MistNet: Measuring historical bird migration in the US using archived weather radar data and convolutional neural networks. Issue 11 (27th August 2019)

• Record Type: Journal Article
• Authors: Lin, Tsung‐Yu; Winner, Kevin; Bernstein, Garrett; Mittal, Abhay; Dokter, Adriaan M.; Horton, Kyle G.; Nilsson, Cecilia; Van Doren, Benjamin M.; Farnsworth, Andrew; La Sorte, Frank A.; Maji, Subhransu; Sheldon, Daniel
• Editors: Freckleton, Robert
• Is Part Of: Methods in ecology and evolution. Volume 10:Issue 11(2019)
• Page Start: 1908
• Page End: 1922
• Publication Date: 2019-08-27
• Subjects: aeroecology -- bird migration -- convolutional neural networks -- deep learning -- machine learning -- movement ecology -- ornithology -- weather radar; Ecology -- Periodicals; Evolution -- Periodicals; 577
• DOI: 10.1111/2041-210X.13280
• Languages: English
• ISSNs: 2041-210X

Abstract

Abstract: Large networks of weather radars are comprehensive instruments for studying bird migration. For example, the US WSR‐88D network covers the entire continental US and has archived data since the 1990s. The data can quantify both broad and fine‐scale bird movements to address a range of migration ecology questions. However, the problem of automatically discriminating precipitation from biology has significantly limited the ability to conduct biological analyses with historical radar data. We develop Mist Net, a deep convolutional neural network to discriminate precipitation from biology in radar scans. Unlike prior machine learning approaches, Mist Net makes fine‐scaled predictions and can collect biological information from radar scans that also contain precipitation. Mist Net is based on neural networks for images, and includes several architecture components tailored to the unique characteristics of radar data. To avoid a massive human labelling effort, we train Mist Net using abundant noisy labels obtained from dual polarization radar data. In historical and contemporary WSR‐88D data, Mist Net identifies at least 95.9% of all biomass with a false discovery rate of 1.3%. Dual polarization training data and our radar‐specific architecture components are effective. By retaining biomass that co‐occurs with precipitation in a single radar scan, Mist Net retains 15% more biomass than traditional whole‐scan approaches to screening. Mist Net is fully automated and can be applied to datasets of millions of radar scans to produce fine‐grained predictions that enable a range of applications, from continent‐scale mapping to local analysis of airspace usage. Radar ornithology is advancing rapidly and leading to significant discoveries about continent‐scale patterns of bird movements. General‐purpose and empirically validated methods to quantify biological signals in radar data are essential to the future development of this field. Mist Net can enable large‐scale, long‐term, and reproducible measurements of whole migration systems. 摘要: ⼀、⼤規模的氣象雷達網路是研究⿃類遷徙的全⽅位⼯具。US WSR‐88D 氣象雷達網路 覆蓋美國⼤陸，存擋從 1990 年代⾄今的雷達數據。這些數據可被⽤來量化⼤尺度到細尺 度上的⿃類活動，從⽽回答⼀系列⿃類遷徙的⽣態問題。然⽽，過去雷達數據分析仰賴⼤ 量的⼈⼒以區分雷達影像上的降⾬及⽣態活動。這個缺點侷限了利⽤歷史數據以分析⽣ 態活動的可能性。 ⼆、我們開發了 MISTNET，⼀個基於深度卷積神經網路的機器學習模型來分辨雷達數據 中的降⾬和⽣物訊號。不同於傳統的機器學習模型，MISTNET 可以做到細尺度的辨識， 並能從同時存在降⾬和⽣物訊號的雷達掃瞄中收集⽣物資訊。MISTNET 的設計基於影 像辨識的深度神經網路，並包含了針對雷達數據的特性所設計的架構。為了避免標記影 像所需的⼤量⼈⼒，我們使⽤從雙極化的雷達數據中獲得的帶有雜訊的影像標籤來訓練 MISTNET。 三、在歷史和近期的 WSR‐88D 雷達數據中，MISTNET 能辨識出⾄少 95.9% 以上的⽣ 物量，並且僅有 1.3% 假陽性錯誤率。透過保留和降⾬同時存在的⽣物訊號，MISTNET ⽐傳統過濾整張影像的⽅法多保留 15% 的⽣物量。MISTNET 實現了全⾃動化，能夠處 理多達百萬幅的雷達掃瞄數據集來產⽣空間上細尺度的辨識。這些特性成就 MISTNET 可廣泛⽤於各種應⽤，包含從⼤陸尺度到區域性的空域分析。 四、雷達⿃類學的發展迅速，並已經獲得了⼤陸尺度上⿃類活動的重⼤認識。通⽤於⼀般 ⽤途且可量化驗證的雷達⽣物訊號量化⽅法是這個領域未來發展的關鍵。MISTNET 可 ⽤於⼤規模且⾧時間的量測整體遷徙系統，且測量的結果是可以重複實現的。