Comparison of Remote Sensing Time-Series Smoothing Methods for Grassland Spring Phenology Extraction on the Qinghai–Tibetan Plateau
Accurate evaluation of start of season (SOS) changes is essential to assess the ecosystem’s response to climate change. Smoothing method is an understudied factor that can lead to great uncertainties in SOS extraction, and the applicable situation for different smoothing methods and the impact of sm...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2020-10-01
|
| Series: | Remote Sensing |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2072-4292/12/20/3383 |
| _version_ | 1797550797146292224 |
|---|---|
| author | Nan Li Pei Zhan Yaozhong Pan Xiufang Zhu Muyi Li Dujuan Zhang |
| author_facet | Nan Li Pei Zhan Yaozhong Pan Xiufang Zhu Muyi Li Dujuan Zhang |
| author_sort | Nan Li |
| collection | DOAJ |
| description | Accurate evaluation of start of season (SOS) changes is essential to assess the ecosystem’s response to climate change. Smoothing method is an understudied factor that can lead to great uncertainties in SOS extraction, and the applicable situation for different smoothing methods and the impact of smoothing parameters on SOS extraction accuracy are of critical importance to be clarified. In this paper, we use MOD13Q1 normalized difference vegetation index (NDVI) data and SOS observations from eight agrometeorological stations on the Qinghai–Tibetan Plateau (QTP) during 2001–2011 to compare the SOS extraction accuracies of six popular smoothing methods (Changing Weight (CW), Savitzky-Golay (SG), Asymmetric Gaussian (AG), Double-logistic (DL), Whittaker Smoother (WS) and Harmonic Analysis of NDVI Time-Series (HANTS)) for two types of different SOS extraction methods (dynamic threshold (DT) with 9 different thresholds and double logistic (Zhang)). Furthermore, a parameter sensitivity analysis for each smoothing method is performed to quantify the impacts of smoothing parameters on SOS extraction. Finally, the suggested smoothing methods and reference ranges for the parameters of different smoothing methods were given for grassland phenology extraction on the QTP. The main conclusions are as follows: (1) the smoothing methods and SOS extraction methods jointly determine the SOS extraction accuracy, and a bad denoising performance of smoothing method does not necessarily lead to a low SOS extraction accuracy; (2) the default parameters for most smoothing methods can result in acceptable SOS extraction accuracies, but for some smoothing methods (e.g., WS) a parameter optimization is necessary, and the optimal parameters of the smoothing method can increase the <i>R</i><sup>2</sup> and reduce the <i>RMSE</i> of SOS extraction by up to 25% and 331%; (3) The main influencing factor of the SOS extraction using the DT method is the stability of the minimum value in the NDVI curve, and for the Zhang method the curve shape before the peak of the NDVI curve impacts the most; (4) HANTS is the most stable method no matter with (fitness = 35.05) or without parameter optimization (fitness = 33.52), which is recommended for QTP grassland SOS extraction. The findings of this study imply that remote sensing-based vegetation phenology extraction can be highly uncertain, and a careful selection and parameterization of the time-series smoothing method should be taken to achieve an accurate result. |
| first_indexed | 2024-03-10T15:34:31Z |
| format | Article |
| id | doaj.art-6eec7372958746b484d7b0fa52c18c64 |
| institution | Directory Open Access Journal |
| issn | 2072-4292 |
| language | English |
| last_indexed | 2024-03-10T15:34:31Z |
| publishDate | 2020-10-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Remote Sensing |
| spelling | doaj.art-6eec7372958746b484d7b0fa52c18c642023-11-20T17:18:54ZengMDPI AGRemote Sensing2072-42922020-10-011220338310.3390/rs12203383Comparison of Remote Sensing Time-Series Smoothing Methods for Grassland Spring Phenology Extraction on the Qinghai–Tibetan PlateauNan Li0Pei Zhan1Yaozhong Pan2Xiufang Zhu3Muyi Li4Dujuan Zhang5State Key Laboratory of Remote Sensing Science, Jointly Sponsored by Beijing Normal University and Institute of Remote Sensing and Digital Earth of Chinese Academy of Sciences, Beijing 100875, ChinaState Key Laboratory of Remote Sensing Science, Jointly Sponsored by Beijing Normal University and Institute of Remote Sensing and Digital Earth of Chinese Academy of Sciences, Beijing 100875, ChinaState Key Laboratory of Remote Sensing Science, Jointly Sponsored by Beijing Normal University and Institute of Remote Sensing and Digital Earth of Chinese Academy of Sciences, Beijing 100875, ChinaState Key Laboratory of Remote Sensing Science, Jointly Sponsored by Beijing Normal University and Institute of Remote Sensing and Digital Earth of Chinese Academy of Sciences, Beijing 100875, ChinaState Key Laboratory of Remote Sensing Science, Jointly Sponsored by Beijing Normal University and Institute of Remote Sensing and Digital Earth of Chinese Academy of