陳國(guó)華,郭品文

南京信息工程大學(xué) 氣象災(zāi)害教育部重點(diǎn)實(shí)驗(yàn)室,江蘇 南京 210044

?

中國(guó)地面氣溫統(tǒng)計(jì)降尺度預(yù)報(bào)方法研究

陳國(guó)華*,郭品文

南京信息工程大學(xué) 氣象災(zāi)害教育部重點(diǎn)實(shí)驗(yàn)室,江蘇 南京 210044

2013-02-06收稿，2013-05-16接受

公益性行業(yè)(氣象)科研專項(xiàng)(GYHY201006017)

摘要利用中國(guó)752個(gè)基本、基準(zhǔn)地面氣象觀測(cè)站2000—2010年地面溫度日值數(shù)據(jù),采用具有自適應(yīng)特征的Kalman濾波類型的遞減平均統(tǒng)計(jì)降尺度技術(shù),對(duì)中國(guó)地面溫度進(jìn)行精細(xì)化預(yù)報(bào)研究。分析該方案的降尺度效果,并與常用插值降尺度方法進(jìn)行比較。結(jié)果表明:1)遞減平均統(tǒng)計(jì)降尺度技術(shù)相比插值方法有較大的提高,顯著減小東西部預(yù)報(bào)效果差異,1～3 d預(yù)報(bào)的均方根誤差減小了1.4 ℃;2)該方案1～3 d預(yù)報(bào)的均方根誤差為1.5 ℃,預(yù)報(bào)誤差從東南地區(qū)(均方根誤差為1.4 ℃)向西北地區(qū)(均方根誤差為1.8 ℃)逐漸增大,并且預(yù)報(bào)效果夏季優(yōu)于冬季。因此,遞減平均統(tǒng)計(jì)降尺度技術(shù)對(duì)中國(guó)地面溫度進(jìn)行精細(xì)化預(yù)報(bào)是可行的。

關(guān)鍵詞

精細(xì)化預(yù)報(bào)

統(tǒng)計(jì)降尺度

插值

遞減平均法

實(shí)現(xiàn)精細(xì)化天氣預(yù)報(bào)最關(guān)鍵和有效的技術(shù)手段就是降尺度技術(shù),主要有嵌套高分辨率數(shù)值模式的動(dòng)力降尺度技術(shù)和基于現(xiàn)有大尺度數(shù)值產(chǎn)品的統(tǒng)計(jì)降尺度技術(shù)(劉永和等,2011)。動(dòng)力降尺度的物理意義明確,是長(zhǎng)期發(fā)展的根本方向,但它的缺點(diǎn)是消耗的計(jì)算量大,模式的性能受所提供的邊界條件的影響很大,并且預(yù)報(bào)效果還達(dá)不到精細(xì)化預(yù)報(bào)和服務(wù)的需求(陳豫英等,2005)。就上述缺陷,目前各國(guó)學(xué)者主要使用統(tǒng)計(jì)降尺度方法解決。統(tǒng)計(jì)降尺度通過建立不同尺度間的關(guān)系來實(shí)現(xiàn)尺度間的轉(zhuǎn)換,其優(yōu)點(diǎn)在于它可以和現(xiàn)有數(shù)值模式產(chǎn)品相配合,能較好地解決現(xiàn)有動(dòng)力降尺度法的不足,計(jì)算量小、容易實(shí)現(xiàn),是提高氣象要素精細(xì)化預(yù)報(bào)水平更為直接有效的手段。Kidson and Thompson(1998)和Mearns et al.(1999)均用統(tǒng)計(jì)降尺度和動(dòng)力降尺度兩種方法進(jìn)行氣候預(yù)測(cè),認(rèn)為動(dòng)力降尺度雖有更好的物理基礎(chǔ)作為支撐,但統(tǒng)計(jì)降尺度消耗的計(jì)算量小,很多方面更為可取。

常用的統(tǒng)計(jì)降尺度方法主要概括轉(zhuǎn)換函數(shù)法、環(huán)流分型技術(shù)和天氣發(fā)生器(李江萍和王式功,2008)。應(yīng)用最多是轉(zhuǎn)換函數(shù)法,它包括了多元線性回歸、SVD等線性轉(zhuǎn)換函數(shù)法(Oshima et al.,2002;范麗軍等,2007)和人工神經(jīng)網(wǎng)絡(luò)、支持向量機(jī)等非線性轉(zhuǎn)換函數(shù)法(Mpelasoka et al.,2001;Anandhi et al.,2009)。Huth et al.(2008)通過對(duì)比線性回歸與神經(jīng)網(wǎng)絡(luò)兩種方法進(jìn)行溫度降尺度,證明了線性方法也可以取得較好的降尺度效果。Lian et al.(2010)運(yùn)用PCA、CCA等統(tǒng)計(jì)降尺度方法,利用1960—1990年的月平均海平面氣壓和溫度值成功預(yù)測(cè)中國(guó)沿海地區(qū)氣溫。賈小龍等(2010)使用變形的典型相關(guān)分析方法對(duì)中國(guó)冬季溫度進(jìn)行了可預(yù)報(bào)性研究,很好地揭示了大氣環(huán)流和溫度的內(nèi)在聯(lián)系,并且預(yù)測(cè)效果明顯高于模式輸出結(jié)果。但目前而言,上述研究方法主要運(yùn)用于氣候預(yù)測(cè)方面(陳威霖等,2012)。

