Marcin Lawenda

—

Unclaimed · AI-structured profileORCID OpenAlex

ML researcher applying interpretable deep learning to wildfire spread prediction from ensemble simulations

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-20) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

No output score is available: the linked bibliometric record lacks a publication count for this author. This axis would measure total scholarly production — papers over time and venue spread — and stays unscored until fuller data is linked.

Impact—

No impact score is available: citation and influence metrics for this author are absent from the provided data. This axis would measure how often the work is cited and its reach within wildfire and machine-learning research; it stays unscored, not estimated.

Focus100/100

Scores 100 on focus. The raw field distribution contains exactly one field — machine learning for geoscience (wildfire / natural-hazard modeling) — and the single cataloged 2026 paper sits in it, so maximal concentration follows by construction. Because the score rests on one paper, it signals a deliberately narrow niche so far rather than proven long-term specialization; additional cataloged work would test whether that focus holds.

Collaboration32/100

Scores 32 on collaboration, with 4 distinct coauthors in the raw data: Aleksandra Krasicka, David Caballero, Luis Torres, and Łukasz Szustak. The record shows a small, stable team around the 2026 first-author paper rather than a broad network. The low-moderate score reflects limited breadth so far; whether this core group persists or expands cannot be judged from one cataloged paper.

Activity—

No activity score is available: the data include no multi-year publication timeline. This axis would measure publishing cadence and recency — whether output is sustained or episodic — and a single 2026 paper cannot support that judgment.

Reproducibility55/100

Scores 55, the mean over exactly 1 paper (reproPapers: 1), which Evolve rates at medium reproducibility readiness. The reviewed study is a controlled, honest comparison of U-Net, physics-informed, and Swin transformer surrogates on ensemble simulations — a design that aids replication — but synthetic-only data and fuel-load circularity temper external reuse for real-world wildfire prediction. A single-paper basis makes this score indicative only.

Research summary

Marcin Lawenda (ORCID 0000-0003-4844-3655) is a researcher working at the intersection of machine learning and geoscience, focused on wildfire and natural-hazard modeling. The Evolve catalog contains one reviewed paper: the 2026 arXiv preprint 'Interpretable Patch-Based Deep Learning for Wildfire Spread Prediction from Ensemble Simulations', on which Lawenda is first author. The study, reviewed as controlled and honest, compares deep learning surrogates — including U-Net, physics-informed neural network, and Swin transformer variants — as emulators of wildfire-simulator ensembles, and finds small U-Nets best reproduce the simulated spread. Its stated limits are synthetic-only training data and fuel-load circularity, which temper real-world claims; Evolve rates its reproducibility readiness as medium (repro mean 55). The paper's interpretable, patch-based framing targets hazard-relevant quantities such as burn probability, situating the work in surrogate modeling for geoscience. The bibliometric snapshot is correspondingly narrow: a single cataloged field, four distinct coauthors (Aleksandra Krasicka, David Caballero, Luis Torres, Łukasz Szustak), a maximal focus score of 100, and a collaboration score of 32. Output, impact, and activity axes remain unscored because the underlying bibliometric data are insufficient.

Collaboration pattern

The cataloged record points to compact teamwork: 4 distinct coauthors appear across the reviewed data, and the single cataloged 2026 paper is first-authored. Recurring collaborators are Aleksandra Krasicka, David Caballero, Luis Torres, and Łukasz Szustak. A collaboration score of 32 (0-100 scale) reflects a small, concentrated coauthor pool rather than an extended network; with only one cataloged paper, the breadth and durability of these collaborations cannot yet be judged.

Research keywords

wildfire spread predictiondeep learning surrogatesU-Netphysics-informed neural networksSwin transformerburn probabilityensemble simulationspatch-based deep learninginterpretable deep learningmachine learning for geosciencenatural-hazard modeling

Highlights

Small U-Nets best reproduce wildfire-simulator ensembles
Evolve's review of the 2026 arXiv preprint (first-authored by Lawenda) reports that among the tested deep learning surrogates — U-Net, physics-informed neural network, and Swin transformer variants — small U-Nets best reproduce the wildfire-simulator ensembles; the paper's keyword set spans wildfire spread prediction, deep learning surrogates, and burn probability.
Interpretable, patch-based design for ensemble emulation
The preprint's framing — interpretable, patch-based deep learning trained on ensemble simulations — positions the work as surrogate modeling for hazard prediction, in the single field Evolve classifies as machine learning for geoscience (wildfire / natural-hazard modeling).
Honest limits: synthetic-only data and fuel-load circularity
The same review flags that training relies on synthetic data only and that fuel-load circularity limits real-world claims; the paper's reproducibility readiness is rated medium (repro mean 55 across 1 paper), so independent validation on real data remains open.

Timeline

2026Posts first-authored arXiv preprint 'Interpretable Patch-Based Deep Learning for Wildfire Spread Prediction from Ensemble Simulations'.
2026Evolve's review calls the study controlled and honest, finding small U-Nets best reproduce wildfire-simulator ensembles.
2026The paper enters the Evolve catalog rated medium for reproducibility readiness (repro mean 55 across 1 paper).
2026Work is cataloged in one field — machine learning for geoscience (wildfire / natural-hazard modeling) — with 4 distinct coauthors.

Papers in the Evolve catalog

2026-09-17first author
Interpretable Patch-Based Deep Learning for Wildfire Spread Prediction from Ensemble Simulations

Frequent co-authors

Aleksandra Krasicka×1 David Caballero×1 Luis Torres×1 {\L}ukasz Szustak×1

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

Bibliometric dimensions are computed from public bibliometric data (OpenAlex-sourced); institutional affiliation and country could not be resolved.
Output, impact, and activity axes are unscored because publication-volume, citation, and cadence data are insufficient for this author.
The reproducibility score (55) averages a single cataloged paper, so it is a weak, early-stage signal.
The profile is name-resolved from public bibliographic data (ORCID 0000-0003-4844-3655) and rests on one reviewed paper; treat it as early-stage.