Scholarlink

<p>Accurate climate data are essential for advancing ecological research and guiding effective conservation planning in mountainous regions. However, they remain scarce in arid mountain environments. This study addresses this gap by developing a high-resolution monthly climate dataset for Djebel Aïssa National Park (southwestern Algeria) through the downscaling of temperature and precipitation variables from WorldClim v2.1. The methodology involved four key steps. First, historical and contemporary climate data were used to refine elevation-dependent gradients by adjusting lapse rates. Second, these adjusted lapse rates were applied to extrapolate monthly climate variables using the ALOS PALSAR digital elevation model at a spatial resolution of 12.5&#xa0;m. Third, a two-phase hybrid downscaling approach was implemented: Generalized Additive Models (GAMs) were used to capture the primary orographic effects of elevation (physical component), while Random Forest (RF) models were trained on the residuals to model fine-scale spatial variability using terrain attributes such as slope, aspect, and ruggedness (stochastic component). Finally, the hybrid models were bias-corrected using quantile mapping: gamma-based adjustments for precipitation and empirical quantile mapping for temperature, both referenced to lapse rate–adjusted observations. The hybrid models demonstrated strong predictive performance, with R² values ranging from 0.978 to 0.994 for precipitation and exceeding 0.997 for temperature across all months. Bias correction significantly enhanced model reliability, particularly in high-altitude areas where climate variability was previously underestimated due to sparse meteorological records in WorldClim. The resulting bias-corrected dataset shows strong spatial coherence with lapse-rate–derived climate patterns, marking a substantial improvement over existing products. This work demonstrates the effectiveness of integrating statistical and machine learning frameworks for downscaling climate data in topographically complex and data-scarce regions. The resulting high-resolution dataset provides a robust climate baseline to support ecological modelling, biodiversity assessments, and conservation planning in Mediterranean arid mountain ecosystems.</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

Modelling high-resolution climate datasets for Djebel Aïssa National park (Southwestern Algeria) by downscaling worldclim data using GAM and RF

  • 期刊论文

  • 发表日期：

    2025年9月30日

查看全文
  • Imene Habibi,
  • Hammadi Achour,
  • Abdelkrim Benaradj,
  • Farid Bounaceur

摘要

Accurate climate data are essential for advancing ecological research and guiding effective conservation planning in mountainous regions. However, they remain scarce in arid mountain environments. This study addresses this gap by developing a high-resolution monthly climate dataset for Djebel Aïssa National Park (southwestern Algeria) through the downscaling of temperature and precipitation variables from WorldClim v2.1. The methodology involved four key steps. First, historical and contemporary climate data were used to refine elevation-dependent gradients by adjusting lapse rates. Second, these adjusted lapse rates were applied to extrapolate monthly climate variables using the ALOS PALSAR digital elevation model at a spatial resolution of 12.5 m. Third, a two-phase hybrid downscaling approach was implemented: Generalized Additive Models (GAMs) were used to capture the primary orographic effects of elevation (physical component), while Random Forest (RF) models were trained on the residuals to model fine-scale spatial variability using terrain attributes such as slope, aspect, and ruggedness (stochastic component). Finally, the hybrid models were bias-corrected using quantile mapping: gamma-based adjustments for precipitation and empirical quantile mapping for temperature, both referenced to lapse rate–adjusted observations. The hybrid models demonstrated strong predictive performance, with R² values ranging from 0.978 to 0.994 for precipitation and exceeding 0.997 for temperature across all months. Bias correction significantly enhanced model reliability, particularly in high-altitude areas where climate variability was previously underestimated due to sparse meteorological records in WorldClim. The resulting bias-corrected dataset shows strong spatial coherence with lapse-rate–derived climate patterns, marking a substantial improvement over existing products. This work demonstrates the effectiveness of integrating statistical and machine learning frameworks for downscaling climate data in topographically complex and data-scarce regions. The resulting high-resolution dataset provides a robust climate baseline to support ecological modelling, biodiversity assessments, and conservation planning in Mediterranean arid mountain ecosystems.