對(duì)于短期天氣預(yù)報(bào)的統(tǒng)計(jì)降尺度研究,目前依然比較匱乏。國(guó)內(nèi)主要采用雙線性插值、Kriging插值、反距離權(quán)重插值等空間內(nèi)插方法(李新等,2000;鈔振華,2011;彭彬等,2011;徐振亞等,2012)和滑動(dòng)訓(xùn)練法等(智協(xié)飛等,2013)進(jìn)行統(tǒng)計(jì)降尺度研究,此外還有學(xué)者使用一些簡(jiǎn)單的轉(zhuǎn)換函數(shù)(郭虎等,2008),但均有一定局限性。近年來國(guó)外主要采用Kalman濾波、滑動(dòng)窗等偏差估計(jì)方法(Engel and Ebert,2011;Monache et al.,2011)對(duì)氣象要素進(jìn)行統(tǒng)計(jì)降尺度研究。Cui et al.(2011)等提出了一種具有自適應(yīng)特征的Kalman濾波類型的遞減平均統(tǒng)計(jì)降尺度技術(shù),對(duì)北美集合預(yù)報(bào)系統(tǒng)的輸出值做統(tǒng)計(jì)后處理,獲得精細(xì)化預(yù)報(bào)產(chǎn)品,取得明顯效果。

本文將上述遞減平均統(tǒng)計(jì)降尺度技術(shù)運(yùn)用于中國(guó)進(jìn)行地面氣溫的精細(xì)化預(yù)報(bào)試驗(yàn),探索該方案在中國(guó)的可用性,以及研究其與插值降尺度方法預(yù)報(bào)能力的差異。

1　資料和方法

1.1資料

本文所用資料為中國(guó)752個(gè)基本、基準(zhǔn)地面氣象觀測(cè)站的地面日平均溫度數(shù)據(jù),資料時(shí)段為2000—2010年,選取其中缺測(cè)較少的600站進(jìn)行預(yù)報(bào)試驗(yàn)。

1.2降尺度方案設(shè)計(jì)

1.2.1預(yù)報(bào)場(chǎng)和分析場(chǎng)的建立

數(shù)值預(yù)報(bào)產(chǎn)品自身帶有系統(tǒng)性偏差和隨機(jī)誤差,使得基于數(shù)值產(chǎn)品的精細(xì)化預(yù)報(bào)的誤差不僅源于降尺度過程,也來源于數(shù)值產(chǎn)品本身,最終導(dǎo)致難以評(píng)價(jià)該降尺度方案的優(yōu)劣以及效果的好壞。出于上述考慮,本文將使用地面氣象觀測(cè)站的站點(diǎn)觀測(cè)數(shù)據(jù)構(gòu)建“預(yù)報(bào)場(chǎng)”和“分析場(chǎng)”,以去除數(shù)值預(yù)報(bào)產(chǎn)品帶來的誤差。首先,在上述所選的600站中較均勻地隨機(jī)挑選100站作為低分辨率“預(yù)報(bào)場(chǎng)(F100)”,代替業(yè)務(wù)中所使用的低分辨率的大尺度模式產(chǎn)品,剩余的500個(gè)觀測(cè)站作為高分辨率的“精細(xì)化分析場(chǎng)(A500)”。然后,將低分辨率“預(yù)報(bào)場(chǎng)”的日平均溫度使用空間內(nèi)插的方法插值到“精細(xì)化分析場(chǎng)”的站點(diǎn)上,作為精細(xì)化初估場(chǎng)(F500)。具體臺(tái)站分布見圖1,“·”表示低分辨率“預(yù)報(bào)場(chǎng)(F100)”,“+”表示高分辨率的“精細(xì)化分析場(chǎng)(A500)”。

圖1　預(yù)報(bào)試驗(yàn)所選的600個(gè)站點(diǎn)分布Fig.1　Distribution of the selected weather stations for the forecasting test

