Climate-driven declines in arthropod abundance restructure a rainforest food web

Abstract

A number of studies indicate that tropical arthropods should be particularly vulnerable to climate warming. If these predictions are realized, climate warming may have a more profound impact on the functioning and diversity of tropical forests than currently anticipated. Although arthropods comprise over two-thirds of terrestrial species, information on their abundance and extinction rates in tropical habitats is severely limited. Here we analyze data on arthropod and insectivore abundances taken between 1976 and 2012 at two midelevation habitats in Puerto Rico's Luquillo rainforest. During this time, mean maximum temperatures have risen by 2.0 °C. Using the same study area and methods employed by Lister in the 1970s, we discovered that the dry weight biomass of arthropods captured in sweep samples had declined 4 to 8 times, and 30 to 60 times in sticky traps. Analysis of long-term data on canopy arthropods and walking sticks taken as part of the Luquillo Long-Term Ecological Research program revealed sustained declines in abundance over two decades, as well as negative regressions of abundance on mean maximum temperatures. We also document parallel decreases in Luquillo's insectivorous lizards, frogs, and birds. While El Niño/Southern Oscillation influences the abundance of forest arthropods, climate warming is the major driver of reductions in arthropod abundance, indirectly precipitating a bottom-up trophic cascade and consequent collapse of the forest food web.

Keywords: arthropods; bottom-up cascade; climate warming; food web; rainforest.

Fig. 1.

Air temperature trends within the Luquillo forest. MnMaxT vs. year for the (A) El Verde (1978–2015) and (B) Bisley Tower (1994–2012) meteorological stations. The stations are ∼8.2 km apart at elevations of 350 and 352 m, respectively. The least-squares regression lines have been drawn through the data points. The difference between the slopes of the two regression lines is not significant (t = 0.281, P = 0.78).

Fig. 2.

Mean dry-weight arthropod biomass per 100 sweeps taken in the same sample area in the Luquillo rainforest during July 1976, January 1977, July 2011, and January 2013. One SE around the mean biomass is shown for each bar. Total sweeps taken in each period was 800, except for July 1976, when 700 sweeps were taken. Data for 1976 and 1977 are from Lister (22).

Fig. 3.

Comparison of total dry-weight biomass for the major arthropod taxa captured in sweep samples taken during the summer (A, C, and E) and winter (B, D, and F) seasons 1976–1977 and 2011–2013, within the same Luquillo forest study area. Arn, Areneida; Col, Coleoptera; Dip, Diptera; For, Formicidae; Hem, Hemiptera; Hom, Homoptera; Hym, other Hymenoptera; La, Lepidoptera adults; LI, Lepidoptera larvae; Ort, Orthoptera.

Fig. 4.

Comparison of the average dry-weight biomass of arthropods caught per 12-h day in 10 ground (A) and canopy (B) traps within the same sampling area in the Luquillo rainforest. Numbers above the bars give the mean daily catch rate in dry weight of arthropods per day for the respective dates. Data for 1976 and 1977 are from Lister (22).

Fig. 5.

Trends in the abundance of canopy arthropods and walking sticks in the Luquillo forest El Verde study area. (A) Linear regression of the total number of canopy arthropods captured per foliage weight sampled at El Verde against the period when the samples were taken. (B) Cubic regression for the total number of canopy arthropods captured per foliage weight sampled against the MnMaxT during the period when the samples were taken. (C) Quasi-Poisson regression of total number of walking sticks vs. the period when the population was sampled. (D) Quasi-Poisson regression of total number of walking sticks vs. the MnMaxT during the period when the population was sampled. The 95% confidence intervals are shown around the best-fit regression lines. For the Poisson regressions, Pr(χ) is the result of a likelihood-ratio χ² test of whether the independent variable improves the Poisson model beyond an intercept-only model. P < 0.05 indicates a statistically significant regression.

Fig. 6.

Comparison of anoles censuses conducted in the same Luquillo rainforest study area during July 1976 and January 1977, with censuses conducted during July 2011 and January 2012. The total number of anoles in each census, estimated by the Schnabel multiple capture–recapture method (27), is given above each histobar. Brown represents male and female A. gundlachi. Green represents A. evermanni. Gray represents A. stratulus. SI Appendix, Table S1 gives the 95% confidence intervals around estimated abundances for A. gundlachi and A. evermanni.

Fig. 7.

Population trends for E. coqui and birds near the El Verde Field station. (A) Quasi-Poisson regression of estimated total number of E. coqui individuals against time from censuses conducted by Stewart (28) in the Activity Transect. (B) Quasi-Poisson regression of the estimated number of E. coqui individuals against MnMaxT during the time periods when Stewart’s censuses were conducted. (C) Quasi-Poisson regression for the total number of birds captured during equal length, 4-d sessions of mist netting (31) near the El Verde Field Station against the period when the mist netting was conducted. (D) Quasi-Poisson regression of the total number of birds captured during Waide’s (31) 4-d sessions vs. MnMaxT during the year of mist netting. The 95% confidence intervals are shown around the best-fit regression lines. Pr(χ) is the result of a likelihood-ratio χ² test of whether the independent variable improves the model beyond an intercept-only model. P < 0.05 indicates a statistically significant regression.