Inversion of net primary productivity in the arid region of Northwest China based on various regressions
JIAOWei1,2,, CHENYaning1,, LIZhi1, LIYupeng1,2, HUANGXiaoran1,2, LIHaixia1,2 1. State Key Laboratory of Desert and Oasis Ecology,Xinjiang Institute of Ecology and Geography,Chinese Academy of Sciences,Urumqi 830011,China2. University of Chinese Academy of Sciences,Beijing 100049,China 通讯作者:通讯作者:陈亚宁,E-mail:chenyn@ms.xjb.ac.cn 收稿日期:2016-08-22 修回日期:2017-01-18 网络出版日期:2017-03-20 版权声明:2017《资源科学》编辑部《资源科学》编辑部 基金资助:新疆维吾尔自治区自然科学基金(2015211A048) 作者简介: -->作者简介:焦伟,女,山东泰安人,硕士生,主要从事生态水文方面的研究。E-mail:jiaoweisdnu@163.com
关键词:净初级生产力;逐步多元回归;主成分回归;偏最小二乘回归;岭回归;西北地区 Abstract Vegetation Net Primary Productivity(NPP)is an important parameter when evaluating terrestrial ecosystems and provides a significant reference for research into global carbon cycles. Based on MODIS data and meteorological station data from 2000 to 2014,we used multi-stepwise regression,principal components regression,partial least-squares regression and ridge regression and estimated vegetation NPP and temporal-spatial distribution patterns in the arid region of Northwest China. We found that the multi-stepwise regression model was superior and the model simulation results were coordinated with measured values in the spatial distribution. A multi-stepwise regression model can be used to retrieve NPP in arid and semi-arid areas and reflects vegetation growth and distribution in the study area. There were obvious regional differences in NPP distribution in the arid region of Northwest China; vegetation NPP in the mountains was increased and decreased in the plains. High values were found in the north,northwest and southeast of the arid region and low values in the south and southeast. At an annual scale,NPP in Northwest China slightly increased from 2000 to 2014 at a rate of 0.40 gC/(m2·a). Trends in annual vege-tation NPP are different among areas,vegetation NPP in 58.66% areas increased,13.64% of areas remained relatively stable,and 27.7% of areas slightly decreased from 2000.
Keywords:net primary productivity;multi-stepwise regression;principal components regression;partial least-squares regression;ridge regression;Northwest China -->0 PDF (985KB)元数据多维度评价相关文章收藏文章 本文引用格式导出EndNoteRisBibtex收藏本文--> 焦伟, 陈亚宁, 李稚, 李玉朋, 黄晓然, 李海霞. 基于多种回归分析方法的西北干旱区植被NPP遥感反演研究[J]. , 2017, 39(3): 545-556 https://doi.org/10.18402/resci.2017.03.16 JIAOWei, CHENYaning, LIZhi, LIYupeng, HUANGXiaoran, LIHaixia. Inversion of net primary productivity in the arid region of Northwest China based on various regressions[J]. 资源科学, 2017, 39(3): 545-556 https://doi.org/10.18402/resci.2017.03.16
