Palimpsest

New journal article: optimizing your genetic archive

Tue, 08 Sep 2026 20:15:30 +1200

Breeding programs often keep an ex-situ “backup” of their breeding material. In tree breeding, that archive is also used for creating new genetic combinations (matings) instead of working with the original genotypes that are in trials in the boonies, the boondocks, the woops, the middle of nowhere.

In the same way that you should not keep a single backup, the breeding program keeps multiple copies of the genetic archives. After a while, this occupies more and more room and it turns out to be more and more expensive.

This article explores optimization of the archive (using genomic data) so we maintain as much as variability with as few individuals as possible. The article is Open Access and is still subject to minor edits.

Duncan McLean, Mark F. Paget, Luis A. Apiolaza, Ken G. Dodds & Sai Krishna Arojju. 2026. Genomic diversity, trait–ancestry associations and core subset optimization of Pinus radiata germplasm in New Zealand. Scientific Reports DOI: 10.1038/s41598-026-69338-0

This work was funded by the NZ Radiata Pine Breeding Company. Thanks to SAi Arojju for the final push to get this article over the line.

The article I didn't write

Thu, 03 Sep 2026 13:50:46 +1200

I am writing an article¹ on genetics of wood properties and slowly doing a literature review when I can set a little time aside. After looking in academic databases (Scopus and Google Scholar²) I did a Google search, thinking “perhaps people wrote something outside journals and conferences”. Besides the typical junk results, there was the AI summary, which I usually ignore. However, this time I was curious, clicked on it and ended up in a Gemini session.

I tried to describe what I wanted and Gemini replied explaining me the topic. No, I want journal articles. I got a link to one useful one, but with the wrong authors, and three made up ones. Nothing new here: LLM are designed to make up answers. However, the following article caught my attention, because I did not remember writing it and I was supposed to be the first author!

Strange how software can undermine reality. I remember all the promise of the late 1990s Google search field; now it’s a dumpster fire of misinformation and disinformation.

I take ages writing articles, I think of the problem, meditate it and it may take 5 years to see the light. On top of that, I don’t like to write using a template. ↩︎
“Academic” between quotes, of course. It’s full of repeated papers, wrong attributions, etc. ↩︎

Jetsam 74: commuter selfie

Wed, 03 Jun 2026 23:13:22 +1200

How does it fit?

Wed, 03 Jun 2026 22:09:18 +1200

There is a question, or group of questions, that often comes up when chatting with colleagues and students: How does X fit in the breeding program? or How do you keep track of X, Y and Z in the breeding program? or, even, How do you cope with so much stuff that keeps on coming to the breeding program? By the way, when I say breeding program, I mostly mean the genetic evaluation part of the program, which I consider the central (dare I say most important?) part.

Now, to answer the question I have to acknowledge that I’ more of a mathematics than a biology person; I like to learn a few basics from which I can derive everything else. Therefore, unsurprisingly, I think of only two relationships when talking about new(ish) technologies, drumroll…

the breeder’s equation (gain = intensity x accuracy x variability / time) and the mixed model equations (y = Xb + Za + … + e). Boring, I know.

I turn whatever new element in two questions:

How will it affect gain by potentially changing variability, accuracy, selection intensity or time to selection? and
Where does it enter the mixed model equations?

I have talked about the breeder’s equation before, and keep in mind that I often think of the multivariate version, because we are selecting for a selection index, and working on improving a multitrait breeding objective.

When talking about the mixed models, I think at two levels, single environment and multienvironment. At the single location, the model basically accounts for an experimental design, a mating design (via pedigree/genomics) and a potentially very long etc. Say we get LiDAR from drones then we can get more Xs in the model (topographic indices from the DTM, competition from other individuals from the crown height model, etc) but sometimes we get more Ys (tree heights, for example). Or we get new assessments from Near Infrared Spectroscopy from core samples: more Ys. Or we get hyperspectral imaging which can produce more Xs describing aspect of the terrain, and more Ys as selection criteria for drought or disease, etc.

