More than 75 percent decline over 27 years in total flying insect biomass in protected areas

If you care about the fabric of life on land, you have to care about insects. They stitch together pollination, they chew plants into soil, and they feed birds, bats, and fish. Roughly eight out of ten wild plants depend on insect pollinators. About six out of ten birds, at least for some stage of their lives, rely on insects as food. In the United States alone, the ecosystem services wild insects provide have been pegged at around fifty-seven billion dollars a year. That's the backdrop. Now imagine asking a very simple, very unsettling question: in places we set aside to protect nature, how much flying insect life is actually left? Hallmann and colleagues in Germany decided not to guess. They built a monitoring effort that was as standardized as a laboratory protocol and ran it for nearly three decades. Starting in nineteen eighty-nine and running through twenty sixteen, they set up Malaise traps—tent-like interceptors for flying insects—inside sixty-three protected areas in the country's lowlands. Across those years, they accumulated ninety-six location-year combinations and, when you add up every bottle the traps filled, one thousand five hundred and three samples. Taken together, those jars held fifty-three point five four kilograms of invertebrates. That's not a back-of-the-envelope estimate; that's literal, weighed mass. The care they took in making those numbers comparable is the quiet hero of this story. Every trap was the same make and size, oriented the same way, sealed the same way, and operated the same way from March through October, day and night. Crews emptied them roughly every eleven days. In the lab, the team weighed insects wet—after standardizing ethanol concentration—using a protocol strict enough that a validation set showed about four tenths of a percent deviation between repeated weighings. Weather was not treated casually either. They pulled data from one hundred sixty-nine nearby meteorological stations and interpolated temperature and precipitation to each site. Habitat was cataloged right around the traps, and the broader land use—arable fields, grasslands, forests, and water—was mapped within two hundred meters. So what changed over time? In a word: almost everything you would not want to change. After adjusting for the fact that insect biomass rises and falls within each season, the simplest trend says biomass declined about six point one percent per year. When the team summed a typical season from April first to October thirtieth—so you're not over-weighting early spring or late fall—that compounded to a seventy-six point seven percent drop over twenty-seven years. Let that sink in. More than three quarters gone. And in midsummer, when flying insects should be at their peak, the loss was sharper still, reaching about eighty-one point six percent. Protected areas were not protected from this. To understand how they reached those numbers, it helps to peek under the hood of their statistical engine, because it's built to respect how the data were actually collected. Each jar in the dataset covers a window of days when a trap was open. The team imagined a latent, or hidden, daily biomass at each site—call it z—that ebbs and flows with the season, changes with weather and habitat, and trends over the years. The observed weight in a jar is then modeled as the sum of those daily z's across the exposure window, plus measurement noise. On the log scale, expected daily biomass is the intercept plus a bundle of covariates, plus a term that grows or shrinks by the same factor each year. That annual factor is the exponential of log lambda; when log lambda is negative, you have a steady decline. They also let each site have its own baseline, a random effect that catches local quirks. In practice, they fit a series of Bayesian mixed-effects models and compared how well they captured the patterns. The bare-bones model—seasonal curve plus an annual slope—returned that log lambda of about minus zero point zero six three, which is where the six point one percent per year decline comes from. When they layered in the variables they could actually measure—temperature, precipitation, land cover around the traps, and plant community indicators—the fit improved and the estimated trend got even more negative: log lambda around minus zero point zero eight one. In other words, after accounting for weather and habitat, the underlying decline was steeper. The final model explained about two-thirds of the variation in biomass on the log scale, which is good in ecological field data, and it did so with chains that converged nicely by standard diagnostics. Seasonality matters here because it tells you whether the peak of the year is sliding around or shrinking. The team captured that with a simple curve—day of year and its square—which lets biomass rise to a hump and fall. They even allowed the shape of that hump to shift across years. The striking thing is that, regardless of these seasonal wiggles, the whole curve sank over time. That's why those midsummer numbers are so stark: the peak itself is lower. What about weather? You'd expect warm days to boost flying activity and growth rates, and rain to ground insects. That's basically what Hallmann and colleagues saw. In their final model, higher mean temperature during trapping windows was associated with higher biomass; the coefficient there was around zero point three zero four on the standardized scale. More precipitation pushed the other way, with a negative coefficient of about zero point zero seven one. The count of frost days and winter precipitation—lagged variables meant to capture carry-over effects—didn't have strong predictive power for the next season's biomass. These relationships make biological sense. They just don't make the long-term decline go away. Land cover