1 引言
植被净初级生产力(Net Primary Productivity,NPP)是生态系统生产能力和碳汇能力的反映[1,2],是陆地生态系统和大气之间碳循环的重要组成部分[3-5]。大量研究表明,随着气温升高,植被生产力不断增加,北半球高纬度地区森林线开始向高纬地区移动[6-8];但也有研究发现美国西南部地区呈现退化的趋势[9],表明植被生产力的变化具有空间异质性。因此在区域尺度上研究自然和人为因素扰动作用对陆地生态系统碳循环的影响具有重要的意义。西北干旱区作为对气候变化响应最敏感的地区之一,得到国内外****的广泛关注,过去半个多世纪,气温上升速率高达0.39℃/10a,是全球气温上升速率(0.14℃/10a)的2.78倍[10,11],因此,准确反演西北干旱区生物量对全球碳循环的影响以及国家气候谈判中关于碳源汇等问题具有重要的意义。 近年来,全球及区域尺度上植被NPP的反演模型不断发展,从传统的估测方法(包括站点实测法[12,13]、气候相关统计模型[14]、生态系统过程模型[15])发展到基于遥感手段的遥感模型估测法[16,17]。鉴于站点实测方法在数据采集时的局限性,基于遥感手段的模型方法成为研究植被NPP反演的主要途径,能够快速及时获取全球及区域尺度上的陆地生态系统碳储量。早期的植被生物量研究多采用归一化植被指数来检测植被的生长,随着遥感数据的不断完善,更多****利用植被指数、地形以及气象因子与生物量之间建立回归模型,以寻求更为简洁、快速反演植被生物量的方法。杨存健等对云南省西双版纳森林生物量与气象因子建立回归模型并反演生物量,模型的R2为0.589[18]。范文义等通过构建逐步回归模型、神经网络模型来反演森林生物量并对模型的精度进行评价[19]。张旭琛等以实测数据为基础结合遥感气象数据进行回归建模,反演并分析了伊犁地区草地植被生物量的空间分布特征[20]。但是以上研究没有考虑自变量之间的相关性,这在一定程度上会影响模型模拟的可靠性。 在回归分析中,解决自变量之间共线性问题的主要回归分析方法有逐步回归分析[19,21]、主成分回归分析[22]、偏最小二乘回归[23],以及岭回归分析[24]。Fan W Y等、Liu Y M等采用逐步多元回归模拟了中国太阳总辐射和森林生物量数据,并证明了模型的可靠性[19,21]。胡馨月等运用主成分回归进行了温度分布的反演[22];Bin等将偏最小二乘回归运用在多元光谱分析,相对误差的预测精度为1.88%[23]。高思远等运用岭回归建立ET0的多元反演研究,提高了模型精度和可靠性[24]。但是关于植被生物量的模拟及模型的综合比较分析方面的研究还比较少。 本研究采用归一化植被指数(Normalized Difference Vegetation Index,NDVI)、MODIS NPP数据、气象站点数据以及样点实测数据,建立生物量与各影响因子之间的逐步回归模型、主成分回归模型、偏最小二乘模型以及岭回归模型,比较分析了这4类模型在生物量反演方面的精度,以寻求高效、准确反演植被NPP的最优模型,并对西北干旱区植被NPP遥感反演进行初步尝试,为探讨西北干旱区植被NPP的分布规律提供参考,实现西北干旱区植被资源的合理利用与生态环境的可持续发展。
2 研究区概况与数据来源
2.1 研究区概况
西北干旱区位于欧亚大陆腹地,介于N34˚54 -N49˚19 和E73˚ 44 -E106˚ 46 之间,贺兰山、乌鞘岭以西,昆仑山以北包括新疆全部的广大区域,面积约250万km2(图1)。干旱少雨的气候特点、山盆相间的地貌形态、特殊的土壤植被特征,使得这一地区形成了与中国其他地区迥异的植被类型和生态系统[25]。这一地区山地生态系统垂直分异明显、绿洲景观特色鲜明,荒漠类型复杂多样,形成了独具一格的山地、绿洲、荒漠共存的地理景观;由于干旱区水资源短缺且蒸发强烈,在水分胁迫作用下,该区域植被覆盖率低,土地荒漠化问题严重,形成了以浅根系植被为主体,存在大量旱生、耐盐碱、抗风沙的干旱植被,区域生态环境极其脆弱。 显示原图|下载原图ZIP|生成PPT 图1研究区示意 -->Figure 1Sketch of study area -->
2.2 数据来源及预处理
2.2.1 模型变量选择 MOD17A3数据是利用BIOME-BGC模型进行模拟,得到年NPP累积量,时间尺度是一年,空间分辨率是1km。该数据产品目前已在全球和区域NPP与碳循环研究中得到验证和广泛应用[3,26,27]。但在官网下载的MOD17A3原始数据在沙漠区是空值,因此对于西北干旱区来说,数据缺失严重(图2),进而无法直接利用MOD17A3数据产品对西北干旱区的植被NPP进行相关方面的研究。因此,本研究以NDVI、气温、降水、海拔高度、潜在蒸散(Potential Evapotranspiration,PET)及太阳辐射等作为自变量,MOD17A3产品数据作为因变量进行回归建模,模拟研究区植被NPP的空间分布规律及特征。 显示原图|下载原图ZIP|生成PPT 图22014年MODIS 17A3数据空间分布 -->Figure 2Distribution of MODIS 17A3 data of 2014 in Northwest China -->
将456个和114个站点自变量数据分别代入上述得到的4种模型中,计算植被NPP,并与因变量(MOD17A3)数据进行对比分析,利用模型指标进行评估。模型拟合结果散点见图4,4种回归模型预测结果(多元线性回归、主成分回归、偏最小二乘回归、岭回归)的变化斜率为0.81、0.55、0.74、0.78,R2分别为0.83、0.48、0.81、0.81;针对模型预测结果,从模型评估指标分析(表3),4种模型的NS系数分别为0.85、0.40、0.81、0.81;精度分别为84.43%、60.32%、82.35%、82.63%,模型模拟值与MOD17A3产品值的PBIAS分别为-0.77、14.02、-1.24、-0.87,R2分别为0.85、0.50、0.81、0.81,SRMSE分别为0.39、0.77、0.44、0.44。综合模型评估指标以及拟合预测结果散点图发现,4种回归模型中主成分回归模型的模拟精度较低,回归结果不理想。鉴于此,通过阅读大量文献并深入分析模型机理发现,主成分回归方法虽然有效地降低了数据的维数,但是无法避免使新变量受到原始变量中重叠信息的影响,且无法辨识噪音与信息,影响模型的精度;另外在提取主成分时是撇开因变量进行的,造成主成分对因变量的解释能力会变弱,达不到理想的结果。 Table 3 表3 表3不同建模方法的评估指标比较 Table 3Comparisons of evaluation indexes by different models
