| 杨慧,李玥,武凌,吴兵,高玉红,剡斌,周慧,唐洁,赵永伟.MODIS干旱监测指数结合PSO_RBFNN反演甘肃省胡麻出苗期土壤湿度[J].干旱地区农业研究,2025,(1):267~277 |
| MODIS干旱监测指数结合PSO_RBFNN反演甘肃省胡麻出苗期土壤湿度 |
| MODIS drought monitoring index combined with PSO_RBFNN for soil moisture retrieval at the seedling stage of flax in Gansu Province |
| |
| DOI：10.7606/j.issn.1000-7601.2025.01.28 |
| 中文关键词: 遥感 土壤湿度 干旱指数 中分辨率成像光谱仪 径向基函数神经网络 粒子群优化算法 |
| 英文关键词:remote sensing soil moisture drought index MODIS RBFNN PSO |
| 基金项目:国家自然科学基金（32460443，32060437）；甘肃省科技计划-自然科学基金重点项目（23JRRA1403） |
|
| 摘要点击次数: 0 |
| 全文下载次数: 0 |
| 中文摘要: |
| 针对目前单一的遥感干旱监测指数难以全面反映作物生长期土壤湿度的动态变化等问题，以甘肃省胡麻出苗期田间土壤相对湿度为研究对象，选取作物形态及绿度、冠层温度、冠层含水量等指标作为遥感干旱监测指数。利用MODIS遥感干旱监测指数和农田土壤相对湿度实测数据，结合经过粒子群优化的径向基函数神经网络（PSO_RBFNN），构建了农田土壤相对湿度反演模型，并基于BP_NN、RBFNN、PSO_RBFNN人工神经网络和逻辑回归（LR）4种机器学习方法，利用土壤相对湿度实测数据对不同模型反演结果的精度进行了验证和对比分析。结果表明：与其他3种模型相对比，MODIS遥感干旱监测指数与PSO_RBFNN结合反演甘肃省胡麻出苗期的农田土壤相对湿度效果较好；模型对10 cm和20 cm深度土壤相对湿度的反演结果平均精度分别达到89.91%和91.71%。与RBFNN、LR和BP_NN模型相比，PSO_RBFNN模型10 cm土层平均预测精度分别提高8.69个百分点、4.94个百分点、4.76个百分点，20 cm土层平均预测精度分别提高6.91个百分点、6.86个百分点、9.32个百分点。模型回归分析显示，相对于1∶1斜线，PSO_RBFNN模型偏差最小，与10 cm和20 cm深度土壤相对湿度的相关系数分别达到0.68和0.74，说明利用PSO_RBFNN模型反演有效，可为区域农田土壤湿度遥感监测反演提供新的案例借鉴。 |
| 英文摘要: |
| Given the limitations of a single remote sensing drought monitoring index in fully capturing the dynamic changes in soil moisture during crop growth, this study focuses on field soil relative moisture during the flax seedling stage in Gansu Province. The selected remote sensing drought monitoring indices include crop form and greenness, canopy temperature, and crop canopy water content. Utilizing the MODIS drought monitoring index and measured soil relative humidity data, combined with a radial basis function neural network (PSO_RBFNN) optimized by particle swarm optimization, an inverse model of soil relative humidity in farmland was constructed. The accuracy of inversion results of different models—BP_NN, RBFNN, PSO_RBFNN artificial neural network, and logistic regression (LR)—was verified and compared using measured soil relative humidity data. The results indicated that the combination of the MODIS drought monitoring index and PSO_RBFNN outperformed the other three models in retrieving the soil relative humidity at the flax seedling stage in Gansu Province. The average inversion accuracy for soil relative humidity at 10 cm and 20 cm depths was 89.91% and 91.71%, respectively. Compared with the RBFNN, LR, and BP_NN models, the average accuracy improved by 8.69, 4.94, 4.76 percentage points in 10 cm soil depth and 6.91, 6.86, 9.32 percentage points in 20 cm depths, respectively. Regression analysis showed minimal deviation relative to the 1∶1 slash line, with correlation coefficients for soil relative humidity at 10 cm and 20 cm depths reaching 0.68 and 0.74, respectively. This study highlights the effectiveness of the PSO_RBFNN model and offers a valuable reference for remote sensing monitoring and the inversion of regional farmland soil moisture. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| | |