Sentinel-2和GF-1影像结合提取苜蓿空间分布

及时准确地获取苜蓿空间分布信息有利于对草业生产发展和管理提供科学数据支撑。该研究基于GF-1/WFV和Sentinel-2遥感影像，以甘肃省金昌市作为研究区，构建了苜蓿的归一化植被指数（Normalized Difference Vegetation Index，NDVI）数据集，并结合苜蓿光谱反射率随生育期的变化规律，提出一种利用MATLAB寻峰函数（Findpeaks）提取苜蓿遥感特征的方法，通过确定最小峰值突出（Minimum Peak Prominence，MPP）值实现金昌市苜蓿空间分布信息的提取。研究结果表明，基于Sentinel-2遥感数据的识别苜蓿精度优于GF-1/WFV，识别精度和Kappa系数在85%和0.7以上，主要是由于Sentinel-2数据的NDVI时间序列曲线密度较GF-1/WFV大，可以更好地识别苜蓿刈割前后的关键时间点；寻谷法的苜蓿提取总体精度、Kappa系数、用户精度、制图精度指标均比寻峰法高，基于Sentinel-2影像的寻谷法苜蓿遥感识别总体精度为92.25%，Kappa系数为0.81，位置精度为86.44%；2019年金昌市苜蓿空间分布整体呈现从北到南逐渐增多的趋势，统计得到苜蓿种植面积为15 449.07 hm2，其中金川区的苜蓿面积为1 353.42 hm2，占金昌市苜蓿总面积的8.76%；永昌县的苜蓿面积为14 095.65 hm2，占总面积的91.24%。研究结果证实，基于Sentinel-2遥感数据的寻谷法可以有效识别苜蓿空间分布，对于实现草牧场精准化管理和草牧业生产信息精准监测具有重要意义。

Spatial distribution extraction of alfalfa based on Sentinel-2 and GF-1 images

Abstract: Alfalfa is a key feed variety for the development of herbivorous animal husbandry, which is of great significance to the development of animal husbandry and food safety in China. Timely and accurate acquisition of alfalfa spatial distribution information is helpful to provide scientific data support for the development and management of grass production. This study was based on high-resolution GF-1/WFV and Sentinel-2 remote sensing images in Jinchang City, Gansu Province, a Normalized Difference Vegetation Index (NDVI) dataset was constructed, combined with the change of spectral reflectance of alfalfa with the growth stage, an algorithm of extracting the remote sensing characteristics of alfalfa using the Findpeaks function of MATLAB was proposed. By determining the value of Minimum Peak Prominence (MPP), the limitation of automatic identification and area extraction of alfalfa was solved, and the spatial distribution information of alfalfa in Jinchang City was extracted. Firstly, the time series of NDVI of alfalfa was analyzed, and it found that NDVI of alfalfa increased many times as peak value decreased in one year. The NDVI time series curve had many peaks and troughs, among which the peak represented the high value of NDVI in a growing period, which was the flourishing period of alfalfa growth and development, and the trough reflected the alfalfa from the peak period to the cutting state. According to the field investigation, the number of peaks and troughs of alfalfa was determined, and the number of troughs on the NDVI time series curve was 3-4, and the number of peaks was 3-5. Combined with the verification results of position accuracy found that the classification accuracy increased when MPP was in the range of 0.30-0.40 and reached the maximum value when MPP was 0.40, while the classification accuracy tended to decrease with the increase of MPP. Therefore, 0.40 was the most reasonable value of MPP to extract the information of alfalfa in Jinchang City, and the potential spatial distribution results of alfalfa were obtained by using the Findpeaks function of MATLAB software. The spatial distribution data set of alfalfa planting area in the study area was established by masking the terrain characteristics and the spatial distribution data of cultivated land, removing forests, and other land objects. Finally, the spatial distribution of alfalfa in Jinchang City in 2019 was obtained by using ENVI software to carry out classification post-processing such as multiplicity filtering and fragment elimination. The results show that: 1) The recognition accuracy of Sentinel-2 remote sensing data was better than that of GF-1/WFV remote sensing data. The recognition accuracy and Kappa coefficient of Sentinel-2 data were more than 85% and 0.7, mainly because the density of the NDVI time series curve of Sentinel-2 data was larger than that of GF-1/WFV data, which could better capture and identify the key time points of alfalfa. 2) The find troughs method has more advantages in alfalfa identification in the study area, and the overall accuracy, Kappa coefficient, user accuracy and mapping accuracy of alfalfa identified by remote sensing were higher than those of find peaks method. 3) The find troughs method based on Sentinel-2 image was the best method for alfalfa remote sensing recognition, with an overall accuracy of 92.25%, the Kappa coefficient of 0.81, and the position accuracy of 86.44%. It had a good monitoring effect in terms of spatial location. 4) The spatial distribution of alfalfa in Jinchang City showed a trend of increasing gradually from north to south, most of the continuous areas were mainly concentrated in the south-central and southwest of Jinchang City, and there was only sporadic distribution in the north. The alfalfa planting area of Jinchang City identified by find troughs method based on Sentinel-2 images was 15 449.07 hm2 in 2019, of which the alfalfa area of Jinchuan district was 1 353.42 hm2, accounting for 8.76% of the total alfalfa area of Jinchang City, and the alfalfa area of Yongchang county was 14 095.65 hm2, accounting for 91.24% of the total area. The research results confirmed that the find troughs method based on Sentinel-2 remote sensing data could effectively identify alfalfa, which had important practical significance for the refined and precise management of pastures and the precise monitoring of pasture production information.