逐步回归模型
主成分回归
偏最小二乘回归
岭回归
拟合
预测
拟合
预测
拟合
预测
拟合
预测
NS
0.83
0.85
0.40
0.40
0.80
0.81
0.79
0.81
Precision
79.94
84.43
55.06
60.32
77.83
82.35
78.11
82.63
PBIAS
-0.68
-0.77
9.42
14.02
0.23
-1.24
-0.35
-0.87
R2
0.82
0.85
0.45
0.50
0.80
0.81
0.80
0.81
SRMSE
0.42
0.39
0.78
0.77
0.45
0.44
0.45
0.44
新窗口打开 偏最小二乘回归是一种同时具备主成分分析、典型相关性分析和最小二乘回归的优点,预测精度较高;岭回归分析是一种有偏估计方法;逐步回归法尽可能地将回归效果显著的自变量引入方程,这3种模型的NS系数均大于0.75,模型的R2在0.8左右,拟合程度较高。但是从模型的简易程度以及模型需要的变量来看,逐步线性回归模型较为简单高效,且容易实现。在此基础上本研究利用逐步线性回归模型模拟2012-2014年西北干旱区38个气象站点的植被NPP值,并用MATLAB提取MOD17A3产品相应像元数据进行对比分析,发现模拟值与产品数据变化趋势基本一致(图5a),拟合程度较好(图5b),R2=0.827(p<0.01)。 显示原图|下载原图ZIP|生成PPT 图5逐步回归模型模拟结果的对比与验证分析 -->Figure 5The comparison and verification of the simulated results by Multi-stepwise regression model -->
3.3.1 植被NPP的空间分布特征 在MATLAB、ArcGIS等软件的基础上,利用逐步回归模型反演西北干旱区2000-2014年的植被NPP(图6a,见第552页)。植被NPP高的地区主要分布在水热条件较好的山区,如阿尔泰山、天山北坡森林区、塔里木河流域、伊犁河谷、昆仑山西段及祁连山等地区,总体上呈现出北部及西北部高,南部及东南部低的特征。伊犁河谷地区的植被NPP值最高,阿尔泰山区、塔河地区、昆仑山西段等高山地区植被NPP的量次之,南疆的塔克拉玛干沙漠地区及河西走廊,植被覆盖率极低,NPP低,甚至低于10.00gC/(m2a)。 显示原图|下载原图ZIP|生成PPT 图62000-2014年西北干旱区植被NPP时空变化格局及气温、降水的变化趋势分析 -->Figure 6The spatial-temporal change pattern of vegetation NPP and the changing trend of precipitation and temperaturein Northwest China from 2000 to 2014 -->
本文采用逐步线性回归、主成分回归、偏最小二乘回归以及岭回归等四种方法构建模型,探讨了西北干旱区NPP时空变化,并对其进行了评估,得到如下结论: (1)综合比较4种回归模型的模型指标及拟合散点图可知,逐步回归模型被认为是一种可以简单、高效、准确地反演西北干旱区植被NPP的模型,且模拟结果与MOD17A3产品及实测值具有很强的相关性。 (2)西北干旱区NPP自2000年以来呈波动增加趋势,线性增长率为0.40gC/(m2a)。其中,在2000-2009年间,植被NPP变化趋势较为平稳,线性增长斜率为0.22gC/(m2a);2009年以后呈明显增加趋势,线性增长斜率为0.45gC/(m2a)。 (3)西北干旱区NPP变化的空间分布表现为周边山区(阿尔泰山、伊犁河谷、天山北坡、昆仑山西段)的生物量较高,而平原盆地及塔克拉玛干沙漠地区植被覆盖率较低,尤其塔克拉玛干沙漠低于10gC/(m2a)。在过去的10余年间,西北干旱区58.66%区域的植被NPP呈现增长趋势,13.64%的地区保持不变,27.7%的地区表现为退化趋势。退化的区域主要集中在沙漠边缘及戈壁地带。 The authors have declared that no competing interests exist.
Pan SF,Tian HQ,DangalS,et al.Impacts of climate variability and extremes on global net primary production in the first decade of the 21st century [J]. ,2015,25(9):1027-1044. [本文引用: 1]
[2]
LiX,ZhuZ,ZengH,et al.Estimation of gross primary production in China(1982-2010)with multiple ecosystem models [J]. ,2016,324:33-44. [本文引用: 1]
[3]
LiZ,Chen YN,WangY,et al.Dynamic changes in terrestrial net primary production and their effects on evapotranspiration [J]. ,2016,20(6):2169-2178. [本文引用: 2]
[4]
GuoQ,HuZ,LiS,et al.Contrasting responses of gross primary productivity to precipitation events in a water-limited and a temperature-limited grassland ecosystem [J]. ,2015,214-215(3):169-177. [本文引用: 1]
