Models generated from laboratory-based thermal performance experiments predict that arthropod crop pest densities will escalate under rising temperatures. Conversely, natural enemies are predicted to decline under warming, exacerbating pest outbreaks. We tested these predictions using 141,562 field-year observations of 43 arthropod populations across spatial and temporal temperature gradients. Pests exhibited remarkable heterogeneity of responses to elevated temperatures, with some populations increasing and others decreasing. Natural enemies also showed variable responses to elevated temperatures, with partial support emerging for the hypothesis that natural enemies are more vulnerable to warming than pests. Laboratory-measured thermal performance and life-history traits failed to explain the variability of responses across taxa. Our findings challenge predictions of universal pest proliferation, highlighting the urgent need for species-specific monitoring approaches in agricultural climate adaptation.

Study system

Our study was conducted in two major agricultural regions at similar latitudes: Andalusia, Spain (approximately 37.5°N, 4.6°W) and California's San Joaquin Valley (approximately 36.6°N, 120.2°W). Both regions are critical agricultural production areas, with each generating billions of dollars in agricultural value annually. Regions share Mediterranean climate patterns with wet, mild winters and dry, hot summers, creating comparable seasonal growing conditions and arthropod emergence patterns.

During our study period (2006 – 2019), mean annual temperatures across all years were 17.9°C in Andalusia and 17.7°C in the San Joaquin Valley, California, with mean seasonal temperature ranges of 10.6 – 26.4°C and 10.3 – 26.9°C, respectively (Fig. S1). The crop systems analyzed varied between regions based on available data. In Andalusia, we analyzed cotton, olive, citrus, grapes, and rice, while the San Joaquin Valley sites included cotton and citrus. Both regions were selected for their exceptional agricultural monitoring infrastructure and long-term pest surveillance programs, providing the high-quality, standardized data essential for our ecoinformatics approach.

Pest monitoring databases

We analyzed three comprehensive ecoinformatics databases containing arthropod abundance, pesticide application, and agricultural management data. The primary source was the Red de Alerta e Información Fitosanitaria (RAIF) database from Andalusia, Spain, a government-operated monitoring network that has tracked over 130 pest species and 40 natural enemies across Mediterranean crops since 2006. We complemented this with the Citrusformatics and Cottonformatics databases from California's San Joaquin Valley. These databases provide comparable temporal coverage to RAIF. Data from California were reported by private pest control advisors (PCAs) and provide high-resolution population assessments following standardized protocols established by the University of California Cooperative Extension. The two California databases collectively monitor 14 pest and natural enemy species across commercial production sites, with data collection protocols designed to guide integrated pest management decisions. Our analyses focused on 43 populations representing 39 distinct species (28 pest species and 11 natural enemy species; see below for details on species selection). Our final dataset encompassed 13,308 field sites across five crops (cotton, olive, citrus, grapes, and rice) over 13 years (2006 – 2019). This comprised 141,562 field-year observations for abundance analysis and 66,663 field-year observations for pesticide analysis.

Data processing, statistical analyses, and software

Data processing was primarily conducted using ftfy (Python) for Unicode correction. All analyses were conducted in R version 4.3.2 (R Core Team, 2023). We used dplyr and tidyverse for data manipulation, lme4 (version 1.1.35.1) for generalized linear mixed models, MuMIn for calculating marginal and conditional R², and ggplot2 for visualization. Model convergence was assessed using lme4 default criteria; models that failed to converge were excluded from analysis. Statistical significance was assessed at α = 0.05 unless otherwise noted.

Processing arthropod abundance data

To prepare the arthropod abundance data for our analyses, we performed a series of preprocessing steps. We first established a set of inclusion criteria to select focal species. Initial taxonomic and sampling criteria for species selection prior to data processing included: a) taxa must be identified to species (with exceptions for natural enemies due to limited data availability), b) abundance measures or percent infestation metrics must be available for each focal species (to ensure methodological comparability), and c) species must have a minimum of 300 raw field-year observations to ensure sufficient statistical power in downstream modeling.

