by Brian Tomasik
First written: 2 Nov. 2016; last update: 8 May 2017


This piece discusses the findings of one study on how humanity has affected global terrestrial net primary productivity. Depending on the interpretation of the trends, humanity seems to have either slightly decreased or slightly increased the amount of food energy available to feed non-large animals (and thereby, to fuel animal suffering). The magnitude of humanity's net impact either way is smaller than what one would conclude from other studies that measure changes in indexes of wild-animal abundance.

Defaunation studies vs. HANPP

Net primary productivity (NPP) is a crucial metric for those concerned about wild-animal suffering, because it roughly approximates, given certain assumptions, the total amount of morally relevant brain activity that an ecosystem brings into existence.

The literature on human appropriation of net primary production (HANPP) aims to quantify the percent of Earth's potential NPP that humans extract for their own uses, such as food, livestock feed, fuel, and fiber. This is useful because it provides information on the aggregate impact of human activity on the biosphere, which is relevant to analyzing questions like whether it's good or bad to increase the human population size.

This page cites studies that find substantial decreases in indexes of wild-animal populations in recent decades. This suggests that humanity has, on the whole, significantly reduced wild-animal suffering. However, defaunation studies are probably noisy, and I suspect(?) they may not represent well many small and "boring" animals like krill, springtails, and rotifers, which are some of the most numerous animals on Earth, both in terms of numbers of individuals and even in terms of aggregate neuron counts. In my opinion, NPP may be a more stable measure of wild-animal suffering than an index of animal populations, because unless there's a dramatic shift in the fraction of NPP that's consumed by fungi, bacteria, other non-animal heterotrophs, or abiotic fire, then NPP should be roughly proportional to animal metabolism, which is a rough approximation of sentience-weighted animal suffering.

As an analogy to see why NPP may be a more reliable metric, consider the following. It's easier to measure changes in the total number of employed individuals in the USA by a single, large-scale survey than it is to measure changes in the numbers of doctors, lawyers, teachers, mechanics, secretaries, etc. separately and then try to aggregate them into an index of total employment changes. Likewise, it's probably easier to measure changes in the total food supply of all animals combined than to measure (perhaps noisy) changes in individual animal populations and then aggregate those. Of course, it's great to be able to use both approaches as somewhat independent lines of evidence.

Krausmann et al. (2013) results

In order to examine humanity's net impact on global NPP, I read the following study, which I'll discuss in detail: "Global human appropriation of net primary production doubled in the 20th century" (2013) by Krausmann et al. Looking into other HANPP studies would enrich the analysis.

The most important figure for my purposes is the following, which is taken from the "Supporting Information" to the main paper. I focus on Figure S2A on the left side, but Figure S2B is included for reference as well. I'll explain the abbreviations below.

As you can see, NPP here is measured based on mass of carbon, not based on Joules of energy. However, if we assume that the distribution of energy content of biomass hasn't changed significantly over time, then NPP should also roughly track Joules of energy made available for heterotrophic metabolism.


NPPpot is the potential NPP in the absence of human harvesting of biomass or land-use changes. I drew the blue circle on upper left corner of Figure S2A to show approximately what Earth's NPP would look like if humans hadn't existed.

You'll notice that NPPpot rises over time. Why? NPPpot is calculated based on the "Lund Potsdam Jena managed Lands (LPJmL) dynamic global vegetation and water balance model" (p. 2). This model predicts that Earth's potential NPP increased over the 20th century, "mainly due to the so called CO2 fertilization effect" (p. 3), i.e., increases in plant growth due to higher atmospheric carbon-dioxide concentrations. This is a surprisingly large effect, and I wonder how much of it is accurate and how much of it results from idiosyncrasies of the model and input data. The authors agree that one should have "caution when interpreting the trajectory of NPPpot" and that "Many uncertainties [...] relate to this effect" (p. 3).

