Evidence of hidden hunger in Darwin’s finches as a result of non-native species invasion of the Galapagoes cloud forest

Invasive species pose a major threat to forest biodiversity, particularly on islands, such as the Galapagos. Here, invasive plants are threatening the remnants of the unique cloud forest and its iconic Darwin’s finches. We posit that food web disturbances caused by invasive Rubus niveus (blackberry), but also the management measures used to control it, could contribute to the rapid decline of the insectivourous warbler finch (Certhidae olivacea). We compared changes in long-term management, short-term management and unmanaged areas. We measured C:N ratios, δ15N-nitrogen and δ13C-carbon signatures in bird blood and arthropods, as indicators of resource use change, in addition to mass abundance and diversity of arthropods. We reconstructed the bird’s diets using isotope mixing models. The results revealed that finches in (Rubus-invaded) unmanaged areas foraged on abundant yet low quality arthropods and had shorter tarsi. Is this the first evidence of hidden hunger in degraded terrestrial ecosystems in Galapagos?


17
Abstract. Invasive species pose a major threat to forest biodiversity, particularly on islands, 18 such as the Galapagos. Here, invasive plants are threatening the remnants of the unique cloud 19 forest and its iconic Darwin's finches. We posit that food web disturbances caused by invasive 20 Rubus niveus (blackberry), but also the management measures used to control it, could 21 contribute to the rapid decline of the insectivourous warbler finch (Certhidae olivacea). We 22 compared changes in long-term management, short-term management and unmanaged 23 areas. We measured C:N ratios, δ 15 N-nitrogen and δ 13 C-carbon signatures in bird blood and 24 arthropods, as indicators of resource use change, in addition to mass abundance and diversity 25 of arthropods. We reconstructed the bird's diets using isotope mixing models. The results 26 revealed that finches in (Rubus-invaded) unmanaged areas foraged on abundant yet 27 low quality arthropods and had shorter tarsi. Is this the first evidence of hidden hunger in 28 degraded terrestrial ecosystems in Galapagos? 29 Introduction: 30 Invasive species are a major threat to biodiversity globally, even more to endemic island 31 species which are particularly vulnerable, as host species gene pools and "escape" strategies 32 are more restricted on insular island ecosystems (Atkinson, 1989). Species invasions at the 33 primary producer level can cause massive ecosystem level changes (Szabo et al., 2012). In 34 times of rapidly dwindling biodiversity, intensive habitat management is often the only option 35 available to tackle invasive species and save the threatened focal species and/or their 36 ecosystems (Moser et al., 2018). However, intensive management such as physical or 37 chemical plant removal can also cause ecosystem disturbances and have detrimental effects 38 on non-target species. Assessing the direct and peripheral effects of management measures 39 is difficult and labour-and time-intensive. Here we present a novel stable isotope approach 40 that detects and diagnoses ecosystem degradation, allowing for rapid response actions. 41 Disturbance of food web structures and niche structure degradation are implicitly preserved in 42 the isotopic signature of the focal species, as isotopic ratio of an organism is the result of all 43 trophic pathways making up that individual, reflecting the trophic niche (Layman et al., 2012). the most, with a decline of up to 50% in the forested and 75% in agricultural areas (Mauchamp 52 and Atkinson, 2010). The primary habitat of the warbler finch is cloud forest that has 53 experienced a 99% reduction in its area since the middle of the last century (Dvorak et al.,54 2012) due to human agricultural activities but also due to invasion of introduced plant species 55 such as the invasive blackberry Rubus niveus (Rentería et al., 2012). The cloud forest is 56 dominated by the endemic tree species Scalesia pedunculata (Rentería et al., 2012) with max. 57 height of 12 m and a breast height diameter of up to 30 cm (Jäger, unpubl. data). Scalesia is 58 a member of the daisy family Asteraceae and grows in dense stands in the humid zones of the 59 major Galapagos islands. Although on a decadal scale, Scalesia canopy-cover is not affected 60 by the Rubus invasion, at the invaded sites, understory plant composition is dramatically 61 altered; with impenetrable dense thickets of Rubus with an above ground biomass of up to 10t 62 ha -1 (Rentería et al., 2012). We hypothesised that areas invaded by Rubus act as a plentiful 63 food resource for the primary consumers, mainly arthropods and that the subsequent 64 consumption by the finches of the available abundant "low quality" primary consumers, leads 65 to trophic disturbances with physiological consequences for the insectivorous birds. 