Materials and Methods Fieldwork and sample processing Field collection and sample processing has been described previously by Cuff, Tercel, et al., (2022), but is briefly described in Supplementary Information 1. In short, money spiders (Araneae: Linyphiidae) and wolf spiders (Araneae: Lycosidae) were collected from occupied webs and the ground in barley fields between April and September 2018. Linyphiids occupying webs (n = 78) were prioritised for collection, but ground-active linyphiid and lycosid spiders were also collected. For each linyphiid taken from a web, the height of the web from the ground (mm) and its approximate dimensions were recorded, the latter calculated as approximate web area (mm2). To obtain data on local prey density, ground and crop stems were suction sampled using a ‘G-vac’ for approximately 30 seconds at each 4 m2 quadrat from which spiders were collected. Extraction, amplification and sequencing of DNA, and bioinformatic analysis is described by Cuff, Tercel, et al. (2022) and Drake et al. (2022), and is also detailed in Supplementary Information 2. Amplification was carried out using two complementary PCR primer pairs: one targeting invertebrates generally, and one intended to exclude amplification of spider DNA to reduce the prevalence of ‘host’ reads in the data output (Cuff et al. 2023). Amplicons were sequenced via Illumina MiSeq V3 with 2x300 bp paired-end reads. The resultant sequencing read counts were converted to presence-absence data of each detected prey taxon in each individual spider. Given the prevalence of sequencing reads associated with each spider analysed and the impossibility of disentangling these from detections of intraspecific predation (i.e., cannibalism), all such reads were removed (Cuff et al. 2023), although intrageneric and intrafamilial predation were still detected. Weather data Weather data were taken from publicly available reports from the Cardiff Airport weather station (6.6 km from the study site) via “Wunderground” (Wunderground, 2020), to represent local weather conditions. This does not necessarily reflect smaller-scale effects (e.g., microclimate-scale; Bell, 2014; Holtzer et al., 1988), but the timescale of detection for dietary metabarcoding reduces the value of that resolution given that spiders may forage across multiple microclimates. We collated data from 1st January 2018 to 17th September 2018 (the last field collection). Weather data were also separately extracted for the week preceding each of the two 2017 collection dates (3rd to 9th August and 29th August to 4th September 2017). Specifically, daily average temperatures (°C), daily average dew point (°C), maximum daily wind speed (km h-1), daily sea level pressure (hPa) and day length (min; sunrise to sunset) were recorded. Precipitation data were downloaded via the UK Met Office Hadley Centre Observation Data (UK Met Office, 2020) as regional precipitation (mm) for South West England & Wales. Weather data were converted to mean values for seven days preceding the collection of spider samples to correspond with the longevity of DNA in the guts of spiders (Greenstone et al., 2014). Statistical Analysis All analyses were conducted in R v4.0.3 (R Core Team, 2020). To assess how weather affects spider trophic interactions over time, we analysed dietary changes across weather gradients using multivariate models. To identify whether this was likely to be driven by changes in prey abundance, we assessed the corresponding changes in the prey communities and then used null models to ascertain whether spiders were responding to prey abundance changes through prey choice. Given the dependence of linyphiid spiders on webs for foraging, we also compared web height and area over weather gradients to assess whether this may be a component of adaptive foraging. To assess the inter-annual consistency of prey choices in response to weather conditions, we also assessed whether prey preference data could be used to improve the predictive power of prey choice models. For this, we generated null models for 2017 data with prey abundance weighted by prey preferences estimated with the 2018 data. This allowed us to assess the consistency of prey choice under similar conditions, but also provides insight as to whether this framework can be used to predict predator responses to diverse prey communities under dynamic conditions. We detail the specific stages of this analytical framework in the below sections. Sampling completeness and diversity assessment To assess the diversity represented by the dietary analysis and the invertebrate community sampling, and the completeness of those datasets, coverage-based rarefaction and extrapolation were carried out, and Hill diversity calculated (Chao et al., 2014; Roswell, Dushoff, & Winfree, 2021). This was performed using the ‘iNEXT’ package with species represented by frequency-of-occurrence across samples (Chao et al., 2014; Hsieh et al., 2016; Figures S4 & S6). Relationships between weather, spider trophic interactions and prey community composition Prey species that occurred in only one spider individual were removed before further analyses to prevent outliers skewing the results. Spider trophic interactions were related to temporal and weather variables in multivariate generalized linear models (MGLMs) with a binomial error family (Wang, Naumann, Wright, & Warton, 2012). Trophic interactions were related to temporal variables and their pairwise interactions (including spider genus to account for any confounding effect), weather variables and their pairwise interactions, and weather variables and their interactions with spider genus and time (to account for any confounding effects) in three separate MGLMs. These variables were separated into different models (Temporal model, Weather interaction model and Confounding effects model) to improve model fit and reduce singularity. Invertebrate communities from suction sampling were related to temporal and weather variables in identically structured MGLMs (excluding the spider genus variable) with a Poisson error family. All MGLMs were fitted using the ‘manyglm’ function in the ‘mvabund’ package (Wang et al., 2012). ‘Temporal model’ independent variables were calendar day (day), mean