HANPP is calculated relative to NPPpot, which can change over time due to factors like atmospheric CO2 concentration. As a result, HANPP doesn't exactly track the overall reduction in "NPP left over for use by wildlife" due to humanity's existence relative to a counterfactual where humanity doesn't exist, because the maximum possible global NPP can (and does) increase as a result of humans. If we're seeking to measure humanity's total impact, we should include the effect of CO2 emissions, which means we should actually compare against the NPP level of my blue circle.


NPPeco is the amount of NPP left over for wildlife, i.e., the NPP not used by humans. Interestingly, you can see that it's almost unchanged between 1910 and 2005. And in fact, NPPeco has increased slightly since ~1970, which is difficult to square with findings that indices of both vertebrate and invertebrate abundance have declined by roughly half since 1970. Probably there's significant noise on both sides of this antinomy.

That said, NPPeco is always less than the circled blue point in 1910 that approximates a world without humans altogether. So it might still be the case that humanity's existence overall has reduced wild-animal metabolism, even if population growth since 1910 has not enlarged the reduction. If this is the case, it suggests an interesting divergence between humanity's total impact vs. the marginal impact of additional humans, which may weaken our ability to reason from humanity's total net impact to implications for how big the human population should be.

It's also possible that NPPeco will decline going forward because NPPpot might stop increasing as much. The "Supporting Information" explains that "In general fertilization by CO2 (and also N) is expected to saturate at high levels" (p. 3). So maybe marginal humans from our current point would in fact reduce NPPeco?

HANPP and NPPact

By definition, HANPP := NPPpot - NPPeco (p. 2), so the red HANPP curve is just the gap between the purple and green curves.

HANPP can be partitioned (p. 2) as HANPPluc + HANPPharv, where HANPPluc (I think "luc" stands for "land-use change") is NPPpot - NPPact, with NPPact being the actual global NPP that exists. In other words, humans appropriate potential NPP both by reducing the productivity of land (HANPPluc) as well as by harvesting biomass from land (HANPPharv). You can see these two components in the graph by looking at the space between the purple and green curves: HANPPluc is the difference between purple and yellow, while HANPPharv is the difference between yellow and green.

NPPact has increased, and in fact, its 2005 value looks slightly higher than NPPpot in 1910 (which is my approximation of a human-free world). Assuming that the fraction of all NPP that's eaten by animals is roughly constant between 1910 and 2005, then it looks as though the metabolism of all animals has slightly increased due to humanity's existence. Of course, as noted before, this estimate is fairly noisy, and it's possible there has actually been a slight decrease in total animal metabolism if, for example, the CO2 fertilization effect was in fact weaker than what's modeled here.

What if NPPpot levels off?

The caption of Figure S2 (not shown in the above screenshot) includes the following numbers:

Value in 1910 (GtC/y) Value in 2005
NPPpot 52 59
NPPact 49 55

Since NPPpot increased by 7 GtC/y over this period, NPPact was probably increased by almost 7 GtC/y as well over what it would have been without CO2 fertilization and other factors increasing NPPpot. In other words, without NPPpot increases from 1910 to 2005, NPPact probably would have declined from 49 GtC/y to roughly 55 - 7 = 48 GtC/y, which is a 2% decrease. And assuming that a human-free world would have had an NPP value of 52 GtC/y (the NPPpot value in 1910), then ignoring NPPpot changes, humanity's existence decreased NPP from a hypothetical 52 GtC/y to 48 GtC/y, an 8% decrease. This decrease was due to a population of ~7 billion humans, which implies an average decrease of NPP of about 8%/7 = 1% per billion humans if we ignore NPPpot changes. And if we assume that NPPpot won't continue to increase going forward, then a 1% decrease in NPPact per billion humans might be a reasonable estimate of the marginal impact of additional births.

