Cold waves crossing the Amazon rainforest are a rare phenomenon predicted to increase in intensity under climate change. We here describe an extensive cold wave occurring in June 2023 in Amazonian-Andean forests, compared environmental temperatures to experimentally tested thermal tolerances and its impact on lowland animal communities (insects and wild mammals). While we found strong reductions in abundance of all animal groups under the cold wave, tropical lowland animals showed thermal tolerance limits below the lowest environmental temperatures measured during the cold wave, and abundances of most studied taxa recovered over the next season; nevertheless, small thermal safety margins suggest that an increased intensity of cold waves in the future could imperil animal communities in the Amazon.

Temperature data

Air temperature was measured at each plot along the elevation gradient with iButton sensors (Analog Devices, Inc, Wilmington, USA) at 1.5 m height every four hours, hanging from a horizontal branch. The sensors were shielded with white plastic dishes (diameter ca. 18 cm) to protect them from rain and direct sunlight [13]. In addition, each plot was equipped with a TOMST4-temperature and soil humidity logger (TOMST s.r.o., Prague, Czech Republic), continuously measuring temperature and soil humidity at 6 cm depth, as well as temperature at the surface and in 15 cm height.

Insect data

(1) Malaise: At each plot in each field season, one malaise trap was operated for seven days. Malaise traps were based on the Townes Malaise trap model, albeit with a black roof and a slightly smaller size (dimensions of the capture area: height front: 0.90 m; height rear: 0.60 m; length: 1.60 m); Ethanol (96 %) was used as the capture fluid to ensure the preservation of specimens. For each malaise trap sample, the wet, in ethanol preserved,insect biomass was determined by weighingthe insects usinga metal sieve. When the time between two drops of ethanol reached 10 s, the wet weight of the sample (biomass) was measured using a precision scale.

(2) Baited pitfall traps: At each plot in each field season, three differently baited pitfall traps (human dung, chicken carrion, fermented banana) were set up in a triangle with ~50 m distance and left open for two days. Traps consisted of two stacked cups (ca. 500 ml) buried with the upper rim being even with the surface, filled halfway with water and a drop of unscented detergent. Baits wrapped in gauze were hanging in a 45-degree angle over the cup. The traps were covered with a triangular plastic tarp to protect them from rain. All collected insects were counted, and the wet biomass was determined by placing all organisms of a sample individually with forceps on a microscale.

(3) Measurements of thermal tolerance limits: We determined critical thermal minima (CT_min) and maxima (CT_max) of insects using an Eppendorf Thermostat C [16]. In total, 414 insects (Coleoptera (n = 136), Diptera (n = 80), Hymenoptera (n = 93), Lepidoptera (n = 37), Hemiptera (n = 41) and Orthoptera (n = 27)), which were collected by hand-netting on the study plots, were placed inside plastic tubes in the Eppendorf Thermostat C and exposed to gradually increasing or decreasing heat (acclimation temperature 28 °C for 10 min, ramping rate 0.5 °C per minute). We tested for adaptation to colder temperatures by exposing a subset of insects (n = 64) to an acclimation temperature of 14 °C for 10 min before following the normal protocol. CT_max and CT_min were defined as the temperature at which insects lost their locomotive ability, at which the insect was removed from the Thermostat. For both CT_max and CT_min temperatures, we also noted if insects were dead or still alive after a waiting time of 60 minutes (after the end of the experiment) at room temperature.

Mammal data

Mammal data werecollected by camera trap monitoring within each of the three field seasons. At each plot, four motion- and infrared-operated camera traps (Bushnell Trophy Cam HD Essential) were attached to a tree at ~1 m height, if available near trails, for seven days, summing up to 6048 camera trap hours for the three plots. Cameras were programmed to record 10 second videos with an inactive period of one minute. For analyses of mammal abundance, we considered two videos of the same species as independent with one hour time difference. All species were identified using the field guide from Emmons and Feer [19] and taxonomy was updated according to the ASM database (ASM Mammal Diversity Database). Abundance was calculated by summing up all observed individuals of each species (for each plot in each season), and species richness by counting the number of recorded species per plot and season.