A downscaling method for gravity satellite derived groundwater storage changes based on a feature-weighted CatBoost model

doi:10.11947/j.AGCS.2026.20250359

Abstract

Abstract:

The GRACE and GRACE-FO gravity satellite missions have provided a new approach for monitoring global groundwater storage (GWS) anomaly. However, their coarse spatial resolution limits their effectiveness in supporting fine-scale regional water resource management. Focusing on the issue of GWS monitoring in the North China Plain, this study proposes a downscaling method that integrates multi-source satellite data, variable weighting, and machine learning modeling to enhance the spatial resolution of GRACE (GRACE-FO) data. The Bayesian Three-Cornered Hat (BTCH) method is first applied to weight and fuse six GRACE-FO solutions, producing a robust regional GWS baseline dataset. Variable Importance in Projection (VIP) scores are then calculated using Partial Least Squares Regression (PLSR) to establish a feature-weighting mechanism that enhances the model's responsiveness to key variables. Finally, a CatBoost model with ordered target encoding and ordered boosting is employed to downscale the GWS data from 1° to 0.25° resolution. The results were validated through comparison with in-situ well water level measurements. Compared with the original GWS data before downscaling, the downscaled GWS data exhibit significantly higher correlation with well water level observations and reveal much richer spatial details. Compared with traditional Random Forest and XGBoost methods, the proposed approach demonstrates superior performance in spatial trend consistency and physical reliability, significantly enhancing the robustness and practical applicability of GRACE (GRACE-FO) GWS inversion results.

Key words: GRACE (GRACE-FO), groundwater storage anomaly, data fusion, feature weighting, downscaling

CLC Number:

P228

Wentao HOU, Yun XIAO, Jie CAO, Yukang WANG, Chunting CAO, Han WANG. A downscaling method for gravity satellite derived groundwater storage changes based on a feature-weighted CatBoost model[J]. Acta Geodaetica et Cartographica Sinica, 2026, 55(3): 490-501.