That said, a lot of the decrease in NPP was due to the first 1 to 2 billion humans of population size (the world human population in 1900 was 1.56 to 1.71 billion), who decreased NPP from a hypothetical 52 GtC/y without humans to 49 GtC/y. And then the remaining 5 to 6 billion humans of population size only decreased NPPact from 49 GtC/y to 48 GtC/y when we subtract off NPPpot changes.

This calculation assumes the only change from 1910 to 2005 was to increase the size of the human population. This ignores important changes in technology, consumption habits, etc. (Of course, future increases in the human population will also speed up changes in future technology.)

Another complication with the above calculations is that I'm assuming that reductions in NPPpot only depend on the size of the current human population, ignoring history. In fact, some losses in NPPpot were due to past ecosystem changes that still have lingering effects.

Animals created due to HANPPharv

Part of HANPPharv is the biomass that humans eat directly as food, supporting their metabolism. But many human lives are pretty good, so except for lower-level "suffering subroutines" within human brains and bodies, we may not feel as though there's huge amounts of intense suffering from human metabolism, relative to the scale of intense suffering from animal metabolism. (Surely the absolute amount of human suffering is immense. Right now, as you read this, people around the world are being tortured, raped, killed, and so on. ~1.8 humans die every second, often in moderately to extremely painful ways.)

Another big component of HANPPharv is the biomass fed to cattle and other big farm animals. While animal advocates feel that these animals do endure significant suffering, these animals are relatively big, so it's plausible that we would judge them to suffer less per unit of metabolism than smaller, shorter-lived animals do.

Some of HANPPharv consists of biomass that will mostly be eaten by bacteria/archaea, such as perhaps a decent portion of food waste that ends up in landfills.

And some of HANPPharv consists of biomass that humans burn by fire, which doesn't create significant sentience (although it might cause intense pain to those animals accidentally immolated).

However, other parts of HANPPharv will end up feeding appreciable numbers of small animals. Here's an incomplete list of such components of HANPPharv:

  • Feed given to smaller farm animals like chickens.
  • Consumer food waste that's eaten by rats in garbage cans or flies in landfills.
  • Excrement, toilet paper, and food scraps that go to secondary wastewater treatment, which often gives rise to some populations of rotifers, nematodes, etc.
  • Rotting lumber and paper products that may feed invertebrates.
  • "On-site backflows to nature", which I discuss further in Appendix 1 at the bottom of this piece.

So, a decent portion of HANPPharv is actually made available to small animals, many of them non-domesticated animals. Appendix 2 discusses two more minor ways in which HANPPharv may be slightly too high for my purposes. Thus, in my opinion, a decent portion of HANPPharv should really be counted as NPPeco.

In light of this discussion, I think the correct measure of NPP for approximating animal suffering falls somewhere between the yellow and green curves in Figure S2A. If we think the actual suffering-relevant NPP curve is very close to the yellow NPPact, then humanity may have slightly increased the metabolism of non-large animals overall (compared against the blue circle). However, if a nontrivial fraction of HANPPharv is not counted, then humanity may have slightly decreased the metabolism of non-large animals overall. Either way, this net impact of humanity's existence on animal suffering is plausibly within the level of noise in Krausmann et al. (2013)'s calculations, and it's much less than the declines of 50% or more seen in defaunation studies.

How noisy are these results?

Krausmann et al. (2013) compile a sizeable collection of data sources and should be commended for their thoroughness. However, as they admit, there are many gaps in the data, and they made up many assumptions. Even the error bars used for sensitivity analysis mostly "are based on an educated guess of the likely uncertainty range" ("Supporting Information", p. 12). So the numbers presented here should not be taken as gospel.

