面向綠色基礎設施的城市信息學：紐約市行道樹數據收集、分析與公眾科學的綜合研究

著：來源 譯：王鈺 校：林添懌

0 引言

大數據和物聯網的興起產生了觀察、測量、量化和分析城市的新方法，為研究城市動態系統提供了新的契機。在城市環境中收集的大批量、高精度、高質量的數據使我們能夠從不同尺度和維度對城市時空屬性進行研究。新的數據整合和分析進一步揭示了城市景觀與人類活動之間復雜的相互作用關系，以及其對社會—生態—經濟產生的長期影響。數據分析和人工智能技術已在城市基礎設施的各個子系統中得到廣泛應用。先前的相關研究已對城市環境中的交通運輸、能源利用、商業零售、人居活動以及其他社會經濟活動的時空動態進行了廣泛討論。

相比之下，城市樹木所具有的時空特征仍有待學者進行更加廣泛和深入的研究。在大部分城市中，有關城市綠色基礎設施（尤其是樹木）的數據挖掘、集成和分析相對較少，而且數據難以獲取。行道樹是城市景觀系統的重要組成部分，具有豐富的生態、環境和美學價值。然而，目前許多城市中行道樹的信息缺失限制了研究人員對當地城市林業進行準確的現狀評估和調查。具體體現在哪些因素（例如地理條件、城市形態、城市設計和當地社會經濟條件）影響了城市樹木的布局、位置，以及相應的空間格局在社區尺度上如何進一步影響當地的環境條件、公眾健康、歸屬感和生活質量。對城市景觀進行數字化分析有助于進一步探索“量化場所”（quantifying place）的方法，理解如何通過整合多源且異構的城市數據來獲取具有高時空分辨率和高維度的地點特征[1]。這種量化方法可以進一步將普適計算過程中產生的數據轉化成可以應用于具體地點的超本地化智能（hyper-local intelligence），從而通過數據分析和機器學習來實現城市環境中的地域與情境感知（locational and situational awareness）[2]。

本研究介紹了現今與城市樹木相關的數字化研究和實踐。文獻綜述總結了與行道樹密切相關的研究與實踐，例如行道樹在生態韌性、環境健康、城市設計質量、社區文化建設等方面如何發揮作用，以及城市信息學如何將樹木數據應用于城市設計規劃與數據驅動的運營管理。本研究介紹了公眾科學（citizen science）的概念，其作為一種參與性的數據收集方法如何能夠豐富城市林業的數字化過程和數據利用，并創造額外的社會效益和教育意義。隨后的案例研究聚焦于紐約市的樹木普查計劃，詳細說明了該城市樹木數據項目的動機、規劃、實施過程以及后續的研究應用。考慮到城市系統的生態—社會—技術復雜性，筆者提出了一個包含數據集成、協作和公民參與的城市信息框架。該框架總結了新數據和技術如何擴展我們對于城市林業時空動態的理解。

1 文獻綜述

1.1 作為綠色基礎設施的城市林業

行道樹作為城市綠色基礎設施的重要組成部分，具有改善局部小氣候、促進環境健康、定義社區景觀文化、提升生活品質的作用。從生態與環境健康角度出發，城市樹木可通過降低溫室氣體排放、調節溫度、凈化空氣等產生長期的環境效益、促進身心健康[3-4]。研究證明，行道樹的種植和管理是維持城市綠色基礎設施的重要組成部分[5]。據統計，紐約市城市林業對包括雨洪控制（每年總計34億L）和空氣凈化（每年總計減少 2 200t污染物）在內的生態環境影響，每年大約可產生1.22億美元（約7.97億人民幣）的經濟效益[6-7]。從城市設計的角度出發，樹木作為街景的重要組成部分，在車輛交通和行人之間具有建立視覺秩序、進行物理緩沖的作用，同時還可以提升公共空間的品質[8]。

行道樹通常是由規劃師和風景園林師共同設計的城市景觀。城市林業的空間格局不僅受到地理和氣候因素影響，還受到城市形態、規劃政策和設計決策的影響。隨著城市中精細粒度數據的增多，越來越多的研究開始對城市樹木的空間格局以及相應的生態環境影響進行量化和比對分析。例如，有研究收集了來自美國58個城市的數據以調查不同地區影響城市樹木覆蓋率的潛在因素。結果表明，城市局部的樹木覆蓋率主要取決于該區域具體的土地利用特征[9]。而城市樹木分布的空間差異會進一步形成不同的小氣候條件，具體體現在城市中不同地點的溫度差異。例如，一項對美國多個城市中樹木分布與局部地區溫度的比對分析表明，同一時間里，城市中不同地區的溫度差異最大高達12 ℃，溫度差具體取決于土地利用和樹木覆蓋率[10]。遺憾的是，在大多數情況下，這種空間差異并不是隨機偶然的，而是由之前的城市規劃決策和景觀設計造成的。樹木的空間差異也引發了規劃決策更深層次的問題，例如環境公平（environmental justice）倫理爭議和潛在的城市設計偏見。在美國城市中，這種空間差異通常反映出在城市發展歷史中由政策、規劃和設計偏見形成的景觀公平問題。先前對美國城市的研究表明，空氣污染、土地使用、林木覆蓋與弱勢社群（包括移民、少數族裔和低收入社區）的長期健康之間存在潛在聯系[11-12]。

城市林業對人居健康的影響，以及其在時間、空間和類型方面的分布特征帶來了更多復雜的問題和爭議。傳統觀點認為行道樹是寶貴的公共設施，可通過凈化空氣對環境產生積極的影響，同時也能促進更多的步行活動和運動鍛煉，以鼓勵積極的生活方式。然而，近年來也有一些研究質疑城市林業對當地環境的影響是否為絕對積極的，并考查行道樹作為花粉重要來源對城市生態環境和公民健康的影響。例如，2013年的一項研究在花粉高峰季節對紐約市45個地點的行道樹花粉散播情況進行了監測，結果表明，樹冠覆蓋區域的500m半徑輻射范圍內可接觸到的花粉量為當地總量的39%[13]。而在紐約市進行的另一項研究還證明，青年人群接觸樹木花粉（半徑為250 m）與過敏的風險在統計學上存在顯著相關性[14]。不止在紐約，其他研究在北美多個城市進行了專項調查，發現某些樹種可能會加劇人對花粉的過敏反應，從而增加哮喘的發病風險[15-16]。以上的爭議的解決需進行更全面的數據采集和分析研究，從而進一步揭示城市中人、生態和建成環境之間的復雜關系。

1.2 城市信息學

遙感技術、物聯網、計算機視覺、大數據和機器學習等技術的發展創造了新的數據源，并為城市科學研究開拓了新的方向。在討論智慧城市時，Ratti總結了城市系統中信息流的3個關鍵組成部分：感知、分析和執行[17]。感知是指通過傳感器、虛擬參與或人機交互中獲取實時數據的能力和過程[18]。分析是指通過數據分析、建模、模擬和可視化等手段來解決城市問題，通常強調應用程序和決策制定。與“商務分析”利用數據進行投資優化和業務運營相似，城市分析利用數據來論證與城市治理、規劃、設計、開發和管理相關的政策、決定和運營模式。執行是指利用安裝有傳感器的組件（例如自動電源開關）來執行的自動化控制或信息反饋行為。在智慧城市中，這套智能系統通常被稱為“傳感器和執行器網絡”。

