Imagine a global weather and climate forecasting system that collects data regularly in just a handful of countries, and takes measurements elsewhere only during extreme weather events. That is what today's global flu-surveillance system mostly looks like.

The shortcomings of flu surveillance have long been recognized (seeNature 440, 6–7; 2006), but they are attracting renewed attention following the creation in labs of strains of the H5N1 avian influenza virus that can spread between mammals. The main cited public-health benefit of the research is that it will allow for monitoring for such mutations in the wild, and give a remote chance of containing an emerging pandemic.

It is certainly urgent to monitor wild flu strains for mutations that might make them transmissible between mammals (see Nature 482, 439; 2012). But as Malik Peiris, a flu virologist at the University of Hong Kong, says, detection of a breaking pandemic is “a very ambitious goal, and this is where vastly enhanced global surveillance is needed”.

Current surveillance can barely identify threats, let alone track them.

Current surveillance can barely identify threats, let alone track them. The precursor to the H1N1 virus that caused a pandemic in 2009 had been circulating worldwide for years in pigs, and the pandemic virus had been infecting humans in Mexico for months, before either was detected. That virus is also a reminder that threats come from many flu subtypes other than H5N1.

An analysis by Nature shows that timely, continued and representative global surveillance of the genetic sequences of flu isolates from pigs and poultry just isn't happening (see page 520). From 2003 to 2011, most countries collected few or no sequences, and genetic surveillance of flu in pigs was and is almost non-existent. There is typically a lag of years between collection of viruses and the release of their sequences into public databases, so there are very few data on their recent evolution.

Yet the analysis gives hope that this situation could be rectified, given political will, modest funding and international coordination. Hong Kong has collected the most flu sequences from pigs after the United States and China, and most of those come from labs at the University of Hong Kong, including Peiris's; this shows what a few dedicated centres can achieve. Similarly, the Influenza Genome Sequencing Project of the US National Institute of Allergy and Infectious Diseases, which was launched in 2004 and sequences whole flu genomes from isolates collected globally, accounts for around half of sequences generated worldwide. And in the past decade, many nations affected by H5N1 have greatly improved their surveillance, often despite limited resources and poor veterinary and health infrastructure.

More sequencing alone is not enough. Sequences tend currently to come in fits and starts, in response to an outbreak, one-off projects or as funding allows, and there is little sustained passive surveillance. Global, scientific and representative sampling is needed, from multiple outbreaks and diverse populations, taking into account risk factors such as the size of livestock populations, husbandry practices and proximity to waterfowl reservoirs.

Funding is not the only problem. Few countries, for example, compensate for culled animals to encourage farmers to report outbreaks; and some might conceal, or not actively look for, flu infections for trade reasons. Nations can be reluctant to share viral isolates if they do not get anything in return, although the World Health Organization's Pandemic Influenza Preparedness Framework, published last year, should help to ensure that they do get appropriate benefits, including access to vaccines.

Surveillance makes sense even without the promise of tracking a pandemic. Detecting outbreaks in livestock allows control through culling or vaccination to avoid crippling losses, and limits the opportunities for viruses to mutate, outpace vaccines and possibly turn pandemic. Surveillance also generates crucial data for epidemiology and drug-resistance monitoring, yet it remains a low priority. Sequencing costs can fall all they like, but without greater, and more sustained, routine surveillance efforts, there will be few samples to sequence.