Figures/Tables 14

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Tab. 1

Fig. 6

Fig. 7

Tab. 2

Fig. 8

Fig. 9

Fig. 10

Tab. 3

Tab. 4

References 26

[1]	冉将军, 闫政文, 吴云龙, 等. 下一代重力卫星任务研究概述与未来展望[J]. 武汉大学学报(信息科学版), 2023, 48(6): 841-857.
	RAN Jiangjun, YAN Zhengwen, WU Yunlong, et al. Research status and future perspectives in next generation gravity mission[J]. Geomatics and Information Science of Wuhan University, 2023, 48(6): 841-857.
[2]	刘东旭, 胡立堂, 孙建冲, 等. 基于GRACE/GRACE-FO数据降尺度方法反演库尔勒东区地下水储量变化[J]. 测绘学报, 2024, 53(7): 1265-1277. DOI: . doi: 10.11947/j.AGCS.2024.20230354
	LIU Dongxu, HU Litang, SUN Jianchong, et al. Retrieval of groundwater storage anomalies in eastern region of Korla by downscaling GRACE/GRACE-FO data[J]. Acta Geodaetica et Cartographica Sinica, 2024, 53(7): 1265-1277. DOI: . doi: 10.11947/j.AGCS.2024.20230354
[3]	FOWLER H J, BLENKINSOP S, TEBALDI C. Linking climate change modelling to impacts studies: recent advances in downscaling techniques for hydrological modelling[J]. International Journal of Climatology, 2007, 27(12): 1547-1578.
[4]	WILBY R L, WIGLEY T M L, CONWAY D, et al. Statistical downscaling of general circulation model output: a comparison of methods[J]. Water Resources Research, 1998, 34(11): 2995-3008.
[5]	黄上富, 段光耀, 和继军, 等. 基于GWR模型的海河流域地下水储量降尺度研究[J]. 水电能源科学, 2023, 41(11): 39-42, 30.
	HUANG Shangfu, DUAN Guangyao, HE Jijun, et al. Downscaling of groundwater storage change in Haihe River Basin based on GWR model[J]. Water Resources and Power, 2023, 41(11): 39-42, 30.
[6]	褚江东, 粟晓玲, 张特, 等. 基于随机森林模型的GRACE数据3种空间降尺度对比[J]. 湖泊科学, 2024, 36(3): 951-962.
	CHU Jiangdong, SU Xiaoling, ZHANG Te, et al. Comparison of three spatial downscaling concepts of GRACE data using random forest model[J]. Journal of Lake Sciences, 2024, 36(3): 951-962.
[7]	MIRO M E, FAMIGLIETTI J S. Downscaling GRACE remote sensing datasets to high-resolution groundwater storage change maps of California'scentral valley[J]. Remote Sensing, 2018, 10(1): 143.
[8]	SAHOUR H, SULTAN M, VAZIFEDAN M, et al. Statistical applications to downscale GRACE-derived terrestrial water storage data and to fill temporal gaps[J]. Remote Sensing, 2020, 12(3): 533.
[9]	SEYOUM W M, KWON D, MILEWSKI A M. Downscaling GRACE TWSA data into high-resolution groundwater level anomaly using machine learning-based models in a glacial aquifer system[J]. Remote Sensing, 2019, 11(7): 824.
[10]	曹杰, 肖云, 龙笛, 等. 联合重力卫星和水井资料监测华北平原地下水储量变化[J]. 武汉大学学报(信息科学版), 2024, 49(5): 805-818.
	CAO Jie, XIAO Yun, LONG Di, et al. Combined gravity satellite and water well information to monitor groundwater storage changes in the North China Plain[J]. Geomatics and Information Science of Wuhan University, 2024, 49(5): 805-818.
[11]	RODELL M, HOUSER P R, JAMBOR U, et al. The global land data assimilation system[J]. Bulletin of the American Meteorological Society, 2004, 85(3): 381-394.
[12]	WAN Z, HOOK S, HULLEY G. MOD11C3 MODIS/Terra land surface temperature/emissivity monthly L3 global 0.05 deg CMG V006.[EB/OL]. [2025-4-15]. https://lpdaac.usgs.gov/products/mod11c3v006/.
[13]	费宇红, 苗晋祥, 张兆吉, 等. 华北平原地下水降落漏斗演变及主导因素分析[J]. 资源科学, 2009, 31(3): 394-399.
	FEI Yuhong, MIAO Jinxiang, ZHANG Zhaoji, et al. Analysis on evolution of groundwater depression cones and its leading factors in North China Plain[J]. Resources Science, 2009, 31(3): 394-399.
[14]	朱菊艳, 郭海朋, 李文鹏, 等. 华北平原地面沉降与深层地下水开采关系[J]. 南水北调与水利科技, 2014, 12(3): 165-169.
	ZHU Juyan, GUO Haipeng, LI Wenpeng, et al. Relationship between land subsidence and deep groundwater yield in the North China Plain[J]. South-to-North Water Transfers and Water Science & Technology, 2014, 12(3): 165-169.