66 The Galapagos National Park Directorate has pursued a policy of intensive Rubus removal, 67 with machetes and subsequent herbicide control since 2003, to protect the remaining Scalesia 68 forest. The invasive species management leads to the temporary removal of the understory in 69 the controlled areas and a reduced availability of arthropods (Cimadom et al., 2019). In the 70 immediate aftermath of control measures, significant reductions in the warbler finch's breeding 71 success have been observed (Cimadom et al., 2014) suggesting a major disruption of the 72 warbler finch's food web structure (Cimadom et al., 2019). We posit that both the invasion of 73 Rubus and the management of Rubus cause major ecosystem level changes in resource 74 structures, which have significant implications for our focal species, the warbler finch 75 (Certhidea olivacea). 76 We sought to determine whether measuring the isotope signatures and stoiciometry of blood 77 samples from the focal species could be used as a metric to indicate habitat degradation, thus 78 quantitatively characterizing trophic structures. Laymann (Layman et al., 2007) suggested, 79 δ 13 C-δ 15 N niche space is a representation of the total extent of trophic diversity within a food 80 web. This is based on the premise that organisms, consumers and prey species reflect the 81 consequences of changes in their environmental conditions and habitat structure, revealing 82 shifts in their diet through the isotopic signatures in their blood (Fry, 2006 Furthermore, using the stable isotope signature, it is possible to determine the trophic position 107 of different organisms in the food web. In general, there is a slight discrimination in the isotopic 108 components (C and N) in animals with respect to their diet Epstein, 1981, 1978). 109 Trophic fractionation of nitrogen is generally higher than that of carbon and is caused by the 110 preferential metabolism of light nitrogen compounds (Podlesak and McWilliams, 2006 Even if the three study areas have similar overall arthropod productivity, the proportion of 134 arthropod species which are suitable as prey could be lower. Thus, we asked. Does the 135 predominant prey consumed differ between the invaded and managed areas and is there a 136 difference in prey quality? To address these questions, we used traditional gravimetric and 137 abundance analysis, in combination with stable isotope and stoichiometric signatures. 138 Specifically, we measured whether there is an overlap of carbon and nitrogen isotopic 139 signatures between available prey and the warbler finch's blood (after accounting for trophic 140 fractionation) to gain an understanding of feeding pathways and total niche space. In addition, 141 we obtained information on the prey quality i.e. carbohydrate/fats versus protein (C:N). 142 We hypothesized that the near-complete removal of the forest understory leads to a decrease 143 in the quantity of arthropod prey available, and the presence of Rubus leads to an increase in 144 available low quality arthropod prey, resulting in a decrease in bird-body mass index (BBMI), 145 weight and tarsus length, as indicators of finches' overall condition. 146

Arthropod biomass 149
Biomass was measured in two sampling rounds in 2015; one at the beginning of the breeding 150 season in late January (round 1) and one in the middle of the breeding season in mid-April 151 (round 2). The wet and warm season usually begins in early December and precipitation tails 152 off, usually finishing by the end of May, with the drier and cooler weather dominating for the 153 rest of the year. Overall, arthropod biomass (dry weight) was highest in the long-term 154 management area (LTM) and lowest in the recently controlled area (RC). When comparing the 155 arthropod biomass across forest strata or layers, the canopy had significantly higher arthropod 156 biomass than the other layers, moss and understory, in all cases (F(2,162)꞊25.323 P<0.001), 157 ( Figure 1). In the middle of the breeding season (round 2, mid-April) arthropod biomass was 158 significantly higher than at the beginning of the breeding season (round 1, late January), 159 (F(1,162)꞊5.104, P꞊0.025). Overall, recently controlled areas (RC) in round 1 had the lowest 160 arthropod biomass and long-term managed areas (LTM) in round 2 had the highest (Figure 2). 