大多數(shù)研究認(rèn)為,數(shù)據(jù)密度、數(shù)據(jù)分布和空間異質(zhì)性是影響插值精度的主要因素,插值方法的選擇不是絕對(duì)的、唯一的(Lam,1983)?？紤]到是在不規(guī)則的站點(diǎn)間進(jìn)行插值計(jì)算,雙線性插值和樣條插值等空間內(nèi)插方法不適用,本文選用反距離插值(文中以IDW表示,下同)、距離平方反比權(quán)重插值(IDS)、Cressman客觀分析(Cressman)和Kriging插值(Kriging)4種方法(林忠輝等,2002)對(duì)低分辨率“預(yù)報(bào)場(chǎng)(F100)”進(jìn)行空間內(nèi)插,獲取精細(xì)化初估場(chǎng)(F500)。

將4種插值結(jié)果與“精細(xì)化分析場(chǎng)(A500)”進(jìn)行比較分析可以發(fā)現(xiàn)(圖2),整體而言不同插值方法的效果隨時(shí)間的變化基本一致,春、冬季誤差較大,秋季最小。Kriging插值和IDS插值的準(zhǔn)確度明顯優(yōu)于IDW插值和Cressman插值,前兩種插值的均方根誤差的年平均值接近3.0 ℃,較后兩者減小了0.1 ℃左右;相比之下,Kriging插值效果略優(yōu)于IDS插值,但兩者十分接近?？紤]到IDS插值比Kriging插值更為實(shí)用和簡(jiǎn)潔,且效果相差無幾,故本文選擇距離平方反比權(quán)重插值(IDS)進(jìn)行空間內(nèi)插,獲取精細(xì)化預(yù)報(bào)的初估場(chǎng)(F500)。

圖2　不同插值方法的均方根誤差Fig.2　RMSE of different interpolation methods

1.2.2降尺度矢量(DV)的構(gòu)建

降尺度矢量DV體現(xiàn)了精細(xì)化初估場(chǎng)(F500)與精細(xì)化分析場(chǎng)(A500)之間的統(tǒng)計(jì)關(guān)系,定義為同一時(shí)刻F500與A500間的差值場(chǎng),即DV(t)=F500(t)-A500(t),t為某一時(shí)刻。

DV(t0)=(1-w)×DV(t-1)+w×(F500(t0)-A500(t0))。

(1)

如公式(1),DV通過遞減平均法進(jìn)行更新,t0表示起報(bào)時(shí)刻;t-1表示前一個(gè)起報(bào)時(shí)刻;w為自適應(yīng)系數(shù),在0～1之間,反映不同時(shí)刻的DV對(duì)于t0時(shí)刻的貢獻(xiàn)大小,即過去多長(zhǎng)時(shí)間的信息將對(duì)預(yù)報(bào)當(dāng)天的訂正有影響。

前人的研究成果(王輝贊等,2006)表明,Kalman濾波方法對(duì)于連續(xù)性變量較其他統(tǒng)計(jì)預(yù)報(bào)方法的自適應(yīng)能力更強(qiáng),能夠更好地跟蹤預(yù)報(bào)對(duì)象。本文所使用的遞減平均統(tǒng)計(jì)降尺度技術(shù)即為Kalman濾波類型的方法,通過提取歷史預(yù)報(bào)和實(shí)況之間的誤差信息,估算系統(tǒng)偏差,使得預(yù)報(bào)更為接近觀測(cè)值。遞減平均法初始化偏差的方法分為偏差“熱啟動(dòng)”和偏差“冷啟動(dòng)”兩種。所謂偏差“熱啟動(dòng)”,就是利用過去一定天數(shù)的F500(t)與A500(t)差值的時(shí)間平均值作為偏差的初始值;出于計(jì)算方便或者資料不足等原因,將偏差的初始值直接設(shè)為0,即為偏差“冷啟動(dòng)”。本文將使用后者進(jìn)行偏差初始化(李莉等,2011)。

1.2.3降尺度預(yù)報(bào)值DF

DF(t)=F500(t)-DV(t0)。

(2)

公式(2)中,t為預(yù)報(bào)時(shí)效,將t時(shí)效的高分辨率的精細(xì)化初估場(chǎng)(F500)與起報(bào)時(shí)刻的降尺度矢量DV作差,即可得到t時(shí)效下精細(xì)化的降尺度預(yù)報(bào)值DF,本文的預(yù)報(bào)時(shí)效t分別取1、2、3 d。

1.2.4自適應(yīng)參數(shù)w的獲取

通過IDS插值獲取初估場(chǎng)后,利用公式(1)和(2),采用偏差“冷啟動(dòng)”方式估計(jì)插值偏差,進(jìn)行降尺度預(yù)報(bào)。

圖3給出了對(duì)w進(jìn)行敏感性試驗(yàn)后,所得預(yù)報(bào)誤差的均方根誤差隨w的變化趨勢(shì)。發(fā)現(xiàn),當(dāng)w在0.4附近時(shí),均方根誤差達(dá)最小,也就是說此時(shí)預(yù)報(bào)試驗(yàn)的總誤差最小;w向兩端趨近時(shí),均方根誤差逐漸變大,尤其當(dāng)w趨向于0時(shí)變化十分顯著。最終,本文取w=0.43作為遞減平均的自適應(yīng)參數(shù)。