通常，城市林業數據的生成基于高清衛星圖像或遙感影像，通過高光譜和激光雷達技術定位樹冠的位置。例如，有學者結合高光譜影像數據和高分辨率LiDAR數據來繪制美國加利福尼亞州圣塔芭芭拉市的樹木分布圖[19]。另一類研究方法基于街景圖像數據，使用計算機視覺和算法識別來實現更加復雜的圖像檢測分類技術。例如，Seiferling等使用計算機視覺算法來量化街道圖像（Google街景視圖）中的綠色像素，以估算街道樹木覆蓋程度[20]。這種新方法以人的視角模擬了行人在城市環境中對樹木的視覺感知。與此同時，最近的研究也顯示了這種技術的應用前景以及局限性。例如，一項研究探索了在美國加利福尼亞州的5個城市中利用街景圖像和深度學習進行街道樹木自動檢測的可行性[21]。盡管初步的結果表明自動識別記錄樹木地理位置的準確率較低（38%），但是此項探索性研究測試了一種新的城市信息收集方式，即通過圖像識別（例如衛星圖像或街景圖像）來自動更新現有的城市樹木數據庫。

城市信息學是一個新興的跨學科研究領域，旨在結合城市系統中的人文因素、環境因素和技術因素從而創造新的城市問題解決方案[22]。城市信息學也激發了城市景觀和綠色基礎設施相關的前沿探索研究。從技術的角度出發，城市信息學關注于數據的收集、挖掘、集成、分析和應用過程，并通過定量統計和數據科學來研究城市現象[23]。“開放數據”是指可公開使用和發布，且不涉及隱私、機密或安全性問題的數據集[24]。例如，紐約市于 2012年通過了《第11號地方法》（NYC Local Law 11，通常稱為《開放數據法》），要求市政府機構通過稱為“紐約市開放數據”的通用數字門戶網站提供可公開的城市數據[25]。通常，一個城市中的不同機構都會收集、管理和發布有關城市基礎設施系統和公共資產（包括土地使用、建筑物、街道網絡、行道樹和交通設施）的清單列表。盡管城市開放數據的最終目的是為了提升政務信息的透明度，促進數字創業和公眾參與，目前大多數此類信息資源仍是自上而下由市政府機構發布的。這些可供下載的城市數據大多與公共服務、政策管理和業務運營等方面相關。與此同時，為了推動政府開放場景，實現市民在智慧城市項目中的參與程度，并充分發揮城市開放數據的潛在價值，自下而上的社區項目和公眾參與對于支持、補充并最終主導一些數據收集決策至關重要。

1.3 公眾科學

除了遙感和計算機視覺技術外，基于眾包過程的城市數據收集也開始流行起來。盡管對于應用傳感技術和人工智能進行城市數據挖掘已經開展了廣泛探索，但此類數據收集方式的可行性和可靠性仍需要更加深入地論證。由于機器自動采集的數據準確率較低，而實際的城市樹木的信息系統和數據管理工作又會同時涉及人工參與和機器自動化學習，僅靠機器進行采集既不可行也不可取，所以公眾在城市景觀數字化、智能化的過程中扮演著不可或缺的角色。一項研究通過對美國和瑞典的多個城市的行道樹數據收集過程進行記錄分析并得出結論：經過6 h培訓的志愿者，可以相當準確地報告樹木的屬和種，準確率分別達到90.7%和84.6%[26]。這項研究還發現在所選擇的城市中，所有常見樹種中的槭樹屬、七葉樹屬、山楂屬、皂莢屬、懸鈴木屬和椴樹屬的報告的準確性較高。因此，當地居民的參與作為人工數據收集的一部分，有助于進一步檢驗現有數據庫的準確性。

除了提供技術支持外，此類參與過程還可以作為公共教育的一部分，以提升社區意識，并產生積極的社會效益。在公眾科學應用的進程中，當地居民能夠參與科學研究或試點項目以促進科學知識普及，并向研究人員提供本地經驗和反饋。此過程加強了市政單位、研究機構、利益相關者和社區成員之間的協同關系[27]。在行道樹數據收集過程中，公眾科學還可以讓群眾參與城市綠色基礎設施建設和社區韌性建設，培養社區意識。不過先前美國城市林業公眾科學項目相關的記錄表明，參與人群大多是受教育程度和收入水平相對較高的中產階級[28]。參與群體的不均衡性可能會引發規劃決策的公平性爭議，尤其是與開發用地規則、房屋所有權政策和設計決策相關的社區發展與貧富差異問題。因此，基于環境公平的考量，城市規劃部門鼓勵更多的少數族群或弱勢群體積極參與當地的公眾科學項目，從而使社區規劃和設計決策過程更加平等公正[29]。

2 案例研究：紐約市的城市樹木數據

紐約市是美國人口最多的城市，也是全球最大的城市之一。紐約市包括曼哈頓區、布魯克林區、皇后區、布朗克斯區和斯塔滕島區5個行政區，總人口超過830萬人。得益于其高密度的建成環境、街道網絡和公共交通設施，紐約市被評為北美最適宜步行的城市之一。在促進可持續發展與健康公平的城市生活方面，紐約市是城市規劃和設計實踐創新的先驅。自20世紀70年代以來，紐約市開展了許多始于草根基層或由社區主導的植樹運動，旨在促進少數族裔和邊緣人群享有更加公平的城市資源和更好的生活質量[30]。2007年，紐約市公園與游憩局推出了“植樹百萬在紐約”（Million Trees NYC）城市計劃。這是一項在全市范圍內開展的長期行動，計劃在5個行政區增種100萬棵樹木。在社區組織、當地合作伙伴以及大約50 000名公民和志愿者的支持下，這一計劃最終在2015年順利完成[31]。“植樹百萬在紐約”計劃的成功使其成為城市林業和景觀都市主義的實踐案例。與此同時，紐約市還是城市開放數據以及城市科學（urban science）的先驅。2012年，該市頒布了一項城市公共數據開放法案，以提高市政信息的透明度、問責制和可獲得性。隨后紐約市啟動了一個城市開放數據門戶“OpenNYC”，市民可以在該平臺查找、瀏覽和下載城市公共數據集。截至2019年底，該平臺已發布并管理多達2 167個數據集，包含建筑、交通、公共基礎設施、環境、社區和住房、人口、社會經濟活動等與城市相關的信息資源[32]。

紐約市于1995年發起了第一次樹木普查（“NYC TreesCount!”）。與人口普查類似，紐約市每10年通過調查來更新全市范圍的樹木信息。在2015年進行的第3次樹木普查中，紐約市公園與游憩局首次利用了手機應用程序“TreeKIT”，并邀請公眾參與樹木普查，包括在線教學、專業人員的現場培訓以及樹木實地調查等多個模塊。此次普查還為每位志愿者（被紐約市稱為“公眾科學家”）提供了一個工具包，其中包含用于測量樹木位置和標準周長的測量輪和卷尺，以及用于報告樹木信息的樹種識別圖冊等工具[33]。該活動自啟動以來，歷經一年的時間完成了樹木數據的收集、清理和驗證。最初的原始數據收集包括了666 134棵行道樹，其中的225 595棵是由 2 000多名居民參與，通過對當地社區樹木進行地圖標注和數字化記錄[33]所收集的（圖1）。

紐約樹木普查活動最終于2016年在當地數千名居民的參與下完成。最終692 892棵行道樹被繪制在地圖上，并在確認其所屬的 233個樹種后被納入正式的城市樹木普查數據庫[34]。通過對樹木普查得到的數據質量進行比對檢查，研究人員發現與受過專業培訓的樹木學家相比，以公民為主導的數據收集準確率達到了大約97%[35]。紐約市將最終的樹木普查數據通過網站“紐約行道樹地圖”（NYC street tree map）進行發布，并利用交互式數據可視化工具將全市5個行政區的行道樹統計結果進行共享[36]（圖2）。網站訪問者可以選擇特定的區域來進一步獲取更詳細的行道樹信息，例如地理位置、種類和大小（按樹干直徑）。搜索窗口允許用戶查詢特定街道地址周圍的行道樹情況。網站訪問者還可以注冊成為用戶，并對一些特定區域的樹木進行監測（例如用戶住宅附近的樹木）。據統計，截至2017年5月，紐約市行道樹地圖已經為10 000多項活動提供了數據支持，包括涉及特定的樹木維護、樹木普查數據更新以及其他城市環境生態監測項目[33]。