161 We discounted Diplopoda from our analysis, as observations had shown, from focal follows 162 monitoring bird foraging, as well as bird stomach contents analysis, that due to their large size, 163 Diplopoda were never eaten by the finches. Following convention, once it has been established 164 that birds do not feed on a specific species, herein Diplopoda, it is reasonable that they can be 165 excluded from the investigation (Wolda, 1990) 166 Relative arthropod biomass data showed similar relative abundance patterns within the forest 167 layers across management areas. When comparing dominant arthropod orders between 168 rounds and forest layers, we found that in each sampling round, each forest layer had 169 consistently similar dominant arthropod orders ( Figure 2). However, the dominant orders 170 (ranked by dry weight biomass) differed between forest layers. In the canopy, the two dominant 171 orders were Araneae and Coleoptera in all three study areas. In the two areas (RC & LTM), 172 the third most important order was Hemiptera, which were almost absent in the unmanaged 173 area (NC). In the unmanaged area, the third most important order was Lepidoptera. 174 In the moss layer, the dominant arthropod orders differed between the three study areas: In 175 the recently controlled area (RC), the most dominant orders were Lepidoptera, followed by 176 Hymenoptera and Araneae. In the long-term-managed area (LTM), dominant orders were 177 Coleoptera and Hymenoptera followed by Araneae. In the un-managed (NC) the most 178 dominant order was Acari, followed, by Lepidoptera and Araneae. 179 In the understory from the recently controlled area (RC), the dominant order was Araneae, 180 followed by Diptera and Orthoptera. In the long-term-managed area (LTM), the dominant order 181 was Hemiptera, followed by Orthoptera and Araneae. In the unmanaged area (NC), the order 182 of the dominant orders was the same as in the long-term management area (LTM) but the 183 relative mass abundance of Araneae was much lower than in the other two areas (less than 184 10%). 185

Primary producer isotope signatures 186
Scalesia pedunculata was set as the isotopic dietary baseline. The δ 15 N isotopic signatures of 187 both the Scalesia and the Rubus leaves were not significantly different across the three 188 different management areas but nitrogen isotopes signatures of the two species were 189 significantly different from one another (Scalesia mean: 5.7‰ and Rubus mean: 1.2‰, 190 F(1,57)=139 p<0.001). There were significant differences in δ 13 C of Scalesia across sampling 191 events of the different rounds, but not between management areas, in the drier-latter part of 192 the breeding season-round 2 values were more enriched (F(1,24)=18.56 p<0.0001 Figure S2), 193 this was attributable to differences in seasonal plant water availability and had no influence on 194 the consequent dietary reconstructions. There was no significant difference between the 195 molecular C:N ratios of Rubus and Scalesia leaves. Canopy C:N ratios were significantly lower than the other two forest layers, moss (P<0.001) 207 and understory (P꞊0.047). Nitrogen densities (mg N m -2 ) were significantly different across 208 forest layers (F(2,158)꞊20.280, P<0.001). Canopy arthropods had significantly higher nitrogen 209 densities than the other two forest layers, moss and understory (P<0.001 in both cases). 210 Although whole system mean arthropod nitrogen densities ranged between 3.97 and 8.31 mg 211 N m -2 for the three management areas) significant differences were not detected, possibly a 212 consequence of high variation and compounding measurement uncertainties. 213 The trophic structures of the Scalesia forest persisted across the management types based on 214 the isotope signatures, warbler finches occupied the highest position in the trophic web 215 sampled. As predicted, arthropods, in general, occupied the lower levels, the herbivorous 216 arthropods consistently had the lowest δ 15 N values and the carnivorous arthropods "a few per 217 mil" higher. Clear trophic isotopic enrichment was observed in the finch blood in the long-term 218 management area (LTM), with the highest δ 15 N of and δ 15 N-range (of all the areas, figure 5) 219 with less distinct differences between secondary consumers observed in the recently 220 controlled area (RC) and unmanaged area (NC). components