day length in minutes for the preceding week (day length), spider genus (for dietary models only, to ascertain any effect of spider taxonomic differences on dietary differences over time and day lengths) and all two-way interactions between these variables. ‘Weather interaction model’ independent variables were mean temperature, precipitation, dewpoint, wind speed and pressure for the preceding week, and pairwise interactions between weather variables. ‘Confounding effects model’ independent variables were day (to investigate the interaction between time and weather), spider genus (for dietary models only, to ascertain any effect of spider taxonomic differences on dietary differences over time and day lengths), mean temperature, precipitation, dewpoint, wind speed and pressure for the preceding week, and two-way interactions of each weather variable with day and genus. Trophic interaction and community differences were visualised by non-metric multidimensional scaling (NMDS) using the ‘metaMDS’ function in the ‘vegan’ package (Oksanen et al., 2016) in two dimensions and 999 simulations, with Jaccard distance for spider diets and Bray-Curtis distance for invertebrate communities. For the dietary NMDS, outliers (n = 21; samples containing rare taxa) obscured variation on one axis and were thus removed to facilitate separation of samples and achieve minimum stress. For visualization of the effect of continuous variables against the NMDS, surf plots were created with scaled coloured contours using the ‘ordisurf’ function in the ‘ggplot’ package (Wickham, 2016). Relationships between web characteristics and weather variables Web area and height were compared against weather and temporal variables using a multivariate linear model (MLM) with the ‘manylm’ command in ‘mvabund’ (Wang et al., 2012). Log-transformed web area and height comprised the multivariate dependent variable, and day, spider genus, temperature, precipitation, dewpoint, wind, pressure and two-way interactions between each of these and day and genus comprised the independent variables. Variation in spider prey choice across weather conditions To separately represent spiders from different weather conditions in prey choice analyses, sample dates for every spider were clustered based on the mean weather conditions (temperature, precipitation, dewpoint, wind and pressure) of the week before collection (7 days, to align approximately with spider gut DNA half-life; Greenstone et al., 2014). Alongside data from 2018 (n = 24 collection dates), two sampling periods from 2017 were included in the clustering to ascertain similarity of weather conditions for additional inter-annual prey choice analyses described below. The clustering process is described in Supplementary Information 3. Five clusters were generated: High Pressure (HPR), Hot (HOT), Wet Low Dewpoint (WLD), Dry Windy (DWI), Wet Moderate Dewpoint (WMD), and 2017 (2017 sampling periods). Prey preferences of spiders in each of the weather clusters was analysed using network-based null models in the ‘econullnetr’ package (Vaughan et al., 2018) with the ‘generate_null_net’ command. Consumer nodes in this case represented spiders belonging to each of the weather clusters. Econullnetr generates null models based on prey abundance, represented here by suction sample data, to predict how consumers will forage if based on the abundance of resources alone. These null models are then compared against the observed interactions of consumers (i.e., interactions of spiders within each weather cluster with their prey) to ascertain the extent to which resource choice deviated from random (i.e., density dependence). The trophic network was visualised with the associated prey choice effect sizes using ‘igraph’ (Csardi & Nepusz, 2006) with a circular layout, and as a bipartite network using ‘ggnetwork’ (Briatte, 2021; Wickham, 2016). The normalised degree of each weather cluster node was generated using the ‘bipartite’ package (Dormann, Gruber, & Fruend, 2008) and compared against the normalised degree of the same node in the null network to determine whether spiders were more or less generalist than expected by random. Prior to the prey choice analysis, an hemipteran prey identified no further than order level through dietary analysis was removed due to the inability to pair it to any present prey taxa with certainty. Validating and predicting relationships between years To test how generalisable the results are and the extent to which weather drives prey preferences, we used a measure of prey preference (observed/expected values; observed interaction frequencies divided by interaction frequencies expected by null models) from the above prey choice analysis to assess whether we could more accurately predict observed trophic interactions under similar weather conditions for data from a linked study at the same location in 2017. These additional data represent a subset of the spider taxa analysed above (Tenuiphantes tenuis and Erigone spp.) collected using the same methods by the same researchers and in the same locality (Cuff, Drake, et al., 2021). The similarity in weather conditions between the 2017 study period and each of the five 2018 weather clusters was determined via NMDS of the weather data in two dimensions with Euclidean distance. Centroid coordinates for each 2018 weather cluster and the 2017 data were extracted and pairwise distances calculated between weather clusters: In order, the most proximate weather clusters to the 2017 weather data were HPR (mean Euclidean distance = 8.845), HOT (9.290), WMD (13.626), DWI (13.817) and WLD (18.682; Figure S3). To facilitate comparison between the two years, observed/expected values from the 2018 prey choice models were extracted separately for each of the weather clusters and scaled between 0.1 and 1. For this, 0.1 was used as a minimum since 0 would result in interactions being excluded altogether in the null models, and one as a maximum given the limits of econullnetr but also because this is a multiplier applied to the prey abundances, so greater values would skew prey abundances beyond realistic proportions. Scaling was achieved by the following equation: Missing values (e.g., prey that were absent in certain weather