Machine Learning-Based Data Fusion of Laser Ceilometer and ERA5 for Cloud Base Height Characteristics in Tianshan Mountains

Cloud base height (CBH) is a critical parameter in surface radiation and aviation meteorology research. Accurate CBH data with high spatiotemporal resolution is essential for studying cloud radiative effects and improving the accuracy of numerical weather prediction. This study utilizes CBH observations from laser ceilometers at Xinyuan and Yining stations in the Ili River Valley of the western Tianshan Mountains during 2022-2023 to evaluate the ERA5 reanalysis CBH data. Additionally, the Random Forest method is employed to correct the ERA5 dataset. Based on the refined data, the spatial-temporal distribution and trend characteristics of CBH over the Tianshan Mountains from 1960 to 2020 are investigated.The results indicated that the following. (1) The correlation coefficients between laser ceilometer observations and ERA5 CBH are 0.547 at Xinyuan Station and 0.658 at Yining Station. A systematic bias exists between the two datasets, primarily ranging from??2500 to 1500 m. (2) Both datasets consistently demonstrate a unimodal diurnal variation in CBH characterized by lower daytime values and higher nighttime values, along with distinct seasonal patterns showing elevated CBH during summer (June~August) and reduced values in March and December. Comparative analysis reveals systematic biases in ERA5 data, with consistent overestimation of low-level clouds and underestimation of mid-level clouds relative to ground-based ceilometer observations. Vertical distribution analysis further identifies significant discrepancies: ERA5 exhibits a 0.02~6.9% higher frequency of CBH occurrences within the 0~1000 m altitude range compared to ceilometer measurements, while the opposite pattern emerges at higher elevations (2000~4500 m) where ceilometer observations show 0.06~1.6% greater frequency than ERA5 estimates. (3) The correlation coefficient between laser ceilometer observations and corrected ERA5 CBH significantly improved from 0.609 to 0.754. Seasonal analysis reveals that CBH displays a strong negative correlation with elevation during spring, summer, and autumn, progressively decreasing with increasing altitude, while this altitude dependence weakens in winter. Spatially, high CBH values in winter are predominantly distributed along the periphery of the Tarim Basin, with low-value areas concentrated in the southwestern Junggar Basin and high-altitude regions of the Tianshan Mountains above 3500 m. For low clouds (excluding winter), high-frequency zones are mainly located in the Tianshan Mountains above 1500 m, whereas mid-level clouds predominantly occur below 2500 m. Winter exhibits distinct patterns: low clouds show enhanced frequencies in both the southwestern Junggar Basin and Tianshan Mountains above 1500 m, while mid-level clouds maintain widespread high frequencies across all regions except the southwestern Junggar Basin. High clouds remain relatively rare throughout the year (approximately 150~750h·a-1), primarily appearing below 1500 m altitude. (4) The EOF modes exhibit pronounced interannual and decadal variability, with the first mode showing an out-of-phase spatial pattern between regions below and above 2,500 m elevation, while the second mode demonstrates an opposite variation between the southwestern Junggar Basin and the combined Tianshan Mountains-Tarim Basin periphery. Over the past six decades, the central Tianshan Mountains and southwestern Junggar Basin have experienced predominantly significant increases in CBH, whereas other regions showed no substantial changes. Notably, at elevations exceeding 2,500 m in the Tianshan Mountains, the occurrence frequency of low clouds has decreased significantly, mid-level clouds have increased significantly, and high clouds have remained essentially stable.