基于Sentinel-2A影像估算黄土高原光合/非光合植被盖度

以黄土高原为例,基于Sentinel-2A影像和地表实测地物光谱与盖度数据,分别在模拟混合场景和野外实测混合场景中,评估4种NPV植被指数(NPVI):SWIR32(短波红外比值指数)、DFI(干枯燃料指数)、STI(土壤耕作指数)和NDTI(归一化差异耕作指数)估算非光合植被盖度(f_NPV)的有效性,并利用优化法确定线性光谱混合模型的关键参数端元值,估算研究区光合植被盖度(f_PV)和f_NPV.结果表明,在模拟混合场景下,4种NPVI与模拟f_NPV线性关系的R²是0.365~0.750;在野外场景中,其相关性均有一定程度的降低,R²是0.147~0.211.研究构建NDVI-SWIR32像元三分模型,并确定了最优端元值:NDVI_PV=0.80,SWIR32_PV=0.60, NDVI_NPV=0.17,SWIR32_NPV=0.77,NDVI_BS=0.23,SWIR32_BS=0.99.模型对f_PV和f_NPV估算精度R²分别是0.817和0.463,NSE分别是0.806和0.458.利用该模型估算全区2019年4、8和12月的平均f_PV和f_NPV,分别为20.3%和59.2%,48.6%和33.1%,10.7%和59.0%.随时间推移,f_PV从东南向西北不断增加而后减小,f_NPV与之相反. NDVI-SWIR32模型可以用于Sentinel-2A影像数据来监测黄土高原地区f_PV和f_NPV的时空动态变化.

Sentinel - 2A data - derived estimation of photosynthetic and non - photosynthetic vegetation cover over the loess plateau

In this study, we evaluated four non-photosynthetic vegetation indices (NPVI), including Shortwave Infrared Ratio (SWIR32), Dead Fuel Index (DFI), Soil Tillage Index (STI) and Normalized Difference Tillage Index (NDTI) for Non-photosynthetic Vegetation (f_NPV) estimation in the simulated and field mixed scenarios, respectively, and applied them to estimate f_NPV using Sentinel-2A data (10m) over the Loess Plateau. We applied a linear unmixing model to estimate Photosynthetic Vegetation (f_PV) and f_NPV based on the triangular relationship between Normalized Vegetation Difference Index (NDVI) and NPVI (e.g., SWIR32). The NDVI-NPVI endmember values were determined. The results showed that the correlation coefficient (R²) between each NPVI and simulated f_NPV was between 0.365 to 0.750, and 0.147 to 0.211 between each NPVI and f_NPV under the field mixed scenario. Using this approach, we estimated the Loess Plateau’s average f_PV and f_NPV for April, August and December in 2019, being 20.3% and 59.2%, 48.6% and 33.1%, and 10.7% and 59.0%, respectively. The R² of the model for f_PV and f_NPV estimation reached 0.817 and 0.463, respectively, while the NSE was 0.806 and 0.458, respectively. The results also revealed the seasonal variation f_PV from southeast to northwest over time, and the opposite trend for f_NPV. Our study suggests that the NDVI-SWIR32 model can be used with Sentinel-2A data to adequately monitor the spatiotemporal dynamics of f_PV and f_NPV in the Loess Plateau.