Once we move across environments, there is access to a huge amount of environmental datasets (both climate and soil related) that can be used to explain changes of ranking GIVEN that we have good connectedness between environments. If we don’t, my assumption is that we are chasing statistical noise.

After trying this for a while, it turns into a reflex. Hey, what if we get data from “shiny new thing” now that’s a lot cheaper? Use one side of a napkin to figure out how it affects the breeder’s equation, and then the other side of the napkin to check implications on the genetic evaluation system.

In the old days you could have used an envelope or a cigarette pack for writing and figuring out the problem, but who posts stuff or smoke these days?

Fresh from the oven

Tue, 26 May 2026 21:18:20 +1200

As a good cowboy—or is it stockman in Australia?— Laurie Schimleck rounded up a few of us to put together a review article on wood durability. Due to various circumstances, it took a bit longer than expected but finally came out! We just received a free reading author link to share “Heartwood Durability: The Quest for Natural Decay Resistance in a Changing Resource” in Current Forestry Reports (2026) 12:15 (full article link).

Wood durability (from a breeding point of view) picked my attention since around 2008, when we started work with the New Zealand Dryland Forests Initiative (NZDFI). Durable eucalypts are fantastic creatures, with their own trade-offs and complications, lots of Near Infrared Spectroscopy with students and colleagues, and a few published articles and some in preparation.

Thanks to Lauri for pushing us along.

Forestry is about people

Mon, 18 May 2026 17:34:21 +1200

I have heard that many times. I’m sure that you have heard a variant of this phrase in your own work area; simply replace ‘Forestry’ with whatever you do. Truck driving is about people or Nuclear engineering is about people. However, when one works in research rarely are people the subject of study. I work in the intersection of forestry, wood properties, quantitative genetics and a long etcetera BUT I have never seen a study of people doing this stuff. Instead, there are thousands of studies about the weirdest details of that intersection.

Last Wednesday, I was attending six presentations by undergraduate students working in their dissertations (I am coordinator for that work in our School). Remote sensing is very popular, economics and management are always there. There was even a presentation on genetics of wood properties. And then, gloriously, one of the students (Ruby Sumner) shared that she did a Bachelor of Psychology before studying forestry and that her dissertation was on the social impact of forestry workers in their communities (supervised by Trevor Best). I am the first to confess that I’m a quantitative person, a numbers wallah if you will, but I’m also willing to confess that I was very happy to hear about this not-so-quantitative study.

Dissertations are small projects and I believe the subject deserves much more attention. It will be a lot easier to say that forestry is about people when we know what is happening to their lives, particularly forestry workers.

Ah, learning

Sat, 25 Apr 2026 20:35:14 +1200

I keep on forgetting to write down the addresses of interesting resources for writing and hand-coding

HTML for People which is a good reminder of how to put together websites using HTML. I know, on these AI days one is almost automatically supposed to ask an LLM; but I want to know how things work. Better, I want to remember what I started learning 30 years ago.
LazyVim for Ambitious Developers, billed as ‘a course for the editor curious’ and a pretty good introduction to working with a great Neovim distribution.
Influential Engineering Resources with links to books and blogposts on software engineering.

Fixing the pipeline, again

Fri, 13 Mar 2026 11:31:30 +1300

All good with my pipeline until I updated Hugo, and woke up a theme bug. I was not really looking forward to updating Hugo, but I run brew upgrade and everything was dutifully updated. Isn’t software marvelous? Well, there was a lurking bug in the Congo theme. Really, and the theme has not been fixed yet.

At the end I removed the theme, which was as a git submodule (that’s a whole I hate git subplot), cleaned the repository to remove all vestiges of the submodule, downloaded the latest version of the not-yet-released version of the theme, copied it under themes, deleted the 2 offending lines, git added the new version of the theme and the site works again. Finished at 12:30 am.