圖3　不同自適應(yīng)參數(shù)w下的精細(xì)化預(yù)報(bào)均方根誤差Fig.3　RMSE of fine-scale forecasting with a different self-adapting parameter w

2　精細(xì)化預(yù)報(bào)效果分析

2.1全國(guó)預(yù)報(bào)誤差隨時(shí)效的變化

圖4為預(yù)報(bào)誤差的均方根誤差隨預(yù)報(bào)時(shí)效的變化以及與IDS插值均方根誤差的比較。首先,關(guān)注均方根誤差隨時(shí)效的變化,它的基本變化趨勢(shì)與業(yè)務(wù)預(yù)報(bào)一致:隨著預(yù)報(bào)時(shí)效的增加,預(yù)報(bào)誤差也在不斷增大。1 d預(yù)報(bào)全年的均方根誤差為1.4 ℃,2 d和3 d預(yù)報(bào)全年的均方根誤差稍大,分別為1.6 ℃、1.7 ℃;并且,1 d預(yù)報(bào)與2 d預(yù)報(bào)之間的差異要大于2 d預(yù)報(bào)與3 d預(yù)報(bào)的差異。其次,經(jīng)訂正以后的預(yù)報(bào)與IDS插值結(jié)果相比,均方根誤差均顯著減小,尤其是1 d預(yù)報(bào)時(shí)年平均均方根誤差減小了約1.5 ℃,減小幅度接近50%。

圖4　不同預(yù)報(bào)時(shí)效下不同季節(jié)的預(yù)報(bào)均方根誤差Fig.4　Forecasting RMSE in different seasons on different forecast days

2.2全國(guó)預(yù)報(bào)準(zhǔn)確率

在精細(xì)化預(yù)報(bào)過程中,較為關(guān)注的是每日全國(guó)的預(yù)報(bào)能力,以預(yù)報(bào)誤差的絕對(duì)值小于1.5 ℃作為預(yù)報(bào)準(zhǔn)確的標(biāo)準(zhǔn),計(jì)算平均每日全國(guó)預(yù)報(bào)準(zhǔn)確的站數(shù),并與IDS插值效果進(jìn)行比較。

由圖5可見,降尺度預(yù)報(bào)的效果與插值結(jié)果相比有很大的提高,前者預(yù)報(bào)準(zhǔn)確率在70%～80%,而插值預(yù)報(bào)平均每日絕對(duì)誤差小于1.5 ℃的站數(shù)在240站左右,準(zhǔn)確率尚不足50%。降尺度預(yù)報(bào)使得平均每日預(yù)報(bào)準(zhǔn)確的站數(shù)較插值預(yù)報(bào)提高了近100站,也就是使預(yù)報(bào)準(zhǔn)確率提高了近20%,其中,在1 d預(yù)報(bào)情況下,預(yù)報(bào)準(zhǔn)確率提高了30%之多。

圖5　平均每日預(yù)報(bào)的絕對(duì)誤差小于1.5 ℃的站數(shù)Fig.5　Daily number of stations whose absolute error was less than 1.5 ℃

2.3預(yù)報(bào)誤差的時(shí)空變化

我國(guó)是一個(gè)氣候多變、地形復(fù)雜的國(guó)家,不同的地區(qū)插值效果必然有所差異,那么全國(guó)各地的降尺度預(yù)報(bào)效果,以及遞減平均法究竟在不同的地區(qū)有多大訂正效果。

圖6　降尺度預(yù)報(bào)和IDS插值的均方根誤差的空間分布(單位:℃)　　a.IDS插值;b.1 d預(yù)報(bào);c.2 d預(yù)報(bào);d.3 d預(yù)報(bào)Fig.6　Distribution of the RMSE of fine-scale forecasting and IDS interpolation:(a)IDS;(b)1 d forecast;(c)2 d forecast;(d)3 d forecast

圖6為降尺度預(yù)報(bào)和IDS插值的均方根誤差的空間分布?？偨Y(jié)四幅圖的共同特征可知,預(yù)報(bào)誤差的基本趨勢(shì)是從東南地區(qū)向西北地區(qū)逐漸變大。均方根誤差在西南和西北的山區(qū)為一個(gè)明顯的大值區(qū),而我國(guó)東南部和東北地區(qū)則是一個(gè)相對(duì)的小值區(qū)。這主要是由于北方邊疆和西部地區(qū)相對(duì)復(fù)雜的地形和稀疏的站點(diǎn)分布,使得插值過程中產(chǎn)生了較大的誤差,所設(shè)計(jì)的降尺度方案很難完全對(duì)該類地區(qū)進(jìn)行訂正。

