The Global Body Size Biomass Spectrum is Multimodal

Recent research provides an unprecedented account of the diversity and biomass of life, but the data also suggest unexplained patterns such as the co-dominance of very different life forms. We compile the planetary body size biomass spectrum across all taxa and investigate possible underlying forces. We find that small (10 -14 g) and large (10 6 g) organisms vastly 5 outweigh other sizes. The global spectrum reveals an allometric power exponent close to zero, 6 with the marine spectrum in particular showing multiple closely packed modes that are compatible 7 with metabolic food webs. All habitat realms share two distinct size modes that correspond well to 8 the evolutionary innovations of unicellular and complex multicellular life forms, plus a smaller third mode representing unicellular endosymbiotic life. Each mode contains both producers and 10 consumers. These findings show both differences and similarities across habitat realms and point 11 to a size-based synthesis of microevolution, macroevolution, and macroecology.


13
Body size is a fundamental metric in biodiversity and evolutionary sciences 1 . Size 14 spectra-be they uniform, unimodal, or multimodal-help identify governing forces. Organisms 15 range from 10 -17 (Nanoarchaeum equitans) to 10 9 g (Sequoiadendron giganteum) in carbon 16 weight, but the global size-biomass distribution, or size spectrum, remains to be comprehensively 17 explored across taxa 2 . Previously, empirical studies showed that biomass appears relatively 18 equal across broad but still limited body size ranges either within habitat realms 3 or when 19 averaged over species 4 . In other words, biomass has so far appeared to be nearly scale-invariant 20 across many taxa.

21
Theories for the body size biomass spectrum have also been limited to specific taxa or 22 focused on specific realms, but they can be roughly divided into evolutionary and ecological.

23
Within taxa such as terrestrial mammals, macroevolutionary theory focusing on variations 24 between species predicts a unimodal and right skewed spectrum of species richness 5 and hence 25 a right skewed body size spectrum when combined with roughly equal species biomass across 26 size 4 (high biomass at small sizes and some rare, very large sizes). Across taxa, the evolution of 27 complexity 6,7 identifies several distinct life forms (simple unicellular, unicellular endosymbiotic, 28 and complex multicellular) that each occupy different body sizes separated by rare and abrupt 29 evolutionary transitions 8,9,6 . Within each life form, larger sizes may be selected for 10 up to an 30 upper limit 11-13 , consistent with macroevolutionary theory. However, there is no quantitative theory 31 for whether or how size modes may evolve -an open question that may determine their debated 32 status as major transitions. The evolutionary perspective also does not explicitly account for 33 ecological trophic interactions, unlike the following theories. In the marine realm, where primary 34 producers tend to be small and trophic interactions are often governed strongly by size 3,14 , 35 metabolic food web theory predicts a broad-scale power law with a small negative or zero 36 exponent (called scale invariance). In addition, food web theory also predicts finer-scale modes 15-37 17 separated by predator-prey mass ratios of approximately three orders of magnitude, a 38 4 separation that increases with predator size 18

49
A lack of cross-taxa data integration and coarse empirical size resolution has prevented 50 further theoretical refinement and synthesis relevant to understanding the global distribution of 51 body sizes on Earth. Further, size-biomass distributions are either non-existent or heavily biased 52 even within well-studied taxonomic groups. Given that population census remains prohibitively 53 expensive even for a few species, here we obtain a first global picture by pairing previous 54 estimates of cross-taxa biomass with size distribution estimates based on key species within 55 groups. We then examine the cross-taxa size-biomass patterns, offer preliminary mechanistic 56 interpretations, and extensively test the sensitivity of the results to different methodological 57 approaches.

60
To compile the global aggregate body size biomass spectrum among biological groups 61 defined by habitat and taxonomy, we used global biomass (gigatons [Gt] in carbon content) 62 assessments and minimum, mean, and maximum body sizes (grams [g] in carbon content) within 63 groups (Tables S1-3). We relied on the most comprehensive existing synthesis of global biomass 64 5 estimates, which incorporate uncertainties within and between multiple studies 24 , and then 65 complemented these with improved estimates available for cryptogamic phototrophs 25 , hard 66 corals 26,27 , mangroves 28 , and subterranean prokaryotes 29 . We allocated biomass as a skew 67 normal distribution within each of the 36 groups shown in Figure 1    individual groups (Tables S1-S3) was large enough in some cases that the ranking of groups by 84 biomass for a given body size (as depicted in Figure 1) is itself uncertain. For example, according 85 to the best available estimates, terrestrial microbes outweigh marine microbes by five-fold, but 86 their data uncertainties (2-to 6-fold) make it possible that the true ranking is the opposite (Tables 87 S1-S2).

88
The most diverse body size (10 4 g with 53% of groups represented) was close to but did 89 not coincide with the large body size peak (10 6 g with 38% of groups represented), whereas the 6 small (10 -14 g) and medium (10 -9 g) body size peaks contained 19% and 44% of groups 91 represented respectively ( Figure 1C). This diversity pattern is likely influenced by the tendency to 92 name organismal groups at finer resolution near our own size, illustrating a strong size bias in 93 naming that does not correspond to the macroecological distribution of biomass.