conditions) were represented as 1 to prevent transformation of their abundances in the null models; this treats prey for which data were absent naively, but could increase perceived preferences for them. The scaled values were used to weight the abundance of prey available to the spiders in the 2017 data using the weighting option in econullnetr, whereby values less than 1 proportionally reduce the probability of that taxon being predated in the null models. This effectively redistributes the 2017 relative prey abundance data according to the preference effect sizes generated for each of the 2018 weather clusters. If prey preferences are similar between the 2017 spiders and those from the weather cluster being used to weight the model, the composition of simulated diets should more closely resemble observed diets and fewer significant deviations from the null model should be found. Null models were generated as above (Variation in spider prey choice across weather conditions) but based on the prey availability and trophic interactions from 2017 samples. Three types of model were run: i) a conventional model based on observed prey abundances; ii) a model with prey abundances set to be equal across all prey taxa; and iii) observed prey abundances weighted by prey preferences determined for each of the weather clusters in the 2018 prey choice analysis. A separate model was run for each 2018 weather cluster with abundances weighted by the corresponding scaled observed/expected values. The unweighted conventional model was compared against weighted models to ascertain whether the prey preference weightings from 2018 improved the predictive power of the null models. To compare effect sizes between the unweighted and each other null model for each resource taxon, mean standardised effect size (SES) values were calculated from the paired ‘pre-harvest’ and ‘post-harvest’ data from each model, and paired t-tests were carried out with these between the unweighted and each weighted model. The SES values were plotted for each model and joined between taxa to visualise these paired differences using ‘ggplot’ (Wickham, 2016). Null model-predicted trophic interactions were generated via a modified ‘econullnetr’ function (generate_null_net_indiv) which produces outputs at the individual level to generate simulated diets for individual spiders to compare dietary composition between null model predictions and observed data. These models were run with 2300 simulations to represent 50 simulations per individual spider in the 2017 dataset (n = 46). Null diets were associated with sample IDs by aggregating the 50 simulations per sample and retaining a mean incidence of prey (i.e., mean occurrence across all 50 simulations). A visualisation of the per-sample differences in null model and observed data was generated via NMDS. Mean centroid coordinates for the observed 2017 data and the predicted diets of each model were extracted and the Euclidean distance between the observed data centroid and that of each model was calculated (as above for weather conditions). Supplementary Information 1: Field collection and sample processing Money spiders (Araneae: Linyphiidae) and wolf spiders (Araneae: Lycosidae) were visually located along transects in two adjacent barley fields at Burdons Farm, Wenvoe in South Wales (51°26'24.8"N, 3°16'17.9"W) and collected from occupied webs and the ground, between April and September 2018 (five visits per week of which spiders from 24 collection dates were used). Transects were randomly distributed across the entire field. Along these transects, 64 separate 4 m2 quadrats, at least 10 m apart, were searched and all observed linyphiids and lycosids were collected. Spiders were placed in 100 % ethanol using an aspirator, regularly changing meshing to limit potential cross-contamination. Linyphiids occupying webs were prioritised for collection, but ground-active linyphiid spiders were also collected. For each spider taken from a web, the height of the web from the ground (mm) and its approximate dimensions were recorded, the latter calculated as approximate web area (mm2). Spiders were taken to Cardiff University, transferred to fresh ethanol, adults identified to species-level and juveniles to genus, and stored at -80 °C in 100 % ethanol until DNA extraction. To obtain data on local prey density, ground and crop stems were suction sampled using a ‘G-vac’ for approximately 30 seconds at each 4 m2 quadrat (n = 64) from which spiders were collected. The collected material was emptied into a bag, any organisms immediately killed with ethyl-acetate and material frozen for storage before sorting into 70 % ethanol in the lab. All invertebrates were identified to family level to match the resolution of the least resolved of the metabarcoding-derived trophic interaction data, and due to difficulties associated with identification to finer taxonomic resolution for many taxa. Exceptions included springtails of the superfamily Sminthuroidea (Sminthuridae and Bourletiellidae were often indistinguishable following suction sampling and preservation due to the fine features necessary to distinguish them) which were left at super-family, mites (many of which were immature or in poor condition) which were identified to order level, and wasps of the superfamily Ichneumonoidea which were identified no further due to obscurity of wing venation due to damage. Supplementary Information 2: Molecular analysis and bioinformatics Extraction and high-throughput sequencing of spider gut DNA Given their prevalence in field collections, dietary analysis was carried out for the linyphiid genera Erigone, Tenuiphantes, Bathyphantes and Microlinyphia (Araneae: Linyphiidae), and the Lycosidae genus Pardosa. Spiders were transferred to and washed in fresh 100 % ethanol to reduce external contaminants prior to identification via morphological key (Roberts, 1993). Abdomens were removed from spiders and again transferred to and washed in fresh 100 % ethanol. DNA was extracted from the abdomens via Qiagen TissueLyser II and DNeasy Blood & Tissue Kit (Qiagen) as per the manufacturer protocol, but with an extended lysis time of 12 hours to account for the complex and branched gut system in spider abdomens (Krehe