2 紐約行道樹地圖網站NYC street tree map website

為了進一步宣傳和推廣樹木普查數據，紐約市公園與游憩局于2016年6月舉辦了城市樹木數據大賽“TreesCount! Data Jam”。該活動的贊助商和合作伙伴包括了眾多私有企業、公共機構、教育研究機構以及倡導城市科技和公民數字化參與的非營利性組織，例如紐約市市長技術創新辦公室、NYC OpenData、微軟、地圖公司CartoDB（現更名為CARTO）、Civic Hall和BetaNYC。活動關注以下5個研究任務：1）統計1995年、2005—2015年的城市林業的變化；2）利用紐約市行道樹普查數據可視化開展城市環境意識的公共教育；3）調查行道樹與其他環境或社會經濟因素之間的潛在聯系；4）開發樹木數據在城市可持續發展和綠色基礎設施管理方面的創新應用；5）探索如何將社區管理和公眾參與更好地與城市林業實踐相結合[37]。

紐約市2015年的樹木普查項目歷時超過一年，如果加上前期準備工作和后期的推廣活動時間，整個計劃甚至需要更長的時間。正如本文文獻綜述部分提到的那樣，新的信息技術和人工智能可以從街景圖像中快速收集城市數據，這種創新方法也被應用到紐約市行道樹數據自動收集的測試之中。例如，一個研究小組利用谷歌街景（Google Street View, GSV）收集的336 998張紐約街景圖像開發了一種樹木檢測算法[20]。該團隊使用多步圖像分割和分類算法，根據街景圖像像素對各種街道對象進行了識別。根據對樹冠數據的驗證，該方法在街區或社區等尺度下相對準確（R2≈0.7），而其中產生的系統性誤差是由于從圖像提取的數據質量問題引起的，例如重疊的樹木和建筑物、樹木陰影以及曝光不足或曝光過度等問題。除了技術上的不確定性外，缺乏人為參與也限制了城市數據收集過程中所產生的額外社會效益。盡管存在上述種種缺陷，這種新穎的行道樹檢測方法是對傳統樹木普查的補充，為快速廣泛地收集城市林業數據提供了獨特的參考價值。

行道樹數據為日常的城市綠色基礎設施管理創造了長期價值，并為與城市規劃、景觀設計、生態環境可持續性以及公共衛生有關的研究提供了新的數據來源。盡管先前研究已經針對城市樹木的生態、環境和美學效益開展了廣泛的調查，但樹木的種植位置及其種類如何影響人類健康仍需更深入的研究。來源等[38]最近進行的一項研究調查了行道樹密度和樹種對呼吸系統健康的潛在影響，尤其是哮喘住院率與接觸性樹木花粉過敏之間的關系。該研究廣泛整合了城市開放數據，用以量化局部地區空氣質量（以PM2.5密度衡量）、呼吸道疾病患病率（以郵政編碼區的哮喘就醫率衡量）、土地利用、住房條件、鄰里人口統計學特征，以及按樹種分類的當地林木密度和花粉過敏原嚴重程度。地理加權回歸模型的結果顯示：在復雜的環境、建筑（室內居住條件）和社會經濟因素的驅動下，呼吸健康存在顯著的空間分布差異。回歸模型還表明：具有嚴重致敏花粉的樹木密度可能會增加局部地區尤其是弱勢社群中的哮喘住院率。

傳統研究主要關注城市林業作為城市綠色基礎設施的一部分對生態的貢獻，而新的數據則能夠進一步探索人居活動與城市樹木間的互動關系及其所呈現出的時空差異。例如，紐約市設有城市熱線311（NYC311）用于響應非緊急服務請求和市民投訴[39]。在2019年，該市共計約有4 400萬次市民熱線呼叫，包括了與鄰里環境、公共安全、噪音擾民和其他生活質量狀況有關的投訴或反映市政的服務需求[40]。其中多項市政服務與行道樹相關，例如報告受損或死亡樹木、要求修剪樹木或請求在特定位置種植新樹等。這樣的服務為居民參與當地社區環境建設提供了平臺，而在此過程中產生的數據可作為公眾對城市林業關注度的數字跟蹤。例如，筆者從2010—2020年，市民報告的超過2 200萬次 （n=22 131 777）的服務請求中提取了所有本地社區希望種植新樹木的請求。圖3是紐約4個主要行政區（曼哈頓區、布魯克林區、布朗克斯區和皇后區）按月匯總的市政服務需求與種植新樹木數量，按時間序列展開的數據可視化圖。市民服務熱線呼叫作為一種集體社會行為，反映了公民參與城市林業活動的季節性規律（圖3）。服務請求在不同地區顯示出類似的時間模式，尤其是在布魯克林區和布朗克斯區。每年，該市在 5月左右收到的種植樹木相關的服務請求最多，而在2月收到的請求最少。這可能是因為居民在5月進行戶外休閑或園藝活動的時間最長，從而對城市林業相關的工作關注度最高。由于物資運輸和工作準備，紐約市園林部門將2個種植季節定為3—5月和10—12月[41]，因此在旺季期間種植新樹木的市民服務請求通常可能需要等待半年才能得到解決。

3 與紐約街頭樹木有關的市民投訴和服務請求的季節性變化模式Seasonal patterns of citizen complaints and service requests related to street trees in NYC

行道樹普查與其他城市開放數據的收集整合提供了一種以高空間分辨率深入了解社區的物質、環境和社會方面的新視角。例如，圖4-1展示的是南布朗克斯的一個低收入社區——莫特黑文街區，2015年，紐約市衛生局在全市59個社區進行了人口健康調查，并發布了《紐約市社區健康概況》。該報告指出，莫特黑文是紐約市哮喘住院率最高的街區（兒童住院率最高，成年人住院率為第3高）[42]。之前對該社區的調查發現居民有超過66%的概率得哮喘住院，該社區健康問題或與鄰近土地的污染有關[43]。盡管鄰近的工業用地和土壤污染是影響當地環境空氣質量的關鍵原因，筆者的研究揭示了另一個可能的潛在原因——花粉過敏原暴露風險。通過對城市數據挖掘和整合，筆者基于不同樹種花粉的致敏程度（即引起花粉過敏的嚴重程度），在單棵樹木尺度上根據樹種繪制了城市行道樹花粉致敏程度地圖。結果顯示，由于莫特黑文社區的大多數行道樹為高致敏性（紅色），該街區是紐約市的花粉過敏多發地之一（圖4-2），筆者同時也根據NYC311收到的居民對于清理枯樹（圖4-3）或修剪樹木（圖4-4）的服務請求將數據進行了可視化分析。

4-1 紐約市布朗克斯區莫特黑文街區的行道樹 Street trees in Mott Haven, a neighborhood in Bronx, NYC4-2 莫特黑文街區的行道樹花粉致敏程度地圖Street trees allergenic map in Mott Haven4-3 居民對于清理枯樹的服務請求可視化分析圖Visualization of residents reporting dead trees according to NYC 311 service requests data4-4 居民對于修剪樹木的服務請求可視化分析圖Visualization of residents reporting overgrown trees (for pruning) according to NYC 311 service requests data