94
The two highest peaks appear to correspond to major transitions in complexity. The

103
We also characterized size spectra allometrically using power functions with exponent

119
The first and last represent the highest biomass peaks, consistent with the cross-realm pattern.

120
Additionally, the first four peaks roughly correspond to green algae, protists, arthropods, and fish, 121 resembling a trophic chain. These separations of 3 to 6 orders of magnitude increase with body 122 size and roughly correspond to, but are larger than, empirical predator-prey size ratios 18 and 123 trophic expectations 19 . This multimodal food chain pattern is anchored in the green algae that 124 access sunlight efficiently as microscopic plankton near the ocean surface, but algae do not 125 comprise the largest biomass peak because high ecological turnover and efficiencies mean they 126 are rapidly consumed and converted to biomass in higher trophic organisms 19,23 . In addition, the 127 marine realm also has some large producers and small consumers that complicate a simple food 128 chain perspective. Producers like macroalgae, seagrass and mangroves, as well as corals that 129 host symbiotic producers, represent a strategy that is different from planktonic primary production 130 and that contributes to the largest body size peak. The most notable of these are mangroves, 131 which are terrestrial-like producers that grow attached to the shallow benthos and become large 132 to compete for light 36 . In addition, decomposer microbes near the top of food chains contribute to 133 the smallest body size peak along with unicellular producers. Thus, likely for both trophic and 134 non-trophic reasons, biomass spectra in the ocean are relatively flat but with the highest 135 aggregate biomasses in the small and large size peaks.

136
The terrestrial realm has more strongly dominant peaks at sizes of 10 -14 and 10 6 g ( Figure   137 2B), as well as a minor peak at 10 -9 g, which are similar to the global pattern and to predictions 138 based on evolutionary transitions. Ecological efficiencies are lower on land versus in the ocean, 139 which allows producers to dominate. Terrestrial biophysics dictate that producers grow large to 140 compete for light, in contrast to marine producers that can be small or large. However, animal 141 consumers also share the large size peak. Prokaryotic decomposers dominate the smallest size

144
Gaussian mixture models also identified three modes as the best description for terrestrial life 145 (R 2 =0.94, Figure S1B). Moreover, the three terrestrial size modes roughly line up with three of the 146 more prominent marine size modes (prokaryotes, protists, and plants/animals), suggesting that 147 even in the ocean where trophic-size structure is important, evolutionary transitions remain 148 evident.

149
The size spectrum can also be divided by the major routes of energy acquisition:

152
The producer exponent largely reflects the dominance of large terrestrial trees and grasses, and 153 the Gaussian mixture model identified only one mode ( Figure S1E). In contrast, the consumer 154 exponent is strongly shaped by trophic interactions and is compatible with metabolic food web 155 theories that predict a slightly negative trend 15-17 with multiple modes ( Figure S1F).

156
We note that compiling the global body size biomass spectrum required numerous

220
We fit skew normal distributions (Eq. (1)) to three observed reference sizes for each 221 organismal group compiled from the literature: minimum, mean, and maximum sizes (Tables S1-222 3), with biomass and fold uncertainty (mean x fold and mean/fold corresponding to 95% CI of a biological group by dividing biomass by abundance when available 24 ; otherwise, we used the size 232 of the most representative species from an independent literature search (those mentioned as 233 most "common" or "widespread"). In a body size-biomass graph and in subsequent discussions,

234
"biomass" is short for biomass density, i.e., the biomass that is expected to be found within one 235 log-biomass unit at a given body size.

236
The body sizes of some species were reported in units of grams carbon, but for many 237 species we needed to extrapolate from wet or dry mass. When size estimates in the literature 238 were reported in wet mass, we first searched the literature for a species-specific wet weight to 239 grams carbon conversion. When a species-specific conversion was not available, we used the 12 conversion from the closest relative within the taxon (see online repository tables). When taxon-241 specific conversions were not available, we assumed 30% dry mass per wet mass unit, and 50% 242 carbon per dry mass unit following previous conventions 24 . In some cases, body size was 243 reported in units of length (particularly among annelids, nematodes, and fishes). For these taxa,

244
we found existing length to weight conversions for the species or the closest relative within the 245 taxon. If body size was reported in diameter, as was the case for most unicellular species, we 246 found the volume assuming that the organism was either spherical 47 or tubular 48 , and then found 247 existing biovolume to biomass conversions for the species or the closest relative within the taxon.

248
For hard corals, since each corallite or colony is often tightly packed among other units, we 249 estimated that volume as the cube of the reported diameter.

250
We excluded from our size datapoints non-free-living disease organisms, which are  incorporates both observational and systematic assumption uncertainties that can be qualitative.

267
We then summed the size-biomass distributions of all groups to generate a single cross-taxa size 268 spectrum for each bootstrap. This process was repeated for each of the 1000 sample sets. The 269 2.5 th and 97.5 th percentiles at each size bin were recorded as the confidence bounds.   (Table S4). In particular, the original large size range for soil fungi was  Figure S2A shows the cross-taxa spectrum with body size measured 288 at the ramet scale. size and biomass. To avoid this conundrum and following previous works 24 , we chose to include 294 skeletal mass in the main text to decouple biomass from metabolism. However, in Table S5 we 295 provide new estimates for biomass and body sizes when the portion of biomass with low 296 metabolism is excluded. This resulted in an alternative cross-taxa spectrum ( Figure S2B).

297
3) and 4) We tested the sensitivity of the size spectrum to the assumption of uniform or 298 normal biomass distributions within groups. For a uniform distribution, biomass was identical 299 across log sizes within groups. The results are in Figure S2C and D.

441
Marine. Subterranean prokaryotes are excluded. See Figure 1 for color reference.

443
A B