融合地理加權回歸與多目標空間最佳化之作物區位配置研究;Integrating Geographically Weighted Regression with Multi-Objective Spatial Optimization for Crop Allocation

NCUIR > College of Earth Sciences > Graduate Institute of Hydrological and Oceanic Sciences > Electronic Thesis & Dissertation > Item 987654321/97445

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/97445

Title:	融合地理加權回歸與多目標空間最佳化之作物區位配置研究;Integrating Geographically Weighted Regression with Multi-Objective Spatial Optimization for Crop Allocation
Authors:	宋冠毅;Sung, Guan-Yi
Contributors:	水文與海洋科學研究所
Keywords:	非支配排序基因演算法;地理加權回歸;多目標最佳化;作物區位配置;NSGA-II;GWR;Multi-objective Optimization;Crop allocation
Date:	2025-08-20
Issue Date:	2025-10-17 11:19:11 (UTC+8)
Publisher:	國立中央大學
Abstract:	本研究旨在探討氣候變遷情境下，如何透過融合地理加權回歸（GWR）與非支配排序基因演算法（NSGA-II），建立一套具空間解釋力與決策導向性的多目標作物區位配置模式。研究範圍涵蓋臺灣西部13縣市，選取稻米、花生與甘藷為示範作物，於466個5公里解析度網格中評估四個目標：作物總產量、農家收益、適栽度與空間緊湊度，並設定產量不得低於2018–2022年國內平均值之限制條件。氣候資料來源為臺灣氣候變遷推估資訊與調適知識平台(Taiwan Climate Change Projection Information and Adaptation Knowledge Platform, TCCIP)提供之26項1960–2022年指標，經主成分分析（PCA）降維後擷取7個主成分，累積解釋變異量達82%。GWR模型結合主成分與作物生產力建構空間回歸模型，修正R^2提升達157.8至989.4%，AICc下降5.7–9.9%，並有效消除殘差空間自相關（Moran’s I趨近於0），顯示模型能準確捕捉氣候與生產力間的空間異質性。進一步以GWR預測結果作為NSGA-II輸入資料，於800個種群與600世代中進行多目標演化求解，結果顯示GWR-NSGA在超體積（HV）與解多樣性指標上優於不含GWR的基準模式（TC-NSGA），其中稻米配置比例穩定占據主導地位（約64–67%），花生占比則較初始值顯著提升（16%至約24%）。本研究證實氣候主成分與空間回歸模型能有效輔助作物區位規劃，所建立之整合模式具備良好解釋力與實務應用潛力，適合作為未來氣候調適與農業空間規劃之決策支援工具。未來建議進一步提升空間解析度、引入多尺度GWR與非氣候因子，以及優化演算法參數與比較其他演化法，以增進模型精度與穩定性。;This study proposes an integrated framework combining Geographically Weighted Regression (GWR) and the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to optimize multi-objective crop allocation under climate change scenarios. The study area includes 13 counties in western Taiwan, with rice, peanuts, and sweet potatoes selected as target crops. A total of 466 feasible 5-km spatial grid units were established, and four objectives were evaluated: total crop production, farm profit, land suitability, and spatial compactness. Constraints were imposed to ensure each crop’s production met or exceeded the national average from 2018 to 2022. Climate data, including 26 variables from 1960–2022 provided by TCCIP, were reduced to seven principal components (PCs) using Principal Component Analysis (PCA), explaining 82% of the total variance. GWR models incorporating these PCs significantly improved model performance over ordinary least squares (OLS), with adjusted R² increasing by 157.8%–989.4%, AICc decreasing by 5.7%–9.9%, and residual Moran’s I approaching zero. The GWR-estimated crop productivity was then integrated into the NSGA-II optimization process (800 population size, 600 generations). Results show that the GWR-based NSGA-II model outperforms the baseline (TC-NSGA) in hypervolume (HV) and diversity metrics. Rice maintained a dominant allocation proportion (64–67%) across all optimal solutions, while the peanut share increased from 16% to approximately 24% post-optimization. The proposed method demonstrates strong spatial explanatory power and practical value in supporting climate-resilient crop planning. Future work should focus on finer-scale spatial data, incorporation of multi-scale GWR, inclusion of non-climatic factors, and algorithm benchmarking to further enhance model accuracy and robustness.
Appears in Collections:	[Graduate Institute of Hydrological and Oceanic Sciences] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	25	View/Open

社群 sharing

Loading...