3 討論

行道樹的數據表明，城市作為一個復雜的社會—技術—生態系統，其自然生態、建筑環境、技術和社會經濟各方面都存在動態變化。本文文獻綜述部分通過總結比較不同方法的優缺點，闡釋了單純由人工主導的眾包（crowdsourcing）或者完全依靠人工智能進行數據收集都無法實現城市樹木數據收集管理的全部目的。例如，使用高分辨率的航空圖像數據或遙感數據在大尺度上統計樹木很容易實現，但是要根據特定的地理位置從人體尺度來對樹木進行精準定位則準確性較低。相比之下，以人為主導進行樹木普查得到的數據空間分辨率較高，但由于這種方式屬于重復勞動和時間密集型工作，人工與時間成本高，且僅依靠公眾參與難以保證數據質量。

考慮到上述不同技術各自的利弊，筆者認為應當將以人為主導和以機器為主導生成的城市樹木數據進行融合。首先，可以使用自動航空影像數據或LiDAR數據對樹冠基準數據進行收集，以刻畫城市樹木集群的總體格局。其次，可以根據計算機從街景圖像中提取的數據來修正基準數據，此過程可以進一步確定特定的樹干位置，尤其在建筑物密度極高的城市地區。接著，通過類似于紐約市樹木普查項目的眾包流程進一步豐富帶有地理標簽的信息，并增加社區參與度。最后，可以發布基于網絡或移動設備的具有可視化和其他應用程序的數據門戶，用來協助城市職能部門的規劃決策，也可利用該信息資源推廣城市環境公共教育并帶來其他社會福利。這種面向公眾的數字工具可以讓本地居民、社區團體和公共服務供應商通過更新和使用數據，使數據“實時”發揮關鍵作用。

筆者近期的另一項城市健康數據整合研究闡述了在數據項目中取得成功的基本原則和要素，包括清晰的技術方法、基于社會倫理爭議的考量，機構間相互協調的溝通模式、跨學科的合作研究以及可持續的伙伴關系[44]，這些原則同樣適用于城市林業數據項目。紐約市行道樹普查項目之所以取得成效，是因為其對城市數據基礎設施進行了長期的投資，通過有意義的市民參與使其與當地社區進行互動，并在數據發布后推廣使用和鼓勵后續研究。圖5展示了一個高層框架，該框架將數據、計算、環境、公民和城市管理一體化，成為一個互聯互通的城市信息系統。該系統基于信息反饋回路機制，將數據收集、分析、產品、用戶反饋聯系起來，從而幫助規劃、設計和與政策相關的決策論證與實際執行。這種互聯互通的內部連接實現了3個關鍵的城市信息反饋循環，即城市系統中的物質屬性、社會屬性以及科技屬性間的聯通。傳統的城市設計、建筑和風景園林學科廣泛研究了城市形態、街道景觀和公共空間的物理形態如何影響社會行為和公眾健康。“物質—科技”聯系代表了數字信息如何顯示和實體空間的規劃、設計和運營。這種連接類型主要是指用于決策支持系統、數據驅動操作和自動控制物理環境的智能城市技術。而“科技—社會”聯系是指數據和技術與人和社區之間產生的動態交互，尤其是如何平衡、協調信息技術對于教育、安全、健康和其他社會方面的勞動密集型服務帶來的積極效益和潛在風險。在用戶層面，通過這樣的人機交互實用工具或應用程序可以提供有效信息來支持人們的日常生活，并且建立新的信息反饋和數據聯通機制。例如，由于哮喘患者的高分辨時空數據（具有患者定位和時間的哮喘病例）的缺失，目前在更微觀尺度上對與健康相關的城市分析仍十分困難，而未來的研究將探索公眾參與如何作為用戶反饋數據被應用于公共健康領域。長久而言，持續的城市數據迭代更新以及信息應用產品將全面支持可持續城市管理和公眾科學。

紐約市行道樹普查的案例在美國并非個例。其他許多城市，包括舊金山、洛杉磯、西雅圖、丹佛、波士頓和芝加哥，也開展了類似的行道樹數據收集項目[45]（表1）。盡管許多城市已開展行道樹數據收集，但目前通常是由特定的城市機構獨立進行數據收集工作，缺乏國家級別的統一的城市樹木信息標準。因此，各個城市之間的數據無法直接進行整合和比較，未來城市樹木數據的收集管理工作仍需要更加明確的采集方法和通用標準。數據集的數量取決于城市的規模以及基于地理和氣候條件的城市樹木覆蓋范圍。各個城市的樹木數據由不同的部門機構收集和管理，其中包括公園與游憩局、交通運輸局和公共事務局。而這些數據標準和格式上的差異不可避免地限制了更大范圍的數據集成和跨城市信息交換[44]。這種跨城市的實踐不僅包括數據收集，還包括針對公共利益而開發的應用技術。城市樹木普查除了將數據進行可視化分析之外，也為不同城市測試這種新技術的可靠性創造了機會。新技術的使用必須要借鑒試點項目的成功經驗，并通過在其他城市進行推廣來進一步擴大應用范圍。例如，自從紐約市樹木普查獲得成功以來，TreeKIT的開發團隊開始在全球其他城市積極推廣類似的技術和產品，包括美國波士頓、美國阿爾伯克基、古巴哈瓦那和中國深圳[35]。

表 1美國部分城市樹木數據Tab. 1 A summary of all urban tree data in U.S. cities

中國城市林業在緩解空氣污染和熱島效應方面起著至關重要的作用。2015年，美國自然保護協會（The Nature Conservancy）與C40國際城市氣候領導聯盟（C40 Cities Climate Leadership Group）合作，對全球城市樹木的環境效益、社會效益和經濟效益進行了研究[46]。這項報告中有2項研究結果與中國城市高度關聯。首先，一項對全球城市植樹的投資回報率比較分析的結果表明，中國城市樹木規劃的投資回報率更高。其次，報告強調了以使用高空間分辨率的數據來確保在合適的地方植樹造林從而讓城市林業能夠服務最需要的人群社區的重要性。盡管之前的研究通過ALOS（Advanced Land Observation Satellite）和SPOT-5（Satellite pour l’observation de la Terre）等地球觀測衛星數據發現了中國9個主要城市的城市林業空間格局及其變化[47]，但大多數城市尚未公開提供全市范圍內帶有地理坐標的樹木數據。由于缺乏高空間分辨率的城市林業數據，城市范圍的科學研究和應用分析受到了一定限制。而在選定的研究區域或特定社區內，城市樹木對環境和生態的影響仍然有限。例如，最近一項研究通過在封閉的校園（以北京林業大學為例）內收集數據，檢測了不同樹種對空氣中顆粒物的吸收效果[48]。另一項研究是在北京城區采用空間隨機抽樣方法，以行道樹碳儲量為研究重點，使用基于道路網絡的分層隨機抽樣技術收集了12種、共計2 040棵行道樹的數據[49]。盡管這些項目使用了合理有效的采樣和分析方法，但由于目前城市尺度數據的缺乏，其研究范圍有限，未來全市樹木普查數據將為上述研究提供重要支持并使其研究模型得以在全市范圍內得到驗證。因此，全面而高質量的城市數據開放共享對于將特定地點的研究擴展到“城市科學”范圍至關重要。例如，近期的另一項研究在中國245個城市中分析了超過100萬張騰訊街景圖像來驗證大數據應用于城市林業的可行性，結果證明了街景圖像分類是一種便捷可行的城市林業數據收集和分析方法[50]。與上文提到的紐約樹木普查案例一樣，使用街景圖像的樹木測算方法很有可能將與以市民為主導的樹木數據收集活動相輔相成，在大規模城市數據收集的同時也充分發揮公眾科學的社會效益。而公眾參與城市數據收集過程所發揮的特殊作用也很好地回應了上文提出的結合數據、計算和公眾科學一體化的城市林業信息整體框架（圖5）。

5 結合數據、計算和公眾科學一體化的城市林業信息整體框架A framework integrating data, computation, and citizen science for urban forestry

