Scholarlink

<p>This study examines the effects of climate change on maize (Zea mays L.) cultivation in Turkey’s Bartın and Zonguldak provinces. The objective was to assess current and future land suitability for maize under changing climatic conditions to support adaptive agricultural planning. We developed an integrated approach combining Coupled Model Intercomparison Project Phase 6 (CMIP6) climate model data with a fuzzy logic system to incorporate expert knowledge and manage environmental uncertainties. Environmental variables included climate parameters, soil properties, and topographic factors. Two machine learning (ML) algorithms, Random Forest and LightGBM, were employed to analyze land suitability for three periods: current (1981–2023), mid-century (2040–2070), and late-century (2070–2100). Current climate analysis reveals that approximately 50% of the study region is moderately suitable for maize cultivation, with less than 1% classified as highly suitable. Under the CMIP6 SSP5-8.5 scenario, projections indicate marginally suitable areas will expand by mid-century due to rising temperatures and altered rainfall patterns. By late-century, a dramatic shift is expected, with approximately 71% of the area becoming highly suitable, demonstrating significant changes in cultivation potential. Random Forest outperformed LightGBM, achieving 97.4% accuracy compared to 82.9%, with superior precision, recall, and F1-scores across all suitability classes. This research provides critical insights for developing adaptive agricultural strategies in response to climate change. The integration of ML techniques with high-resolution climate models offers robust predictions essential for policymakers and land managers to optimize agricultural productivity and ensure food security.&#xa0;</p> Graphical Abstract <p></p> <p>Based on the graphical snapshot, the map foregrounds Bartın and Zonguldak (Türkiye) with 260 geo-referenced top-soil samples and 30-arc-second CHELSA-CMIP6 temperature-and-precipitation rasters for the baseline (1980–2023) and future SSP5-8.5 periods (2040–2070, 2070–2100). A Mamdani fuzzy-logic workflow converts nine edaphic and climatic layers into FAO land-suitability scores; Recursive Feature Elimination selects the least-redundant covariate set, and ten-fold grid search tunes hyper-parameters for downstream classifiers. Ensemble tree learners—Random Forest (RF) and LightGBM—are trained on the fuzzy-derived dataset. RF ranks growing-season temperature (59%), precipitation (24%) and slope (18%) as the three most influential predictors and attains 97.4% overall accuracy, outperforming LightGBM (82.9%). Suitability maps reveal a climate-driven transition from a baseline mosaic of 50% moderately suitable (S2) + 50% highly suitable (S1) land to a late-century pattern dominated by 71% S1 and 14% S2. Concurrently, mean growing-season temperature rises from 33&#xa0;°C to 38.4&#xa0;°C. Integrating fuzzy logic with high-resolution CMIP6 projections and RF classification shows that warming under SSP5-8.5 is likely to expand prime maize land in northern Black-Sea provinces, but temperature becomes the critical limiting factor—highlighting an urgent need for heat-tolerant cultivars and adaptive agronomic planning.</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

Machine Learning-Based Smart Prediction of Future Suitability of Maize in Northern Regions of Turkey Under CMIP6 Climate Scenarios

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • Miraç Kılıç,
  • Murat Birol,
  • Hikmet Günal,
  • Betül Bayraklı,
  • Orhan Mete Kılıç,
  • Aqil Tariq

摘要

This study examines the effects of climate change on maize (Zea mays L.) cultivation in Turkey’s Bartın and Zonguldak provinces. The objective was to assess current and future land suitability for maize under changing climatic conditions to support adaptive agricultural planning. We developed an integrated approach combining Coupled Model Intercomparison Project Phase 6 (CMIP6) climate model data with a fuzzy logic system to incorporate expert knowledge and manage environmental uncertainties. Environmental variables included climate parameters, soil properties, and topographic factors. Two machine learning (ML) algorithms, Random Forest and LightGBM, were employed to analyze land suitability for three periods: current (1981–2023), mid-century (2040–2070), and late-century (2070–2100). Current climate analysis reveals that approximately 50% of the study region is moderately suitable for maize cultivation, with less than 1% classified as highly suitable. Under the CMIP6 SSP5-8.5 scenario, projections indicate marginally suitable areas will expand by mid-century due to rising temperatures and altered rainfall patterns. By late-century, a dramatic shift is expected, with approximately 71% of the area becoming highly suitable, demonstrating significant changes in cultivation potential. Random Forest outperformed LightGBM, achieving 97.4% accuracy compared to 82.9%, with superior precision, recall, and F1-scores across all suitability classes. This research provides critical insights for developing adaptive agricultural strategies in response to climate change. The integration of ML techniques with high-resolution climate models offers robust predictions essential for policymakers and land managers to optimize agricultural productivity and ensure food security. 

Graphical Abstract

Based on the graphical snapshot, the map foregrounds Bartın and Zonguldak (Türkiye) with 260 geo-referenced top-soil samples and 30-arc-second CHELSA-CMIP6 temperature-and-precipitation rasters for the baseline (1980–2023) and future SSP5-8.5 periods (2040–2070, 2070–2100). A Mamdani fuzzy-logic workflow converts nine edaphic and climatic layers into FAO land-suitability scores; Recursive Feature Elimination selects the least-redundant covariate set, and ten-fold grid search tunes hyper-parameters for downstream classifiers. Ensemble tree learners—Random Forest (RF) and LightGBM—are trained on the fuzzy-derived dataset. RF ranks growing-season temperature (59%), precipitation (24%) and slope (18%) as the three most influential predictors and attains 97.4% overall accuracy, outperforming LightGBM (82.9%). Suitability maps reveal a climate-driven transition from a baseline mosaic of 50% moderately suitable (S2) + 50% highly suitable (S1) land to a late-century pattern dominated by 71% S1 and 14% S2. Concurrently, mean growing-season temperature rises from 33 °C to 38.4 °C. Integrating fuzzy logic with high-resolution CMIP6 projections and RF classification shows that warming under SSP5-8.5 is likely to expand prime maize land in northern Black-Sea provinces, but temperature becomes the critical limiting factor—highlighting an urgent need for heat-tolerant cultivars and adaptive agronomic planning.