Performing research on the variation in lake water on the Qinghai–Tibet Plateau (QTP) can give the area’s ecological environmental preservation a scientific foundation. In this paper, we first created a high-precision dataset of lake water level variation every 10 days, from July 2002 to December 2022, using multi-source altimetry satellite SGDR data (Envisat RA-2, SARAL, Jason-1/2, and Sentinel-3A/3B SRAL), which integrated the methods of atmospheric path delay correction, waveform re-tracking, outlier detection, position reduction using a height difference model, and inter-satellite deviation adjustment. Then, using Landsat-5 Thematic Mapper, Landsat-7 Enhanced Thematic Mapper, and Landsat 8 Operational Land Imager data, an averaged area series of Lake Qinghai (LQ) from September to November, each year from 2002 to 2019, was produced. The functional connection between the water level and the area was determined by fitting the water level–area series data, and the lake area time series, of LQ. Using the high-precision lake water level series, the fitted lake surface area time series, and the water storage variation equation, the water storage variation time series of LQ was thus calculated every 10 days, from July 2002 to December 2022. When the hydrological gauge data from the Xiashe station and data from the worldwide inland lake water level database are used as references, the standard deviations of the LQ water level time series are 0.0676 m and 0.1201 m, respectively. The results show that the water storage of LQ increases by 11.022 × 10⁹ m³ from July 2002 to December 2022, with a growth rate of 5.3766 × 10⁸ m³/a. The growth rate from January 2005 to January 2015 is 4.4850 × 10⁸ m³/a, and from January 2015 to December 2022, the growth rate is 8.9206 × 10⁸ m³/a. Therefore, the increased rate of water storage in LQ over the last 8 years has been substantially higher than in the previous 10 years.

For the traditional 1-D nadir altimetry measurements, the spatial resolution of the sea surface height (SSH) measurements is restricted. However, the new wide-swath altimeters like Surface Water and Ocean Topography (SWOT) are anticipated to enhance spatial resolution and thereby improve the accuracy of derived marine gravity anomalies, particularly in high-frequency signals. The study investigates the efficacy of L2 KaRIn Beta data of SWOT in recovering high-frequency marine gravity signals. CryoSat-2 data was used for comparison. The first-order difference of shipborne gravity data along ship tracks was used as the reference to investigate the high-frequency accuracy of gravity derived from SWOT and CryoSat-2 at different point spacings, respectively. The research results show that compared to CryoSat-2 satellite data, when the point spacing is less than 2 km or greater than 6 km, the accuracy improvement is relatively small, less than 3.0%. However, within the 2–4 km range, the accuracy improvement ranges from 6.8% to 7.9%. When the point spacing is between 4 and 6 km, the accuracy improvement ranges from 5.0% to 6.7%. Therefore, when the point spacing is greater than 2 km and less than 6 km, the accuracy improvement is obvious. The results show that SWOT wide-range altimeter satellite has obviously improved the accuracy of traditional altimeter gravity inversion in high frequency (especially in the range of 2–6 km).

The quality of the SWOT-derived marine gravity field and its subsequent refinement are investigated in the Western Pacific (120–140°E, 0–30°N). Inverting 41 cycles of the SWOT L3_LR_SSH v2.0.1 product yields a gravity field with RMSEs of 2.476 and 4.321 mGal against independent JAMSTEC and NCEI shipborne gravity, respectively, and the accuracy gain converges as the number of SWOT altimetry repeat cycles increases, showing diminishing marginal returns. To further refine this field, three tree-based ensemble learning methods—Random Forest (RF), histogram-based gradient boosting (HGB) and LightGBM (LGBM)—are trained on 67 multi-source features with the SWOT-minus-shipborne residual as the target, and validated under two spatially isolated strategies: a 2° × 2° spatial block hold-out with a 10 km buffer and a cruise-wise hold-out over seven complete cruises, in which training samples within 10 km of any test point are removed. All three methods reduce the error. On the block hold-out, LGBM performs best (RMSE 2.398 mGal, a 3.6% improvement over the 2.488 mGal baseline), followed by HGB (2.402 mGal, 3.5%) and RF (2.422 mGal, 2.7%); on the cruise hold-out the gains are markedly larger, with HGB best (RMSE 1.703 mGal, a 8.6% improvement over the 1.863 mGal baseline), then LGBM (1.714 mGal, 8.0%) and RF (1.794 mGal, 3.7%). A consistent reversal is found between the two assessment levels: RF leads on the in-training out-of-fold RMSE but is weakest on unseen cruises, indicating that RF is competitive within the training distribution yet weaker than boosting methods when extrapolating to unsurveyed areas. Power spectral density analysis shows that the improvements concentrate at wavelengths of 8–45 km, whereas wavenumber below 0.0221 km⁻¹ are slightly degraded.