不同叶面积指数遥感产品在呼伦贝尔草甸草原的应用对比

叶面积指数(leaf area index,LAI)是植被生理过程模拟的重要参数之一,对植被气候关系、全球气候变化研究等具有重要的意义。近年来LAI产品越来越多,这些产品的精度、区域适用性都不相同。为了选择适用于呼伦贝尔典型草甸草原的LAI产品,为以后在呼伦贝尔展开工作提供便利,该文以内蒙古呼伦贝尔草甸草原为研究区域,利用2013年6-8月6期地面试验数据,以HJ-1A/B CCD高分辨率影像为中间桥梁建立植被指数SR与LAI的统计模型,反演得到LAI参考图像,对研究区域内与地面试验同期的MODIS LAI和GLASS LAI、GEOV1 LAI产品分别进行了直接验证与交叉验证。结果显示,3个LAI产品均存在高估现象,以GLASS LAI最为显著约高估41%,其次是MODIS LAI约高估了32%。GEOV1 LAI产品准确性最高,RMSE=0.289 MAE=0.216。GLASS LAI与GEOV1 LAI产品的相关性最好(R2=0.6465)。通过对比全年LAI产品发现,3个产品具有良好的时序一致性。GLASS LAI呈现为平滑曲线,高估现象主要存在于LAI值较小时。MODIS LAI最不稳定性,波动性较大。GEOV1 LAI产品在第133天至第201天这段时间内LAI值比其他两个产品的LAI值小;在第202天后GEOV1 LAI值与GLASS LAI值相差无几,高于MODIS LAI。根据对比分析结果,GEOV1 LAI产品最适用于呼伦贝尔典型草甸草原。通过提取质量控制层数据,确定云覆盖不是影响LAI异常的原因。

Comparation of different LAI products in hulunber meadow steppe

Abstract: Leaf area index (LAI) is an important parameter in vegetation physiological process model. It is important for global climate change research. Recent years, LAI products is increasing in different characteristics. Different products are suitable for different areas, our research fouced on finding out the most suitable LAI product for typical meadow steppe. The study compared the Moderate Resolution Imaging Spectroradiometer (MODIS), GEOLAND2 Version1 (GEOV1), Global Land Surface Satellite (GLASS) Leaf Area Index (LAI) products and HJ LAI in hulunber China in 2013. 6 measurements of LAI data in 2013 from June to August were acquired. The field measurements achieved by LAI-2000 in 3 km×3 km areas on the temperate meadow steppe in hulunber. The SR (simple ratio index) was calculated from the HJ-1A/B CCD image band reflectance in study area. statistical model between vegetation index SR and LAI ground measurement was established. LAI field measurement were well simulated with the statistical model (R2=0.6042). The HJ-1A/B CCD images with 30m spatial resolution were used on generation of LAI reference map. And others LAI products were 1km spatial resolution. So spatial resolution need transformed to the same scale. The three LAI products HJ LAI were evaluated and cross compared to assess their uncertainty and variability. The results showed that the three LAI products are overestimated meadow steppe LAI the most severe GLASS LAI, over about 41%, followed by MODIS about 32%. GEOV1 LAI product is close to HJ LAI, RMSE=0.289 MAE=0.216. In terms of accuracy, the products were ranked in the following order: GEOV1> MODIS> GLASS.GEOV1 appears to be the most accurate product.Compared to MODIS and GLASS, GEOV2 significantly improved the accuracy in the high LAI area. The neural network (NNT) algorithm, whose training dataset was derived from fused CYCLOPES and MODIS products with varied weights and an additional SWIR band, was the key to the favorable performance of GEOV1. So GEOV1 was the most accurate product for estimates meadow steppe. The differences between LAI products were compared. No significant discrepancies existed between the GEOV1 and GLASS LAI product results (R2=0.6465). By comparing LAI products data in whole year we found that the three products with good timing consistency. GLASS LAI showed a smooth curve, and had significantly overestimated when LAI valuesas wsmall. MODIS LAI was unstable. The LAI of GEOV1 products were smaller than the other two products from the 133rd day to the 201st day. GEOV1 LAI was similar with GLASS LAI, and higher than the MODIS LAI. The numerous pixels that cause obvious LAI retrieval anomalies had the highest quality. So few pixels contaminated by clouds were not a significant cause of the LAI retrieval anomalies in study area.