對(duì)比圖6a、b、c、d發(fā)現(xiàn),圖6a中,IDS插值的誤差較大,全國(guó)的均方根誤差普遍高于2 ℃,部分中西部地區(qū)甚至超過了4 ℃。將圖6b、c、d與圖6a比較可以發(fā)現(xiàn),經(jīng)過遞減平均技術(shù)訂正的降尺度預(yù)報(bào)的誤差顯著減小。原本誤差較大的西部地區(qū),經(jīng)訂正以后均方根誤差平均減小了2 ℃,部分誤差的高值中心已消失;原本誤差相對(duì)較小的東南部地區(qū)預(yù)報(bào)狀況也有所改善。整體而言,遞減平均統(tǒng)計(jì)降尺度技術(shù)對(duì)初估場(chǎng)的訂正是相當(dāng)有效的,不但普遍減小了全國(guó)的預(yù)報(bào)誤差,而且很大程度上改善了東西部預(yù)報(bào)效果“不平衡”的現(xiàn)象。

從上述分析可以發(fā)現(xiàn),不同地區(qū)的預(yù)報(bào)效果存在著較大差異,以下根據(jù)預(yù)報(bào)的誤差特征將我國(guó)簡(jiǎn)單的分為4個(gè)區(qū)域,對(duì)不同地區(qū)進(jìn)行誤差分析,研究偏差訂正的改進(jìn)效果。本文以35°N緯線和105°E經(jīng)線為界,將我國(guó)分為東北、東南、西北、西南4個(gè)地區(qū)。

圖7為預(yù)報(bào)時(shí)效為1 d時(shí)不同地區(qū)預(yù)報(bào)誤差的均方根誤差隨時(shí)間的變化。整體而言,預(yù)報(bào)的均方根誤差在冬季最大(平均為1.6 ℃),春季次之,夏季最小(平均為1.2 ℃),也就是說夏季的預(yù)報(bào)效果最好,而冬季最差。這主要是由于春、冬季節(jié)地面日平均溫度波動(dòng)較大,增加了偏差估計(jì)的難度,使得計(jì)算的降尺度矢量DV不能很好地代表系統(tǒng)偏差,從而導(dǎo)致春、冬季節(jié)精細(xì)化預(yù)報(bào)產(chǎn)生較大的誤差。相反,夏季溫度相對(duì)穩(wěn)定、波動(dòng)小,通過遞減平均估算的偏差較準(zhǔn)確,對(duì)初估場(chǎng)產(chǎn)生了較好的訂正。對(duì)不同季節(jié)而言,春季4個(gè)地區(qū)的預(yù)報(bào)效果差異較小,均方根誤差在0.3 ℃以內(nèi);從夏季開始,差異逐漸變大,到冬季東南地區(qū)和西北地區(qū)相差近0.7 ℃。也就是說,在不同季節(jié),不同地區(qū)的預(yù)報(bào)能力有所差異。并且對(duì)不同區(qū)域而言,東南地區(qū)始終是預(yù)報(bào)效果最好的(均方根誤差為1.2 ℃),向西北方向預(yù)報(bào)效果逐漸變差,并且預(yù)報(bào)效果的季節(jié)差異也相應(yīng)不斷增加。

圖7　1 d預(yù)報(bào)的均方根誤差的時(shí)空變化Fig.7　Spatiotemporal variation of 1 d forecast RMSE

從圖8可以看出,相同時(shí)效下,東南地區(qū)的預(yù)報(bào)準(zhǔn)確率始終高于其他地區(qū),其中1 d預(yù)報(bào)時(shí)準(zhǔn)確率高達(dá)85%;東北和西南地區(qū)的效果比較接近,3 d內(nèi)預(yù)報(bào)準(zhǔn)確率均介于70%～80%;西北地區(qū)的預(yù)報(bào)效果始終是最差的。1 d預(yù)報(bào)和3 d預(yù)報(bào)相比,不同地區(qū)的預(yù)報(bào)準(zhǔn)確率均高出10%以上,其中東南部地區(qū)二者的差異最大。降尺度預(yù)報(bào)和IDS插值結(jié)果相比,預(yù)報(bào)準(zhǔn)確率的提高十分顯著。以3 d預(yù)報(bào)為例,東南地區(qū)提高的幅度最小,為10%;西北地區(qū)的準(zhǔn)確率提高幅度最大,接近30%。也就是插值準(zhǔn)確率越低的地區(qū)改進(jìn)的幅度越大,使得降尺度以后預(yù)報(bào)效果的空間差異顯著減小。

圖8　不同時(shí)效下不同區(qū)域預(yù)報(bào)準(zhǔn)確率Fig.8　Forecasting accuracy in different regions on different forecast days

