[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Next Article in Journal
Remote Sensing and Geospatial Analysis in the Big Data Era: A Survey
Previous Article in Journal
Multi-Dimensional Analysis of Urban Growth Characteristics Integrating Remote Sensing Data: A Case Study of the Beijing–Tianjin–Hebei Region
Previous Article in Special Issue
Matching Vegetation Indices and Tree Vigor in Pyrenean Silver Fir Stands
You seem to have javascript disabled. Please note that many of the page functionalities won't work as expected without javascript enabled.
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 17
Issue 3
10.3390/rs17030549
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Monitoring Mangrove Phenology Based on Gap Filling and Spatiotemporal Fusion: An Optimized Mangrove Phenology Extraction Approach (OMPEA)

by
Yu Hong
1,2,
Runfa Zhou
1,2,
Jinfu Liu
1,2,*,
Xiang Que
1,3,
Bo Chen
1,2,
Ke Chen
1,2,
Zhongsheng He
1,2 and
Guanmin Huang
4
1
Fujian Agriculture and Forestry University, Fuzhou 350002, China
2
Key Laboratory of Fujian Universities for Ecology and Resources Statistics, Fujian Agriculture and Forestry University, Fuzhou 350002, China
3
Department of Computer Science, University of Idaho, Moscow, ID 83844, USA
4
Zhangjiangkou National Mangrove Nature Reserve, Zhangzhou 363300, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2025, 17(3), 549; https://doi.org/10.3390/rs17030549
Submission received: 1 December 2024 / Revised: 23 January 2025 / Accepted: 4 February 2025 / Published: 6 February 2025
(This article belongs to the Special Issue Assessment of Spatial and Temporal Patterns in Forest and Grassland Ecosystems Based on Spectral Metrics)
Browse Figures
Figure 1
<p>Map of study area: (<b>a</b>) the overall distribution of study area; (<b>b1</b>–<b>b4</b>) Zhangjiangkou National Mangrove Nature Reserve (ZNR) in Fujian Province, Qi’ao Island Provincial Nature Reserve (QPR) in Guangdong Province, Beilun Estuary National Nature Reserve (BNR) in Guangxi Province, and Dongzhaigang National Mangrove Nature Reserve (DNR) in Hainan Province.</p> "> Figure 2
<p>Workflow of mangrove phenology extraction based on OMPEA.</p> "> Figure 3
<p>Landsat 8 NDVI (16-day 30 m) and denoised Landsat NDVI (16-day 30 m) generated by OMPEA. Gray pixel indicates pixel with no data.</p> "> Figure 4
<p>MODIS NDVI (1-day 500 m) and denoised MODIS NDVI (1-day 30 m) generated by OMPEA. Gray pixel indicates pixel with no data.</p> "> Figure 5
<p>The OMPEA-generated fused NDVI imagery. Gray pixel indicates pixel with no data.</p> "> Figure 6
<p>Scatter density plots and marginal histograms of fused NDVI and denoised Landsat NDVI.</p> "> Figure 7
<p>Composite scatter plots and line plots of various NDVI time series.</p> "> Figure 8
<p>Fused NDVI time-series curve and phenological parameters.</p> "> Figure 9
<p>Boxplots of mangrove phenological parameters.</p> "> Figure 10
<p>The time-series curves for fused NDVI, precipitation, temperature, and their lagged time-series curves with corresponding lag days.</p> "> Figure 11
<p>The OMPEA-generated fused NDVI in QPR from 17 January 2020 to 24 March 2021. (<b>a</b>) Description of denoised Landsat 8 NDVI in a full-time range. (<b>b</b>) Description of denoised Landsat 8 NDVI across three different time ranges, (<b>c</b>,<b>d</b>) is fused NDVI that using (<b>a</b>,<b>b</b>) as inputs, respectively. Gray pixel indicates pixel with no data.</p> ">
Versions Notes

