AI for environmental due diligence? Why human intelligence still matters in land development

When it comes to environmental due diligence, the role of AI is still up for debate.?

Land developers need an efficient way to gauge project risk, but mistakes can be costly. And given AI’s inherent risks, developers are right to be cautious about relying too heavily on AI.

At Transect, we’ve seen both the promise and limitations of AI in speeding up environmental due diligence. While AI can help signal project blockers — as we’ve seen with our AI-powered community sentiment tool Solar Pulse — it’s not a silver bullet.

Human intelligence and environmental expertise are still critical to environmental due diligence, especially now that most of the easily developable land is already taken.??????

Here’s why land developers shouldn’t completely outsource site selection or environmental due diligence to AI, and what they should keep in mind when vetting AI-based site selection tools.

Reason 1: Humans control data quality, which determines data reliability?

Firstly, we need to recognize that the quality of inputs directly affects the quality of outputs. Put simply: garbage in, garbage out.

It’s the job of humans to shape the quality of data to ensure the output will be reliable.?

In the environmental field, quality data looks like:

• Comprehensive data gathered from thousands of analog sources, not just public datasets (which are often insufficient for making critical decisions)
• High-resolution data that captures granular details, since public datasets tend to be too conservative, broad, or incomplete
• A thorough evaluation of data completeness, with experts identifying and addressing any gaps that could affect development risk assessment

Transect’s database, for instance, was developed over the last 8 years by 10 environmental consultants with 100 years of collective experience. The database now includes approximately 4,000 data sources that have been marked as references across ~1,000 regulations, ~1,000 permits, and ~20,000 state and federal species, waters, and other environmental data.

So, why does data depth matter?

Relying only on limited and broad datasets can lead to both false positives and negatives.?

For example, a tool that relies on the U.S. Fish and Wildlife Service (USFWS) critical habitat data to predict protected species occurrence on a site will be wrong more than 90% of the time. That’s because it’s a limited-scope dataset that shows where the USFWS has identified critical habitat for select species. Since only about 800 species have critical habitat designated, and there are over 2,400 species protected under the Endangered Species Act, critical habitat data alone will not provide insights on the delta of 1,600 species without critical habitat designations.

Additionally, species often occur outside their critical habitat designated area. As a result, a developer might think their project is free and clear from endangered species issues due to a lack of critical habitat, when in fact, it’s quite possible that endangered species are close by.

The lesson here is that human oversight and expertise are essential to shaping these data inputs.

With environmental experts in the driver’s seat, you can be much more confident in the analysis.

Reason 2: Lived experience is necessary to build the rules engine?

Even with a strong database, you need human expertise to shape the rules that drive the assessment.?

To give reliable recommendations using automation, you need a sophisticated decision tree or “rules engine” that can consider these thousands of permitting data points and deliver the right answer.

Let’s take wetland impacts as an example.

For a solar project, you can impact up to half an acre of wetlands before you need a permit. But if you stay below a tenth of an acre, you don’t need a permit — unless there’s potential to impact an endangered species, or you’ve found a cultural resources issue on your site, or the project is in a coastal zone, or the project triggered a USACE regional condition… and on and on and on.?

There are many, many project characteristics that can impact the permit requirements for a project.

As environmental consultants, we’ve lived and breathed these iterations before automation or AI came onto the scene. We know the rules and realities that simply aren’t available on a government website.

As a result, we’ve built an expert-configured rules engine that takes these complicated nuances into consideration. To create the rules that deliver a reliable answer, you need that lived experience.?

Reason 3: Human expertise and lived experience are needed to check and interpret results

We also need environmental expertise for interpretation and actionable insights.?

This is especially true considering we’ll probably never be able to completely eliminate hallucination from AI models, as 2023 research shows.

But it’s not just about checking the accuracy of AI-generated information — it’s also about providing recommendations for how to move a project forward.?

If we go back to the wetland impact scenario, environmental experts can advise on permitting timelines, what counts as a temporary versus a permanent impact, what counts toward your total acres of impact, and how developers can avoid and mitigate critical issues.

We can also clarify misinformation. One issue that comes up often is that wetlands permitting usually needs a good three to six months or more to process, much more than the 60-day window the US Army Corps of Engineers claims on its website. An AI tool doesn’t have access to information like this, unless an environmental expert has fed it the information one way or another.??

AI without access to lived experiences and expertise won’t have this crucial context.

What is AI good for??

We can think of dozens of other reasons why we still need to keep environmental experts as guides throughout the environmental due diligence process.

But a fair follow-up question would be “then what should we use AI for?”

One of the use cases we’ve found at Transect is using AI to answer a more limited set of binary prompts.

In particular, we’re excited to quickly gauge community sentiment about renewables, which is a major bottleneck for utility-scale solar projects today.??

Our AI-powered tool Solar Pulse uses natural language processing to scan local news, project velocity, and ordinances to predict community acceptance.?

The tool can tell us whether it finds this evidence or not, and developers can browse the evidence and request verification to check accuracy.

This helps developers avoid sites that are likely to face opposition and instead pursue sites where the community will be more open to embracing a project.

This is a powerful use case, and I’m sure it won’t be the only one for Transect.?

The key takeaway here is that our deep understanding of environmental permitting allows us to determine when AI should play a key role, and when it needs to stay on the sidelines.?

AI can streamline permitting — with the right environmental expertise

Since our founding in 2016, the secret to Transect’s success has been combining technology with our deep environmental expertise.

This blend of automation and rich consulting intelligence has allowed our customers to evaluate hundreds of millions of acres of land and spare hundreds of thousands of man-hours.?

However, AI is only as powerful as the expertise behind it. So the next time you’re evaluating a site selection tool, don’t just ask whether it uses AI — ask how.

More importantly, ensure the team behind the tool has the environmental expertise necessary to guide your project to success. That way, you can eliminate blockers for your renewable projects, not create new ones.