热带山区高时空分辨率NDVI融合精度及其影响因素分析
高书鹏(1991— ),男,云南文山人,博士生,主要从事热带山地遥感时空融合与植被分类研究。E-mail: gaoshuaa@nwafu.edu.cn |
收稿日期: 2021-05-20
要求修回日期: 2021-06-29
网络出版日期: 2022-04-25
基金资助
国家自然科学基金项目(42001277)
国家重点研发计划项目(2016YFB0501502)
云南省水利厅水利科技项目(2014003)
云南师范大学博士研究生学术新人奖资助项目(01300205020503233)
版权
Study on the Factors that Influencing High Spatio-temporal Resolution NDVI Fusion Accuracy in Tropical Mountainous Area
Received date: 2021-05-20
Request revised date: 2021-06-29
Online published: 2022-04-25
Supported by
National Natural Science Foundation of China(42001277)
National Key Research and Development Program of China(2016YFB0501502)
Water Conservancy Science and Technology Project of Yunnan Provincial Water Resources Department(2014003)
PhD student academic Newcomer Award funded Projects of Yunnan Normal University(01300205020503233)
Copyright
高时空分辨率NDVI时序数据作为遥感应用中的重要数据源,对土地覆被动态变化监测具有重要意义,特别是在地表高程变化显著、气候条件复杂、景观异质性强烈的热带山区。虽然当前学者们提出了诸多时空数据融合模型,但针对这些模型在热带山区的NDVI数据融合精度及其影响因素分析尚不多见。对此,本文选取3类时空数据融合方法(权重函数法、概率统计法和多种混合法)中具有代表性的4个模型:STARFM(Spatial and Temporal Adaptive Reflectance Fusion Model)、RASTFM(Spatial and Temporal Adaptive Reflectance Fusion Model)、ESTARFM(Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model)、BSFM(Bayesian Spatiotemporal Fusion Model) (STARFM、ESTARFM为权重函数法;BSFM为概率统计法;RASTFM为多种混合法),选择位于我国热带山区的纳板河流域作为研究区。对融合模型的数据源选择、研究区的地形及景观空间异质性、融合模型、以及薄云和雾霾等大气条件等影响因素进行分析,研究结果表明:① 数据融合精度随输入影像之间的时间间隔及其相对变化量增加而降低;融合中输入的高、低空间分辨率数据光谱匹配度越高,融合精度越高(OLI优于Sentinel-2; MODIS优于VIIRS);经过BRDF校正的数据能够有效提高各模型的融合精度;② 地形及空间异质性对融合结果精度影响显著,融合精度与空间异质性呈负相关,本研究中融合精度随着坡度的增大而减小,但坡向对融合精度的影响较小;地形对RASTFM的影响较其他模型低;③ 融合模型中输入的高质量影像越多,模型的融合精度往往越高;④ 薄云和雾霾会对融合精度产生显著负面影响。本研究的结果对于改进热带山地地区的高时空数据融合模型,生产热带山区复杂地理环境的高精度高时空分辨率NDVI数据集具有重要的参考价值。
高书鹏 , 刘晓龙 , 宋金玲 , 史正涛 , 杨磊 , 郭利彪 . 热带山区高时空分辨率NDVI融合精度及其影响因素分析[J]. 地球信息科学学报, 2022 , 24(2) : 405 -419 . DOI: 10.12082/dqxxkx.2022.210281
As an important data source in remote sensing application, high spatiotemporal resolution NDVI time series data is of great significance for dynamic change monitoring of land cover, especially in tropical mountainous areas, where the surface elevation changes significantly, climate conditions are complex and spatiotemporally heterogeneous. Many multi-spatiotemporal data fusion models have been proposed by scholars. However, it is rare to analyze the fusion accuracy of these models and their influencing factors in tropical mountainous areas. This study takes the Naban River Watershed in the tropical mountainous area of Southwest China as the study area. Four representative models have been selected from three types of spatiotemporal data fusion methods, namely weight function-based method, Bayesian-based method, and Hybrid method. The four models are Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM), Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM), Spatial and Temporal Adaptive Reflectance Fusion Model (RASTFM), and Bayesian Spatiotemporal Fusion Model (BSFM). Among them, STARFM and ESTARFM are weight function-based method, BSFM is Bayesian-based method, and RASTFM is Hybrid method. This study carries out analysis of data source selection, terrain of the study area, landscape spatial heterogeneity, pixel numerical accuracy of fusion model, and atmospheric conditions such as thin clouds and haze. The results show that, firstly, the fusion accuracy decreases with the increase of time interval and its relative variation. A better match in sensor spectrum between the two input data results in a higher fusion accuracy. OLI is better than Sentinel-2 while MODIS is better than VIIRS. Compared with unadjusted data, data adjusted by the Bidirectional Reflectance Distribution Function (BRDF) can effectively improve fusion accuracy.Secondly, fusion accuracy is negatively correlated with spatial heterogeneity. Fusion accuracy decreases when spatial heterogeneity increases. There is a strong negative correlation between fusion accuracy and spatial heterogeneity at elevations. Fusion accuracy decreases when slope increases. In comparison, slope aspect has little influence on fusion accuracy. The influence of terrain on RASTFM is smaller when compared with models. Thirdly, the more high-quality high-resolution raw data as input data for the model, the higher the fusion accuracy will be. Fourthly, thin clouds and haze have a significant negative impact on the fusion accuracy. The results have important values as references for improving the high spatial-temporal data fusion model in tropical mountainous areas and establishing high spatiotemporal resolution NDVI data sets in complex geographical environment.