were Araneae, Hemiptera, Diptera and Lepidoptera, as predicted by the model 244 ( Table 2). The proportion of Diptera and Hemiptera was higher in the diet than expected from 245 availabity, which shows that birds were clearly avoiding the Coleoptera. In the long-term 246 management (LTM), Araneae and Hemiptera comprised over 70% of the diet, according to the 247 model, and again, the warbler finch was not consuming the Coleoptera (Table 2). In the 248 unmanaged area (NC), Hemiptera and Lepidoptera comprised 80% of the diet, based on the 249 model, which was also a consiberably higher proportion of the diet than expected according to 250 availabilty (Table 2)  There was no significant or predictive correlation between δ 15 N-bird-blood and BBMI or bird Discussion: 281 In this study, arthropod biomass and isotopic data, combined with differences in δ 15 N, but not 282 δ 13 C signatures, of the warbler finch's blood across management areas suggested changes in 283 the underlying food web structure ( Figure 4). As hypothesised, the higher mean arthropod 284 biomass, lower C:N ratio and higher δ 15 N-δ 13 C range in the long term-managed area (LTM) 285 suggested that these areas had recovered or semi recovered their trophic structure, compared 286 to the recently controlled (RC) and unmanaged (NC) areas with compromised niche structures. 287 Classical ecological theory suggests that a sympatric species in established ecosystems have 288 minimal resource use overlap, a consequence of competitive exclusion (Gause, 1934) 289 explaining the broader niche space in the long term-managed area (LTM). This is in line with 290 previous studies in benthic systems demonstrating that invasive species occupy a tighter 291 isotopic niche space than their native counterparts (Jackson et al., 2012). 292 We scaled relative arthropod abundance measurements to absolute arthropod abundance per 293 area measurements, based on dry weight data of the different arthropod orders per plot and 294 layer. Although we did not capture all flying insects in our sampling strategy, we felt our 295 methods enabled a reasonable estimate of the available prey biomass. Warbler finches are 296 typical gleaners, not aerial hunters, which is consistent with observations and feeding 297 frequencies in our study. 298

299
The isotope mixing model-MixSIAR analysis revealed that the bird's dominant dietary 300 components did not match the measured relative mass abundances of available arthropods in 301 the long-term-managed area (LTM) and the unmanaged area (NC). This suggests a higher 302 degree of prey selectivity in those areas and is consistent with the Optimal Diet Theory, which 303 posits that predators chose prey that maximise their fitness (Christiansen et al., 1977;Endler, 304 1986). However, the available and consumed arthropod proportions were most similar in the 305 recently-controlled area (RC). In addition, the lowest arthropod biomass was measured in the 306 recently controlled area (RC), the highest arthropod biomass in the long-term managed area 307 (LTM), indeed LTM biomass was three times that of the recently controlled areas (RC) and 308 double that of the unmanaged areas (NC). These results suggest that low prey availability per 309 se leads to less dietary choice, a phenomenon previously observed in long eared owls in 310 Finland (Korpimäki, 1992). This comparison of isotope modelled-consumed versus available 311 prey data is a useful metric, although rarely explored in the study of terrestrial ecosystems. 312 Arthropod biomass was ten times higher in the canopy compared to the moss and understory 313 layers. It was significantly greater in round 2, than in round 1. This difference was most likely 314 the result of the more humid conditions preceding the second sampling round, which is the LTM and RC, canopy arthropods accounted for more than 95% of the finches' diet but in the 318 unmanaged area (NC), it was only 41%, with a higher proportion coming from the understory 319 (52%). This is corroborated by foraging observations that showed that warbler finches used 320 the understory more frequently in the unmanaged areas (NC) than in the long-term-managed The trophic consequence of the finches feeding predominantly on understory arthropods in the 327 unmanaged area (NC) was detectable in the C:N ratio of the finch blood over both years and 328 rounds. The C:N ratio of the blood is an indicator of diet quality (Gödecke et al., 2018) and was 329 significantly higher in the unmanaged area (NC). These patterns were also observed