Page 214 - 《高原气象》2026年第2期

高 原 气 象 45 卷
514

Research on Daily Maximum and Minimum Temperature Forecasts by

Integrating Terrain Features and Neural Networks

6
5
LU Shu 1， 2， 3， 4 ， XU Lin 1， 2， 3， 4 ， GU Xue ， ZHOU Yue ， DAI Zejun 1， 2， 3， 4 ， TAO Yaqing 1， 2， 3， 4
（1. Hunan Key Laboratory of Meteorological Disaster Prevention and Reduction， Changsha 410118， Hunan， China；
2. Hunan Meteorological Observatory， Changsha 410118， Hunan， China；
3. Dongting Lake National Climatic Observatory， China Meteorological Administration， Yueyang 414000， Hunan， China；
4. Key Laboratory of High Impact Weather（special） China Meteorological Administration， Changsha 410118， Hunan， China；
5. Xiangxi Meteorological Bureau， JiShou 416000， Hunan， China；
6. China Meteorological Administration Basin Heavy Rainfall Key Laboratory/Hubei Key Laboratory for Heavy Rain Monitoring and
Warning Research， Institute of Heavy Rain， China Meteorological Administration， Wuhan 430205， Hubei， China）

Abstract： Accurate forecasting of daily maximum （T ） and minimum （T ） temperatures is essential for mete‐
max
min
orological operational， as enhanced precision safeguards socioeconomic stability in agriculture， transportation
and public health. To address pronounced systematic biases in numerical weather prediction models over complex
terrain-particularly regions with heterogeneous underlying surfaces-this study develops an advanced deep learn‐
ing framework integrating geographic feature clustering. The study focuses on China’s Hunan Province， charac‐
terized by a distinctive concave-shaped， three-tiered topography encompassing mountains， hills， basins， and
plains. A baseline convolutional neural network （CNN） was constructed using ECMWF forecast fields， high-res‐
olution CLDAS （China Land Data Assimilation System） reanalysis data， and multi-dimensional geographic vari‐
ables （elevation， slope， aspect， terrain roughness index）. Three comparative experiments rigorously evaluated
terrain-processing efficacy： Method 1 （K-means clustered geographic variables delineating topographic re‐
gimes）， Method 2 （Conventionally standardized non-clustered geographic variables）， and Method 3 （Geograph‐
ic variable exclusion as terrain-agnostic control）. Validation confirmed Method 1's superiority， achieving 24-
hour mean absolute error （MAE） reductions of 4. 7% for T and 9. 4% for T relative to Method 3， while im‐
min
max
proving forecast skill by 2. 5% （T ） and 1. 4% （T ） compared to Method 2. These statistically significant
min
max
gains validate the benefits of explicit terrain feature clustering. Building on this approach， the CNN-Terrain Cor‐
rection （CNN-TC） model was developed for 72-hour T /T predictions. CNN-TC framework delivers transfor‐
min
max
mative improvements. Versus ECMWF outputs， T MAE decreased by 23. 5%~37. 3% and T MAE by
max
min
20. 8%~26. 9% across 24~72 hour lead times. Compared to operational SCMOC products， errors reduced
18. 7%~27. 6% （T ） and 26. 8%~32. 3% （T ）. Critically， the model compresses spatial error dispersion， nar‐
max
min
rowing 24-hour MAE ranges from 1. 2~5. 8 ℃/0. 8~5. 9 ℃ （ECMWF） to a stable 0. 9~1. 7 ℃/0. 8~1. 7 ℃ for
T /T -demonstrating breakthrough operational stability. Monthly verification confirmed persistent superiority，
max
min
with T MAE reductions of 5. 6%~59. 1% and T improvements of 6. 3% to 47. 8% relative to ECMWF. Dur‐
min
max
ing the November 2022 cold-air outbreak， the model captured intricate spatiotemporal cooling patterns， outper‐
forming ECMWF and SCMOC with over 30% error reduction， underscoring its capability in extreme weather.
This study verifies that deep learning combined with terrain clustering effectively mitigates systematic biases
over complex terrain. The CNN-TC framework establishes a robust solution for refined meteorological services
in mountainous regions. Cross-regional implementation requires localized hyperparameter optimization and clus‐
ter retraining to address spatial heterogeneity in climate regimes and surface properties.
Key words： daily maximum and minimum temperature； geographic variables； Convolutional Neural Network；
K-means algorithm