We corrected typographical inconsistencies, including Unicode encoding errors that predominantly occurred in Spanish data due to inconsistent character encodings. These corrections were performed in two phases: we first corrected mismatched encodings; we then identified potential duplicates by training a linear regression model on qgram, cosine, Smith–Waterman, and LCS metrics to assess string similarity.

To control for within-species variation in sampling timing, we excluded temporal outliers by removing observations below the 5th and above the 95th percentile of sampling dates.

To control for uneven sampling across fields, we excluded sites with fewer than three observations.

To obtain robust and reliable measures of abundance per field, we aggregated the time series data by calculating mean arthropod densities across observations within a field-year location. This temporal consolidation reduced our dataset to a single measure of insect abundance per field-year-crop combination.

To facilitate model fitting, we excluded field locations containing only a single year of observation for five species where this occurred (Stethorus spp., Aonidiella aurantii (Spain), Deraeocoris spp., Hippodamia variegata, and Panonychus citri (Spain)), as mixed-effects models require multiple observations per level of random effects.

After these preprocessing steps, we fitted preliminary generalized linear mixed models for all candidate species to test for convergence. We excluded species with non-convergent models and retained only species with 90 or more independent field-year observations for robust parameter estimation in the final models.

Processing pesticide application data

Of the pest species that met our inclusion criteria for the abundance analysis, we assessed which had sufficient pesticide application data to support parallel analyses of temperature effects on pesticide use. We applied a separate minimum threshold of 300 field-year observations of pesticide applications targeting each species to ensure sufficient statistical power for pesticide trend analysis. This filtering identified 16 of the 30 pest populations with adequate pesticide data: Aonidiella aurantii (Spain), Aonidiella aurantii (California), Bactrocera oleae, Bemisia tabaci, Earias insulana, Helicoverpa armigera, Lobesia botrana, Phyllocnistis citrella, Pectinophora gossypiella, Panonychus citri (Spain), Panonychus citri (California), Prays oleae, Tetranychus urticae, Scudderia furcata, Scirtothrips citri, and Lygus hesperus. We then used the number of field-level pesticide sprays applied across the season to create a metric of pesticide application intensity per field-year. Because insecticide applications may remain effective for several days, we counted applications occurring within a four-day window as a single application to avoid overestimating treatment intensity.

Climate data

We extracted temperature data from the ECMWF ERA5 reanalysis dataset at 37 × 37 km spatial resolution for each field site. Variance partition analysis revealed that temperature variation was driven primarily by spatial differences between sites (72.8 ± 1.1%) rather than temporal changes across the 13-year study period (23.6 ± 1.3%), with residual variation accounting for 3.5 ± 1.2% depending on the temperature predictor, supporting our space-for-time substitution approach. Because arthropods respond to temperature differently throughout the year and across life stages, we calculated three temperature predictors to capture different aspects of arthropod thermal experience. These included: (1) annual mean temperature (Julian days 1 – 365) representing overall thermal conditions, (2) summer mean temperature (Julian days 171– 265, approximately June – September) representing growing season conditions, and (3) winter mean temperature (Julian day 365 of previous year through day 79, approximately January – March) representing overwintering conditions. We also tested extreme temperature metrics (summer maxima and winter minima averaged over the 10 hottest and coldest days, respectively), but these produced results similar to seasonal means and are reported in Tables S1, S2, and S4. Given the similar temperature regimes of both regions, these seasonal boundaries were applied consistently across both study regions (Fig. S4).

For most species, we used the standard temperature predictors based on calendar timing. However, for seven species where more than 50% of sampling occurred in early spring (prior to Julian day 171, before annual temperature maxima), we utilized lagged temperature variables from the preceding year to more accurately capture the thermal conditions experienced during critical developmental periods that influence observed population densities. These species included Calepitrimerus vitis, Euphyllura olivina, Euseius spp., Icerya purchasi, Prays oleae, Scirtothrips citri, and Scudderia furcata.

