon content of forests.

The plan is to scan trees around the world with a lidar sensor, drone and light aircraft. Samples of actual trees will help calibrate the model, which will improve the machine learning process to better infer a forests' carbon content when analyzing satellite images. Gill said he expects the model will show a greater carbon impact of deforestation than previously estimated.

Sylvera markets its services as a subscription that provides quarterly updates about offset projects. It aims to cover between 150 to 200 projects by the end of the year.

Going forward, Sylvera aims to work with project developers to help them set more realistic benchmarks on which to base offset projects. They also plan to help exchanges to provide customers with better information about the quality of listed projects.

This article was provided by Bloomberg News.