2 What Are the Problems?

There is much to be positive about in the biodiversity measurement space. The needs for biodiversity measurements are better defined today than they were a decade ago, with a global biodiversity and monitoring framework in place. In addition, researchers are using AI and trying to bring Indigenous knowledge to bear, and citizen science is making important contributions. At the same time, the massive amount of data available can be paralyzing, with the need to develop a simple system that will measure biodiversity for different users. This is one of the several problems that the forum’s first session discussed.

CHALLENGES OF AND OPPORTUNITIES FOR MEASURING PLANT DIVERSITY

Barnabas Daru, assistant professor of biology at Stanford University, noted that plants occur in every possible ecosystem, provide key ecosystem services, and are under threat from anthropogenic sources. For decades, acquiring data on plants meant going into the field and collecting samples, known as vouchered specimens, that were stored in herbaria. Today, researchers and even citizen scientists are using mobile phones to take pictures of plants and upload the resulting observation-only data to a portal or the cloud, adding to the biodiversity knowledge base.

The challenges of measuring plant diversity, said Daru, include data gaps and biases at global and regional scales. Currently, voucher specimens do a better job of trumping observation-only data in terms of capturing the expected plant species richness, although there are large data gaps in biodiversity-rich regions such as the neotropics, Afrotropics, and Southeast Asia. Even where there has been higher sampling density, such as Europe, North America, South Africa, and Australia, there are data gaps and biases at more granular scales, with 60 percent of herbarium specimens collected within about 2 kilometers of a road. Collectors also have species biases, with those in South Africa and Australia tending to favor plants with grass-like morphology and those in New England favoring trees.

There is an opportunity to overcome some of these geographic and functional trait biases and predict the true diversity of native plants using machine learning systems trained on unbiased data, sampling intensity, topography, habitat, and climate. This approach, said Daru, may allow researchers

to better understand the main patterns of vascular plant diversity and to use machine learning to uncover refuges of hidden plant diversity. In fact, his team is using data from more than 200,000 plant species from around the world to identify unique patterns of native vascular plant diversity in underdocumented biodiversity-rich regions, including South America, Africa, and Southeast Asia.

To overcome trait biases, Daru is extracting information on specific leaf traits using AI trained on features from herbarium samples, such as leaf size, length, and the angles of leaf orientation. He used this Leaf Intelligence system on a dataset of more than 50,000 plant species and measured the distribution of leaf morphological traits for plants around the world. This analysis found that longer leaves, for example, are concentrated in the tropics, longer petioles are found more often in arid regions, and inclination angle increases in the tropics to better capture light.

QUANTIFYING BAT DIVERSITY AND RESPONSE TO ENVIRONMENTAL CHANGE IN THE PALEOTROPICS

Based on a map developed in 1995, the accepted thinking was that bat diversity in Africa was less than in Asia despite both regions being paleotropical. However, a map developed 10 years later showed this was not the case. As Iroro Tanshi, a postdoctoral scholar at the University of Washington, explained, the difference resulted from methodological issues and the type of collection nets the researchers who created those maps used to sample bat populations. “If you are not using the right methodology, you are going to be blind,” said Tanshi. “You are not going to see what you need to see.”

Tanshi noted that mountains are hubs of bat biodiversity in Africa. She explained that understanding the drivers of bat species richness requires looking at prey, resource use, elevation, and vegetation structure. The challenge that climate change is presenting is that bats, which favor places that are warm, with rivers for drinking, tall trees, and plenty of insects, will move to higher elevation more quickly than the vegetation will.

THE FOUR PROBLEMS WITH BIODIVERSITY MONITORING DATA

Idealized monitoring data, said Nick Isaac, head of biodiversity monitoring and analysis at the UK Centre for Ecology & Hydrology, should meet three broad needs. First, they should provide broad taxonomic coverage to represent the biodiversity in a given area. Then they should provide broad spatial coverage to capture driver gradients. Finally, they should cover a long span and be collected frequently to be confident in observed trends. “Unfortunately,” said Isaac, “we have been measuring the wrong things in the wrong places at the wrong times.”

There are four problems with the available biodiversity monitoring data that make those data less than ideal. First, there is not enough data. Historical trends data, for example, are too noisy and uncertain and lack the power needed to detect a change in trends, and this is true for many datasets even with a big increase in sampling effort. Second, the available evidence for many groups, such as insects, is taxonomically unrepresentative. Isaac said the only structured monitoring schemes that deliver spatially replicated measures of the state of biodiversity report on birds, mammals, and lepidoptera, and even for birds, for example, only 10 percent of species have at least one record in one extensive dataset. He noted there is a trade-off between taxonomic breadth and data quality.

