Fisher discriminant analysis method applied in drought detection: an instance in an alpine meadow ecosystem. (13th January 2022)

• Record Type: Journal Article
• Authors: Zhang, Tao; Ji, Ximeng; Tang, Yuanyuan; Xu, Mingjie; Zhang, Yangjian; Zhao, Guang; Chen, Ning; Zhu, Juntao; He, Yongtao
• Editors: Hao, Yanbin
• Is Part Of: Journal of plant ecology. Volume 15:Number 5(2022)
• Page Start: 911
• Page End: 920
• Publication Date: 2022-01-13
• Subjects: soil water -- VPD -- GPP -- alpine meadow -- Tibetan Plateau; 土壤水分 -- 饱和水汽压差(VPD) -- 总初级生产量(GPP) -- 高寒草甸 -- 青藏高原; Plant ecology -- Periodicals; Phytogeography -- Periodicals; 581.705
• DOI: 10.1093/jpe/rtac007
• Languages: English
• ISSNs: 1752-9921

Abstract

Abstract: Fisher discriminant analysis can comprehensively take multiple factors into consideration and effectively conduct separations between two classes. If it can be used to detect the occurrences of drought, drought can be detected more effectively and accurately. Based on 9-year carbon flux and corresponding meteorological data, soil water content (SWC) and vapor pressure deficit (VPD) were selected as the discriminant factors. Drought occurrences were detected by applying the Fisher discriminant analysis method in an alpine ecosystem in Tibet. Fisher discriminant analysis was successfully applied to detect drought occurrence in an alpine meadow ecosystem. The soil water deficit and atmospheric water deficit were comprehensively taken into consideration. Consequently, this method could detect the onset and end date of droughts more accurately and reasonably. Based on the characteristics of drought and non-drought samples, the discriminant equation was constructed as y = 24.46SWC − 4.60VPD. When y > 1, the days were distributed above the critical line. In addition, when y was greater than one for more than 10 days, it was labeled as one drought event. If the interval between two drought processes was less than 2 days, it was considered one drought event. With increasing the study period and continued accumulation of observation data, the discriminant equation could be further optimized in the future, resulting in more accurate drought detection.