around the traps tells another, complementary story. Sites with more grassland in that two hundred meter radius tended to have more flying insect biomass, with a positive effect on the order of zero point eight one nine. Sites with more arable land—the crop fields that dominate many European lowlands—showed much less biomass, with a strong negative association around minus one point zero six three. Forest in the immediate surroundings was also associated with less biomass, roughly minus zero point five two two, and open water had a smaller negative effect as well. None of this means those habitats are intrinsically bad for insects; it's about what these covers likely reflect in this landscape. Grasslands here are often semi-natural, with flowers and structure. Arable fields and closed forests in this context can be less favorable for the kind of diurnal, flower-seeking flyers that Malaise traps are best at catching. Zoom in to the microhabitat and the plant community around a trap, and you see more texture. Places with more tree species nearby had more biomass, with a modest positive coefficient of about zero point one zero four. Indices ecologists use to summarize plant niches—Ellenberg values—also mattered: higher nitrogen and light values were associated with more biomass, while higher temperature values were associated with less. Herb species richness, by itself, wasn't a strong linear predictor in this setup. Again, these are cross-sectional relationships embedded in the model. They explain who tends to be high or low, not whether everyone is rising or sinking together. Did the trend differ by habitat type? The team grouped sites into clusters like nutrient-poor heathlands and sandy grasslands, or more nutrient-rich grasslands, margins, and wastelands. Biomass was higher in the richer grassland group—no surprise there—but declines appeared across the board. When they modeled those clusters independently, the estimated annual decline was roughly seven point five percent in the nutrient-poor cluster and about five point two percent in the richer grassland cluster. Different starting points. Same direction. One of the more illuminating checks asked a counterfactual: suppose the covariates had just stayed put. Same average temperatures, same land cover near the traps, same plant communities. Under that scenario, the projected seasonal biomass would have been roughly stable or even up by about eight percent over the study period. That tells you two things. First, some environmental variables actually drifted in a favorable direction in these protected areas—local arable cover may have decreased in the immediate buffer, for example—so their net contribution was to cushion the fall. Second, because the decline still sharpens when you control for them, the real driver is not captured by the variables measured at that scale. This is where the team's caution and their conclusion meet. They tried a lot of plausible explanations: weather decomposed into long-term means and short-term anomalies, habitat around the traps, land cover in the near buffer, and interactions between year and land use. None of it absorbed the time trend. Landscape configuration and climate change, as captured here, looked unlikely to be the primary culprits within these reserves. The design itself had limits: most sites were sampled intensively within a single year rather than continuously across many years, which reduces the power to track within-site trajectories. Only twenty-six of the sixty-three locations had more than one sampled year. Land-use data came from aerial photographs in two time windows and had to be interpolated. Even with those caveats, the broad pattern held across sites, seasons, and models. So what's left? The big, messy processes operating beyond reserve boundaries. Hallmann and colleagues point out the obvious geography: the vast majority of these protected plots are islands in a sea of agriculture. In fact, they report that ninety-four percent of trap locations are bordered by fields. If the surrounding matrix intensifies—more pesticides, fewer flower resources, more homogeneous crops—then the reserves can't wall themselves off. They start to behave like sinks or ecological traps: insects move in and out across invisible lines, but the broader landscape bleeds them dry. That's a hypothesis consistent with the data, not a proven mechanism in this paper, and the authors are careful about that distinction. The ecological implications are not subtle. If you remove three quarters of the flying insect biomass from a landscape, pollination networks thin out. Food chains that depend on swarms of midges, flies, and moths—not just charismatic bees—become brittle. Birds arrive to breed and find fewer mouth-sized morsels for their chicks. We can feel the cascade even if we don't see every step. There's also a methodological lesson here, and it's oddly hopeful. This study worked because it was standardized, long-term, and regionally coordinated. The team ran Bayesian models not to impress statisticians but to honor what the data structure demanded: jars that sum days, seasons that wax and wane, sites that differ, and a trend that, if it exists, must be extracted gently from that noise. They shared their data and code. And they showed that even in places set aside to protect nature, you have to look out to the horizon if you want to understand what's happening inside. If there's a to-do list, it's short and urgent. Keep measuring, the same way, for the long haul. Link protected areas to their surrounding lands, because the borders on a map don't stop insects. And search, with experiments and more granular land-use and chemical data, for the drivers that can explain a decline of this magnitude. As Hallmann and colleagues argued back in twenty seventeen, protected areas are necessary. But if the broader landscape is hostile, they're not sufficient.