由上述分析可見,遞減平均統(tǒng)計(jì)降尺度技術(shù)具有較強(qiáng)的精細(xì)化預(yù)報(bào)能力,降尺度矢量DV的構(gòu)建是成功的,很好地訂正了插值誤差。

3　結(jié)論

本文利用中國(guó)600個(gè)基本、基準(zhǔn)地面氣象觀測(cè)站2000—2010年地面日平均氣溫?cái)?shù)據(jù)構(gòu)建“預(yù)報(bào)場(chǎng)”和“分析場(chǎng)”,采用具有自適應(yīng)特征的Kalman濾波類型的遞減平均統(tǒng)計(jì)降尺度技術(shù),對(duì)中國(guó)地面溫度進(jìn)行精細(xì)化預(yù)報(bào),對(duì)預(yù)報(bào)效果進(jìn)行了評(píng)估,并與IDS插值結(jié)果進(jìn)行了比較,得如下結(jié)論:

1)遞減平均統(tǒng)計(jì)降尺度技術(shù)對(duì)中國(guó)地面溫度進(jìn)行精細(xì)化預(yù)報(bào)是可行的。精細(xì)化預(yù)報(bào)的誤差從東南向西北增大,預(yù)報(bào)能力夏、秋、春、冬依次減弱,且隨著預(yù)報(bào)時(shí)效的增加預(yù)報(bào)準(zhǔn)確率逐漸降低。3 d內(nèi)預(yù)報(bào)的均方根誤差在1.4～1.7 ℃,預(yù)報(bào)準(zhǔn)確率在70%～80%,基本滿足業(yè)務(wù)要求,適合在基層臺(tái)站推廣。

2)距離平方反比權(quán)重插值(IDS)可提供誤差較小的初估場(chǎng),其操作簡(jiǎn)單易行。自適應(yīng)的偏差訂正降尺度方案相比上述插值方法有明顯的提高,最關(guān)鍵的降尺度矢量DV構(gòu)建合理,很好地估計(jì)了插值產(chǎn)生的精細(xì)化初估場(chǎng)的偏差。3 d內(nèi)預(yù)報(bào)的均方根誤差平均減小1.4 ℃、預(yù)報(bào)準(zhǔn)確率提高20%～30%,尤其對(duì)插值誤差較大的西部地區(qū)產(chǎn)生了很好的訂正,預(yù)報(bào)準(zhǔn)確率的提高幅度超過30%,有效地改善了西部的預(yù)報(bào)效果,使預(yù)報(bào)能力的空間差異顯著減小。

參考文獻(xiàn)(References)

Anandhi A,Srinivas V V,Kumar D N,et al.,2009.Role of predictors in downscaling surface temperature to river basin in India for IPCC SRES scenarios using support vector machine[J].Int J Climatol,29(4):583-603.

Cui B,Toth Z,Zhu Y J,2011.Bias correction for global ensemble forecast[J].Wea Forecasting,27:396-410.

鈔振華,2011.三種再分析氣溫資料在中國(guó)西部地區(qū)的可信度評(píng)價(jià)[J].大氣科學(xué)學(xué)報(bào),34(2):162-169.Chao Z H,2011.The reliability evaluation of three types of air temperature reanalysis data in West China[J].Trans Atmos Sci,34(2):162-169.(in Chinese).

陳威霖,江志紅,黃強(qiáng),2012.基于統(tǒng)計(jì)降尺度模型的江淮流域極端氣候的模擬與預(yù)估[J].大氣科學(xué)學(xué)報(bào),35(5):578-590.Chen W L,Jiang H,Huang Q,2012.Projection and simulation of climate extremes over the Yangtze and Huaihe River Basins based on a Statistical Downscaling Model[J].Trans Atmos Sci,35(5):578-590.(in Chinese).

陳豫英,陳曉光,馬金仁,等,2005.基于MM5模式的精細(xì)化MOS溫度預(yù)報(bào)[J].干旱氣象,2(4):52-56.Chen Y Y,Chen X G,.Ma J R,et al.,2005.Fine MOS temperature forecast based on MM5[J].Arid Meteorology,2(4):52-56.(in Chinese).

Engel C,Ebert E E,2011.Gridded operational consensus forecasts of 2-m temperature over Australia[J].Wea Forecasting,27:301-322.

范麗軍,符淙斌,陳德亮,2007.統(tǒng)計(jì)降尺度法對(duì)華北地區(qū)未來區(qū)域氣溫變化情景的預(yù)估[J].大氣科學(xué),131(15):887-897.Fan L J,Fu Z B,Chen D L,2007.Estimation of local temperature change scenarios in North China using statistical downscaling method[J].Chin J Atmos Sci,131(15):887-897.(in Chinese).