The third problem is that monitoring data are spatially biased in that most spatial data reflect wealth, accessibility, and aesthetic appeal. For example, of the 5 million records on insect populations spanning at least 10 years, 3 million of those 5 million come from the United Kingdom, with almost no data from the tropics. Isaac explained that bias arises when the state of interest is correlated with the sample inclusion probability, a problem opinion pollsters deal with and address using well-developed statistical techniques. Problem four is that most data are from the recent past, making it difficult to separate an underlying trend from fluctuations. In addition, for species with a seasonal life history, such as insects, data must be acquired frequently.

Taken together, problems two through four exacerbate the first problem of not having enough data. Isaac noted that accounting for uneven sampling and relaxing the assumption that data are representative leads to even more uncertainty in trends. The way forward, he said, is to develop complementary lines of evidence.

TWO-EYED SEEING: OPPORTUNITIES FOR RESTORING BIODIVERSITY

Prior to colonial settlement, Douglas fir accounted for approximately 4 percent of the trees in the Pacific Northwest. Today, thanks to clear-cutting and plantation-style planting, they account for 90 percent of the tree population, and they are dying because of heat stress. Cristina Eisenberg, associate dean for inclusive excellence and the Maybelle Clark Macdonald director of tribal initiatives in the College of Forestry at Oregon State University, explained that when these trees are struck by lightning, they ignite, leading to a high-severity fire that can burn hundreds of thousands of acres.

Restoring these forests requires incorporating all types of knowledge, including local and Indigenous knowledge, said Eisenberg. Biodiversity, said Eisenberg, is a central principle of Indigenous knowledge, which is a body of observations based on a long-term relationship with an ecosystem. Along with local and Indigenous knowledge, ecological restoration requires having an ecocentric world view in which humans are embedded in nature, rather than an anthropocentric view in which nature is meant to serve humans. For years, there was an assumed scientific chasm

between Indigenous knowledge and Western science, but science and Indigenous knowledge are both evolving to create a common space that recognizes the value of both. For example, data analysis tools have evolved over the past 30 years to create a more holistic and nuanced approach to data looking at relationships.

Combining Indigenous knowledge and Western science creates two-eyed seeing. Eisenberg has operationalized two-eyed seeing using data-sharing agreements, memoranda of understanding, memoranda of agreement, co-stewardship agreements, co-management agreements, and by instituting heightened data security. She noted that one project involving biodiversity assessment and partnership with Tribal Nations in Southwest Oregon is combining Indigenous knowledge about the historical plant and animal species diversity with Western science methods to measure biodiversity. All activities are being implemented within a two-eyed seeing policy framework and are expected to result in a healthy forest in about 20 years.

GENOMIC DEMOGRAPHY AND THE DYNAMICS OF BIODIVERSITY

James O’Dwyer, associate head and associate professor of plant biology at the University of Illinois Urbana-Champaign, discussed the use of genomic data across populations to inform community dynamics and build a framework to complement time series data from plant communities. The goal is to measure contemporary biodiversity and use it to inform future changes in biodiversity.

In pondering whether life history can predict future changes, O’Dwyer came up with the idea that life history variation across species in a community combined with competition for resources may answer that question. By modeling how plant species compete for resources and their reproductive cycle, he and his colleagues demonstrated that life history is a key driver of temporal fluctuations in tropical tree abundance.

That project drew on an extensive database collected over multiple decades, data that are not available for most any other plant community on Earth. This led O’Dwyer to ask if plant genomes are recording the right information from which to make projections forward in time. He explained that large libraries of polymorphisms in a population can provide a sense of the genomic variation within a given population. When combined with information about linkage disequilibrium—correlations across the genome in which polymorphisms co-occur—the resulting model can estimate per capita effective population sizes, which is a key parameter in a life history model.

O’Dwyer tested this modeling approach using census data and population genomic data across the 8 most abundant of 26 species in a forest dynamics plot in the Wind River Forest just north of the Columbia River in Washington State. The resulting model projected population sizes going forward by looking at the population variance as a function of the initial population size and

population genomic data from a single time point without any fitted parameters or fine tuning. As he explained, “these trees are in effect recording what has been happening to them over time and are an untapped source of data at the community level.” While this study only looked at eight tree species in one forest plot in one U.S. state, it shows that population genetic data, which can be scaled globally, may allow for parameterizing multiple aspects of biodiversity dynamics, said O’Dwyer.