4 結論

城市景觀隨著自然、人與機器之間復雜的相互作用而動態發展。瑞士風景園林師、設計理論家喬治·德斯科姆斯（Georges Descombes）認為：“景觀不存在完成或結束狀態，景觀是隨著事件和故事的積累，不斷發展的產物。”[51]數字化城市景觀或是智慧城市項目亦是如此，其過程是長期而復雜的，涉及定量和定性的分析以及設計方面的論證工作。這樣的過程不會也不應當完全依賴人工智能，亦不能拘泥于傳統的人工收集方式。本研究對用于收集和分析城市林業數據的新技術進行了綜述，并介紹了美國參與范圍最廣的城市林業項目之一——紐約市行道樹普查項目的起因、經過和結果。與NYC311服務請求數據相結合的描述性數據分析則揭示了包括社區人口健康和城市運營在內的復雜環境和社會影響因素。

城市林業中社會—生態—技術多方面的動態變化需要綜合運用信息科學、城市規劃、城市管理和社區參與等跨學科的方法對城市系統進行研究。由于城市林業涉及范圍的廣闊性和復雜性，城市樹木數據收集不應僅僅依靠人工調查或者完全由機器自動檢測。眾包數據收集和公眾參與項目不僅具有社會效益和教育意義，而且能夠提升公眾的數字意識，尤其是對城市大數據的認知。這些長期的努力將不斷促進城市中自然、科技和人文的有機聯系，并為未來城市科學的全面發展做出貢獻。

圖表來源：

圖1由作者使用QGIS創建，數據來源于2015年紐約市樹木普查數據以及紐約市公園與游憩局提供的公園和開放空間數據；圖2引自https://tree-map.nycgovparks.org；圖3數據來源于https://data.cityofnewyork.us/Social-Services/311-Service-Requests-from-2010-to-Present/erm2-nwe9，由作者從NYC311提取和匯總數據后創建的數據可視化圖像；圖4由作者使用NYC Tree Census和NYC 311數據創建的數據可視化圖像；圖5由作者繪制；表1由作者整理繪制。

（編輯/王亞鶯）

Urban Informatics for Green Infrastructure: An Integrated Approach for Street Trees Data Collection, Analytics, and Citizen Science in New York City

LAI Yuan

0 Introduction

The emergence of big data and the Internet of Things (IoT) create new research opportunities to investigate cities as dynamic systems through new observation, measurement, quantification, and analytics. Large volume, high granularity, and better-quality data collected in the urban environment enable us to look into cities’ spatialtemporal attributes at various scales. Novel data integration and analytics further unravel the complex interactions between urban landscape and human activity and the resultant long-term social-ecologicaleconomic impact. Advanced analytics and artificial intelligence have been widely implemented in various sub-systems of urban infrastructure. Previous studies have widely investigated spatial-temporal dynamics transportation, energy utility, retail, human mobility, and other socio-economic activities in the urban environment.

In contrast, the spatial — temporal characteristics of urban trees are still mostly unexplored. Data mining, integration, and analysis on urban green infrastructure, especially trees, are relevant new or not available in most cities. Street trees are critical components of urban landscape systems with ecological, environmental, and aesthetic benefits. However, a lack of urban tree data constrains accurate assessment and datadriven investigation of urban forestry’s current status. In particular, what factors (e.g., geography, urban form, urban design, and local socioeconomic conditions) drive urban trees’ location, and how do such spatial patterns further shape the local environmental condition, population health, sense of place, and quality of life at neighborhood scale? Digitization and analytics of urban landscape contributes to a broader vision of “quantifying place” for gaining a high-dimensional understanding of places at high spatial resolution by integrating heterogeneous data sources[1]. Such quantification can further support hyper-local intelligence for pervasive computing applications, which enables locational and situational awareness through data analytics and machine learning[2].

This article provides an overview of current research and practice related digitalization of urban trees. The introduction first describes the background context of trees as a vital part of urban green infrastructure, regarding their ecological, economic, social benefits, and how new data sources and novel analytical methods can extend our understandings in urban forestry. The literature review section summarizes critical aspects that are highly relevant to street tree data research，including previous findings on how street trees contribute to ecological resilience, environmental health, urban design quality, and social coherence with a sense of community. The author overviews urban informatics extends technical capacity and information resources for quantitative research for landscape design and data-driven operation to support urban management. Citizen science as a participatory data collection approach enriches the digitization process of urban forestry with additional social and educational benefits. A case study follows, focusing on NYC Tree Census Program in New York City. This case study provides detailed explanations on the motivation, plan, conduction process, and research and applications that utilize this dataset. The discussion summarizes the experience learned and critical insights from NYC’s case and relevant controversies and ongoing investigations. With considerations regarding cities’ ecological-social-technical complexities, the author proposes a framework for urban forestry data integration, collaboration, and civic participation. This framework concludes how new data and technology extend the understanding of the spatial-temporal dynamics of urban forestry.

1 Literature Review

1.1 Urban Forestry as Green Infrastructure

Street trees are a vital part of urban green infrastructure contributing to local microclimate, environment health, a sense of place, and qualityof-life necessitates. Ecologically, urban trees provide long-term environmental benefits, including greenhouse gas emission reduction, cooling effect, and air purification to promote physical and mental health[3-4]. Previous research proves the importance of planting and managing street trees as a vital part of urban green infrastructure[5]. It is estimated that in NYC the ecological impact of urban forestry includes stormwater mitigation (3 billion and 4 hundred million liter per year in total) and air purification (2,200tons of pollutants reduction per year in total), representing an estimated $122 million (7.97 RMB) economic benefits[6-7]. From an urban design aspect, trees are significant streetscape components that create visual order and physical buffer between motorists and pedestrian, contributing to the quality of public space at the human scale[8].

Street trees are urban landscape designed by planners and landscape architects. The spatial pattern of urban forestry is not only shaped by geographical and climate factors, but determined by urban form, planning policy, and design decisions as well. With more granular data available, increasing studies start to quantify and compare urban tree patterns and ecological and environmental impact. Previous research collected data from 58 U.S. cities to investigate underlying drivers of urban tree coverage across different regions. Results indicate that within a particular city, local-scale spatial variation of tree coverage mainly depends on land use characteristics[9]. One direct outcome of such spatial variation is the different microclimate conditions, especially the temperature differences at the hyper-local scale. For example, one study measured multiple U.S. cities’ temperature at high spatial-temporal resolution and local trees. The results show that different parts of a city may experience different temperature as large as nearly 12 Celsius Degree, depending on land use and tree canopy coverage[10]. Unfortunately, in most cases, such patterns are neither random nor accidental but derived by previous urban planning decisions and landscape architecture design interventions. The spatial disparity of trees raises questions on planning decisions and deeper issues such as environmental justice and underlying urban design biases. In the U.S. cities, such spatial variation often reflects environmental justice issues shaped by policy, planning, and design biases through urban development history. Recent research reveals the underlying connections between ambient air pollution exposure, land use, street trees coverage, and long-term health differences among vulnerable population including immigrants, ethnic minorities, and low-income communities[11-12].

The spatial, temporal, and typological variation of urban forestry’s impact even brings more complexities and controversies. Traditionally, street trees are considered valuable public amenities that create positive environmental effects and impact local population health by cleaning the air and promoting an active lifestyle with more walking and physical exercises. However, some studies question whether urban forestry always brings positive impacts to the local environment, considering street trees as significant sources of pollen in cities. In NYC, a previous study monitored tree pollen at 45 sites during the pollen season in 2013. Results indicate that tree canopy cover with 500meters radial zone explained 39% local pollen exposure variation[13]. Separate research in NYC also prove a statistically significant relationship between local tree pollen exposure (0.25 km radius) and risk of developing allergic sensitization in the youth population[14]. Not just in NYC, different groups of researchers conducted separate investigations in multiple cities in North America, and find out certain tree species may exacerbate allergic reactions to pollen with increased risk of asthma[15-16]. All the above-unsolved questions and controversies require better data and more robust analytics to unpack the complicated relationship between people, ecology, and built environment in cities.

