2026-08-19 バッファロー大学（UB）
The images show waterlines on staff gauges. From left to right, 1) the initial blurred image, 2) filtering the image for uneven light, 3) binarizing the image to black and white, 4) scanning the image to detect the water line, and then transferring the water line to the original image.
＜関連情報＞
- https://www.buffalo.edu/news/releases/2026/08/ai-enhanced-crowd-hydrology.html
- https://www.mdpi.com/2306-5338/13/5/134
水文学における人工知能を活用した市民科学による水位写真の処理 Leveraging Artificial Intelligence in Hydrology to Process Citizen Science Photos of Water Levels
Abhinna Manandhar and Christopher S. Lowry
Hydrology Published: 14 May 2026
DOI:https://doi.org/10.3390/hydrology13050134
Abstract
Emerging Large Language Model capabilities create opportunities for applying AI reasoning across various domains with minimal technical complexity. Motivated by the development of citizen scientists submitting photos of water levels on staff gauges and the increasing need for hydrologic data in ungauged watersheds, this research develops an artificial intelligence approach to measuring stream stage across an existing citizen science monitoring network. To lower the barrier to entry for professional scientists, this research develops a methodology leveraging a Large Language Model (LLM) to extract water levels from images submitted by citizen scientists, and then follows a human-in-the-loop workflow for validating the final results, leaving space for correcting reasoning errors and hallucinations. Various techniques, such as labeling the input image, are also explored in this research to extract maximum accuracy from the LLM.