overall 330 in the C:N ratios of the arthropod samples, indicating that the arthropods, the warbler finches 331 fed on, from in the unmanaged area (NC), were of significantly lower dietary quality, with higher 332 C:N ratios lower and protein concentrations. There was no significant difference between the 333 C:N ratios of Rubus and Scalesia leaves. This suggests that differences in arthropod C:N 334 ratios, were a consequence of the lack of a secondary-arthropod-consumer in the trophic 335 pyramid, the greater proportional abundance of available lower quality prey. A possible reason 336 for this is that the trophic pyramid had not had sufficient adaptive time to exploit all the available 337 niches in accordance with the classical ecological theory of competitive exclusion (Gause,338 1934). In the unmanaged area (NC), the consistently lower δ 15 N range ( Figure 5), specifically 339 the tighter trophic distance between secondary consumer (δ 15 N carnivorous arthropod) and 340 apex consumer in this system (δ 15 Nbird), as well as the lower standard deviation of δ 15 N of the 341 components therein, suggested a more constrained trophic structure, corroborating this finding 342 (Layman et al., 2007). is required for normal growth as well as the development of brain and visual systems. 358 Differences in finch diet quantity and quality led to significant differences in warbler finches' 359 size. Finches had significantly shorter tarsus length in the unmanaged area (NC) than in the 360 managed areas (RC and LTM), but there were no significant differences between rounds or 361 years (Figure 3). We argue that tarsus length is a more robust measure of long-term nutritional 362 status than bird weight or BBMI, as it is an integrated measure and not subject to daily 363 variations due to environmental or physiological status (Kempster et al., 2007). Evident from 364 the fact that despite the differences in finch diet quantity and quality, there were no significant 365 effects on bird weight or bird-BMI, between areas, years or rounds. 366 Warbler finches also had significantly lower breeding success in the short term management 367 areas (RC), evidently a consequence of lower total arthropod mass (Cimadom et al., 2019). 368 Finch breeding success was higher in both the unmanaged and long-term managed areas 369 (Cimadom et al., 2019). This suggests that despite lower food quality in the unmanaged area, 370 breeding was not affected. The shorter tarsus length, however, could indicate that parents 371 compensated quality with quantity, fulfilling the chick's calorific needs but not necessarily their 372 nutritional requirements, at the expense of the size of the chicks. Low protein conditions and 373 lack of nutrient funnelling may have caused the shift towards smaller finch size and shows 374 parallels to human hidden hunger (Gödecke et al., 2018). An alternative explanation is that 375 smaller finches were competitively driven out from the higher quality habitats. In the short term 376 management area (RC), quantity of food rather than quality appeared to be the dominant 377 constraint. 378 Taken together, our data herein suggest that there is a trophic pyramid collapse, due to the 379 invasion of Rubus; a bottom-up control on ecosystem productivity and quality. This shift to a 380 low quality diet was evident in both the isotopic and stoichiometric signatures of the warbler 381 finch's blood and arthropod biomass we posit that it subsequently influenced warbler finch's 382 size. We suggest that rapid environmental change due to the Rubus invasion did not allow for 383 the finch population or the lower orders in the food web to adapt or adjust to the presence of 384 the novel low quality diet. 385 Our date show that although management and control of invasive Rubus leads to dramatic 386 temporary declines in food availability, these vital food resources can be re-established with 387 persistent control measures and time. In this study, we demonstrate it is logistically feasible 388 and financially possible to provide early warning signals of habitat degradation, using isotope 389 and stoichiometric data, which can then provide management insights for effective ecosystem 390 restoration. 391 392

Materials and methods. 450