1.2 Urban Informatics

Recent technological development in remote sensing, the Internet of Things (IoT), computer vision, big data, and machine learning has created novel data sources and untapped new research opportunities for cities’ scientific research. In a discussion of smart cities, Ratti summarized three critical components of the information flows in urban systems: instrumentation, analytics, and actuators[17]. Instrumentation refers to the capacity (as well as the process) of acquiring new-real-time and real-world data from physical sensors, virtual participations, or human-computer interactions[18]. In cities, analytics refers to the (big) data analysis, modeling, simulation, and visualization for urban problem-solving, often emphasizing application and decision-making. Equivalent to the popular term “business analytics” that focuses on leveraging data for investment optimizations and business operations, urban analytics utilizes data for policy, decisions, and operation relevant to urban governance, planning, design, development, and management. Actuators refer to the physical or digital components (e.g., an automated power switch) in charge of execution according to the instrumented sensors. Such intelligence systems often refer as the “sensor and actuator networks” in smart cities.

Conventionally, the generation of urban forestry data relies on satellite imagery data or remote sensing, such as mapping street trees location via hyperspectral and lidar technology. For example, a previous study utilized combined top-view hyperspectral imagery data and highresolution LiDAR data to map trees’ location and species in Santa Barbara, California, U.S.[19]. Increasing data from or street-level imagery data enable more sophisticated image detection/classification techniques using computer vision. For example, Seiferling et al. used a computer vision algorithm to quantify how “green” in street-level images (Google Street Views) as a proxy of tree cover in cities[20]. These new methods mimic the identification of trees in the urban environment from a human perspective. Recent research proves promising application of such technology as well as current constraints. For example, one study explored possible automated detection utilizing deep learning in five cities of California, USA[21]. Although the preliminary results indicate a low accuracy rate (38%) to correctly identify the documented trees’ geo-location, this exploration tested an algorithm to process alternative data (e.g., satellite images or street views) to update the existing urban trees database automatically.

Urban informatics is an emerging transdisciplinary field that investigates and implements innovative solutions at the intersections of people, places, and technology[22]. New technology also inspired new research investigations relevant to the urban landscape and green infrastructure. From a technical aspect, urban informatics investigates urban phenomena through quantitative and computational approaches involving data collection, mining, integration, analytics, and applications[23]. “Open Data” refer datasets that are publicly available for use and distribution without restrictions regarding privacy, confidentiality, or security concerns[24]. In 2012, NYC passed Local Law 11 of 2012 (often known as the “Open Data Law”) that requires city agencies to make administrative data publicly available through a common digital portal known as NYC Open Data[25]. Each agency collects, manages, and publishes a digital inventory of the physical component of urban infrastructure systems and public assets, including land use, buildings, street network, street trees, and transit facilities. While city open data intend to promote more transparent governance, digital entrepreneurship, and civic engagement, it has been mostly topdown, led by the city agencies publishing publicaccessible records involving public services, policy administration, and business operation. To fulfill the purpose of city open data, bottom-up efforts from citizen science projects are critical for supporting, complementing, and eventually leading some data collection decisions.

1.3 Citizen Science

In addition to remote sensing or computer vision technology, data collection based on a crowdsourcing process starts to gain popularity in cities. While urban data mining through sensing technology and artificial intelligence are widely explored, it is crucial to consider the feasibility and reliability during data collection. Mostly, information systems and data management streams for urban trees involve both human engagement and automation utilizing machine learning. Several reasons make human participation necessary in the urban landscape digitization process. Considering the limitations above, solely relying on automated data collection is neither feasible nor desirable due to its low accuracy. One study reviews the street tree data collection in multiple cities in U.S. and Sweden and concludes that with a 6-hour training session, volunteer participates were able to report tree genera and species quite accurately, with an accuracy rate of 90.7% and 84.6%[26]. This study also finds that Acer, Aesculus, Crataegus, Gleditsia, Platanus, and Tilia report the highest reporting accuracy among all common tree genera in selected cities. Thus, local residents’ involvement as a part of human input contributes to further inspecting, validating existing databases.

Besides a purely technical perspective, such engagement processes could serve as a part of public education to build a sense of community with social benefits. Citizen science is a process that general public and residents participate in scientific research or pilot projects for advancing scientific knowledge, providing local experience and feedback to the researchers, as well as constructing robust synergy between the city agency, research institutions, stakeholders, and community members[27]. Potentially, citizen science projects in street trees can cultivate a sense of community and citizen empowerment involving urban green infrastructure and community resilience. However, the historical records on urban forestry projects indicate the participates are predominately middle-class population with relatively higher education attainment and income[28]. Such imbalanced involvement raises equity issues, especially considering the historical spatial injustice embedded in previous zoning regulations, homeownership policy, and planning decisions. For environmental justice, cities need to proactively engage underrepresented or disadvantaged communities for more meaningful participation in planning and design decision-making process in their neighborhoods[29].

2 Case Study: Urban Trees Data in New York City

New York City (NYC) is the most populous city in the U.S. and one of the largest global cities. The City includes five boroughs, including Manhattan, Brooklyn, Queens, the Bronx, and State Island, with over a total of 8.3 million people. Thanks to its built density, street grids, and public transit network, NYC is often ranked as one of the most walkable North American cities. The City has been pioneering in urban planning and design practices to promote more sustainable, healthy, and equitable urban living. The City has a legacy of urban forestry, especially the grassroot, communityled tree planting movement for more equitable and better quality-of-life in minority and marginalized neighborhood since 1970s[30]. In 2007, the NYC Department of Parks and Recreation launched MillionTressNYC, a citywide long-term program aiming to plant 1 million trees in five boroughs. With supports from the community-based organizations, local partners, and approximate 50,000citizens and volunteers, the City achieved this goal in 2015[31]. The triumphant story of MillionTressNYC becomes one of the best practices in urban forestry and landscape urbanism. Meanwhile, NYC is also a pioneer in collecting and utilizing open data for applied urban science and analytics. In 2012, the city signed a local law 11 (often referred to as “open data law”) to increase the transparency, accountability, and accessibility of city-data. The city has launched OpenNYC, an urban open data portal for citizens to view, explore, and download public city datasets through this website. By the end of 2019, OpenNYC has published and managed 2,167 datasets that related to the buildings, transportation, utility infrastructure, environment, neighborhood and housing, population, and socio-economic activities[32].

NYC launched its first tree census in 1995. Similar to the population census, the City conducts a survey and update its tree inventory decennially. During the third tree census conducted in 2015, the NYC Department of Parks and Recreation implemented a mobile crowdsourcing application for tree census for the first-time (“TreeKIT”). Typical participation includes an online tutorial, an on-site training tour with a professional staff, and an on-site tree survey process. The program also provides each volunteer (named as “citizen scientist” by the City) a toolkit with various tools, including a measuring wheel and a tape measure for measuring the location and standard circumference of the trees, and a tree identification guide for reporting the tree species[33]. Since its first launch, the program took one year for the City to collect, clean, organize and validate the data collected from the tree counting program. The initial data collection identified 666,134 street trees, including 225,595 trees mapped and digitized by more than 2000residents through this participatory process[33]. Fig. 1 visualized all documented trees within the 2015 NYC Tree Census, with different colors by species and parkland area in grey.