Species descriptions, sampling methods, and species codes

Here we provide detailed species descriptions, sampling methods, and species codes for focal arthropods from three pest monitoring databases:

1)     The California red scale (Aonidiella aurantii (Maskell) [Hemiptera: Diaspididae], California N = 793; Andalusia N = 1717; codes Ao_au (CA) and Ao_au (SP)) is a polyphagous herbivore pest of citrus. For the California population, we analyzed data that was collected during harvest. Samples were taken by observing presence/absence of adults on fruit and quantified as the total number of infested fruits per bin sample; ca. 100 fruit checked per bin. The final reported values were the number of infested fruits per bin sample. For Andalusia, we analyzed data that was collected weekly during the period from fruit set until harvest. During each sampling date, four fruits per tree were sampled. The final reported values were the percentages of observed fruits infested, with infestation defined as three or more individual adult red scales per fruit.

2)     The olive fly (Bactrocera oleae (Rossi) [Diptera: Tephritidae], N = 17411; code Ba_ol) is an oligophagous herbivore pest of olive. We analyzed data that was collected weekly from early June until approximately 15 days before harvest. Three yellow sticky traps baited with sex pheromone were set per field, and the total number of adult flies captured on all traps was counted weekly. The final reported values were the mean number of adults per trap per day, calculated as the total captures from all three traps divided by 3 traps and by 7 days.

3)     The silverleaf whitefly (Bemisia tabaci (Gennadius) [Hemiptera: Aleyrodidae], N = 6600; code Be_ta) is a polyphagous herbivore pest of cotton. We analyzed data that was collected weekly from June to October. Samples were taken by walking diagonally across each field and sampling one leaf from the main stem of the upper third of each plant, quantified by counting the total number of pupae per leaf. The final reported values were the mean number of pupae per leaf per field.

4)     The citrus flat mite (Brevipalpus lewisi (Donnadieu) [Trombidiformes: Tenuipalpidae], N = 92; code Br_le) is a polyphagous herbivore pest of citrus. We analyzed data that was collected weekly from when citrus fruit reached a definitive size until harvest. Samples were taken by observing presence/absence of mobile individuals on 4 fruits per 25 randomly selected trees. The final reported values were the percentages of infested fruits per field.

5)     The grape leaf rust mite (Calepitrimerus vitis (Nalepa) [Trombidiformes: Eriophyidae], N = 295; code Ca_vi) is an oligophagous herbivore pest of grape. We analyzed data that was collected weekly during phenological stages B2 (swollen bud), C (green tip), and D (incipient leaves) in early spring, with one additional observation at stage N (maturation) to determine populations for the following season. Samples were taken by observing presence/absence of mites and other symptoms on 25 randomly selected vines per station. The final reported values were the percentages of infested vines per station.

6)     The predatory common green lacewing in the genus Chrysoperla ((Steinmann) [Neuroptera: Chrysopidae], olive N = 16441; cotton N = 4202; citrus N = 185; code Ch_spp (OL), Ch_spp (CO), and Ch_spp (CI)) is a generalist natural enemy in several crops. We analyzed data that was collected weekly throughout the growing season. Samples were taken by placing sheets (1m x width of furrow) across the width of the furrow, shaking plants onto the sheets, and counting larval individuals. The final reported values were the mean number of larval individuals per hectare.

7)     The predatory seven-spotted lady beetle (Coccinella septempunctata (Linnaeus) [Coleoptera: Coccinellidae], N = 1638; code Co_se) is a generalist natural enemy in cotton. We analyzed data that was collected weekly throughout the growing season. Samples were taken by placing sheets (1m x width of furrow) across the width of the furrow, shaking plants onto the sheets, and counting adult individuals. The final reported values were the mean number of adult individuals per hectare.

8)     The predatory eleven-spotted lady beetle (Coccinella undecempunctata (Linnaeus) [Coleoptera: Coccinellidae], N = 373; code Co_un) is a generalist natural enemy in cotton. We analyzed data that was collected weekly throughout the growing season. Samples were taken by placing sheets (1m x width of furrow) across the width of the furrow, shaking plants onto the sheets, and counting adult individuals. The final reported values were the mean number of adult individuals per hectare.

9)     The citricola scale (Coccus pseudomagnoliarum (Kuwana) [Hemiptera: Coccidae], N = 961; code Co_ps) is a polyphagous herbivore pest of citrus. We analyzed data that was collected from May to August. Samples were taken by observing presence/absence of mobile crawler forms on single leaves from randomly selected trees. The final reported values were the percentages of infested leaves per field.

