Scholarlink

<p>This study presents a novel integrated framework combining geostatistical interpolation, terrain optimization, and XGBoost machine learning to improve spatial precipitation estimation in topographically complex, data-scarce regions. Through a systematic evaluation, optimal settings were selected for the resampling technique, resolution, interpolation, terrain indices’ calculation parameters, and ML. The best resampling technique was found to be cubic convolution with a 100&#xa0;m DEM resolution, achieving a 92% computational efficiency gain while preserving topographic slope by 62.8%. Moreover, a cross-validation using Leave-One-Out Cross-Validation identified Empirical Bayesian Kriging (EBK) as superior (R² = 0.80, RMSE = 82.49&#xa0;mm, MAE = 67.82&#xa0;mm, PBIAS = 0.81%), outperforming Ordinary Kriging -OK- (RMSE = 87.06&#xa0;mm) and Kernel Interpolation with Barriers (KIB) (RMSE = 94.11&#xa0;mm). Terrain index optimisation across 17 window sizes established an optimal Vector Ruggedness Measure window size of 153 pixels, reducing solo prediction error by 27.6% compared to a window size of 3 pixels (RMSE improved from 156.61&#xa0;mm to 113.32&#xa0;mm; R² increased from 0.35 to 0.66). XGBoost-based prediction revealed that the multivariate model incorporating temperature, TRI, and VRM-153 achieved superior performance (R² = 0.87, RMSE = 70.9&#xa0;mm, MAE = 51.36&#xa0;mm, PBIAS = 0.67%), substantially outperforming univariate approaches using temperature alone as the best univariate (R² = 0.72, RMSE = 103.15&#xa0;mm). Kernel density estimation confirmed optimal distributional correspondence for the three-parameter configuration. This integrated framework establishes a robust, transferable methodology for high-resolution precipitation mapping in data-limited mountainous regions, with direct applications to hydrological modeling and climate adaptation in semi-arid environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced spatial precipitation maps by integrating XGBoost machine learning, terrain indices, and optimal interpolation

  • 期刊论文

  • 发表日期：

    2026年4月10日

查看全文
  • Peshawa Bakhtyar Salih Ahmed,
  • Nawbahar Faraj Mustafa,
  • Miran Hikmat Mohammed Baban

摘要

This study presents a novel integrated framework combining geostatistical interpolation, terrain optimization, and XGBoost machine learning to improve spatial precipitation estimation in topographically complex, data-scarce regions. Through a systematic evaluation, optimal settings were selected for the resampling technique, resolution, interpolation, terrain indices’ calculation parameters, and ML. The best resampling technique was found to be cubic convolution with a 100 m DEM resolution, achieving a 92% computational efficiency gain while preserving topographic slope by 62.8%. Moreover, a cross-validation using Leave-One-Out Cross-Validation identified Empirical Bayesian Kriging (EBK) as superior (R² = 0.80, RMSE = 82.49 mm, MAE = 67.82 mm, PBIAS = 0.81%), outperforming Ordinary Kriging -OK- (RMSE = 87.06 mm) and Kernel Interpolation with Barriers (KIB) (RMSE = 94.11 mm). Terrain index optimisation across 17 window sizes established an optimal Vector Ruggedness Measure window size of 153 pixels, reducing solo prediction error by 27.6% compared to a window size of 3 pixels (RMSE improved from 156.61 mm to 113.32 mm; R² increased from 0.35 to 0.66). XGBoost-based prediction revealed that the multivariate model incorporating temperature, TRI, and VRM-153 achieved superior performance (R² = 0.87, RMSE = 70.9 mm, MAE = 51.36 mm, PBIAS = 0.67%), substantially outperforming univariate approaches using temperature alone as the best univariate (R² = 0.72, RMSE = 103.15 mm). Kernel density estimation confirmed optimal distributional correspondence for the three-parameter configuration. This integrated framework establishes a robust, transferable methodology for high-resolution precipitation mapping in data-limited mountainous regions, with direct applications to hydrological modeling and climate adaptation in semi-arid environments.