By 2016, the NYC TreesCount! project completed along with thousands of local citizens’ contributions. The final Tree Census mapped 692,892 street trees and digitized them into the census database with 233 species identified[34]. A data inspection of tree census data quality concludes that citizen-led data collection has approximate 97% accuracy comparing to a professionallytrained arborists[35]. The City shares the final tree census data by publishing a website (“NYC Street Tree Map”) with interactive data visualization and citywide statistical summary of street trees across five boroughs[36](Fig. 2). A website visitor can select specific districts to zoom in for more detailed street tree information such as geolocation, species, and size (by trunk diameter). The search window allows a user to learn about street trees around the specific street address. Besides, website visitors can register as regular users to monitor specific trees, mostly near their home location. It estimates that by May 2017, the NYC Street Tree Map has supported more than 10,000activities involving specific tree maintenance, census data update, and other urban ecological monitoring programs[33].

In June 2016, the NYC Department of Parks and Recreation further organization TreesCount! Data Jam in the same year to promote this data advocacy and outreach for private, public, nonprofit, and educational organizations for data partnerships. Sponsors of this event include NYC Mayor’s Office of Tech and Innovation, NYC OpenData, Microsoft, CartoDB (now rebranded as CARTO), Civic Hall, and BetaNYC, a non-profit organization advocating urban technology advocacy and civic digital participation in NYC. According to the organizer, this one-day event aims to tackle the following five questions, including: 1) urban forestry change through time between 1995, 2005, and 2015; 2) novel data visualization of NYC street tree census for public education; 3) investigate underlying relationships between street trees and other environmental or socioeconomic factors; 4) create potential applications of tree data for urban sustainability and green infrastructure management; and 5) explore the better practice and use cases for community stewardship and public engagement with urban forestry[37].

The crowdsourcing process for the 2015 NYC Tree Census takes more than one year, and the entire program even takes much longer when taking into account preparation and post-release promotion events. As mentioned in the literature review, recent new data and artificial intelligence enable rapid urban data collection from street view images. Such innovative methods have also been tested in NYC for an automated street tree data collection. For example, a research team developed a tree detection algorithm utilizing 336,998 images collected from Google Street View (GSV) in NYC[20]. Using a multi-step image segmentation and classification method, the team classified various street objects based on street view image pixels. Validation with the tree canopy data proves a moderate accuracy (R2value ≈ 0.7) at the block or community district level. Systemic errors are due to some data quality issues derived from GSV images, including the overlapping trees and buildings, tree shadows, and underexposure or overexposure. Besides the technical uncertainties, a lack of human engagement constrains additional social benefit of community engagement throughout the process of urban data collection. Nonetheless, this novel street tree detection method is complementary to the tree census, providing unique value for rapid and scalable urban forestry data collection.

Street tree data create prolonged values for day-to-day urban green infrastructure management and provides a new data source for research related to urban planning, landscape architecture, ecology, and environmental sustainability, and public health. One recent study conducted by the author investigated the underlying spatial impact of street tree density and species to respiratory health, especially relationships between the prevalence of asthma hospitalization and allergic tree pollen exposure[38]. Although urban trees’ ecological, environmental, and aesthetic benefits have been widely investigated, how specific location and species of trees may have localized health effects is still less explored. Thus, this study first integrates a wide range of city open data to quantify localized air quality (measure by PM2.5density), respiratory disease prevalence (measured by ZIP code level asthma hospitalization rate), land use, housing conditions, neighborhood demographic characteristics, along with local street trees density and pollen allergen severity classified by tree species. The geographically weighted multivariate regression model’s results reveal a significant spatial disparity of respiratory health driven by complex environmental, built (indoor housing condition), and socioeconomic factors. The regression model also indicates that the concentration of tree species with severe allergenic pollen may increase local asthma prevalence, especially in vulnerable populations.

While conventional studies focus on urban forestry’s environmental and ecological contribution, a novel aspect of urban green infrastructure research explores human activities as societal behaviors interacting with street trees. Such citizen engagement with street trees demonstrates spatial and temporal patterns as well. NYC operates a city hotline, NYC311, for responding non-emergency service requests[39]. In 2019, the city reports approximate 44 million citizen interactions with NYC311 in one year, reporting complaints or requests that relate to neighborhood environment, public safety, noise, and other quality-of-life conditions[40]. Multiple service categories are associated with street trees, such as reporting damaged or dead trees, calling for pruning services, or requesting for new trees at a specific location. Such service data provide an additional digital trace of citizens’ engagement with the local community environment and public attention to NYC’s urban forestry. For example, the author extracted all local requests for new trees from more than 22 million (n=22,131,777) service requests reported in recent ten years (2010—2020). Fig. 3 is a time-series data visualization of monthlyaggregated service calls requesting news trees in four major boroughs, including Manhattan, Brooklyn, the Bronx, and Queens. The x-axis represents time by months, and the y-axis represents the monthly total number of requests. As this fig. shows, the temporal patterns service calls, as the digital representation of collective social behavior, reveals the seasonality of citizens’ engagement with urban forestry. Monthly patterns of service requests demonstrate similar temporal patterns in different boroughs, especially in Brooklyn and Bronx. Annually, the city receives most service requests around May and the least requests in February. One assumption is that residents spent the most time outdoor leisure activity or gardening during May, generating the greatest public attention on urban forestry maintenance ace and expansion. However, due to stuff logistics and work preparation, the department scheduled two planting seasons as March-May, and October-December[41]. This indicates that residents who have requested new trees during the peak season may wait for half-year to complete this service request.

Novel data integration among street tree census and other city open data provides additional insights into a neighborhood’s physical, environmental, and social aspects at high spatial resolution. Fig. 4-1 highlights Mott Haven, a low-income neighborhood in South Bronx. In 2015, the NYC Department of Health conducted a citywide population health survey among all 59 neighborhood districts and summarized the research report as NYC Community Health Profiles. The report identified that Mott Haven as the neighborhood with the highest asthma hospitalization rate (highest among children and the third-highest among adults)[42]. Previous investigations in this specific neighborhood find out residents face more than 66% likelihood in asthma hospitalization, possibly due to the approximation to noxious land use areas[43]. While the local ambient air quality close to industrial or toxic land uses is a critical issue, our study reveals another potential cause — pollen allergen exposure. Using data mining and integration, the author classifies pollen allergenicity (i.e., the severity of tree pollen causing raspatory reactions such as asthma) at individual tree level and map street trees based on allergenicity. As Fig. 4-2 shown, most street trees are classified as highly allergenic, making this neighborhood one of the pollen hotspots in NYC. Fig. 4-3～4-4 visualize residents reporting dead trees or overgrown trees (for pruning) according to NYC 311 service requests data.

3 Discussion

Urban street tree data demonstrate that cities as complex social-technical-ecological systems involve nature, built environment, technology, and socioeconomic dynamics. As the literature review summarized the strengths and weaknesses in different approaches, we consider that neither human-led crowdsourcing nor A.I. based data collection can serve the full purpose of urban tree data collection. For example, tree counting based on canopy detection using high-resolution aerial imagery data or remote-sensing data (e.g., LiDAR) can be easily implemented at a large regional scale. However, there is a lack of human-scale and low accuracy for pinpointing trees based on their specific geo-location. In contrast, human-led collection reports data at the high spatial resolution, but it is labor and timeintensive, lacking data quality control, especially when engaging the general public through a participatory crowdsourcing process.

