Work

LakeRINO Concentrates Short-Horizon Forecast Accuracy on Low- Water Anomalies from Irregular Satellite Altimetry across China

Year: 2026

Type: preprint

Abstract: Nationwide lake-level monitoring requires forecasts that preserve exact observation timing and add their greatest value during anomalously low states. We analyzed 58,789 observed monthly water levels from 988 satellite-altimetry measurement objects across China during 2002–2023 and established a physical-waterbody-aware evaluation using only real observations. LakeRINO first learns a Continuous-Time Neural Operator (CTNO) prior under four structured missingness patterns and then combines that held-fixed representation with a 120-month history, historical statistics, 88 static lake and catchment attributes, and a strong MAE-TimeXer reference forecast for 1-, 3-, 6-, and 12-month prediction. On 45 held-out physical waterbodies, CTNO reduced structured-reconstruction macro MAE from 0.5298 to 0.5211, a paired gain of 0.0087 (95% CI, 0.0009–0.0167). On the 2021–2023 temporal test, LakeRINO achieved macro-waterbody MAEs of 0.4639, 0.5517, 0.6109, and 0.6676 across the four horizons and delivered a supported three-month gain over the loss-aware horizon comparator. The defining application result was stronger: at one and three months, comparator-minus-LakeRINO absolute-error gain was largest among targets that subsequently realized the lowest seasonal anomalies. At six months, LakeRINO also reduced macro-waterbody Brier error by 0.00130 relative to TimeXer-MAE (95% CI, 0.00036–0.00227). National burden maps, reliability profiles, and risk-budget curves show how continuous reconstruction, short-horizon low-anomaly accuracy, and seasonal probability information jointly support lake-level monitoring. LakeRINO therefore converts irregular altimetry into queryable lake-level information and concentrates its clearest short-horizon forecast advantage on realized low-water anomalies.

Source: Research Square

DOI: 10.21203/rs.3.rs-10708680/v1

Authors: Dandan Guo, Donghui Zhang, Chi Luo, Xianxiu Zhou, Dianming Xu

Institutions: Xihua University, China Academy of Space Technology, Henan Province Water Conservancy Survey and Design Research

Language: English

Topic: Hydrological Forecasting Using AI

Subfield: Environmental Engineering

Field: Environmental Science

Domain: Physical Sciences

SDGs: Clean water and sanitation, Climate action

Open Access status: green

Funder: Natural Science Foundation of Inner Mongolia