Scenario Generalization and Sector Transfer in Real Surface Irrigation Surrogate Modelling with Tabular Models

Key points

• Surface irrigation planning requires screening many operating conditions and spatial configurations, while high-fidelity shallow-water-equation (SWE) simulations are expensive.1
• To assess deployment feasibility, we evaluate gradient-boosted trees, neural tabular models, AutoML, and tabular foundation models (TFMs) under two regimes: same-sector generalization, where a model predicts unseen scenarios after observing a limited set of configurations in the same sector, and sector-out transfer, where knowledge learned from source sectors is transferred to an unseen sector.1
• The main findings are twofold: a) under the evaluated fixed 8/32 same-sector split, surrogates achieve strong performance, especially with raw spatial coordinates; b) sector-out transfer is substantially harder because the target sector has different coordinates and geometry.1
• Sector-frame normalization of spatial coordinates improves transfer across similarly structured sectors, but does not solve cross-shape generalization.1

Sources (1)

• [1]Scenario Generalization and Sector Transfer in Real Surface Irrigation Surrogate Modelling with Tabular Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 08:30 AM

  We study tabular surrogate models for predicting maximum water height from drone-derived field morphology, terrain, hydrological, spatial, and scenario descriptors, where each row represents a spatial cell under one irrigation scenario.

  Surface irrigation planning requires screening many operating conditions and spatial configurations, while high-fidelity shallow-water-equation (SWE) simulations are expensive.

Extractive summary: sentences quoted from the sources.