We propose a fusion of human-led and machine-lead data generation of urban trees for cities regarding the aforementioned pros and cons related to a different technology. First, a baseline data collection on tree canopy can be conducted using automated aerial imagery data or LiDAR data. This initial data collection enables cities to capture the tree canopy and urban forestry clusters’ general pattern. Further, the baseline data can be improved with data collected from object detection with computer visions from street view images. This process can further identify specific tree trunk locations, especially in urban areas with extremely high building density. Then, geo-tagged information can be further enriched through a crowdsourcing process similar to the NYC Tree Census, with additional community engagement components. Finally, a web- or mobile-based data portal with visualization and other applications can be published for the public education and other social benefits. Such a residentsfacing digital tool plays a crucial role in making the data “live” with updates from local residents, community groups, and public service providers. A recent research on data integration for urban health, the author identified essential principles to succeed in a data project, including methodological clarity, social and ethical awareness, inter-agency coordination, transdisciplinary collaboration, and sustainable partnerships[44]. Such principles also apply to urban forestry data management. NYC Street Tree Census project is successful thanks to a long-term commitment to invest in urban data infrastructure, engage with the local community for meaningful participation, and promote extended use cases and research after releasing the data. Fig. 5 illustrates a high-level framework that integrates data, computation, environment, citizens, and city management as a connected, holistic system. Such a system connects the information feedback loop with data collection, analytics, civic product, user feedback as new input to inform planning, design, and policyrelated decisions.

The inner connections enable three critical information feedback loops as physical-social, cyber- and cyber-social connections in cities. Traditionally, urban design, architecture, and landscape architecture research have extensively investigated how physical configuration of urban form, streetscape, and public space shape collective social behavior and population wellness (i.e., physical-social connection). Some well-known research include studies conducted by William H. Whyte and Jan Gehl, who investigated the interplay between physical space and human behavior. Physical-cyber connections represent how digital information represents and augment physical planning, design, and operation in reality.

This type of connection primarily refers to smart city technology for decision-support systems, data-driven operation, and automated control physical environment. Cyber-social connections refer to the dynamic interactions among data and technology with people and communities. In particular, how to positively aid social benefits while minimizing underlying adverse effects when utilizing information technology and artificial intelligence for human-intensive services involving education, safety, health, and other social aspects. At a user level, practical tools or applications can provide useful information to support people’s daily lives and, hopefully, get feedback and new data input through such user-machine interactions. For example, currently, there is no high-resolution data on asthma patients (e.g., asthma cases with patients’ location and time), which creates barriers for further validating some health-related spatial analysis at a more granular scale. Thus, future research will explore how participatory sensing and public engagement can serve as part of ground-truth validation and user-based data input on population health. Ultimately, such iterative analytical cycles will support sustainable urban data management and holistic citizen science.

The story of the NYC Street Tree Census is not a single case in the U.S. As Tab. 1 shown, many other cities, including San Francisco, Los Angeles, Seattle, Denver, Boston, and Chicago, have conducted similar projects to collect street trees data[45]. While many cities started to invest data collection on street trees, these efforts are often siloed and solely determined by specific city agencies without a generalizable data standard specifying common variable and spatial resolution. The datasets’ volume varies depending on the size of the city and urban forestry coverage based on local geography and climate condition. Urban tree data are collected and managed by different departmental agencies in cities, ranging from the Department of Parks and Recreation, Department of Transportation to the Department of Public Works. Such differences in data standards and format inevitably create constraints for greater data integration efforts and cross-city information exchange protocols[44]. Such cross-cities practice does not just limit to data collection but also involves developing and deploying technology for civic purpose and the public good. Besides data visualization and analytics, the urban tree census provides unique opportunities for testing civic technology implementation in different cities. The novel technology deployment must continue the pilot project’s success and extend the scope by testing and deploying in other cities. Since the success of the NYC Tree Census, the developer of TreeKIT has been actively deploying similar technology and product in other cities worldwide, including Boston, Havana, Shenzhen, and Albuquerque[35].

Reflecting on China’s cities, urban forestry plays a critical role in alleviating air pollution and heat island effects. In 2015, The Nature Conservancy, in a partnership with the C40 Cities Climate Leadership Group, conducted research on the environmental, social, and economic benefits of urban trees in global cities[46]. This report provides two insights that are highly relevant to Chinese cities. First, a comparative estimation of the return on investment (ROI) of tree planting in cities worldwide indicates a much higher ROI for tree planning in Chinese cities. Second, the report highlights the importance of targeting urban forestry at the high spatial resolution to ensure planting trees in the right places to benefit the most-needed population. While previous studies using remote sensing or satellite images to discover urban forestry’s spatial patterns and its changes in nine major Chinese cities using ALOS (Advanced Land Observation Satellite) and SPOT-5 (Systeme Probatoire d’Observation de la Terre) data[47], citywide geo-tagged trees data is not yet publicly available in most cities in China. A lack of urban forestry data at high spatial resolution create constrain for conducting scientific research and applied analytics at an urban scale. The environmental and ecological impact of urban trees is still limited within a selected study area or a specific neighborhood. For example, recent research measured airborne particulate matter mitigation effectiveness by different tree species, based on data collected within an enclosed campus site (in Beijing Forestry University)[48]. An alternative approach is to adopt a randomized spatial sampling technique in the city area. One study focusing on carbon storage of street trees in Beijing, as another example, uses a stratified random sampling technique based on the road networks and collects 2040 street trees in 12 species[49]. While these studies conducted research properly with valid methods and analysis, a lack of city-scale data constrains extending the work scope. Citywide trees census will significantly support the above studies with scaled-up spatial analysis and model validation. Thus, a comprehensive and high-quality city open data is essential for scaling up research from a site-specific study into “urban science” as more extensive investigations with broader impact. Recent research also explored the feasibility of measuring urban forestry using over 1 million street view images from Tencent Street View in 245 major cities in China[50]. The results prove that street view image classification is a feasible and scalable method for rapid urban forestry data collection and analytics. Like the NYC case mentioned above, the tree detection algorithm using street view images is most likely to be a complementary approach to integrate with citizen-led tree data collection regarding the social benefit of the urban data crowdsourcing program. The special role of human engagement in urban data collection process resonates with the proposed integration framework (Fig. 5) as well.

4 Conclusion

Urban landscape dynamically evolves with complex interplays among nature, humans, and machines. Georges Descombes, a Swiss landscape architect, and design theorist describes that “landscape is never finished or completed, like a can of preserves; it is an accumulation of events and stories, a continuously unfolding inheritance.”[51]Digitizing the urban landscape involves a complicated process and requires quantitative, qualitative, and design expertise. Such a process will and shall not be solely relied on human labor nor entirely by A.I. automation. This article provides an extensive overview of recent technological innovations for collecting and analyzing urban forestry data. Using the NYC Street Tree Census as an example, the author describes the motivation, process, and results from one of the largest participatory urban forestry projects in the U.S. A descriptive data analysis integrating with NYC 311 service request data reveals the complex environmental and social factors involving neighborhood population health and city operations.

The multi-faceted social-ecologicaltechnical dynamics in urban forestry require a transdisciplinary view on urban systems and an integrated approach that connects information science, urban planning, city management, and community engagement. Due to the vast scope of urban forestry and complexities mentioned above, urban tree data collection should not solely rely on manual surveys or automated detection entirely conducted by the machine. The participatory nature of crowdsourcing data collection and citizen science projects brings social and educational benefits, raises digital awareness, and makes big data tangible and usable for the general public. Such efforts eventually connect nature, technology, and people in cities and contribute to holistic urban science.

Sources of Figures and Table:
Fig. 1 ? the author using QGIS with 2015 NYC Tree Census data and Parks and Open Space data provided by the City of New York Department of Parks and Recreation; Fig. 2 ? the NYC Department on NYC Street Tree Map, retrieved from https://tree-map.nycgovparks.org; Fig. 3 ? the author after extracting, aggregating, and visualize data from NYC311, data source from https://data.cityofnewyork.us/Social-Services/311-Service-Requests-from-2010-to-Present/erm2-nwe9; Fig. 4 ? the author using both NYC Tree Census and NYC 311 data; Fig. 5 ? the author; Tab. 1 organized by the author.

(Editor / WANG Yaying)