表1 研究选用的数据Tab. 1 Data used in this study |
数据 | 空间 分辨率/m | 时间 分辨率/d | 波长范围/μm | DOY(年份) | 数据共享机构 |
---|---|---|---|---|---|
Landsat OLI | 30 | 16 | Red:0.630~0.680 NIR:0.845~0.885 | 318(2016) 016(2017) 064(2017) 096(2017) 025(2020) 073(2020) | 美国地质调查局(United States Geological Survey, USGS) (http://glovis.USGS.gov/) |
Sentinel-2 | 10 | 10 | Red:0.650~0.680 NIR:0.785~0.900 | 326(2016) 020(2017) 040(2017) 070(2017) | 欧洲航天局(European Space Agency, ESA, ) (https://scihub.copernicus.eu/) |
MODIS | 500 | 1 | Red:0.620~0.670 NIR:0.841~0.876 | 001(2013)~001(2018) | 使用MOD09GA、MCD43A4、MOD09A1产品。(http://reverb.echo.NASA.gov/)。 |
VIIRS | 500 | 1 | Red:0.600~0.680 NIR:0.846~0.885 | 001(2013)~001(2018) | 使用VNP09GA、VNP43IA4、VNP09A1产品。(http://reverb.echo.NASA.gov/) |
DEM | 30 | (http://gdem.ersdac.jspacesystems.or.jp/) |
注:DOY(Day of Year)为儒略日,表示一年中的第几天,例如2016年DOY 318表示2016年的第318天。 |
表2 OLI融合NDVI与观测NDVI的比较Tab. 2 Comparison of the predicted NDVI and the observed OLI NDVI |
模型 | 间隔天数/RCV | OLI融合MOD09GA | OLI融合VNP09GA | OLI融合MCD43A4 | OLI融合VNP43IA4 | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|
R2 | RMSE | R2 | RMSE | R2 | RMSE | R2 | RMSE | ||||
STARFM | 48/0.0155 | 0.80 | 0.08 | 0.78 | 0.08 | 0.82 | 0.07 | 0.80 | 0.07 | ||
64/0.0184 | 0.72 | 0.09 | 0.72 | 0.08 | 0.79 | 0.07 | 0.77 | 0.07 | |||
RASTFM | 48/0.0155 | 0.81 | 0.07 | 0.80 | 0.07 | 0.84 | 0.06 | 0.82 | 0.07 | ||
64/0.0184 | 0.76 | 0.07 | 0.75 | 0.07 | 0.80 | 0.07 | 0.79 | 0.08 | |||
ESTARFM | 56/0.0170 | 0.88 | 0.06 | 0.85 | 0.06 | 0.90 | 0.06 | 0.89 | 0.06 | ||
72/0.0214 | 0.84 | 0.07 | 0.82 | 0.07 | 0.87 | 0.07 | 0.85 | 0.06 | |||
BSFM | - | 0.95 | 0.04 | 0.93 | 0.05 | - | - | - | - |
注:加粗数值表示对应模型的最高融合精度。 |
表3 Sentinel-2融合NDVI与观测NDVI的比较Tab. 3 Comparison of the predicted NDVI and the observed Sentinel-2 NDVI |
模型 | 间隔天数 /RCV | Sentinel-2融合MOD09GA | Sentinel-2融合VNP09GA | Sentinel-2融合MCD43A4 | Sentinel-2融合 VNP43IA4 | ||||
---|---|---|---|---|---|---|---|---|---|
R2 | RMSE | R2 | RMSE | R2 | RMSE | R2 | RMSE | ||
STARFM | 20/0.0082 | 0.89 | 0.09 | 0.87 | 0.09 | 0.90 | 0.09 | 0.89 | 0.09 |
60/0.0145 | 0.81 | 0.09 | 0.81 | 0.09 | 0.88 | 0.09 | 0.87 | 0.09 | |