[15]	LOOMIS B D, RACHLIN K E, LUTHCKE S B. Improved earth oblateness rate reveals increased ice sheet losses and mass-driven sea level rise[J]. Geophysical Research Letters, 2019, 46(12): 6910-6917.
[16]	ARGUS D F, PELTIER W R, DRUMMOND R, et al. The Antarctica component of postglacial rebound model ICE-6G_C (VM5a) based on GPS positioning, exposure age dating of ice thicknesses, and relative sea level histories[J]. Geophysical Journal International, 2014, 198(1): 537-563.
[17]	WAHR J, MOLENAAR M, BRYAN F. Time variability of the Earth's gravity field: hydrological and oceanic effects and their possible detection using GRACE[J]. Journal of Geophysical Research: Solid Earth, 1998, 103(B12): 30205-30229.
[18]	CHAMBERS D P. Evaluation of new GRACE time-variable gravity data over the ocean[J]. Geophysical Research Letters, 2006, 33(17): 2006GL027296.
[19]	李婉秋, 王伟, 章传银, 等. 利用GRACE卫星重力数据监测关中地区地下水储量变化[J]. 地球物理学报, 2018, 61(6): 2237-2245.
	LI Wanqiu, WANG Wei, ZHANG Chuanyin, et al. Monitoring groundwater storage variations in the Guanzhong area using GRACE satellite gravity data[J]. Chinese Journal of Geophysics, 2018, 61(6): 2237-2245.
[20]	CAO Yanping, NAN Zhuotong, CHENG Guodong. GRACE gravity satellite observations of terrestrial water storage changes for drought characterization in the arid land of northwestern China[J]. Remote Sensing, 2015, 7(1): 1021-1047.
[21]	黄飞龙, 谷延超, 何祖建, 等. 贝叶斯三角帽法的GRACE/GRACE-FO组合模型及其不确定性[J]. 大地测量与地球动力学, 2024, 44(4): 417-422.
	HUANG Feilong, GU Yanchao, HE Zujian, et al. GRACE/GRACE-FO combined model and its uncertainty based on Bayesian three-cornered hat method[J]. Journal of Geodesy and Geodynamics, 2024, 44(4): 417-422.
[22]	姚朝龙, 李琼, 罗志才, 等. 利用广义三角帽方法评估GRACE反演中国大陆地区水储量变化的不确定性[J]. 地球物理学报, 2019, 62(3): 883-897.
	YAO Chaolong, LI Qiong, LUO Zhicai, et al. Uncertainties in GRACE-derived terrestrial water storage changes over China's mainland based on a generalized three-cornered hat method[J]. Chinese Journal of Geophysics, 2019, 62(3): 883-897.
[23]	GAO Shengjun, HAO Weifeng, FAN Yi, et al. A multi-source GRACE fusion solution via uncertainty quantification of GRACE-derived terrestrial water storage (TWS) change[J]. Journal of Geophysical Research: Solid Earth, 2023, 128(11): e2023JB026908.
[24]	MA Mingde, MA Xuejuan, XIE Yingzhong, et al. Analysis the relationship between ecological footprint (EF) of Ningxia and influencing factors: partial least-squares regression (PLS)[J]. Acta Ecologica Sinica, 2014, 34(3): 682-689.
[25]	刘梦然. 基于机器学习的黄河流域地下水储量降尺度研究[D]. 开封: 河南大学, 2024.
	LIU Mengran. Downscaling of groundwater storage in yellow river basin based on machine learning[D]. Kaifeng: Henan University, 2024.
[26]	PROKHORENKOVA L, GUSEV G, VOROBEV A, et al. CatBoost: unbiased boosting with categorical features[C]//Proceedings of 2018 Advances in Neural Information Processing Systems 31. New York: Curran Associates, Inc., 2018: 6638-6648.

参数	含义	取值
task type	训练的器件GPU	—
loss function	损失函数RMSE	—
iterations	迭代次数	450
depth	树身	4
learning_rate	学习率	0.01
l2_leaf_reg	L2正则系数	0.05

数据	2018-06—2020-12	2021-01—2022-12
降尺度前	－4.5±1.0	8.1±1.2
降尺度后	－4.4±1.0	7.9±1.2

数据	负相关（－∞，0]	正相关范围
数据	负相关（－∞，0]	（0，0.2]	（0.2，0.6]	（0.6，1.0]
降尺度前	15.03%	10.40%	34.69%	39.88%
Random Forest	15.02%	6.36%	36.42%	42.20%
XGBoost	13.87%	5.20%	35.84%	45.09%
PLSR-CatBoost	12.14%	5.78%	24.28%	57.80%

地区	降尺度前	降尺度后
保定市	0.23	0.45
沧州市	0.49	0.69
邯郸市	0.64	0.65
衡水市	0.45	0.50
廊坊市	－0.04	0.19
秦皇岛市	0.33	0.51
石家庄市	0.09	0.20
唐山市	0.49	0.61
邢台市	0.61	0.62
雄安新区	－0.15	0.07