Most of the calculations use point estimates that have some validity but that could easily be off by some amount. Here are a few examples from a single section (p. 9) of the "Supporting Information", to give a flavor of what I mean:

This biomass flow was assessed by multiplying estimates on the extent of area burnt each year with the average biomass density and a burning efficiency of 54% (46). [...] The number of people living mainly from swidden agricultural systems in 1950 were derived from ref. (47). For South-East Asia and Sub Saharan Africa we assumed the fraction of swidden to regional population to be constant prior to 1950. For Latin America we assumed a decline from 75% of the total population in 1900 to 48% in 1950. [...] Subsequently, the area requirement was calculated as a function of population numbers, per-capita energy demand (assuming that two thirds of a total calorific intake of 4 gigajoule per capita and year (GJ/cap/yr) is derived from swidden agriculture), and agricultural yields. The length of the rotation cycle (fallow periods) was assumed to decrease linearly from 15 years in 1900 to 8 years in 2010 and as an effect of declining fallow periods (50), yields were assumed to decline in parallel linearly from 24 GJ/ha/yr in 1900 to 17 GJ/ha/yr in 2010.

Another perspective on climate effects: Nemani et al. (2003)

The magnitude of CO2 fertilization effect in Krausmann et al. (2013) is surprising, but at least one other study has found a similar trend: Nemani et al. (2003).

Nemani et al. (2003) explain as introduction (p. 1560):

Several regional studies have reported increases in NPP (4–10), but a globally comprehensive analysis of the impacts of climatic changes on NPP is lacking. For the northern mid-latitudes and high latitudes, these studies suggest that multiple mechanisms (e.g., nitrogen deposition, CO2 fertilization, forest regrowth, and climatic changes) have promoted increases in NPP, whereas increases in the tropics have been primarily attributed to CO2 fertilization.

From their own results, the authors find (p. 1561): "Globally, NPP increased [...] by 6.17%, 3.42 PgC over 18 years (P < 0.001), between 1982 and 1999. Ecosystems in all tropical regions and those in the high latitudes of the Northern Hemisphere accounted for 80% of the increase." When I eyeball the NPPact curve in Krausmann et al. (2013)'s Figure S2A above, it looks to me like global NPP increased from ~53.6 GtC/y in 1982 to ~55 GtC/y in 1999 (just my guesses based on measuring the graph), which is a ~3% increase -- half the percent increase estimated by Nemani et al. (2003) over the same period.

This figure from Nemani et al. (2003) shows percent increases or decreases per year in NPP by region:

It's worth remembering that not all of this effect was due to CO2 fertilization. For example (p. 1562):

An increase in NPP of only 0.2% per 1-ppm increase in CO2 could explain all of the estimated global NPP increase of 6.17% over 18 years and is within the range of experimental evidence (27). However, NPP increased by more than 1% per year in Amazonia alone, which accounts for 42% of the global NPP increase between 1982 and 1999. This result cannot be explained solely by CO2 fertilization. We suggest that increases in solar radiation, owing to declining cloud cover in these predominantly radiation-limited forests, is the most likely explanation for the increased tropical NPP (28, 29).

Oceanic NPP

Krausmann et al. (2013) only address terrestrial NPPa. Therefore, when I've said "NPP" above, I've meant "terrestrial NPP". NPP on land accounts for about half of global NPP. So how about changes in oceanic NPP?

Unless we count fishing, it seems as though humans do less appropriation of oceanic NPP. (Fishing is harvesting of secondary, tertiary, etc. production, not primary production.) Smil (2013) says (p. 7) of aquatic photosynthesizers that "direct harvests by humans are rather limited" and that harvests of terrestrial autotrophs "dominate the photosynthate that is used by humans."

In some cases like eutrophication, humans may increase aquatic NPP. On the other hand, several studies suggest that human influence may have decreased phytoplankton in the ocean, though there's high variance in such estimates, with some authors concluding that humans have increased or will increase phytoplankton. More exploration of the oceanic side of things is needed.


Carl Shulman first pointed me to the HANPP literature in 2013.