10) The predatory plant bugs in the genus Deraeocoris ((Knight) [Hemiptera: Miridae], N = 245; code De_spp) are generalist natural enemies in cotton. We analyzed data that was collected weekly throughout the growing season. Samples were taken by placing sheets (1m x width of furrow) across the width of the furrow, shaking plants onto the sheets, and counting individuals of all life stages. The final reported values were the mean number of individuals per hectare.

11) The Egyptian bollworm (Earias insulana (Boisduval) [Lepidoptera: Nolidae], N = 6,543; code Ea_in) is an oligophagous herbivore pest of cotton. We analyzed data that was collected weekly from June to October. Samples were taken by observing areas of ca. 1.25m² and counting the number of small larvae (<1 cm) present. The final reported values were the mean number of larvae per hectare.

12) The carob moth (Ectomyelois ceratoniae (Zeller) [Lepidoptera: Pyralidae], N = 884; code Ec_ce) is a polyphagous herbivore pest of citrus. We analyzed data that was collected weekly from September through October. Samples were taken by observing presence/absence of larval individuals on 4 fruits per 25 randomly selected trees. The final reported values were the percentages of infested fruits per field.

13) The olive psyllid (Euphyllura olivina (Costa) [Hemiptera: Psyllidae], N = 1867; code Eu_ol) is an oligophagous herbivore pest of olive. We analyzed data that was collected weekly from March to May. Samples were taken by walking diagonally across each field and sampling 10 shoots from 20 trees, quantified as the number of cottony masses observed on inflorescences. The final reported values were the mean number of infested inflorescences per shoot.

14) The predatory mites in the genus Euseius ((Congdon) [Acarina: Phytoseiidae], N = 335; code Eu_spp), which predominantly comprise the species Euseius tularensis in the Central Valley, are generalist natural enemies in citrus. We analyzed data that was collected from February to November. Samples were taken by counting the number of mobile mites on five leaves from each branch terminal. Five terminals were sampled per 20-acre block. The final reported values were the mean number of mites per terminal rounded to the nearest integer value.

15) The Texas citrus mite (Eutetranychus banksi (McGregor) [Trombidiformes: Tetranychidae], N = 1084; code Eu_ba) is a polyphagous herbivore pest of citrus. We analyzed data that was collected weekly from April through December. Samples were taken by observing presence/absence of mobile individuals on 4 leaves per 25 randomly selected trees. The final reported values were the percentages of infested leaves per field.

16) The western flower thrips (Frankliniella occidentalis (Pergande) [Thysanoptera: Thripidae], N = 6377; code Fr_oc) is an omnivorous pest of cotton. We analyzed data that was collected weekly from approximately May to September. Samples were taken by walking diagonally across each field and sampling one leaf from the main stem of the upper third of each plant, quantified by counting the total number of larvae per leaf. The final reported values were the mean number of larvae per leaf.

17) The predatory big-eyed bugs in the genus Geocoris ((Fallén) [Hemiptera: Geocoridae], N = 648; code Ge_spp) are generalist natural enemies in cotton. We analyzed data that was collected weekly throughout the growing season. Samples were taken by placing sheets (1m x width of furrow) across the width of the furrow, shaking plants onto the sheets, and counting individuals of all life stages. The final reported values were the mean number of individuals per hectare.

18) The cotton bollworm (Helicoverpa armigera (Hübner) [Lepidoptera: Noctuidae] N = 6819; code He_ar) is a polyphagous herbivore pest of cotton. We analyzed data that was collected weekly from first bud to harvest. Samples were taken by observing areas of ca. 1.25m² and counting the number of small larvae (<1 cm) present. The final reported values were the mean number of larvae per hectare.

19) The variegated lady beetle (Hippodamia variegata (Goeze) [Coleoptera: Coccinellidae], N = 160; code Hi_va) is a generalist natural enemy in cotton. We analyzed data that was collected weekly throughout the growing season. Samples were taken by placing sheets (1m x width of furrow) across the width of the furrow, shaking plants onto the sheets, and counting adult individuals. The final reported values were the mean number of adult individuals per hectare.