Symptoms: it all started when I run hugo serve and got “ERROR error building site: render: [en v1.0.0 guest] failed to render pages” as an answer in Terminal. Tried a few times and the first fail was always different, but always pointed to .../themes/congo/layouts/_partials/functions/warnings.html:10:6. This led me to the theme error.

Complications: The theme was installed as a git submodule that, first, would not update. And, even worse, even the dev release didn’t include the fix. Then the question was How to remove a submodule? According to a search, I was supposed to deinitialize the module, whatever that means, using git submodule deinit -f themename. This failed miserably, so I first deleted the theme rm -rf themes/congo and then run git submodule deinit -f themes/congo, which worked.

After that, I think I removed the submodule from git using rm -rf .git/modules/themes/congo, followed by removing the submodule from .gitmodules by using git config -f .gitmodules --remove-section submodule.themes/congo.

I added the whole shebang to git and commited the change. Later, I copied the latest version of Congo under themes (just plain copy), deleted the offending lines, git add and git commit the theme and, you know, it worked!

Thanks to this page for most of the fix.

Influences 1

Wed, 11 Mar 2026 19:29:19 +1300

I’ve used the notes on “Estimadores del valor genético aditivo en animales. I. Una fuente de información” on and off for the last 34 years.

Prof Ximena García gave me a copy when I was doing her course in 1992, my last year of undergraduate forestry. Funny thing is that this was a postgraduate course, taught in the Department of Animal Production, which was part of Agronomy, not Forestry. I took the course for no credits, just for the fun of learning quantitative genetics. Yes, it sounds weird when I put it that way. We were only two students: a postgrad and myself, and we learnt a lot.

There was a lot of doing his small things by hand, like the regression of additive value on phenotype. Typically the slope of a regression of Y on X is Cov(X,Y)/V(X), so

$$ \frac{Cov(P,A)}{V(P)} = \frac{Cov(A+D+I,A)}{V(P)} = \\ \frac{Cov(A,A)+Cov(D,A)+Cov(I,A)}{V(P)} = \\ \frac{Cov(A,A)+0+0}{V(P)} = \frac{V(A)}{V(P)} = h^2 $$

which helped me understand the idea of heritability; cool, I think.

The cover has lost good part of its colour, and the pages are typewritten, no fancy fonts for the huge number of equations.

Thanks for the book, Ximena, which still is in my office bookshelf after all these years.

Ubuntu update

Mon, 02 Mar 2026 22:49:12 +1300

Today I installed R, r2u (using the script for 24.04 LTS), and RStudio. I also installed neovim and lazyvim, trying to keep basic feature parity between my Ubuntu desktop and macbook.

Talking about parity, dealing with apt (or nala when dealing with larger programs) works really well with system-wide stuff. However, installing smaller, newer stuff feels very hacky (download this, move it there, adjust whatever path, etc).

I find using brew much easier in this situation, with the advantage that I end up with the same versions in both desktop and laptop.

Why neovim? It’s part of my going back to the terminal and reducing contact with Microsoft (sorry, VS Code). Lately I’ve felt this constant nagging from some companies and living in a non-Microsoft environment feels peaceful for a few hours a day.

MKinfer: inferential statistics

Sun, 01 Mar 2026 18:30:49 +1300

R package that includes the MKinfer::perm.t.test() function which performs one-sample and two-sample permutation t-tests, including support for paired tests and confidence intervals.

Are you a cauliflower person

Sun, 01 Mar 2026 14:05:19 +1300

I’ll eat most Brassica oleracea derivatives but kale. You say broccoli? Cockily. Brussels sprouts? In great bouts! Cauliflower? I’ll devour it.

I enjoy reading good stories about breeding as much as, I don’t know, my next door neighbour, but they have to be fun. Meet your greens: How a single unappetizing shrub became dozens of different vegetables kept me reading until the end, so I thought more people who are into breeding things would appreciate it.

