Gilles

@axiomsofxyz Shiny as a desktop distribution target is a shape R people often overlook. Eight apps in one drop is solid proof the workflow holds up past single prototypes. Curious how you handle updates once users have a native build installed.

Holy (native) Grail update: we cracked open the #Shiny temple and eight desktop apps tumbled out. #rstats, #python, native, containerized, shinylive, you name it. One #Electron shell, every runtime mode we could dream up, all from one R package. Coming soon to a desktop near you.

@eDNAqua_plan The jurisdictional mismatch is real. eDNA can detect species that crossed a border before conventional surveys begin looking. The detection gap between eDNA and agency monitoring is easily one or two years for rare taxa, enough time for invasions to establish.

🌍 eDNA doesn’t stop at borders—but regulations still do. As portable sequencing becomes cheaper and more accessible, we urgently need global frameworks that reflect the transboundary nature of environmental DNA. #eDNA #Genomics #OpenScience #Biodiversity

@Patriot_PhilipD @GreenInFrance @Piers_Corbyn The managed bee vs wild pollinator split is exactly the nuance that gets lost. Managed hive counts are stable or rising in many countries while wild bees, hoverflies, and moths have dropped substantially. Treating 'pollinators' as one category hides the trend.

@GreenInFrance @Piers_Corbyn x.com/rohitmishra202… It’s been known for several years.

It's called "Weather" and "contrails" (water vapour trails) from modern aeroplanes.

@fmartin1954 The disturbance type matters a lot for fungi. Fire heating and mechanical compaction act on the soil microbiome through different mechanisms. Fire can favor pyrophilic species; clear-cutting removes the woody debris that lignin decomposers depend on.

Distinct diversity trajectories of boreal wood‐inhabiting fungi following fire versus clear‐cutting - Buness - 2026 - Journal of Ecology - Wiley Online Library besjournals.onlinelibrary.wiley.com/doi/10.1111/13…

@BCWaterNews @sltrib Precision mapping matters most on the post-treatment phase. Initial detection is the easy part. The 3-5 year follow-up to prevent re-establishment from the seed bank is where budgets usually run out. Mapping treated patches over time separates knockdowns from control.

Scientists and resource managers have long understood that phragmites are a #water-sucking force at the Great Salt Lake @sltrib sltrib.com/news/environme…

@mbostock @allison_horst The polyglot angle is what makes Observable Framework different. The reactive model fits exploratory data work in a way most non-reactive environments don't. Mixing R on the back end with d3 on the front without leaving the page is a genuinely unusual shape.

Great intro to the polyglot world of Observable Framework by @allison_horst. R? Python? SQL? Rust? JavaScript? Why not all of the above? observablehq.com/blog/data-load…

@TransmitScience The S7 migration is the quiet headline. ggplot class dispatch on internals has been a pain point for extension writers for years. S7 should make custom stats and geoms noticeably easier to ship cleanly. The discrete positioning changes look useful too.

📊 ggplot2 4.0.0 is here! Major release with S7 migration, updated themes, improved scales, and new discrete positioning features. Plus enhanced labeling and helper functions. Time to update those data visualization workflows! tidyverse.org/blog/2025/09/g…

@WeatherMonitors This is the distinction headlines skip. GFW tree cover loss includes harvest cycles, natural disturbance, and fire, all of which can recover. Permanent deforestation requires pairing the loss signal with land-use classification to rule out regeneration.

Indonesia is on track to become the world champion of tropical deforestation. Under the Prabowo administration, over 433,000 hectares were cleared in 2025, nearly double the previous year. Driven by mining and palm oil, the loss equals six times the size of Singapore.

@KanduriC @GenomeBiology One underappreciated subtlety: BH FDR assumes tests are roughly independent. When tests share structure (same samples, correlated features), the effective count drops and BH over-corrects. The correction still matters, the calibration just shifts.

1/12 🧵 Quick thread: a heads-up for anyone working with high-dimensional omics data and multiple testing — FDR correction can sometimes lie to you. A thread based on our recent publication in @GenomeBiology : doi.org/10.1186/s13059…

@MethodsEcolEvol Abundance SDMs carry a data-quality dependency that presence-only methods hide. Unusual abundance combinations propagate into fits and bias the extreme cells. BORG flags multivariate outliers on community matrices so you can check those sites first: cran.r-project.org/package=BORG

📖Published📖 Authors present the adm R package, developed to support the construction of abundance-based species distribution models , including data preparation, model fitting, prediction and model exploration. Read the article here 👇 buff.ly/sXjdv8o

@LauraDBertola Diet composition inside vs outside a PA gets at a less obvious measure of effectiveness. Species counts react first to habitat protection. The quality signal, whether animals access preferred resources, takes longer to surface. Looking forward to reading it.

We have a new paper out! On 🐯🍽️🧬, discussing diet composition inside and outside a protected area and implications for conservation. Diet and Prey Preference of Tigers (Panthera tigris) in and Around Chitwan National Park, Nepal onlinelibrary.wiley.com/doi/10.1002/ec…

@CitSciOZ @ConversationEDU @inaturalist @Team_eBird @reefecologic @CQU @jcu @LyrebirdDream @Stephen48474730 10x is a big revision. The framing challenge is that casual contributors count the same as dedicated repeat observers in total-participant counts. For effort-weighted estimates the ratio between the two groups matters as much as the headcount.

