山东省耕地质量遥感评价指标分析及预测研究

Analysis of evaluation indicators and prediction for farmland quality based on remote sensing in Shandong Province

关键词:  耕地质量  遥感评价指标  空间分析  预测模型

Key words:  farmland quality  remote sensing evaluation indicators  spatial analysis  prediction model

摘要:

【目的】在全球化背景下，粮食安全面临人口增长及气候变化等多重挑战，耕地质量直接影响粮食生产的可持续性。山东省作为重要的农业产区，受土地开发、工业化进程及气候变化影响，传统实地监测手段难以满足耕地质量动态变化的实时监测需求，如何利用遥感数据对大尺度耕地质量进行分析与预测成为当下研究的重点。【方法】为系统分析山东省耕地质量的空间分布特征、探讨其影响因素，并构建有效的监测模型以支持区域农业的可持续发展，本研究以山东省为研究区，结合Sentinel-2遥感影像数据与地形数据，运用压力-状态-响应框架，构建耕地质量评价指标体系，分析各指标的统计特征与分布特征，并以每个地级市为单位建立独立的随机森林与CART耕地质量预测模型。【结果】（1）基于压力-状态-响应框架的耕地质量评价指标体系能够有效反映山东省不同耕地质量等级的特征，其中生产压力指标中的坡度与耕地质量负相关，RVI与耕地质量正相关；耕地状态指标中NDVI、REP、LSWI、SAVI均与耕地质量等级呈正相关；社会响应指标中除NDWI外均呈现相同的变化趋势；（2）评价指标分布与耕地质量等级的空间分布直接相关；（3）随机森林预测模型表现优于CART模型，特征重要性从高到低依次为坡度、RVI/NDVI、REP/LSWI、SAVI/NDWI/NDRE、MSAVI/GNDVI；社会响应指标中NDWI、MSAVI、GNDVI分别反映灌溉压力、幼苗生长压力及作物衰老压力对耕地质量的影响，因三者覆盖作物全生育周期且关乎生长基础与最终产量，特征重要性高于仅表征作物成熟阶段光合能力的NDRE；（4）近五年山东省耕地质量的空间分布格局总体保持稳定，各年份间局部有变化但整体态势未发生显著改变。【结论】遥感技术与机器学习算法的结合提高了耕地质量评估的效率和准确性，有助于制定有针对性的耕地保护策略和促进农业可持续发展。本研究为耕地质量等级的快速监测提供了有效方法，且为农业生产与生态保护决策提供了科学依据。

Abstract:

[Purpose] Under the backdrop of globalization, food security faces multiple challenges such as population growth and climate change, and farmland quality directly impacts the sustainability of grain production. As a critical agricultural region in China, Shandong Province is affected by land development, industrialization, and climate change. Traditional field monitoring methods struggle to meet the demand for real-time dynamic monitoring of farmland quality changes. Consequently, utilizing remote sensing data for large-scale farmland quality analysis and prediction has become a focal point of current research. [Method] To systematically analyze the spatial distribution characteristics of farmland quality in Shandong Province, explore its influencing factors, and construct effective monitoring models to support sustainable regional agricultural development, this study integrates remote sensing technology and machine learning algorithms for comprehensive evaluation and prediction. Using Shandong Province as the study area, this research combined Sentinel-2 remote sensing imagery and topographic data to establish a farmland quality evaluation index system based on the pressure-state-response (PSR) framework. The system comprises three criterion layers: production pressure, farmland state, and social response. Statistical and spatial distribution characteristics of each indicator were analyzed. Independent Random Forest and Classification and Regression Tree prediction models were developed for each prefecture-level city to assess farmland quality. [Result] (1) The PSR-based evaluation index system effectively characterized farmland quality grades in Shandong Province. Among production pressure indicators, slope exhibited a negative correlation with farmland quality, while the RVI showed a positive correlation. For farmland state indicators, NDVI, REP, LSWI, and SAVI all demonstrated positive correlations with farmland quality grades. All social response indicators except for NDWI exhibit consistent trends in their variations. (2) The distribution of evaluation indicators was directly linked to the spatial distribution of farmland quality grades. (3) The Random Forest prediction model outperformed the CART model across all cities. Feature importance rankings revealed commonalities among cities: slope was the most critical determinant of farmland quality, followed by RVI/NDVI, REP/LSWI, SAVI/NDWI/ NDRE, and MSAVI/GNDVI. This hierarchy highlights the dominant roles of terrain and vegetation health in quality assessment. In the Social Response Layer, NDWI, MSAVI, and GNDVI reflect the impacts of irrigation stress, seedling growth stress, and crop senescence stress on cultivated land quality, respectively, with higher characteristic importance due to their roles in capturing stressors that affect the entire crop growth cycle, foundational growth, and final yield, in contrast to NDRE, which only characterizes photosynthetic capacity in the crop maturity stage and thus has lower significance for assessing cultivated land quality. (4) Over the past five years, the spatial distribution pattern of farmland quality in Shandong Province remained generally stable, with localized variations between years but no significant overall shifts. [Conclusion] The integration of remote sensing technology and machine learning algorithms enhances the efficiency and accuracy of farmland quality assessment, facilitating targeted farmland protection strategies and promoting sustainable agricultural development. This study provides an effective method for rapid monitoring of farmland quality grades and offers a scientific basis for decision-making in agricultural production and ecological conservation.