郭虎,楊建捷,楊波,等,2008.北京奧運(yùn)演練精細(xì)化預(yù)報(bào)方法及其檢驗(yàn)評(píng)估[J].氣象,6(34):17-25.Guo H,Yang J J,Yang B,et al.,2008.Refined forecast techniques and related verification during 2007 Olympic Drilling[J].Meteor Mon,6(34):17-25.(in Chinese).

Huth R,Kliegrova S,Metelka L,2008.Non-linearity in statistical downscaling:Does it bring an improvement for daily temperature in Europe[J].Int J Climatol,28(4):465-477.

賈小龍,陳麗娟,李維京,等,2010.Bp-cca方法用于中國(guó)冬季溫度和降水的可預(yù)報(bào)性研究和降尺度季節(jié)預(yù)測(cè)[J].氣象學(xué)報(bào),68(3):398-410.Jia X L,Chen L J,Li W J,et al.,2010.Statistical downscaling based on BP-CCA:Predictability and application to the winter temperature and precipitation in China[J].Acta Meteorologica Sinica,68(3):398-410.(in Chinese).

Kidnson J W,Thompson C C,1998.A comparison of statistical and model-based downscaling techniques for estimating local climate variations[J].American Meteorological Society,4(11):735-753.

Lam N S,1983.Spatial interpolation method:A review[J].The American Cartographer,10:129-149.

李江萍,王式功,2008.統(tǒng)計(jì)降尺度在數(shù)值預(yù)報(bào)產(chǎn)品使用中的應(yīng)用[J].氣象,34(6):41-45.Li J P,Wang S G,2008.Application of statistical downscaling method to numerical weather forecast[J].Meteor Mon,34(6):41-45.(in Chinese).

李莉,李應(yīng)林,田華,等,2011.T213全球集合預(yù)報(bào)系統(tǒng)性誤差訂正研究[J].氣象,37(1):31-38.Li L,Li Y L,Tian H,et al.,2011.Study of bias-correction in T213 global ensemble forecast[J].Meteor Mon,37(1):31-38.(in Chinese).

李新,程國(guó)棟,盧玲,2000.空間內(nèi)插方法的比較[J].地球科學(xué)進(jìn)展,25(13):261-265.Li X,Cheng G D,Lu L,2000.Comparison of spatial interpolation mrthods[J].Advances in Earth Science,25(13):261-265.(in Chinese).

Lian S M,Wang K,Cui M C,et al.,2010.Exercise in downscaling on sea surface temperature along Chinese coast[J].Chinese Journal of Oceanology and Limnology,18(2):110-116.

林忠輝,莫興國(guó),李宏軒,等,2002.中國(guó)陸地區(qū)域氣象要素的空間插值[J].地理學(xué)報(bào),27(1):47-56.Lin Z H,Mo X G,Li H X,et al.,2002.Comparison of three spatial interpolation methods for climate variables in China[J].Acta Geographical Sinica,27(1):47-56.(in Chinese).

劉永和,郭維棟,馮錦明,等,2011.氣象資料的統(tǒng)計(jì)降尺度方法綜述[J].地球科學(xué)進(jìn)展,26(8):837-847.Liu Y H,Guo W D,Feng J M,et al.,2011.A summary of methods for statistical downscaling of meteorological data[J].Advances in Earth Science,26(8):837-847.(in Chinese).

Mearns L O,Bogardi I,Giorgi F,et al.,1999.Comparison of climate change scenarios generated from regional climate model experiments and statistical downscaling[J].J Geophys Res,3(104):6604-6621.

Monache L D,Nipen T,Liu Y B,et al.,2011.Kalman filter and analog schemes to post-process numerical weather predictions[J].Mon Wea Rev,139:3354-3570.

Mpelasoka F S,Mullan A B,Heerdegen R G,2001.New Zealand climate change information derived by multivariate statistical and artificial neural networks approaches[J].Int J Climatol,21:1415-1433.

Oshima N,Hisashi K,Shinji K,2002.An application of statistical downscaling to estimate surface air temperature in Japan[J].J Geophys Res,107(D10):1401-1410.

彭彬,周艷蓮,高蘋,等,2011.氣溫插值中不同空間插值方法的適用性分析—以江蘇為例[J].地球信息科學(xué)學(xué)報(bào),13(4):539-547.Peng B,Zhou Y L,Gao P,et al.,2011.Suitability assessment of different interpolation methods in the gridding process of station collected air temperature:A case study in Jiangsu province,China[J].Geo-Information Science,13(4):539-547.(in Chinese).

王輝贊,張韌,王彥磊,等,2006.基于Kalman濾波的副熱帶高壓數(shù)值預(yù)報(bào)誤差修正[J].熱帶氣象學(xué)報(bào),22(6):661-666.Wang Z H,Zhang R,Wang Y L,et al.,2006.Errors revised for numerical forecast products of subtropical high based on Kalman filtering[J].J Trop Meteor,22(6):661-666.(in Chinese).