[5]
Flanagan LB,Sharp EJ,Gamon JA.Application of the photosynthetic light-use efficiency model in a Northern Great Plains grassland [J]. ,2015,168(6):239-251. [本文引用: 1]
[6]
Beck P SA,JudaygP,AlixC,et al. Changes in forest productivity across Alaska consistent with biome shift [J]. ,2011,14(4):373-379. [本文引用: 1]
[7]
Van BR,HanecaK,HoogestegerJ,et al.A century of tree line changes in sub-Arctic Sweden shows local and regional varia-bility and only a minor influence of 20th century climate warming [J]. ,2011,38(5):907-921.
[8]
Mekonnen ZA,Grant RF,Schwalm C. Contrasting changes in gross primary productivity of different regions of North America as affected by warming in recent decades [J]. ,2016,218-219(1-2):50-64. [本文引用: 1]
[9]
Williams AP,Allen CD,Millar CI,et al.Forest responses to increasing aridity and warmth in the southwestern United States [J]. ,2010,107(50):21289-21294. [本文引用: 1]
[Chen YN,LiZ,Fan YT,et al.Research progress on the impact of climate change on water resources in the arid region of North-west China [J]. ,2014,69(9):1295-1304.] [本文引用: 3]
[Luo TX,Li WH,Leng YF,et al.Estimation of total biomass and potential distribution of net primary productivity in the Tibetan plateau [J]. ,1998,17(4):337-344.] [本文引用: 1]
[Zhu WQ,Pan YZ,Long ZH,et al.Estimating net primary productivity of terrestrial vegetation based on GIS and RS:A case study in Inner Mongolia,China [J]. ,2005,9(3):300-307.] [本文引用: 1]
[17]
LiangW,Yang YT,Fan DM,et al.Analysis of spatial and temporal patterns of net primary production and their climate controls in China from 1982 to 2010 [J]. ,2015,204:22-36. [本文引用: 1]
[Yang CJ,Liu JY,Zhang ZX.Study on the estimation of the tropical forest vegetation biomass by using remote sensing data,meteorological data and topographical data [J]. ,2004,20(6):22-25.] [本文引用: 1]
[19]
Fan WY,Zhang HY,YuY,et al.Comparison of three models of forest biomass estimation [J]. ,2011,35(4):402-410. [本文引用: 4]
[Gao SY,Cui CF,Fan YP.Remote sensing inversion of ET0 in eastern agricultural area of Qinghai Province based on ridge estimation [J]. ,2016,31(4):693-702.] [本文引用: 3]
[ ZhanJ,Pan XL,Gao ZQ,et al.Satellite estimates and change detection of net primary productivity of oasis-desert based on ecosystem process with remotely sensed forcing in Arid Western China [J]. ,2006,61(1):15-25.] [本文引用: 1]
[Cui LL,Du HQ,ShiJ,et al.Spatial and temporal pattern of vegetation NPP and its relationship with climate in the Southeastern China [J]. ,2016,36(5):787-793.] [本文引用: 1]
