Online multi-resolution image fusion: an application for water mapping
- Haoqing Li,
- Bhavya Duvvuri,
- Ricardo Borsoi,
- Tales Imbiriba,
- Edward Beighley,
- Deniz Erdogmus,
- Pau Closas
Abstract
Fresh water is a vital resource for all aspects of agricultural and
industrial production. Monitoring the variation of surface water level
allows resource managers to detect perturbations, predict long-term
trends in water availability, and set consumption guidelines
accordingly. Satellite imaging data has been increasingly used to map
surface water at global scales. Improving the performance of current
water mapping strategies requires high-resolution image data with low
revisit times. However, imaging devices on board of existing satellites
face a trade-off between their spatial resolution and revisit period,
which limits the applicability of those methods. In this work, a
multimodal image fusion methodology is developed for water mapping. By
combining data from multiple instruments, high-resolution image
sequences with low revisit times are generated, leading to improved
water mapping results. The proposed methodology was based on Bayesian
filtering and smoothing theory, and is able to combine each observed
images recursively for a reduced computation complexity. Experiments
with real data acquired by Sentinel and Landsat instruments showed that
the proposed strategy can lead to significant improvements in water
mapping results with compared to competing methodologies.