There may be 10 times as many citizen scientists in Australia as we thought – and that’s great news for science! theconversation.com/there-may-be-1… via @ConversationEDU #citscitc #CitSciMonth HT @inaturalist @Team_eBird @reefecologic @CQU @jcu @LyrebirdDream @Stephen48474730 et al.

@chercher_ai Accumulation curves have a hidden property: the shape depends on the order you add samples in. Randomize that order and the curve flattens. Nearby sites share species, so sequential addition inflates the climb through spatial autocorrelation.

In ecology there's a concept called a species accumulation curve or a collector's curve. As you collect more samples in an area, you tend to find fewer new species the more you sample, and you start reaching saturation. You can do the same thing with thoughts: "I"-Naturalist!

@MathonLaetitia One thing that moves you past the tutorial plateau fastest: pick a small task from your real work and do it in R instead of the tool you know. Real tasks make the learning stick in a way practice problems never do. R for Data Science (free online) covers the mechanics.

Hey ! 🔎Would anyone have recommendations of online courses or resources to learn R for beginners ? (Data handling and cleaning, basic data analyses..) 💻📊 Thanks ! #R #DataScience #rstats

@jorgemfassis Trait choice is a hidden degree of freedom that shapes the prediction as much as the climate layers. Different trait sets weight different environmental axes, and that's before hybridization adds parameter uncertainty from both parent distributions.

Does trait choice change the outcome of hybrid Species Distribution Models under climate change? We tested this and fond that correlative & hybrid models converged, both pointing to future range shifts. Read at: doi.org/10.1002/ecog.0… #Marinebiology #climatechange

@JoachimSchork The nested legend case is the one that bites hardest. Multiple colors grouped by category, and every manual positioning fix breaks the next time the data changes. Anything that expresses the structure declaratively saves a lot of late-stage plot reworking.

Struggling with messy legends in your plots? The legendry package for ggplot2 helps you create cleaner and more effective visualizations in R. With better control over legends, it makes your plots easier to understand. Here’s why legendry is a game-changer: ✔️ Customizable Legends: Fine-tune legend positions, titles, and layouts effortlessly. ✔️ Enhanced Readability: Improve plot clarity by adjusting legend elements to fit your audience’s needs. ✔️ Consistency Across Plots: Maintain uniform legend styles when working with multiple data sets. ✔️ Simplify Complex Visualizations: Organize information better by segmenting legends. ✔️ Seamless Integration: Works smoothly with ggplot2 and the tidyverse for an efficient workflow. The visualization shown below is taken from the package website: teunbrand.github.io/legendry/ If you want to learn more about data visualization in R using ggplot2 and its extensions, you might check out my online course on "Data Visualization in R Using ggplot2 & Friends"! Further details: statisticsglobe.com/online-course-… #R4DS #DataViz #RStats #Rpackage #DataAnalytics #StatisticalAnalysis #DataScientist

@WasiqYasir VIF flags the problem. What it leaves open: which variables to drop. For the same correlation threshold there are often many valid uncorrelated subsets, and the one you pick changes the model. corrselect enumerates them all: cran.r-project.org/package=corrse…

Multicollinearity can quietly ruin your regression model 📉 Use Variance Inflation Factor (VIF) to detect it: ✔️ <5 Safe ⚠️ 5–10 Moderate ❌ >10 Problem Fix it with PCA, feature selection, or Ridge regression. X1 & X3 in the plot? 🚨 High VIF #DataScience #MachineLearning

@KVishal1012 For density work, equal-area is the only projection that doesn't silently rescale your numbers. Rates per cell stay comparable across latitudes only when area is constant. hexify generates equal-area hex grids with ISEA tiling for this case: cran.r-project.org/package=hexify

Every projection preserves something and distorts everything else. There is no neutral map. Equal-area projections get area right but distort shape. Conformal projections preserve local angles but distort area. Equidistant projections get distance right from a specific point. You can't have all three. The right CRS for your analysis depends entirely on what property your question actually requires. Most people never ask that question.

@adilbhatti_ The expansion is measurable at the decadal scale. What's harder to pin down is whether it tracks climate alone or interacts with host availability and land use change. Those three move together in most temperate regions, which makes attribution statistically tricky.

Human-biting ticks in temperate zones are expanding with climate change, land use shifts & biodiversity loss. Range expansion, invasive species & new pathogens make tick-borne diseases a growing threat. pubmed.ncbi.nlm.nih.gov/40862142/?utm_…

@epistatistic Congrats. INLA-based DLNMs are a nice addition. The lag-dimension prior specification is the part I'd be most curious about: whether bdlnm lets you set priors separately on exposure and lag, or couples them through the cross-basis structure.

🚀 New R package {bdlnm} now on CRAN! Bayesian DLNM with INLA, extending {dlnm} to a Bayesian framework. Includes prediction, visualization, optimal exposure estimation, attributable risk, and uncertainty from posterior samples. 🔗 pasahe.github.io/bdlnm/ #Rstats #DLNM #INLA