⚡ electrical engineering

Software Uncertainty in Integrated Environmental Modelling: the role of Semantics and Open Science

This paper proposes that semantic transparency and open-source software are essential strategies for mitigating the often-overlooked risk of software errors in large-scale transdisciplinary environmental modeling, thereby ensuring more reliable science-based support for policy-making under conditions of deep uncertainty.

Original authors: Daniele de Rigo

Published 2026-06-04

📖 5 min read🧠 Deep dive

CC BY 3.0

Original authors: Daniele de Rigo

Original paper licensed under CC BY 3.0 (http://creativecommons.org/licenses/by/3.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to bake a massive, world-changing cake to decide how to protect our forests and rivers. This isn't just a recipe; it's a "transdisciplinary" masterpiece, meaning it mixes ingredients from climate science, economics, geography, and biology into one giant, complex batter.

The author, Daniele de Rigo, argues that while we spend a lot of time worrying about whether the ingredients (the data) are fresh or if the recipe theory (the science) is sound, we are completely ignoring a hidden danger: the software itself might be broken.

Here is the breakdown of the paper's main points using simple analogies:

1. The "Black Box" Problem

In environmental modeling, scientists often use long chains of computer programs to turn raw data into final answers.

The Analogy: Imagine a giant, sealed vending machine. You put raw ingredients (data) in the top, and a finished cake (the policy recommendation) pops out the bottom.
The Issue: Most of these machines are "black boxes." You can't see inside. You have to trust the label on the box (a scientific paper) that says, "This machine works perfectly." But you can't see the gears, the wires, or the code running the machine.
The Risk: If there is a tiny, silent glitch in the gears (a software error), the machine might still spit out a cake that looks perfect but tastes terrible. Because the error is hidden, no one notices until it's too late. These are called "silent faults."

2. The Invisible "Software Uncertainty"

We usually think of uncertainty as "we don't know enough about the climate" or "our data is messy." The paper introduces a new kind of uncertainty: Software Uncertainty.

The Analogy: Even if you have the perfect recipe and the freshest eggs, if your oven has a broken thermostat that randomly adds 50 degrees of heat, your cake will burn. That broken thermostat is the software error.
The Claim: In complex environmental models, these software glitches can subtly change the results. They might make a flood look like a drought, or a safe forest look like a fire hazard. Because the software is so complex, even the people who built it might not fully understand how the errors spread through the system.

3. The Solution: "Open Source" and "Semantic Transparency"

The paper suggests two main ways to fix this, comparing them to opening the kitchen doors and labeling the ingredients.

A. Open Science (The "Open Kitchen" Approach)
Instead of keeping the recipe and the machine locked away, scientists should share the actual source code (the instructions for the machine).

The Analogy: Instead of just showing you the finished cake, the chef says, "Here is the recipe, here is the list of ingredients, and here is the code for the oven. You can try to bake it yourself to see if you get the same result."
Why it helps: If the code is free and open, other experts can look inside the "black box," find the broken gears, and fix them. This is the first step toward "reproducible research."

B. Semantics (The "Labeling" Approach)
This is about making sure the computer understands what the data means, not just what it is.

The Analogy: Imagine a robot chef that only knows numbers. If you give it a number "5," it doesn't know if that's 5 cups of flour or 5 cups of poison.
The Solution: "Semantic" programming adds labels to the data. It tells the computer, "This number is temperature, and it must be between -20 and 50."
The Benefit: If the computer tries to use a temperature value as a weight for a cake, the "semantic check" acts like a smart alarm. It screams, "Wait! You are using temperature as weight! That makes no sense!" This catches errors before they ruin the final result.

4. The "Design Diversity" Safety Net

Finally, the paper suggests that for the most critical decisions, we shouldn't rely on just one machine.

The Analogy: If you are flying a plane, you don't rely on just one compass. You have three. If one is broken, the other two tell you the truth.
The Claim: Scientists should build different versions of the same model using different software teams and different coding styles. If all three different "machines" give the same answer, you can trust it. If they give different answers, you know something is wrong with the software, and you need to investigate.

Summary

The paper is a warning: We are building complex environmental models on top of software that might be full of hidden errors.

To fix this, we need to stop treating software as a mysterious "black box." We need to:

Open the code so anyone can check it (Open Science).
Label the data so the computer knows what it's doing (Semantics).
Build multiple versions of the models to cross-check them (Design Diversity).

By doing this, we ensure that the "cakes" we bake for environmental policy are actually safe to eat, rather than being poisoned by invisible software bugs.