Sciences, Beijing 100875, ChinaState Key Laboratory of Remote Sensing Science, Jointly Sponsored by Beijing Normal University and Institute of Remote Sensing and Digital Earth of Chinese Academy of Sciences, Beijing 100875, ChinaAccurate evaluation of start of season (SOS) changes is essential to assess the ecosystem’s response to climate change. Smoothing method is an understudied factor that can lead to great uncertainties in SOS extraction, and the applicable situation for different smoothing methods and the impact of smoothing parameters on SOS extraction accuracy are of critical importance to be clarified. In this paper, we use MOD13Q1 normalized difference vegetation index (NDVI) data and SOS observations from eight agrometeorological stations on the Qinghai–Tibetan Plateau (QTP) during 2001–2011 to compare the SOS extraction accuracies of six popular smoothing methods (Changing Weight (CW), Savitzky-Golay (SG), Asymmetric Gaussian (AG), Double-logistic (DL), Whittaker Smoother (WS) and Harmonic Analysis of NDVI Time-Series (HANTS)) for two types of different SOS extraction methods (dynamic threshold (DT) with 9 different thresholds and double logistic (Zhang)). Furthermore, a parameter sensitivity analysis for each smoothing method is performed to quantify the impacts of smoothing parameters on SOS extraction. Finally, the suggested smoothing methods and reference ranges for the parameters of different smoothing methods were given for grassland phenology extraction on the QTP. The main conclusions are as follows: (1) the smoothing methods and SOS extraction methods jointly determine the SOS extraction accuracy, and a bad denoising performance of smoothing method does not necessarily lead to a low SOS extraction accuracy; (2) the default parameters for most smoothing methods can result in acceptable SOS extraction accuracies, but for some smoothing methods (e.g., WS) a parameter optimization is necessary, and the optimal parameters of the smoothing method can increase the <i>R</i><sup>2</sup> and reduce the <i>RMSE</i> of SOS extraction by up to 25% and 331%; (3) The main influencing factor of the SOS extraction using the DT method is the stability of the minimum value in the NDVI curve, and for the Zhang method the curve shape before the peak of the NDVI curve impacts the most; (4) HANTS is the most stable method no matter with (fitness = 35.05) or without parameter optimization (fitness = 33.52), which is recommended for QTP grassland SOS extraction. The findings of this study imply that remote sensing-based vegetation phenology extraction can be highly uncertain, and a careful selection and parameterization of the time-series smoothing method should be taken to achieve an accurate result.https://www.mdpi.com/2072-4292/12/20/3383phenologysmoothing methodsStart of season (SOS)Qinghai–Tibetan Plateau (QTP)grassland |
| spellingShingle | Nan Li Pei Zhan Yaozhong Pan Xiufang Zhu Muyi Li Dujuan Zhang Comparison of Remote Sensing Time-Series Smoothing Methods for Grassland Spring Phenology Extraction on the Qinghai–Tibetan Plateau Remote Sensing phenology smoothing methods Start of season (SOS) Qinghai–Tibetan Plateau (QTP) grassland |
| title | Comparison of Remote Sensing Time-Series Smoothing Methods for Grassland Spring Phenology Extraction on the Qinghai–Tibetan Plateau |
| title_full | Comparison of Remote Sensing Time-Series Smoothing Methods for Grassland Spring Phenology Extraction on the Qinghai–Tibetan Plateau |
| title_fullStr | Comparison of Remote Sensing Time-Series Smoothing Methods for Grassland Spring Phenology Extraction on the Qinghai–Tibetan Plateau |
| title_full_unstemmed | Comparison of Remote Sensing Time-Series Smoothing Methods for Grassland Spring Phenology Extraction on the Qinghai–Tibetan Plateau |
| title_short | Comparison of Remote Sensing Time-Series Smoothing Methods for Grassland Spring Phenology Extraction on the Qinghai–Tibetan Plateau |
| title_sort | comparison of remote sensing time series smoothing methods for grassland spring phenology extraction on the qinghai tibetan plateau |
| topic | phenology smoothing methods Start of season (SOS) Qinghai–Tibetan Plateau (QTP) grassland |
| url | https://www.mdpi.com/2072-4292/12/20/3383 |
| work_keys_str_mv | AT nanli comparisonofremotesensingtimeseriessmoothingmethodsforgrasslandspringphenologyextractionontheqinghaitibetanplateau AT peizhan comparisonofremotesensingtimeseriessmoothingmethodsforgrasslandspringphenologyextractionontheqinghaitibetanplateau AT yaozhongpan comparisonofremotesensingtimeseriessmoothingmethodsforgrasslandspringphenologyextractionontheqinghaitibetanplateau AT xiufangzhu comparisonofremotesensingtimeseriessmoothingmethodsforgrasslandspringphenologyextractionontheqinghaitibetanplateau AT muyili comparisonofremotesensingtimeseriessmoothingmethodsforgrasslandspringphenologyextractionontheqinghaitibetanplateau AT dujuanzhang comparisonofremotesensingtimeseriessmoothingmethodsforgrasslandspringphenologyextractionontheqinghaitibetanplateau |