RASTFM | 20/0.0082 | 0.89 | 0.09 | 0.88 | 0.09 | 0.91 | 0.08 | 0.90 | 0.08 |
60/0.0145 | 0.83 | 0.09 | 0.81 | 0.09 | 0.88 | 0.08 | 0.87 | 0.09 | |
ESTARFM | 40/0.0114 | 0.90 | 0.06 | 0.89 | 0.07 | 0.93 | 0.06 | 0.93 | 0.06 |
55/0.0185 | 0.86 | 0.09 | 0.85 | 0.09 | 0.89 | 0.08 | 0.88 | 0.09 | |
BSFM | - | 0.94 | 0.07 | 0.92 | 0.07 | - | - | - | - |
注:加粗数值表示对应模型的最高融合精度。 |
图5 OLI与不同低空间分辨率数据的融合精度(STARFM模型,间隔48 d)Fig. 5 Scatterplots of the predicted and the observed OLI (used STARFM and the interval of days is 48 d) |
表4 融合精度与地形因子SHI的相关性Tab. 4 Correlation between fusion accuracy and SHI of terrain factors |
地形因子 | 模型 | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
STARFM | RASTFM | ESTARFM | BSFM | |||||||
MOD09GA | MCD43A4 | MOD09GA | MCD43A4 | MOD09GA | MCD43A4 | MOD09GA | ||||
海拔 | -0.82 | -0.76 | -0.78 | -0.69 | -0.88 | -0.74 | -0.89 | |||
坡度 | -0.86 | -0.84 | -0.87 | -0.86 | -0.85 | -0.83 | -0.89 | |||
坡向 | -0.64 | -0.53 | -0.60 | -0.54 | -0.61 | -0.51 | -0.55 |
感谢朱孝林博士提供的STARFM、ESTARFM开源代码,黄波教授团队提供的RASTFM融合程序。
[1] |
|
[2] |
|
[3] |
|
[4] |
|
[5] |
张立福, 彭明媛, 孙雪剑, 等. 遥感数据融合研究进展与文献定量分析(1992—2018)[J]. 遥感学报, 2019, 23(4):603-619.
[
|
[6] |
李树涛, 李聪妤, 康旭东. 多源遥感图像融合发展现状与未来展望[J]. 遥感学报, 2021, 25(1):148-166.
[
|
[7] |
|
[8] |
|
[9] |
|
[10] |
|
[11] |
|
[12] |
|
[13] |
|
[14] |
|
[15] |
黄波. 时空遥感影像融合研究的进展与趋势[J]. 四川师范大学学报(自然科学版), 2020, 43(4):427-434,424.
[
|
[16] |
|
[17] |
|
[18] |
刘建波, 马勇, 武易天, 等. 遥感高时空融合方法的研究进展及应用现状[J]. 遥感学报, 2016, 20(5):1038-1049.
[
|
[19] |
|
[20] |
|
[21] |
|
[22] |
|
[23] |
|
[24] |
平博, 孟云闪, 苏奋振. 面向GF-1 WFV数据和MODIS数据的时空融合算法对比分析[J]. 地球信息科学学报, 2019, 21(2):157-167.
[
|
[25] |
纳板河流域国家级自然保护区管理局. 保护区概况[EB/OL]. https://nbhbhq.xsbn.gov.cn.html,2016-09-26.
[National Nature Reserve Administration of Naban River Watershed.2016. Overview of the nature reserve[EB/OL]. https://nbhbhq.xsbn.gov.cn.html,2016-09-26.
|
[26] |
|
[27] |
宫世贤, 凌升海. 西双版纳雾在减少[J]. 气象, 1996(11):11-15.
[
|
[28] |
|
[29] |
|
[30] |
|
[31] |
|
[32] |
张雪红. 基于高分辨率遥感的桉树林空间异质性与尺度效应研究[D]. 南京:南京大学, 2012.
[
|
[33] |
|
[34] |
|
[35] |
|
[36] |
|
[37] |
|
[38] |
|
[39] |
|
[40] |
|
[41] |
|
/
〈 |
|
〉 |