Can a Public Database Detect a Plant Disease Outbreak?

Xylella fastidiosa is one of the most damaging plant pathogens in Europe. It killed millions of olive trees in southern Italy and is under active quarantine across the continent. Meanwhile, anyone can download every public record of it from GBIF, the Global Biodiversity Information Facility, in about a second.

So here is a question worth answering: if a new outbreak started tomorrow, would that public database notice?

I pulled all 1,170 records and tested them against outbreaks we already know happened.

Test one: did the database notice?

The European incursions are a good test case because they are recent, well documented, and independently dated by plant-health authorities. If GBIF records carry real signal, the first public record for a country should appear close to the documented first detection.

The result is a split, and the split is the finding. The two largest and most publicised epidemics, Italy and Spain, show up in the public record almost immediately. The two smallest incursions do not. Germany's case was a single glasshouse plant that was destroyed, and Portugal's outbreak, confirmed in 2019, has left no trace in this dataset seven years later.

France runs the other way, with a record three years before the Corsica outbreak. That is worth flagging rather than smoothing over: it may be earlier undetected presence, a research collection, or a record whose date and place do not refer to the same event. A single record cannot distinguish those.

So the database notices, but it notices selectively. Detection appears to track how large and how publicised an outbreak is, not whether the pathogen is present. That is a usable rule with a hard limit: these records can corroborate a major incursion, and they cannot be used to argue a country is clean.

Test two: correcting the bias instead of complaining about it

The obvious objection to any country ranking here is that the United States contributes 543 of the 1,170 records, about 48 percent. The usual response is to note the bias and move on. It is more useful to correct it.

Countries differ enormously in how much biodiversity data they publish at all. The US has 1.3 billion total GBIF records; Italy has 14 million. Dividing each country's Xylella records by its total record count converts raw counts into a rate, which is comparable across countries.

This is the part I find most convincing, because the correction can be checked against what we independently know. Before correction, the ranking mostly reproduced which countries publish the most data. After correction, Italy and Brazil rise to the top, and both have documented major Xylella epidemics, olive quick decline and citrus variegated chlorosis respectively. The adjusted ranking agrees better with known epidemiology than the raw one did.

One caveat belongs right next to that result. Iran ranks first on this measure from only 10 records, and a rate built on a small numerator is unstable. I would treat the top of this list as a shortlist worth checking, not a finding in itself.

What the raw data still cannot say

Only 309 of the records carry coordinates, and they cluster where surveillance programmes operate. Blank regions on this map are mostly places nobody has sampled and uploaded, not places the pathogen is known to be absent. Prevalence, rate of spread, and severity all remain out of reach here, because those need surveillance with a known sampling frame.

Where this leaves things

Two conclusions survive the analysis. The public record reliably captures large outbreaks and misses small contained ones, which makes it a corroboration tool rather than an early-warning system. And correcting for recording effort is worth the small amount of work it takes, since it moved this ranking measurably closer to known epidemiology.

The next step I would take is to add climate suitability, which is well characterised for this pathogen, and compare it against the effort-corrected rates. Countries with suitable climate, meaningful recording activity, and near-zero Xylella records would be the most interesting places to look, because that combination suggests a genuine surveillance gap rather than a data gap.