PL EN
Evaluating reservoir-release information for short-term dissolved oxygen forecasting and risk-calibrated early warning in two regulated rivers
 
More details
Hide details
1
PVKK Institute of Technology, Anantapuramu, A.P, India
 
2
Department of Computer Applications PVKK Institute of Technology, Anantapuramu, India.
 
3
Department of Computer Science and Engineering PVKK Institute of Technology, Anantapuramu, India.
 
4
Department of Mathematics, Dayananda Sagar University, Bengaluru, Karnataka, India
 
5
Department Electronics and Communication Engineering, RMK Engineering College, Gummidipoondi, Tamil Nadu, India
 
 
Corresponding author
M.Mallikarjuna Rao   

PVKK Institute of Technology, Anantapuramu, A.P, India
 
 
 
KEYWORDS
TOPICS
ABSTRACT
Reservoir releases are among the few levers available before low dissolved oxygen (DO) develops downstream, but whether release information improves short-term DO warnings has rarely been tested. Two regulated rivers receiving treated wastewater were studied: the Tualatin below Henry Hagg Lake (Oregon) and the Chattahoochee below Buford Dam and Atlanta (Georgia). Daily and 15-min records for 1 January 2019 to 17 October 2026 came from USGS, Hagg Lake outflow from the Bureau of Reclamation Hydromet network, and weather reanalysis and archived forecasts from Open-Meteo, with every request logged and checksummed. The daily DO minimum 1–3 days ahead was forecast with persistence, climatology, ridge regression and gradient-boosted point and quantile models. Thresholds, travel times and input screening were fixed using pre-test data only. Online-adapted conformalised quantile regression calibrated the intervals; conformal risk control set the alarm margin for a 10% target miss rate. Over the 655 test days, release inputs did not improve forecasts at any lead time on either river (Diebold–Mariano p ≥ 0.079). With archived weather forecasts and today’s release replacing target-day reanalysis and the actual schedule, ridge regression reduced one-day mean absolute error relative to persistence by 5.9% (p = 0.066) on the Tualatin and 4.9% (p = 0.18) on the Chattahoochee; beyond one day, all learned models remained 11–14% better (p ≤ 0.003). The 90% intervals covered 90.5% and 89.7% of one-day observations. At the pre-specified 6.0 mg·L−1 threshold, the Tualatin alarm missed 6.0% of 134 low-DO days (95% CI 2.3–13.0%) at an 8.7% false-alarm rate. The Chattahoochee’s pre-specified 5.0 mg·L−1 threshold yielded only two low-DO test days, too few to evaluate the alarm; post-hoc 6.6 and 7.2 mg·L−1 thresholds gave 21.5% and 12.7% miss rates. A sustained threefold release changed the forecast warm-season minimum by at most 0.27 mg·L−1, and a 0.8 mg·L−1 sag during a wastewater overflow was not anticipated. Short-term DO risk on these reaches was governed by oxygen persistence, temperature and weather rather than release volume, and the stated miss rate held only where calibration and test conditions were similar.
Journals System - logo
Scroll to top