Fixing my pipeline

Sat, 28 Feb 2026 23:01:59 +1300

I want to like Hugo—or any static site generator + git system for that matter—but it doesn’t quite flow; it doesn’t feel like my own. So I checked what I was missing.

My first obstacle was starting with the empty page, so I needed to use a proper archetype:

---
title: "{{ strings.Replace (strings.Substr .Name 11) "-" " " | strings.FirstUpper }}"
date: {{ .Date }}
draft: true
author: "Luis"
tags:
  - 
---


![Caption here](/assets/images/image-here.jpeg)

But I also only wanted to type something like newpost some-name and still get the date at the beginning of the file name, as in 2026-02-28-some-name.md. This needed to create a shell function (in my case in zsh), so I included the following in ~.zshrc:

newpost() {
  name=post/$(date -I)-$1.md
  hugo new $name
  code content/$name
}

By the way, code calls Visual Studio Code from the terminal. I still need to fix other inconveniences, but now it is a little less annoying to write posts.

Keeping up with research

Fri, 20 Feb 2026 18:00:00 +1200

It is impossible to be up to date in research. Some people try, but I gave up a long time ago. Even if my whole job was reduced to reading new material, I could not keep up with the volume of journal articles. The first time I was aware of this, it produced a sense of defeat or existential dread; it was overwhelming. Then I had to dig into my (theoretical) inner Buddhist and just let go: there is no point on keeping with the news.

I convinced myself that it was much better to understand a small core of ideas—mostly the breeder’s equation, mixed linear models, context specific information about my crops—and the rest was anchoring new stuff into the core. Say that I got NIR spectra relationships; that’s new responses (y) in my model. Or genomic data, that’s a bunch of X, or simply a corrected pedigree or enriched numerator relationship matrix. Perhaps we got hyperspectral drone data; we have some new tree crown metrics that may end up as X, competition indices that may enrich the right-hand side of the models, etc.

I believe one feels ’less panicky’ about new stuff if one can link it somewhere in the model, or graft it to one’s knowledge model, to sound like dealing with trees. I don’t need to keep on pressing Ctrl+R on new research, but only figure out it fits in my well-understood abstraction of the world.

Less willing

Wed, 11 Feb 2026 20:30:00 +1200

One effect of AI, at least on me, is that I am less willing to write public posts.

I have always liked making up stories and writing ’explanations to myself’, so I do not forget how I fixed a problem. At the beginning, it was on paper, but then I had my first web page around 1996. Remember Geocities? This was followed by my first domain names in 2002. Posting on the web was good for me; I even got job offers from that.

Lately, however, writing has been twisted into something else. The whole web, including this site, has been (and is being) absorbed by AI companies, so people can quickly regurgitate millions of blog posts, images, videos, … content. Training material for placeholders to be sandwiched between ads.

AI makes me question sharing and the idea of contributing to open culture, which is the complete opposite of what we wanted to do at the beginning of the internet. Anathema. Irony: LinkedIn offering me to rewrite this post with AI.

Why the old is not yet new

Mon, 15 Dec 2025 19:30:00 +1200

The idea of using economically defined breeding objectives is 80 years old.BLUP is 70 years old, R and Python are over 30 years old, techniques for analyzing large numbers of markers are 25 years old. And still you’ll find many breeding programmes that are not using some or all of these tools.

What is relevant is when these tools are economically and technically feasible to be used. For some breeding programmes this is just happening now. Moreover, even now your use can be suboptimal because of software issues. Most software can’t integrate the complex models we need for MET and genomics when we have large numbers of genotyped individuals.