Technical Summary: Software Uncertainty in Integrated Environmental Modelling

Problem Statement
The paper addresses a critical, yet frequently underestimated source of uncertainty in Wide-Scale Transdisciplinary Modelling for the Environment (WSTMe): software uncertainty. While environmental sciences increasingly rely on computational aspects to model complex systems, the structural role of software errors is often neglected in favor of theoretical modeling assumptions.

The core problem is the prevalence of "black-box" executables in Integrated Natural Resources Modelling and Management (INRMM). These systems often consist of chains of nontrivial data-transformation models (D-TM) packaged as monolithic units where the internal mechanisms are hidden. End-users rely on peer-reviewed publications to describe these mechanisms, but the actual software code is rarely accessible. This creates a paradox where software is vital to computational science (CS) but remains "invisible" in published research.

The paper argues that software errors in these complex chains can subtly alter outcomes without obvious symptoms (silent faults), potentially affecting collective environmental decision-making. This is particularly dangerous in contexts of "deep uncertainty," where statistical analysis of modeled quantities is insufficient, and policy decisions rely heavily on model outputs.

Methodology and Conceptual Framework
The paper does not present a new experimental dataset but rather proposes a conceptual and methodological framework to mitigate software uncertainty through Semantic Transparency and Open Science.

Distinction between Theoretical and Real D-TMs:
The author distinguishes between a "Theoretic D-TM" ( $Y = f^*(X)$ ), where the algorithm is described in literature and parameters ( $\theta^*$ ) are optimized, and a "Real D-TM" ( $Y = f_\zeta$ ), where software uncertainty ( $\zeta$ ) affects both the function family and the optimality of parameters.
Semantic Enhancement:
The proposed methodology involves moving from black-boxes to Semantically Enhanced D-TMs. This approach utilizes Semantic Array Programming (SemAP) and its extensions to geospatial problems.
- Modularization: Breaking down monolithic models into concise, free-software modules.
- Semantic Checks: Applying modal/deontic logic operators ( $\square_{sem}$ ) to inputs, outputs, and the modules themselves. These checks act as pre-conditions, post-conditions, and invariants to ensure data is used within its expected semantic context.
- Goal: To catch "silent faults" that occur when intermediate data is misused outside its expected semantic context, which traditional error handling (exceptions, crash messages) often misses.
Design Diversity and Ensemble Modelling:
To address the unpredictability of software error propagation, the paper suggests adapting ensemble modelling techniques. Instead of relying on a single implementation, the framework advocates for:
- Design Diversity: Independently re-implementing critical D-TM modules (partially independent re-implementations) to compare supposedly equivalent algorithms.
- Transparency: Making the chain of free-software modules transparent, exposing not just final outputs but key intermediate layers of data and information.

Key Contributions

Identification of Software Uncertainty: The paper explicitly categorizes "software uncertainty" as a distinct and critical component of the overall uncertainty in environmental modelling, separate from incomplete scientific knowledge or modeling assumptions.
The "Invisibility" Paradox: It highlights the disconnect between the necessity of coding in scientific research and the lack of code sharing in publications, arguing that this hinders reproducibility.
Semantic Modularization: It proposes semantic array programming as a technical solution to make code understandable and verifiable, moving beyond simple code availability to semantic clarity.
Mitigation Strategy: It outlines a strategy where the "chain of free-software modules" should be transparent, allowing for the detection of silent faults and the application of ensemble techniques to gauge software-induced uncertainty.

Results and Claims
As a position paper and abstract, the work does not report specific experimental results or quantitative performance metrics. Instead, its "results" are theoretical arguments supported by citations to existing literature on software engineering and environmental modelling.

Claim on Silent Faults: The paper claims that semantic modularization can help detect a subset of silent faults that would otherwise corrupt scientific results.
Claim on Reproducibility: It posits that the free availability of source code is the "first step" from non-reproducible research toward reproducible research.
Claim on Decision Making: It asserts that software errors in WSTMe can have consequences on collective environmental decision-making, particularly when deep uncertainty prevents standard statistical validation.

Significance
The paper argues that the structural role of scientific software uncertainty is "dramatically underestimated." Its significance lies in shifting the focus of environmental modelling from purely theoretical uncertainty (e.g., climate scenarios, tipping points) to the practical, often hidden uncertainty introduced by the software implementation itself.

By advocating for Open Science and Semantic Transparency, the paper suggests that the scientific community must move away from black-box models. It proposes that only through the availability of free software, semantically enhanced modularization, and design diversity can the field adequately address the "unpredictable" behavior of software errors in complex, large-scale environmental systems. This approach is presented as essential for providing robust, science-based support to policy-making in the face of deep uncertainty.