基于SIF的东北春玉米干旱动态阈值构建

为实现对春玉米全生育期干旱的动态监测，消除分段式干旱灾害阈值在跨越发育期的灾情监测中的跳跃性，该研究构建了东北春玉米全生育期干旱动态阈值，实现了对玉米全生育期灾情的连续性监测。该研究以东北地区春玉米为研究对象，基于时间序列日光诱导叶绿素荧光（solar-induced chlorophyll fluorescence，SIF）数据，结合2000—2020年东北地区春玉米实际干旱灾情资料，采用多项式、高斯拟合法构建东北春玉米不同干旱等级曲线，得到最优拟合模型，以相邻干旱等级拟合曲线的平均值作为玉米全生育期各干旱等级动态临界阈值，并用独立样本和典型干旱事件验证动态临界阈值。结果表明，高斯拟合模型在描述东北春玉米不同干旱等级时间序列叶绿素荧光指数值时较多项式拟合效果更佳；该研究构建的东北春玉米不同干旱等级动态阈值能较好地反映东北地区春玉米干旱情况，干旱等级识别结果与实际灾情等级基本符合的占91.03%，其中完全符合的占82.76%，验证精度较高；基于典型干旱事件的时空验证结果也表明该干旱等级动态阈值能较好地反映东北地区春玉米干旱时空演变特征和干旱灾情的发生发展动态。

Constructing the drought dynamic threshold of spring maize in Northeast China using SIF index

Spring maize is suffering from the ever-increasing drought under global warming in Northeast China in recent years. Most previous studies focused on the disaster level threshold of segmented agrometeorological drought, according to the crop development stage. However, it cannot fully meet the practical application under complex conditions at present. This study aims to construct the drought dynamic threshold of spring maize in Northeast China from 2000 to 2020. Time series solar-induced chlorophyll fluorescence (SIF) index with the actual data of drought disaster was utilized to construct the drought and drought-free sample sets. Binomial, trinomial and Gaussian fitting were selected to determine the different drought level curves, according to the curve change of SIF value in the study period. The optimal fitting model was obtained using the coefficient of determination (R²), Akaike information criterion (AIC) and Bayesian information criterion (BIC). The average of the fitting curves for the adjacent drought levels was then taken as the dynamic critical threshold of each drought level during the whole growth period of maize. The optimal dynamic critical threshold was finally verified by independent samples and typical drought events. The results showed that the Gaussian fitting model was more effective in representing the SIF values of different drought levels, compared with the binomial and trinomial fitting. The dynamic thresholds of different drought levels better represented the actual drought situation of spring maize. The drought level identification was fully consistent with the actual disaster level in 82.76% of cases, and basically consistent with the actual disaster level in 91.03% of cases, indicating the high verification accuracy. Taking the typical drought event in Liaoning Province as an example, the drought process was verified by the threshold value. The times of drought occurrence and end were completely consistent with the actual disaster records. Spatially, the drought-affected areas were also consistent in the actual disaster records. Taking Chifeng City as a typical drought site, the drought events with threshold values were verified to be consistent in both the time and level of drought occurrence. There was basically 100% matching between the actual disaster records and threshold verification in the whole drought, indicating a high verification accuracy. Therefore, the dynamic drought threshold can be expected to better reflect the spatiotemporal evolution characteristics of spring maize drought, including the occurrence and development dynamics of drought disasters. Better recognition of drought level was achieved in the dynamic drought threshold, compared with the segmented threshold in the development stage. But some challenges also remained. The SIF data makes it difficult to capture the phenomenon of sudden drought, indicating the drought duration less than 8 days. In addition, there is a certain impact on the identification, because the phenology of maize varies in the climate change, leading to the advanced or postponed seedling stage.