Most crop breeding programmes don’t have economic weights for their traits. Many are not yet using animal model BLUP. Even fewer have operational access to genomic data. There is a lot to be done. 😉

Optimal

Tue, 23 Sep 2025 18:30:00 +1200

I think of myself as a “numbers wallah”^*, a guy who deals with problems by turning them into maths of various kinds. During the last 35 years, my kind of maths has mostly been statistics mixed with quantitative genetics, which is a very useful combination. But maths has been present in my life for longer, at least since high school.

Thirty six years ago I was sitting an Optimization course, as part of my forestry degree (sporting longer and darker hair, equally nerdy, though). Linear programming, dynamic programming and simulation, that kind of stuff. I was fascinated by the subject and was going to do my undergraduate thesis in some forest transportation problem. But then came tree breeding and changed the direction of my life (thanks, Roberto!).

Last week, I was talking with a colleague visiting from Chile, and we were talking about a harvesting optimization problem with spatial constraints. I mean, he was doing all the work, I was just throwing opinions about his objective function… And I remembered how much I liked this type of problems. Together with working with forest mensuration students and linking all this work to financial evaluations, I have been rediscovering parts of forestry work that I haven’t touched in a long time and that made me feel excited. Now I am starting to read, a little bit, about optimization and thinking of how to connect it to the rest of my work.

^*In the same way there are chaiwallahs or dabbawallahs.

Rain or shine

Wed, 20 Aug 2025 21:30:00 +1200

During the last week, we had 3 field activities with forestry students of the University of Canterbury: two trips to Bottle Lake Forest with first years doing Forest Mensuration and a trip to Amberley with second years doing Introduction to Silviculture.

The weather was spectacular for a winter day in Amberley, where Shaf van Ballekom, manager of Proseed, showed us the seed orchard and their propagation work. It’s a great connection between Ngāi Tahu Farming and the School of Forestry.

In contrast, the weather was subdued during our first visit to Bottle Lake when learning to establish prism plots. Later, it was cold (near 0°C) with some rain and even hail during our second visit to practice flexible timber cruising. We had good conversations with students, sorting out why things were working well or not working when compared to the classroom. Kudos to Matt Deering (School of Forestry) for helping me with 48 students in the mensuration activity.

Cute detail: a few students thought that the last activity was going to be cancelled because it was cold and rainy. I pointed out the course outline and first lecture: activities run “rain or shine” (unless it affects the safety of the activity). Warm clothes, rain gear and a thermos with a hot drink go a long way towards enjoyable time in the field.

Barely above sea level

Mon, 11 Aug 2025 18:30:00 +1200

Last Friday I went to Bottle Lake Forest, a multipurpose plantation just at the edge of Christchurch. Being next to the city, recreation is a very important activity and there are plenty of walking and cycling trails. I was there checking a stand for taking the forest mensuration class for an activity, looking at some bearings (road and stand were very neatly orientated) and then I saw: 10 metres above sea level.

Christchurch is a very flat city, next to the ocean, but sometimes one loses the sense of how low we are. I live at around 30 masl and last week we went up to 100 masl (!) while driving 16 km. Low in productivity too; we may need to adjust the age for running the growth models.

It’s very cool to have the chance to practice with a stand that’s pretty much in the city.

100 billion dollars

Fri, 08 Aug 2025 21:09:23 +1200

That is $100 000 000 000. If you have 1 million paying customers, each of them has to pay $100 thousand. If you get 100 million paying customers, each of them has to pay one thousand dollars just to recoup capital costs, and without considering the time value of money. Without upgrading components every few years.

Now multiply all the values by 3. That’s the reported projected investment in data centers for AI in 2025 alone. Three hundred billion for bullshit machines.

You may counter “companies are spending their own money”. But if you have an investment fund or a retirement fund, and I do, most likely you are part-owner of Alphabet, Amazon, Meta, Microsoft, etc. We are part-funding this bubble. I am sure that everyone can think of better uses for these resources, from the betterment of humanity to the creation of thousands of new companies trying new things. Instead, we got bullshit and plagiarism machines based on stealing intellectual property at a massive scale.