Appendix 1: Backflows to nature

Krausmann et al. (2013) include in HANPPharv several kinds of biomass that are actually returned to the original ecosystem rather quickly. For example:

  • Krausmann et al. (2013) count ("Supporting Information", p. 4) the NPP of crop roots as harvested NPP: "All belowground NPP on cropland is assumed to be killed during harvest and is accounted for as HANPPharv." But (except for crops where belowground biomass is harvested, like carrots, potatoes, etc.), belowground NPP is actually made available to wild soil decomposers.
  • I think a similar complaint as in the previous bullet applies to the authors' handling of collateral plant damage from timber harvesting ("Supporting Information", p. 8).

A previous study (Haberl et al., 2007) by many of the same authors as Krausmann et al. (2013) explains this in more depth, in its "Supporting Information":

On-site backflows to nature, i.e., unused crop residues, roots or other harvest losses on cropland and in forestry, and livestock feces dropped during grazing, were calculated as the difference between available and recovered crop residues, or the difference between wood fellings and wood removals including bark. For the calculation of livestock feces, we assumed that [...] two-thirds are dropped by the animals on grazing areas, whereas one-third of these excrements is collected by farmers and, hence, not accounted for under this category.

As the name suggests, "on-site backflows" only include biomass made available to wild animals on the location from which they were taken. Hence, off-site backflows, like human food waste thrown out in the trash, or cattle excrement moved elsewhere, aren't counted.

Haberl et al. (2007) include the following table:

On-site backflows to nature are 2.46 Pg C/yr, which is 30% of the total human harvest of 8.18 Pg C/yr. All backflows to nature, including those not on-site (e.g., food wasted by humans, rotting lumber, etc.), are an even bigger percentage. So we should count at most 70% of HANPPharv as actual harvest, and maybe appreciably less than 70%.

Appendix 2: Other minor complaints with HANPPharv

Prevention of natural fires

HANPPharv includes "biomass killed in human-induced fires" (Krausmann et al., 2013). I agree with counting burned biomass in this way. However, I think(?) human-induced fires generally somewhat suppress natural fires (since it takes some time for burnable material to reaccumulate), and I assume that the authors' calculations didn't reduce HANPPharv to account for natural fires prevented? Of course, this effect is probably small.

On the flip side, HANPPharv from burning might also be an underestimate because HANPPharv "does not include the amount of biomass burned on grasslands (excluded for reasons of data availability)" (Krausmann et al., 2013, "Supporting Information", p. 9).))

Harvest in infrastructure areas

Krausmann et al. (2013) say ("Supporting Information", p. 8) that they assume that half of NPP in parks and "grass strips along roads" is "harvested" for human use. But is that realistic? Apart from burning or throwing out leaves and grass clippings, I don't see how humans are harvesting vegetation and taking it away from wildlife in those cases.

The main text of Krausmann et al. (2013) says: "HANPPharv on built-up land (e.g., mowing lawns and cutting trees) is assumed to be 50% of NPPact (as in ref. 17)." I can see how mowing lawns and cutting trees can be seen as "harvesting", but I would think that most of the biomass is returned to nature, either on-site or off-site. Maybe some fraction of grass and wood is burned, but probably not 50%?

The "ref. 17" source mentioned in the previous paragraph is Haberl et al. (2007). But as far as I can tell, Haberl et al. (2007) don't justify the 50% number, either -- they merely take it as an assumption: "Harvest in infrastructure areas, e.g., biomass harvested during gardening or park and infrastructure maintenance, was assumed to amount to 50% of the aboveground NPPact."


  1. I read the whole Krausmann et al. (2013) paper and supplement and didn't see mention of oceanic NPP. Moreover, Krausmann et al. (2013) say that HANPP "by 2005 reached 14.8 GtC/y". But global NPP is 104.9 GtC/y. So even relative to the world's actual NPP (not potential NPP absent human land-use change), 14.8 GtC/y can't be as high as the 25% HANPP found by the authors. Hence, I infer that Krausmann et al. (2013) is only counting terrestrial NPP.  (back)