Abstract

:
Monitoring mangrove phenology requires frequent, high-resolution remote sensing data, yet satellite imagery often suffers from coarse resolution and cloud interference. Traditional methods, such as denoising and spatiotemporal fusion, faced limitations: denoising algorithms usually enhance temporal resolution without improving spatial quality, while spatiotemporal fusion models struggle with prolonged data gaps and heavy noise. This study proposes an optimized mangrove phenology extraction approach (OMPEA), which integrates Landsat and MODIS data with a denoising algorithm (e.g., Gap Filling and Savitzky–Golay filtering, GF–SG) and a spatiotemporal fusion model (e.g., Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model, ESTARFM). The key of OMPEA is that GF–SG algorithm filled data gaps from cloud cover and satellite transit gaps, providing high-quality input to ESTARFM and improving its accuracy of NDVI imagery reconstruction in mangrove phenology extraction. By conducting experiments on the GEE platform, OMPEA generates 1-day, 30 m NDVI imagery, from which phenological parameters (i.e., the start (SoS), end (EoS), length (LoS), and peak (PoS) of the growing season) are derived using the maximum separation (MS) method. Validation in four mangrove areas along the coastal China shows that OMPEA significantly improves the potential to capture mangrove phenology in the presence of incomplete data. The OMPEA significantly increased usable data, adding 7–33 Landsat images and 318–415 MODIS images per region. The generated NDVI series exhibits strong spatiotemporal consistency with original data (R2: 0.788–0.998, RMSE: 0.007–0.253) and revealed earlier SoS and longer LoS at lower latitudes. Cross-correlation analysis showed a 2–3 month lagged effects of temperature on mangroves’ growth, with precipitation having minimal impact. The proposed OMPEA improves the possibility of capturing mangrove phenology under non-continuous and low-resolution data, providing valuable insights for large-scale and long-term mangrove conservation and management.

1. Introduction

Mangrove forests are one of the most productive and biodiversity-rich coastal ecosystems [1]. Global warming and human activities have gradually degraded mangrove habitats, disrupting key physiological processes such as leaf spreading, flowering, and fruiting [2]. Monitoring these processes is essential for understanding the impacts of climate change and human interference on mangrove ecosystems and has become a focal point in mangrove phenology research [3].
Traditional phenological monitoring of mangrove forests relies on ground-based observations to determine key parameters such as the start (SoS), end (EoS), length (LoS), and peak (PoS) of the growing season [4]. While ground-based observations provide detailed localized insights, they lack the spatial and temporal coverage needed to understand large-scale and long-term changes in mangrove phenology [5]. The introduction of remote sensing has revolutionized phenological studies, offering the ability to monitor mangrove forests at broader spatial and temporal scales [6,7]. For instance, Pastor-Guzman et al. [8] extracted vegetation indices, such as normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI), from MODIS imagery to investigate mangrove phenology and its relationship with environmental factors.
Satellite datasets, including MODIS, Landsat, Sentinel, and AVHRR, are widely used for phenological studies [9]. Selecting data with appropriate spatial and temporal resolutions is crucial, as higher temporal resolution improves the accuracy of phenological event detection [10]. Mangrove growth stages often exhibit rapid changes within two-week intervals [11], and satellites like Landsat and Sentinel, with revisit intervals of 5–16 days, may miss critical phenological events [6]. Although 1-day MODIS data offer higher temporal resolution, cloud cover—prevalent in mangrove regions—reduces the number of valid observations, compromising phenological monitoring reliability [12,13]. Missing data during key periods like SoS or EoS can lead to errors of up to 73 days [14]. Furthermore, low-resolution imagery often fails to capture the heterogeneity of mangrove forests, reducing sensitivity to phenological variations [15]. High-resolution data can better represent mangrove distributions and growth dynamics, improving phenological parameter accuracy [16].
An optimal approach to mangrove phenology monitoring must integrate high temporal and spatial resolutions while addressing data gaps caused by cloud cover. Denoising algorithms, such as Savitzky–Golay (S-G) filtering, Fourier Transforms, and Double Logistic functions, are commonly used for reconstructing vegetation index time series [17,18,19]. Among these, S–G filtering is particularly effective for smoothing noisy data while preserving critical features [20]. To improve its performance, Chen et al. [21] introduced the gap filling and Savitzky–Golay filtering (GF–SG) algorithm, which combines gap filling with S-G filtering, significantly reducing the impact of noise. However, while denoising methods enhance temporal resolution, they do little to improve spatial resolution, which is essential for detailed mangrove monitoring. Spatiotemporal fusion algorithms, such as spatial and temporal adaptive reflectance fusion model (STARFM), flexible spatiotemporal data fusion (FSDAF), and enhanced STARFM (ESTARFM), address this limitation by combining high-spatial, low-temporal resolution data (e.g., Landsat) with low-spatial, and high-temporal resolution data (e.g., MODIS) [22,23,24]. ESTARFM, in particular, excels in handling noise and outliers, making it suitable for complex environments like mangrove-tidal boundaries [25]. However, spatiotemporal fusion requires continuous and complete baseline observations, limiting its applicability in areas with prolonged data gaps or high noise levels [26]. Integrating denoising methods with spatiotemporal fusion can overcome these challenges, enabling high-precision phenological monitoring in mangrove environments.
This study introduces the optimized mangrove phenology extraction approach (OMPEA), which attempts to combine the GF–SG algorithm with ESTARFM to overcome data gaps and generate high-resolution NDVI imagery. Key phenological parameters, including SoS, EoS, LoS, and PoS, are extracted using the maximum separability (MS) method, which identifies phenological transitions in NDVI time series. Furthermore, the cross-correlation function (CCF) is applied to analyze the effects of temperature and precipitation on mangrove phenology. The method is validated in four representative mangrove regions in China, where its performance is evaluated using statistical metrics and visual comparisons. The specific objectives of this paper are as follows: (1) a widely applicable optimization method for mangrove phenology extraction; (2) to generate 1-day 30 m fused NDVI imagery from Landsat and MODIS that can reflect phenology patterns of mangrove forests; (3) to extract and compare mangrove phenological parameters in different regions; and (4) to analyze the effect of temperature and precipitation on mangrove phenology differences. The findings of this study contribute to a more comprehensive understanding of mangrove growth dynamics, which is of significant value for the conservation and management of mangroves, particularly with regard to ecological restoration. Furthermore, the proposed methodology provides a reference scheme for large-scale and long-term plant phenology monitoring in cloudy areas.

