Writing study: MOVEBENCH: A Benchmark for Global-Scale Wildlife Movement Forecasting

NeurIPS overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← NeurIPS 2026 oral writing corpus

Justin Kay, Shir Bar, Ellen O Aikens, Martin Becker, Francesca Cagnacci, Juliet Cohen, Scott W Forrest, Jessica Kendall-Bar, Madeleine Lucas, Macon Overcast, Meredith S Palmer, Will Rogers, Nicholas J Russo, Christian Rutz, Larissa T Beumer, Michael B Brown, Ying-Chi Chan, Sarah C Davidson, Diego E Soto, Anne G Hertel, Roland Kays, Benjamin Koger, Guram Mikaberidze, Thomas Mueller, Ruth Oliver, Thorsten Papenbrock, Robert Patchett, Jared A Stabach, Dane Taylor, Scott W Yanco, Sara Beery · Official NeurIPS 2026 oral record

Observed writing choices, not acceptance rules. This study examines the inspected manuscript version. It does not establish why the paper was accepted or that its presentation choices should be copied unchanged.

Figure, table and section ledger

Editorial contribution type: Datasets, benchmarks & measurement (primary).

Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.

Canonical PDF

72 physical pages

Before references / appendix

≈ 10.09 page-equivalents

Main figures / tables

4 / 2

Attached appendix figures / tables

1 / 8

Attached appendix span

≈ 34.91 page-equivalents

Before references / appendix: ≈ 14.01%References / intervening material: ≈ 37.5%Attached appendix: ≈ 48.49%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Source-bound correction: Source-checked 2026-10-02: remove the false bibliography Appendix A on page 28 and the A-G contents entries on page 38. Keep page 38 as the appendix cover/contents boundary, and measure real section starts on pages 39, 40, 45, 52, 52, 53 and 55. The long references section begins on page 11 and continues through page 37.

Section-by-section lengths and source pages

The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.

Top-level heading	Starts on PDF page	Estimated span	Share of entire PDF
1 Introduction	1	≈ 2.51 pp.	3.49%
2 Related work	4	≈ 0.39 pp.	0.54%
3 Movement forecasting	4	≈ 0.94 pp.	1.31%
4 MOVEBENCH	5	≈ 0.88 pp.	1.22%
5 Metrics	6	≈ 0.96 pp.	1.33%
6 Experiments	7	≈ 2.77 pp.	3.85%
7 Discussion and Conclusion	10	≈ 0.47 pp.	0.65%
Acknowledgments	10	≈ 0.41 pp.	0.57%
References	11	≈ 27 pp.	37.5%
Appendix	38	≈ 1 pp.	1.39%
A Additional related work	39	≈ 1.08 pp.	1.5%
B Trajectory datasets	40	≈ 5.68 pp.	7.89%
C Environmental covariates	45	≈ 6.53 pp.	9.07%
D Metrics	52	≈ 0.29 pp.	0.4%
E Method details	52	≈ 1.19 pp.	1.65%
F Hyperparameters	53	≈ 1.24 pp.	1.72%
G Additional experiment details	55	≈ 17.91 pp.	24.88%

All detected figure and table locations

Main text

• Figure 1 · p. 2
• Table 1 · p. 4
• Table 2 · p. 5
• Figure 2 · p. 6
• Figure 3 · p. 8
• Figure 4 · p. 9

Attached appendix

• Table 3 · p. 40
• Table 4 · p. 46
• Table 5 · p. 47
• Table 6 · p. 48
• Table 7 · p. 49
• Table 8 · p. 49
• Table 9 · p. 49
• Table 10 · p. 55
• Figure 5 · p. 71

Argument and organization

1. Problem and contribution

   The benchmark motivates probabilistic wildlife movement forecasts in continuous environments rather than repackaging road-constrained human trajectories. It explains why individual transfer, species diversity and environmental covariates change the evaluation problem before discussing model scores.

2. Argument sequence

   The paper defines a forecast and its inputs, describes corpus construction, specifies a distributional metric, and only then evaluates baselines under temporal and held-out-individual splits. The appendix carries the extensive source attribution and disaggregated study results needed to inspect such a heterogeneous corpus.