Study site: The study was conducted in the Scalesia forest at "Los Gemelos" (00°37'20" S, 451 90°23'00" W) on Santa Cruz Island, Galapagos, in 2015 and 2016. This area is dominated by 452 the endemic tree Scalesia pedunculata, but most of its understory has been invaded by 453 blackberry Rubus niveus. In some areas, the Galapagos National Park Directorate (GNPD) 454 had controlled Rubus in order to restore the native forest. Within the forest, we defined three 455 study areas, which differed in whether management of invasive Rubus took place and when: 456 (1) the unmanaged area (NC, 8 ha), which was heavily invaded by Rubus and had never been

Sampling campaigns
Data on daily precipitation was provided by the nearest weather station (operated by Rolf 478 Sievers, S 0°39′57.49″ W 90°22′35.04″) located about 4.5 km south and 150-200 m lower than 479 the study site Los Gemelos. Precipitation data were available over the entire study period. 480 For each round in both years, Warbler finch blood samples and arthropod samples were 481 collected from the three management areas. Scalesia pedunculata leaves from the canopy 482 and Rubus niveus leaves from the understory were sampled randomly across management 483 areas (5 replicates per area per round for Scalesia, 10 replicates for Rubus). 484 We collected ten replicate samples of warbler finch's blood per management area (LTM, RM 485 and NC), round (1 and 2) and Year (2015 and 2016), 120 samples in total. Blood samples were 486 obtained by pinpricking the brachial vein of the finches with a lancet (Tebbich et al., 2004). 487 One blood sample per individual was collected on a 5 mm 2 Whatman GFA fibreglass filter 488 discs, which was stored inside a coded test tube for subsequent isotope analysis. 489 Birds were captured with mist nets and ringed to avoid pseudoreplication. Left tarsus length 490 was measured using a calliper (accuracy 0.01 mm). Each tarsus was measured twice, an 491 average of the two values was used in our calculations. Birds were weighed using a field 492 balance (accuracy 0.1 g). We developed a bird body mass index (B-BMI) based on the tarsus 493 length and weight. We used an anologous formula to that of humans (WHO, 1995) weight (kg) 494 /height (m) 2 , substituting height with tarsus length, as we used the tarsus length as a parameter 495 for growth. All units were converted accordingly. 496 We collected the arthropods according to the collection procedure established by Schmidt-497 Yáñez (Schmidt Yáñez, 2016) from three defined microhabitats specifically canopy, moss and 498 understory. Small tree finches and warbler finches mainly forage in the canopy, understory and 499 in the moss growing on tree trunks (Filek et al. 2018) and we sampled arthropod biomass in 500 each of these micro habitats. Canopy samples were taken by branch clipping. For this, a white 501 polyester bag (diameter 50 cm, length 150 cm) was attached with clips to a metal ring (diameter 502 50 cm) at the top of a five-meter bamboo pole. The bag was pulled over a branch in the canopy 503 (3-5 m height). The branch was then immediately cut off with a loppers (GARDENA), so that it 504 fell into the collection bag, which was closed immediately by twisting the pole to prevent 505 arthropods from escaping. Branches and leaves were then examined for arthropods inside the 506 bag. All encountered arthropods were collected with an aspirator and stored in 70% alcohol. 507 The branches and leaves were then put into a separate Ziploc bag for a second examination 508 in the laboratory. The leaves were subsequently dried for 72 hours at ca. 60°C in a drying 509 chamber to determine the dry weight. 510 Arthropods within the moss were collected from the same trees as the corresponding canopy 511 samples. A 50 cm wide band of moss was carefully scratched off from the circumference of 512 the tree trunk at a height of 1.5 m and transferred into a plastic tray. The moss was then briefly 513 searched for larger arthropods that might escape from the tray area and then placed in a Ziploc 514 bag for a second examination in the laboratory. As with the previous samples, all arthropods 515 were stored in 70% alcohol. The moss samples were dried for 72 hours at ca. 60°C in a drying 516 chamber to determine the dry weight. 517 To sample the understory, 5 m long transects with a buffer of 1 m width in each direction 518 amounting to an area of 10 m² were visually searched for 15 min by one person. Arthropods 519 encountered on vegetation up to 1.7 m above the ground were collected either by hand or with 520 an aspirator and stored in 70% alcohol. Flying insects could not be recorded by this method. 521 Standard methods to sample insects from understory vegetation (e.g. using a sweep net) could 522 not be used, as the understory vegetation in our study area was invaded by spiny R. niveus.