20) The cottony cushion scale (Icerya purchasi (Maskell) [Hemiptera: Monophlebidae], N = 750; code Ic_pu) is a polyphagous herbivore pest of citrus. We analyzed data that was collected from March to April. Samples were taken by observing presence/absence of adult females on 60-cm portions of tree trunks and inner branches. The final reported values were the percentages of infested sampling units per field.

21) The European grapevine moth (Lobesia botrana (Denis & Schiffermüller) [Lepidoptera: Tortricidae], N = 1085; code Lo_bo) is a polyphagous herbivore pest of grape. We analyzed data that was collected weekly throughout the growing season. Two delta traps were set per field, and the total number of individuals on in each trap was counted. The final reported values were the mean number of adult moths per field.

22) The western tarnished plant bug (Lygus hesperus (Knight) [Hemiptera: Miridae], N = 1467; code Ly_he) is a polyphagous herbivore pest of cotton. We analyzed data that was collected weekly, primarily throughout June and July. Samples were taken by swinging a sweep net approximately 50 times across the top of the plant canopy to create a single sample; ca. 6-12 of these sweep net samples were completed per sampling date. The final reported values were the mean number of motile Lygus per sweep net sample.

23) The citrus peelminer (Marmara gulosa (Guillèn and Davis) [Lepidoptera: Gracillariidae], N = 774; code Ma_gu) is a polyphagous herbivore pest of citrus. We analyzed data that was collected at harvest. Samples were taken by observing presence/absence of adults on fruit and quantified as the total number of infested fruits per bin sample; ca. 100 fruit checked per bin. The final reported values were the number of infested fruits per bin sample.

24) The maize caterpillar (Mythimna loreyi (Duponchel) [Lepidoptera: Noctuidae], N = 937; code My_lo) is a polyphagous herbivore pest of rice. We analyzed data that was collected weekly from July to August. Traps baited with sex pheromone were set in each field, and the total number of adult individuals in each trap was counted. The final reported values were the mean number of adults per trap per day, calculated as the total captures from traps divided by the number of traps and by 7 days.

25) The predatory damsel bugs in the genus Nabis ((Latreille) [Hemiptera: Nabidae], N = 2333; code Na_spp) are generalist natural enemies in cotton. We analyzed data that was collected weekly throughout the growing season. Samples were taken by placing sheets (1m x width of furrow) across the width of the furrow, shaking plants onto the sheets, and counting individuals of all life stages. The final reported values were the mean number of individuals per hectare.

26) The predatory minute pirate bugs in the genus Orius ((Wolff) [Hemiptera: Anthocoridae], N = 3305; code Or_spp) are generalist natural enemies in cotton. We analyzed data that was collected weekly throughout the growing season. Samples were taken by placing sheets (1m x width of furrow) across the width of the furrow, shaking plants onto the sheets, and counting individuals of all life stages. The final reported values were the mean number of individuals per hectare.

27) The citrus red mite (Panonychus citri (McGregor) [Acari: Tetranychidae], California N = 1350; Andalusia N = 1853; code Pa_ci (CA) and Pa_ci (SP)) is a polyphagous herbivore pest of citrus. For the California population, we analyzed data that was collected from March to June. Samples were taken by observing presence/absence sampling on single tree leaves from randomly selected trees. The final reported values were the percentages of infested leaves per field. For Andalusia, we analyzed data that was collected weekly from March through November. Samples were taken by observing presence/absence of mobile individuals on 4 leaves per 25 randomly selected trees. The final reported values were the percentages of infested leaves per field.

28) The bayberry whitefly (Parabemisia myricae (Kuwana) [Hemiptera: Aleyrodidae], N = 771; code Pa_my) is a polyphagous herbivore pest of citrus. We analyzed data that was collected weekly from May to December. Samples were taken by observing presence/absence of individuals of any life stage on 4 shoots per 25 randomly selected trees. The final reported values were the percentages of infested shoots per field.

29) The pink bollworm (Pectinophora gossypiella (Saunders) [Lepidoptera: Gelechiidae], N = 5663; code Pe_go) is an oligophagous herbivore pest of cotton. We analyzed data that was collected weekly from July to October. Two funnel traps baited with sex pheromone were set per field, and the total number of adult individuals in each trap was counted. The final reported values were the percentages of infested shoots per field.

