Volume and uncertainty estimates of on-farm reservoirs using surface reflectance and LiDAR data

Abstract

Accurate estimates of water volumes are crucial for water management. This study applies an automated methodology to detect small-scale on-farm dams and develops a novel application of Bayesian inference to jointly simulate volumes and uncertainties. A linear relationship was assumed between flooded pixel area images derived from LiDAR datasets and water index rasters derived from Sentinel 2 images. Using a Markov chain Monte Carlo (MCMC) method led to accurate estimates of water elevations and reservoir volumes, with a systematic error of about 2.5% and 10% of the maximum capacity during the study period in Keepit dam and Pamamaroo lake, respectively. Additionally, the method quantifies uncertainties of volumes. The presence of woody vegetation growing at the reservoir walls leads to a deterioration of estimates. This methodology may be used as an auditing tool in water governance schemes or to gain knowledge on water losses at the field scale.