523
A canopy, understory and moss sample were collected in ten randomly selected sampling 524 points in each of the three study areas We chose these microhabitats/forest layers because 525 they were identified as the most important foraging substrates of the warbler finch (Filek et al.,526 2018). In 2015, we collected a total of 180 composite arthropod samples, ten replicates per 527 forest layer (canopy, moss and understory), per management area (LTM, RC and NC), at two 528 times (rounds 1 and 2) at the beginning of the breeding season late January (round 1) and in 529 the middle of the breeding season in mid-April (round 2). The composite samples were created 530 amassing the individual arthropods (all species), sampled at specific layer in a specific 531 management area. All arthropods were collected regardless of their life stages and stored in 532 70% ethanol. 533 Dry mass values of arthropods were obtained by washing off the ethanol three times with 534 deionised water and drying samples at 50°C overnight between each wash. This washing 535 procedure had been tested and shown not to affect either δ 13 C and δ 15 N or nutrient content of 536 the sample (Hood-Nowotny et al., 2016). Once dried, we weighed the samples (each arthropod 537 order separately per field replicate) on a five-figure precision balance. 538 We identified samples from round 1 to arthropod order. We selected a representative group 539 (individuals from one particular order) of each composite sample (from round 1) to be analysed 540 for isotopic signature separately for further use in a diet reconstruction model. We chose the 541 orders with the highest percentage of dry mass per sample, ensuring that there was at least 542 one representative order (sub-sample) per forest layer or one per management area. Orders 543 with less than 5% of dry mass out of the total mass of the composite sample, were not chosen 544 for the individual isotopic analysis. With these representative orders, 162 additional sub-545 samples were created. We reunited the sub-samples with their respective analysed composite 546 sample mathematically, by means of simple isotope based mass balance equations. This 547 procedure was adopted to allow capturing the data in a logistically and economically feasible 548 manner. 549 Once dry mass values were obtained, all composite samples and sub-samples were dried 550 again, milled (Retch, DE) homogenised and a representative aliquot transferred (typically 3 551 mg) into 3.5 X 5 mm tin capsules, for analysis of stable isotopes of carbon and nitrogen, with 552 a full range of standards bracketing all sample values. Subsequently, δ 13 C and δ 15 N of the sub-553 samples were back-calculated mathematically, using a simple mass balance equation and 554 reunited individual sample to the corresponding composite sample to allow statistical analysis. 555 IRMS samples were analysed using a Flash 2000 Elemental Analyser in carbon and nitrogen 556 configuration, linked to a Thermo Scientific Delta V Advantage automated isotope ratio mass 557 spectrometer (IRMS) (Bremmen DE). A full complement of internal in-house and internationally 558 certified standards was run with the samples to calculate isotopic ratios and % C and N values. 559 The isotope ratios were expressed as parts per thousand per mil (‰) and as δ deviation from 560 the internationally recognized standards Vienna Pee Dee Belemnite (VPDB) and AIR. All 561 samples are referred to this scale from herein. 562 563  abundance from the different forest layers to arthropod mass abundance per 10 m 2 plot, which 575 was the area of the sampled understorey plot. This was intended to achieve a better 576 representation of the mass abundance of the available arthropods across the forest layers. 577 Estimates of arthropod mass abundance were obtained for each forest layer, management 578 area and round in the following way: 579 580 For the canopy arthropod samples, we applied a scaling factor according to Kitayama and Itow 581 (Kitayama and Itow, 1999). Aboveground foliage biomass in a montane forest stand on Santa 582 Cruz, Galapagos, was taken 1,482 kg of foliage per hectare, being 1,482 g of foliage in 10 m 2 . 583 The correction factor related the arthropods mass and was scaled to the foliage mass 584 measured of the leaves collected with the arthropods. The same leaf mass dependent scaling 585 factor was used for all three areas since no significant differences were found in canopy cover 586 between the areas in these experiments (Schmidt Yáñez, 2016). 587 For the moss arthropod samples, we applied a correction factor according to the surface area 588 around the trunk, occupied by the moss collected. For this, we used the diameter at breast 589 height (DBH) in meters measured for each sampling point. Knowing the surface (DBH × height) 590 that a given mass of arthropods occupies at that sampling point, we estimated the arthropods 591 mass per 10 m 2 . It was assumed that the percentage of moss cover did not change (between 592 sampling points) and that the mass of moss varied proportionally. We did not apply a scaling 593 factor to the understory data as the whole 10 m 2 plot was sampled (Table S2). 594 595 596