3. Evidence and verification

   The main comparison retains simple motion baselines and distinguishes temporal forecasting from transfer to new individuals. Normalized energy score measures a predicted distribution, not merely a point location. The text reports settings where deep approaches do not consistently beat simpler baselines; Figure 4 adds probability-surface examples rather than only trajectory overlays.

Problem, claim and contribution

The benchmark motivates probabilistic wildlife movement forecasts in continuous environments rather than repackaging road-constrained human trajectories. It explains why individual transfer, species diversity and environmental covariates change the evaluation problem before discussing model scores.

Introduction pp. 1–3; forecasting formulation §3, p. 4; dataset §4, p. 5 · Main paper

Argument and section order

The paper defines a forecast and its inputs, describes corpus construction, specifies a distributional metric, and only then evaluates baselines under temporal and held-out-individual splits. The appendix carries the extensive source attribution and disaggregated study results needed to inspect such a heterogeneous corpus.

§§3–5, pp. 4–6; evaluation pp. 7–9; Appendices B–G, pp. 40–72 · Main paper

Figures, captions and visual explanation

Figure 1 is a coordinated six-panel overview: a world map, covariate layers, species-specific trajectories with scale bars, forecast-task sketch, phylogenetic coverage and movement ranges. Figure 3 uses repeated axes and separate split rows to compare covariate use, taxonomic groups, movement ranges and horizons without hiding subgroup behavior in one mean.

Figure 1, p. 2; Figure 3, p. 8; rendered pages 2 and 8 · Main paper

Tables, comparisons and evidence

The main comparison retains simple motion baselines and distinguishes temporal forecasting from transfer to new individuals. Normalized energy score measures a predicted distribution, not merely a point location. The text reports settings where deep approaches do not consistently beat simpler baselines; Figure 4 adds probability-surface examples rather than only trajectory overlays.

Metric §5, p. 6; Figure 3 and results, pp. 7–8; Figure 4, p. 9 · Main paper

Appendix and reproducibility

Source-study tables include provenance/licensing and curation details; covariate tables record cadence, spatial resolution and licenses. Table 10 is one long continued per-study result table across pp. 55–71, not seventeen independent tables. This is an evidence appendix organized for attribution and heterogeneity, not a target page budget.

Appendix B and Table 3, pp. 40–45; Appendix C and Tables 4–9, pp. 46–51; Appendix G and Table 10, pp. 55–71; compute p. 72 · Main paper

Clarity and writing conventions

The writing connects task choices to biological variation and uses split-specific, horizon-specific language rather than a universal best-model claim. Its extensive citations document source studies; the unusual reference/appendix length should not be mistaken for a typical conference manuscript requirement.

Task and evaluation framing pp. 4–9; references pp. 11–38; study provenance pp. 40–44 · Main paper

What to reuse—and what not to infer

Transferable lesson: For a heterogeneous benchmark, visualize coverage, define the prediction and split contract, report subgroup/horizon behavior, and make the original data sources and licenses independently inspectable.

Boundary: The public 72-page author version is not a verified submission or camera-ready page budget. The included wildlife studies do not constitute a random or exhaustive sample of all wildlife movement.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-10-02

Reading status

Writing-focused manuscript scan + selected visuals inspected

Main text

Writing-oriented scan across manuscript framing, section sequence, figure/table captions, selected argument/evaluation prose and conclusion; not exhaustive technical or proof verification. Main Sections 1–7, pp. 1–10. References span pp. 11–38 in the public 72-page version. Rendered the corpus/task overview p. 2 and grouped performance analysis p. 8. Source is the exact public author version listed below, not a verified accepted/camera-ready revision.

Appendix / supplement

Writing-oriented appendix/documentation scan: Appendices A–G, pp. 39–72: related context, study provenance and curation, environmental covariates, metrics, implementation, hyperparameters, per-study results and compute. Source code, data, model outputs and external video assets were not executed or independently validated.

Visually inspected PDF pages

2, 8

This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.

• Public author arXiv manuscript 2609.15780v1; not verified identical to the accepted or camera-ready OpenReview revision · 72 pages