基于开花期卫星遥感数据的大田小麦估产方法比较

发布时间：2018-11-16 15:01

【摘要】：【目的】卫星遥感具有覆盖范围广、获取速度快、信息量大、动态性强等优势,能够及时准确地获取作物产量信息,反映作物产量空间变化趋势。遥感技术作物估产已成为现代农业生产中研究热点。通过改善遥感估产建模方法,以实现进一步提高大田作物遥感估产精度,为宏观了解不同区域作物产量形成情况及变化趋势提供直观、可靠的参考。【方法】论文结合2011—2012年江苏省大丰、兴化、姜堰、泰兴、仪征5个县区的定点观测试验,以国产卫星产品HJ-1A/1B影像为遥感数据,于小麦开花期开展大田定位观测区卫星遥感植被指数、关键生长指标与收获期单产间的定量分析。通过对产量与小麦生长指标以及植被指数进行定量关系分析,进一步增强遥感反演的机理性和重演性。将卫星遥感变量与小麦产量进行相关关系分析作为遥感估产的直接建模方法,间接建模方法则是选取与产量相关性较好的遥感变量以及与遥感变量相关性较好的主要苗情指标,利用筛选得到的敏感遥感变量,首先监测对应的小麦生长指标,结合该小麦生长指标与产量间的定量关系,进而建立间接估产模型,利用此模型进行小麦遥感间接估产。利用直接和间接建模方法,以相关性最高为原则,筛选估算产量的敏感卫星遥感变量。以2012年试验数据为建模样本,采用线性回归分析方法,分析小麦开花期苗情指标、产量与卫星遥感变量两两之间的相关性,分别构建以遥感植被指数为基础的大田小麦估产模型,与地面实测结果一起建立模型共同分析。以2011年试验数据为验证样本,选取评价指标拟合度(R2)和均方根误差(RMSE),对两类模型的估算精度进行验证和比较,以提高遥感反演的定量化水平和可信度。【结果】分别以差值植被指数(difference vegetation index,DVI)和比值植被指数(ratio vegetation index,RVI)为基础的单因子直接估产模型的均方根误差(root mean square error,RMSE)为918 kg·hm-2和1 399.5 kg·hm-2,以DVI和RVI遥感变量构建双变量估产模型的RMSE为1 036.5 kg·hm-2,以归一化植被指数(normalized difference vegetation index,NDVI)和叶片氮积累量为基础构建的间接估产模型的RMSE为805.5 kg·hm-2,说明开花期HJ-1A/1B影像估算小麦区域产量是可行的,且精度较高;经比较,以NDVI和叶片氮积累量为基础的间接估产模型精度明显高于直接估产模型,相较于DVI直接估产模型RMSE降低了112.5 kg·hm-2,相较于RVI直接估产模型RMSE降低了594 kg·hm-2,相较于双因子模型RMSE降低了231 kg·hm-2。【结论】国产卫星HJ-1A/B可以较好满足估测小麦产量要求,且利用间接方法建立作物遥感估产模型要好于直接方法,研究结果为利用遥感技术更为准确估算大田小麦产量提供了一种新的途径。
[Abstract]:[objective] Satellite remote sensing has the advantages of wide coverage, fast acquisition speed, large amount of information, strong dynamic and so on. It can obtain crop yield information accurately and timely, and reflect the spatial variation trend of crop yield. Crop yield estimation by remote sensing has become a hotspot in modern agricultural production. By improving the modeling method of crop yield estimation by remote sensing, the precision of crop yield estimation can be further improved, which can provide a visual view for the macroscopic understanding of crop yield formation and change trend in different regions. [methods] based on the fixed-point observation experiments in five counties of Dafeng, Xinghua, Jiangyan, Taixing and Yizheng in Jiangsu Province from 2011-2012, the HJ-1A/1B images of domestic satellite products were used as remote sensing data. The quantitative analysis of satellite remote sensing vegetation index, key growth index and harvest yield was carried out in the field location observation area during the flowering period of wheat (Triticum aestivum L.). Through quantitative analysis of yield, wheat growth index and vegetation index, the mechanism and repeatability of remote sensing inversion are further enhanced. The correlation analysis between satellite remote sensing variables and wheat yield was taken as the direct modeling method for estimating yield by remote sensing, while indirect modeling method was used to select remote sensing variables with good correlation with yield and main seedling condition indexes with good correlation with remote sensing variables. Based on the sensitive remote sensing variables obtained by screening, the corresponding wheat growth index was first monitored, and the quantitative relationship between the wheat growth index and the yield was combined, and then an indirect yield estimation model was established, which was used to estimate the yield indirectly by remote sensing. By using direct and indirect modeling methods, sensitive satellite remote sensing variables for estimating yield are screened based on the principle of maximum correlation. Based on the experimental data in 2012, the correlation between seedling growth index, yield and satellite remote sensing variables in flowering stage of wheat was analyzed by linear regression analysis. The field wheat yield estimation models based on remote sensing vegetation index were constructed, and the model was analyzed together with the measured results on the ground. Taking the experimental data of 2011 as the validation sample, the estimation accuracy of the two models is verified and compared with the evaluation index fitting (R2) and root mean square error (RMSE),). [results] based on the difference vegetation index (difference vegetation index,DVI) and the ratio vegetation index (ratio vegetation index,RVI), the root mean square error (root mean square error,) of the single factor direct yield estimation model was obtained. RMSE) for 918 kg hm-2 and 1 399.5 kg hm-2, using DVI and RVI remote sensing variables to construct a bivariate yield estimation model with a RMSE of 1 036.5 kg hm-2, to normalize the vegetation index (normalized difference vegetation index, The RMSE of the indirect yield estimation model based on NDVI) and leaf nitrogen accumulation was 805.5 kg hm-2, which indicated that the HJ-1A/1B image at flowering stage was feasible and accurate in estimating wheat regional yield. By comparison, the precision of indirect yield estimation model based on NDVI and leaf nitrogen accumulation was significantly higher than that of direct yield estimation model. Compared with DVI direct yield estimation model, RMSE decreased by 112.5 kg hm-2,. Compared with RVI direct yield estimation model, RMSE decreased by 594 kg hm-2, compared with two-factor model RMSE by 231 kg hm-2. [conclusion] domestic satellite HJ-1A/B can better meet the requirement of wheat yield estimation. The indirect method is better than the direct method in establishing crop yield estimation model by remote sensing. The results provide a new way to estimate wheat yield more accurately by remote sensing.
【作者单位】：扬州大学江苏省作物遗传生理国家重点实验室培育点/粮食作物现代产业技术协同创新中心;
【基金】：国家自然科学基金(41271415) 江苏高校优势学科建设工程(PAPD) 江苏省农业自主创新资金(CX(16)1042) 苏州市农业科技创新项目(SNG201643) 扬州市科技计划(YZ2016242) 扬州大学科技创新团队
【分类号】：S127;S512.1

【相似文献】