598
We excluded the order Diplopoda (generally, the millipedes) from the total mass data in 2015, 599 as they have never been reported or observed to be consumed by the warbler finch (Filek et  600 al., 2018). This assumption was supported by comparing the isotopic signatures of the 601 Diplopoda with the finch blood data; the Diplopoda were well outside the sphere of 602 Arthropod mass e = Arthropod mass s l ea f mass s *1.482 consumption of the finches ( Figure S3). Therefore, Diplopoda were excluded, since they 603 sometimes dominated the samples in terms of mass (Table in Figure S3) and their presence 604 was preventing us from teasing out the influence of the other orders.   We compared both bi-plot C&N isotope signature data sets similar to Figure S3 (arthropods 629 and bird blood), to evaluate the profile of the nutrients sources used by the birds and to 630 determine whether the diet of the warbler finches reflected the arthropod signatures and which 631 specific arthropod orders were dominant in their diet (data not shown). To determine potential 632 diet components of the warbler finch, we analysed the isotopic signatures of all representative 633 arthropods orders. Subsequently, we analysed management areas and forest layers for the 634 most abundant orders and potential sources of food. We statistically analysed the generated 635 signatures from the representative orders to determine if the orders had significant differences 636 across management areas. 637 to determine the composition of diets based on the dietary isotopic signatures. The warbler 639 finch blood, representing the highest order consumer. The arthropods orders were defined as 640 food sources with their corresponding trophic fractionation factors (Δ)(Hobson and Clark, 641 1992). We calculated diet proportion (f) and from isotopic values using the models. 642 To determine which of the arthropods orders present in the Scalesia forest were consumed by 643 the warbler finch and in what proportion, we used Bayesian mixing models and compared the 644 probabilities of all combinations predicting up to three possible dietary sources, under the 645 creation of Markov Chain Monte Carlo (MCMC) chains. The analysis was conducted using the 646 R-package "MixSIAR . MixSIAR creates and runs Bayesian mixing models 647 to analyze biological tracer data (i.e. stable isotopes, fatty acids), which estimate the 648 proportions of source (prey) contributions to a mixture (consumer). 'MixSIAR' is a framework 649 that allows a user to create a mixing model based on their data structure and research 650 questions, via options for fixed/ random effects, source data types, priors, and error terms 651 . 652 To develop the MixSIAR model, we used the consumers' signatures (warbler finch blood), 653 possible food sources (orders signatures) and a trophic fractionation factor (TFF) for each 654 element (carbon and nitrogen). The TFF used were obtained from the literature based on 655 experimental values from laboratory studies on common quail's blood (Hobson and Clark,656 1992) as there were no equivalent values available for warbler finches. The models were 657 created by establishing informative priors based on the relative abundance of the arthropods 658 by taxonomic order and field observations (Filek et al., 2018). We set the factor managed area 659 as a random effect, as we were analysing whether warbler finches fed on different components 660 in different areas. 661 As diagnostic tools, we used the Gelman-Rubin Diagnostic and the Geweke Diagnostic. The 662 Gelman-Rubin Diagnostic provides a value for each factorial. Less than 10% of those values 663 should be below 1.05. The Geweke Diagnostic provides a standard z-score and 5% per chain 664 and are expected to be outside +/-1.96. 665 We developed the set of food resources from the initial number of arthropod orders that were 666 found in round 1 of 2015. Consequently, based on their relative abundance, we selected nine 667 top orders, representing more than 95% of the total dry mass abundance (excluding 668 Diplopoda). Several attempts were pursued to define priors based on abundance ranking and 669 field observations, also increasing the length of the chain iterations. The best-fit model consited 670 of more than 3000,000 chain iterations for the nine top orders. 671 We also used the MixSIAR package to determine which forest layer was the dominant diet 672 source. We used the warbler finch blood data from 2015 (both rounds) for the consumer and 673 we used the same TFFs. We entered the forest layers as sources. The best-fit model consisted 674 of 1000,000 chain iterations for the three forest layers.