[Wu SS,Yao ZJ,Jiang LG,et al.The spatial-temporal variations and hydrological effects of vegetation NPP based on MODIS in the source region of the Yangtze River [J]. ,2016,31(1):39-50.] [本文引用: 1]
[28]
San BT,Suzen ML.Digital elevation model(DEM)generation and accuracy assessment from ASTER stereo data [J]. ,2005,26(22):5013-5027. [本文引用: 1]
[29]
GuY,HuntE,WardlowB,et al.Evaluation of MODIS NDVI and NDWI for vegetation drought monitoring using Oklahoma Mesonet soil moisture data [J]. ,2008,35(22):1092-1104. [本文引用: 1]
[China Meteorological Data Service Centre[EB/OL].(2015-05)[ 2016-08-22]. .]URL [本文引用: 1]
[32]
Sun JG.A correlation principal component regression analysis of NIR data [J]. ,1995,9(1):21-29. [本文引用: 1]
[33]
GusnantoA,PawitanY,HuangJ,et al.Variable selection in random calibration of near-infrared instruments:Ridge regres-sion and partial least squares regression settings [J]. ,2003,17(3):174-185. [本文引用: 1]
[34]
Pan SF,Tian HQ,DangalS,et al.Modeling and monitoring terrestrial primary production in a changing global environment:Toward a multiscale synthesis of observation and simulation [J]. ,2014,doi:10.1155/2014/965936. [本文引用: 1]
[Zeng JS,Liu XG,Luo SH.Application of principal com-ponent regression and partial least square in blast furnace iron-making [J]. ,2009,36(1):33-36.] [本文引用: 1]
[YiK.Comparison and analysis of the principle of ridge regression in common statistical software [J]. ,2013,30(2):109-112.] [本文引用: 1]
[38]
Jobbagy EG,Sala OE,Paruelo JM.Patterns and controls of primary production in the Patagonian steppe:A remote sensing approach [J]. ,2002,83(2):307-319. [本文引用: 1]
[39]
Yang YT,LongD,Guan HD,et al.GRACE satellite observed hydrological controls on interannual and seasonal variability in surface greenness over mainland Australia [J]. ,2014,119(12):2245-2260. [本文引用: 1]
[Li PF,Sun XM,Zhao XY.Analysis of precipitation and potential evapotranspiration in arid and semiarid area of China in recent 50 years [J]. ,2012,26(7):57-63.] [本文引用: 1]
[Yao JQ,YangQ,Chen YN,et al.Climate change in arid areas of Northwest China in past 50 years and its effects on the local ecological environment [J]. ,2013,32(5):1283-1291.] [本文引用: 1]
[42]
Chen YN,LiZ,Fan YT,et al.Progress and prospects of climate change impacts on hydrology in the arid region of northwest China [J]. ,2015,139:11-19. [本文引用: 1]
[43]
ArredondoT,garcìa-MoyaE,Huber-SannwaldE,et al. Drought manipulation and its direct and legacy effects on productivity of a monodominant and mixed-species semi-arid grassland [J]. ,2016,223:132-140. [本文引用: 1]
[LiX,Li XB,Chen YH,et al.Temporal responses of vegetation to climate variables in temperate steppe of northern China [J]. ,2007,31(6):1054-1062.] [本文引用: 1]