Data and Model Checkpoints from: ExoPrompt: Transformer-based greenhouse climate forecasting with structured conditioning and physics-based simulation

Data and model checkpoints accompanying the paper:<br>Soykan et al. (2026). ExoPrompt: Transformer-based greenhouse climate forecasting with structured conditioning and physics-based simulation. Computers and Electronics in Agriculture, 246, 111673. https://doi.org/10.1016/j.compag.2026.111673<br>--- Data ---<br>new_world_sim.zipSynthetic GreenLight simulation outputs used for pretraining. Covers 9 parameter scenarios across 15 geographic locations and 2 lighting modes (LED/HPS). Each runincludes paired .csv time-series and .json exogenous parameter files.<br>gt.zipGround-truth climate measurements from a tomato greenhouse trial (Bleiswijk, 2009-2010) under LED and HPS lighting, reorganised into CSV format for use withthe ExoPrompt training pipeline.<br>c_leakage_gt_led_conditions_csv.zipSimulation sweeps varying a single exogenous parameter (cLeakage) with aligned ground-truth subsets, used for the controlled robustness experiment in Section 3.3.<br>--- Model Checkpoints ---<br>ckpts_pretrained_main.zipPretrained ExoPrompt and Vanilla backbones evaluated zero-shot on ground truth (Table 4).<br>ckpts_200k_main.zipFine-tuned checkpoints for mixed, LED-only, and HPS-only splits (Figure 5).<br>ckpts_cleakage.zipPretrained and fine-tuned checkpoints for the controlled cLeakage study (Section 3.3-3.4).