30) The citrus leafminer (Phyllocnistis citrella ((Stainton) [Lepidoptera: Gracillariidae], N = 2171; code Ph_ci) is an oligophagous herbivore pest of citrus. We analyzed data that was collected weekly from May through October. Samples were taken by observing presence/absence of larval individuals on 4 shoots per 25 randomly selected trees. The final reported values were the percentages of infested shoots per field.

31) The citrus mealybug (Planococcus citri (Risso) [Hemiptera: Pseudococcidae], N = 1767; code Pl_ci) is a polyphagous herbivore pest of citrus. We analyzed data that was collected weekly from March through September. Samples were taken by observing presence/absence of individuals of any life stage on 4 fruits per 25 randomly selected trees. The final reported values were the percentages of infested fruits per field.

32) The olive moth (Prays oleae (Bernard) [Lepidoptera: Praydidae], N = 11779; code Pr_ol) is a specialist herbivore pest of olive. We analyzed data that was collected weekly from March through early summer and again from late August through November. Two funnel traps baited with sex pheromone were set per field, and the total number of adult males captured in both traps was counted weekly. The final reported values were the mean number of adults per trap per day, calculated as the total captures from both traps divided by 2 traps and by 7 days.

33) The citrus thrips (Scirtothrips citri (Moulton) [Thysanoptera: Thripidae], N = 2205; code Sc_ci) is a polyphagous herbivore pest of citrus. We analyzed data that was collected primarily through the first two months after petal fall (ca. April – July). Samples were taken by observing presence/absence of juvenile individuals on 25 – 150 small developing fruitlets per field. The final reported values were the percentages of infested fruitlets per field.

34) The fork-tailed bush katydid (Scudderia furcata (Brunner von Wattenwyl) [Orthoptera: Tettigoniidae], N = 792; code Sc_fu) is an omnivorous pest of citrus. We analyzed data that was collected over 60 days after petal fall (ca. April/May). Samples were taken by observing presence/absence of individuals in small, defined areas of foliage. The final reported values were the percentages of infested foliage areas per field.

35) The predatory brown lady beetles in the genus Scymnus ((Kugelann) [Coleoptera: Coccinellidae], N = 279; code Sc_spp) are generalist natural enemies in cotton. We analyzed data that was collected weekly throughout the growing season. Samples were taken by placing sheets (1m x width of furrow) across the width of the furrow, shaking plants onto the sheets, and counting adult individuals. The final reported values were the mean number of adult individuals per hectare.

36) The beet armyworm (Spodoptera exigua (Hübner) [Lepidoptera: Noctuidae], N = 6139; code Sp_ex) is a polyphagous herbivore pest of cotton. We analyzed data that was collected weekly from first bud to defoliation. Samples were taken by observing areas of ca. 1.25m² and counting the number of small larvae (<1 cm) present. The final reported values were the mean number of larvae per hectare.

37) The Egyptian cotton leafworm (Spodoptera littoralis (Boisduval) [Lepidoptera: Noctuidae], N = 5472; code Sp_li) is a polyphagous herbivore pest of cotton. We analyzed data that was collected from first bud to defoliation. Samples were taken by observing areas of ca. 1.25m² and counting the number of small larvae (<1 cm) present. The final reported values were the mean number of larvae per hectare.

38) The predatory spider mite destroyers in the genus Stethorus (Weise) [Coleoptera: Coccinellidae], N = 140; code St_spp) are specialist natural enemies in cotton. We analyzed data that was collected weekly throughout the growing season. Samples were taken by placing sheets (1m x width of furrow) across the width of the furrow, shaking plants onto the sheets, and counting adult individuals. The final reported values were the mean number of adult individuals per hectare.

39) The red spider mite (Tetranychus urticae (Koch) [Trombidiformes: Tetranychidae], N = 6841; code Te_ur) is a polyphagous herbivore pest of cotton. We analyzed data that was collected weekly from approximately May to September. Samples were taken by walking diagonally across each field and sampling one leaf from the main stem of the upper third of each plant. Plants were scored as infested if adult female mites were present on the sampled leaf. The final reported values were the percent plants infested per field.