Sentinel-2宽波段光谱指数预测拔节后受冻冬小麦减产率

冬小麦拔节期后的晚霜冻害在区域尺度上表现出一定的空间差异，决定了应对冻害采取分区措施。基于Sentinel-2卫星数据的宽波段光谱指数预测区域冬小麦减产率，对灾情评估和生产管理决策具有重要意义。采用人工模拟霜冻试验，以光谱重采样将ASD FieldSpec® 3光谱辐射计获取的冠层反射率模拟为Sentinel-2传感器对应的波段反射率。采用19个已有光谱指数和以3种形式（简单比值、简单差值、归一化）组合构建的光谱指数构建冬小麦减产率线性回归模型。在每种形式中，筛选出决定系数排名前三的宽波段光谱指数作为候选指数。以商丘地区自然霜冻事件为契机，以Sentinel-2卫星数据计算候选指数，利用地面采样点的实测减产率验证候选指数精度。结果表明：（1）随着处理温度的降低，冬小麦冠层反射率在近红外区呈降低趋势，在可见光区和短波段红外区呈升高趋势；（2）重采样前后反射率数据计算的19个已有光谱指数中，大部分指数与减产率呈显著相关(p<0.001)。筛选出的12个候选光谱指数预测冬小麦减产率的线性回归精度较好，在校正集和验证集中的决定系数均高于0.631；（3）Sentinel-2卫星数据计算的候选光谱指数精度验证结果表明，包含波段B9的3个光谱指数未能通过显著性检验，其他9个光谱指数通过极显著性检验（p<0.001）。基于波段B8，B8a和B12组合的2个光谱指数(B8a-B12和B8-B12)精度较好，决定系数R2分别为0.543和0.492，均方根误差RMSE分别为8.510%和8.971%。该研究构建光谱指数B8a-B12和B8-B12符合简单差值形式，是预测冬小麦减产率宽波段光谱指数的最佳组合。研究成果揭示了冬小麦幼穗发育期冠层反射率在不同低温胁迫下的响应机制，表明Sentinel-2宽波段光谱指数预测冬小麦减产率有良好的精度，在霜冻后区域尺度上的冬小麦减产率预测具有可行性，对于不同地区霜冻灾害的措施制定具有指导作用。

Predicting Yield Reduction Rates of Frost-Damaged Winter Wheat After Jointing Using Sentinel-2 Broad-Waveband Spectral Indices

On the regional scale, the late frost damage to winter wheat after jointing showed a spatial difference which determines that the sub-regional measures against the frost damage should be implemented. Broad-waveband spectral indices based on the Sentinel-2 satellite were proposed in this study to predict yield reduction rates of winter wheat. It is of great significance to disaster assessment and production management decision-making. Based on the artificial frost simulation experiments, the canopy reflectance data measured by ASD FieldSpec® 3 spectroradiometer were simulated to the Sentinel-2 wavebands using spectral resampling. And then, the nineteen published spectral indices and three new forms of wavelength random combinations (simple ratio, simple difference, and normalized difference) were used to construct the linear regression models with winter wheat yield reduction rates. In every form, broad-waveband spectral indices with the top three coefficients of determination were selected as the candidate indices. Aiming at the frost event in the Shangqiu area, all candidate indices were calculated using the Sentinel-2 reflectance data and used to predict winter wheat yield reduction rates, which were validated by the measured yields of the ground sampling points. The results indicated that: (1) With the decrease of the treatment temperatures, canopy reflectance shows a decreasing trend in the near-infrared region, but increased in the visible and short-waveband infrared regions. (2) Most of the nineteen published spectral indices were significantly (p<0.001) correlated with yield reduction rates, regardless of whether the canopy reflectance data were before or after resampling. The twelve candidate spectral indices screened out have good linear regression accuracy for predicting the yield reduction rates of winter wheat and the coefficient of determination above 0.631 in the calibration and the validation datasets. (3) The accuracy of candidate spectral indices calculated from sentinel-2 satellite data showed that the three spectral indices, including the band B9 failed to pass the significance test, and the other nine spectral indices all passed the extremely significant test. The two spectral indices (B8a-B12 and B8-B12) based on the combinations of the B8, B8a, and B12 had good accuracy. The coefficient of determination was 0.543 and 0.492, the root means square error was 8.510% and 8.971%, respectively. Further, B8a-B12 and B8-B12 were found to conform to the simple difference form, which was considered the optimal combination of the broad-waveband spectral indices predicting yield reduction rates of winter wheat. The research results revealed that the response mechanism of canopy reflectance at early spike development stage of winter wheat under different low-temperature stress, indicating that Sentinel-2 broad-waveband spectral indices have good accuracy in predicting the yield reduction rate of winter wheat. It is feasible to predict yield reduction rate at a regional scale after frost and has a guiding role in formulating frost disaster measures in different regions.