Compact pretrained models balance speed and accuracy in flood water mapping
Accuracy and efficiency of compact foundation models for surface-water mapping during floods
Computer Vision and Pattern RecognitionMachine Learning
Summary
Mapping floodwater from satellite images is important but can be slow and requires a lot of data. The authors compared small pretrained AI models with traditional models to see which maps water better and faster. They found that while pretrained models were faster and saved computing power, they weren’t always more accurate than standard models. Both types made mistakes near water edges, but pretrained models offered a good trade-off between speed and accuracy. This suggests small pretrained models can help when quick results matter, though they don’t always outperform classic methods.
What this means in practice
- •For disaster response teams: Process satellite images quickly using compact pretrained models to identify flooded areas with reasonable accuracy during emergency operations.
- •For environmental monitoring groups: Deploy efficient models to monitor changing surface water extent over large areas with reduced computing resources.
Authors
Parastoo Farajpoor, Mohammadreza Narimani
Abstract
Pretrained Earth observation models may reduce the effort needed to map surface water, but their value depends on the alternative against which they are compared. We evaluated Prithvi-EO-2.0-Tiny using the 446 hand-labeled Sentinel-2 chips from 11 flood events in Sen1Floods11. A fine-tuned model, a frozen-encoder model, and an identical randomly initialized architecture were compared with a similarly sized U-Net, a larger Prithvi model, and spectral-index thresholds. Experiments varied the number of training chips, excluded events from model fitting, and measured computational cost and prediction reliability. With 252 training chips, U-Net achieved a water intersection-over-union (IoU) of 0.856 on the official test set, compared with 0.833 for fine-tuned Prithvi-Tiny. Pretraining improved the tiny transformer over random initialization by a reported 0.046 IoU with 25 training chips and 0.023 with 252 chips, but did not reverse the U-Net ranking. Across 11 event-exclusion evaluations, mean IoU was 0.768 for fine-tuned Prithvi-Tiny and 0.777 for U-Net; the paired interval included zero. Fine-tuned Prithvi-Tiny processed a 512x512-pixel chip in 16.1 ms on a GPU and 258 ms on a laptop CPU, compared with 30.6 and 665 ms for U-Net. Errors were concentrated near water boundaries, and high predictive entropy identified many, but not all, erroneous pixels. These results support a specific role for compact pretrained models: lower measured computational demand with competitive accuracy, rather than consistent superiority. Event-dependent performance, shared hyperparameter selection, and the distinction between observed water and newly inundated land delimit the conclusions.