徐振亞,任福民,楊修群,等,2012.日最高溫度統(tǒng)計(jì)降尺度方法的比較研究[J].氣象科學(xué),32(4):395-402.Xu Z Y,Ren F M,Yang X Q,et al.,2012.A comparison study on statistical downscaling methods in daily maximum temperatures[J].J Meteor Sci,32(4):395-402.(in Chinese).

智協(xié)飛,季曉東,張璟,等,2013.基于TIGGE資料的地面氣溫和降水的多模式集成預(yù)報(bào)[J].大氣科學(xué)學(xué)報(bào),36(3):257-266.Zhi X F,Ji X D,Zhang J,et al.,2013.Multimodel ensemble forecasts of surface air temperature and precipitation using TIGGE datasets[J].Trans Atmos Sci,36(3):257-266.(in Chinese).

The statistical downscaling technique based on large-scale numerical forecasting productions is an effective method for fine-scale forecasting.In China,researchers use interpolation methods such as bilinear interpolation and inverse distance interpolation to produce a downscaled forecast.In recent years,the Kalman filter—type self-adapting decaying average downscaling technique has been designed overseas for forecast downscaling,which is better than the MOS method.Based on a daily surface temperature dataset of 752 weather stations for the period 2000 to 2010 in China,a fine-scale prediction test with a low-resolution “forecast field” and fine-scale “analysis field” for daily average temperature,using the self-adapting Kalman Filter—type decaying average statistical downscaling technique,was designed,without the effect of forecasting error in numerical forecasting production.The result of the downscaled prediction was compared with the interpolation method and analyzed for its possibility of application in China.

The decaying average technique in this paper filtered the observational data in order and determined the change of the dynamic system constantly.Then,systematic bias(called the “downscaling vector”,DV) was estimated.Finally,the prediction outcome was then corrected by the bias.This was a kind of self-adapting bias-estimated method,similar to the Kalman filter and a statistical post-processing method.The DV,defined as the difference between the “forecast field” and “analysis field” at the same time,presents the statistical relationship and systematic bias between the forecast and analysis.The DV that is weight-averaged between the last DV and the forecast error at the same time is updated by the decaying average algorithm.Thus,we can extract error information between the forecast and the observational data to estimate the forecast bias.The result show that:

(1)The 1—3 d forecast accuracy rate of the self-adapting Kalman filter—type decaying average statistical downscaling technique was 70%—80%,which is basically satisfactory for professional application.The RMSE of the 1—3 d forecasts was between 1.4 ℃ to 1.7 ℃,and the average value in China was 1.5 ℃.The error increased from Southeast China(RMSE of 1.4 ℃) to Northwest China(RMSE of 1.8 ℃).The forecasting ability decreased with the increase of the forecast limitation.The forecast effect was best in summer and worst in winter.

(2)The IDS interpolation method was able to provide the best estimated field,but the decaying average statistical downscaling technique was better than any interpolation method.The critical DV was structured reasonably,able to estimate the interpolation bias of the estimated field well.The RMSE decreased to 50% on average(approximately 1.4 ℃),and the forecast accuracy rate increased by 20%—30% in the 1—3 d forecasts.In particular,the large gap in forecasting ability between West China and East China was reduced,and the accuracy rate increased more than 30% in West China,where there was a large error.

In conclusion,the forecasting ability of the method was verified by producing a low-resolution “forecast field” and fine-scale “analysis field” without the influence of forecasting error in the numerical forecasting product.Therefore,the decaying average statistical downscaling technique is feasible for operational fine-scale surface temperature forecasting in China.

In future work,we intend to combine numerical forecasting products and reanalysis temperature data to generate a realistic prediction test for demonstrating how the Kalman filter—type self-adapting decaying average downscaling technique performs.

fine-scale forecasting;statistical downscaling;interpolation;decaying average algorithm

(責(zé)任編輯：劉菲)

doi：10.13878/j.cnki.dqkxxb.20130206001

Surface temperature statistical forecasting downscaling research in China

CHEN Guohua,GUO Pinwen

MeteorologicalDisaster,MinistryofEducation(KLME),NanjingUniversityofInformationScience&Technology,Nanjing210044,China

引用格式：陳國(guó)華,郭品文,2016.中國(guó)地面氣溫統(tǒng)計(jì)降尺度預(yù)報(bào)方法研究[J].大氣科學(xué)學(xué)報(bào),39(4):569-575.

Chen G H,Guo P W,2016.Surface temperature statistical forecasting downscaling research in China[J].Trans Atmos Sci,39(4):569-575.doi：10.13878/j.cnki.dqkxxb.20130206001.(in Chinese).

*聯(lián)系人，E-mail:glory_1999@163.com