2. Materials and Methods

2.1. Study Area

This study focuses on four representative mangrove regions along the southeastern coast of China (Figure 1): Zhangjiangkou National Mangrove Nature Reserve (ZNR) in Fujian Province, Qi’ao Island Provincial Nature Reserve (QPR) in Guangdong Province, Beilun Estuary National Nature Reserve (BNR) in Guangxi Province, and Dongzhaigang National Mangrove Nature Reserve (DNR) in Hainan Province. These sites span a latitudinal gradient from tropical (19°51′N) to subtropical (23°56′N), offering diverse ecological conditions and mangrove landscapes.
ZNR has the most extensive natural mangrove community north of the Tropic of Cancer, with dominant species including Kandelia candel (KC), Aegiceras corniculatum (AC), and Avicennia marina [27]. QPR preserves the most complete and concentrated mangrove stands in Zhuhai City, China, with AC and Sonneratia apetala being the predominant species [28]. BNR has the largest contiguous mangrove forest on mainland China, featuring species such as AC, KC, Excoecaria agallocha, and Acanthus ilicifolius [29]. Established as China’s first mangrove wetland reserve, DNR remains the largest, most species-diverse, best-preserved, and most biodiverse mangrove reserve in China [30]. An analysis of remote sensing imagery from 2020 revealed significant cloud cover across all sites. The number of valid values for each pixel varied from 0 to 33 throughout the year, with the majority falling below 15. This persistent cloud cover complicates remote sensing observations and highlights the challenges of monitoring mangroves phenology in these regions.

2.2. Dataset

2.2.1. Remote Sensing Data

Landsat 8 OLI data and MODIS reflectance products for the study area were acquired from the Google Earth Engine (GEE) platform (https://earthengine.google.com/ (accessed on 15 March 2024)). To capture the full mangrove growing season and to support spatiotemporal interpolation, data from 1 October 2019 to 31 May 2021 were utilized. (1) A total of 32, 27, 26, and 60 Landsat 8 OLI images were collected for ZNR, QPR, BNR, and DNR, respectively. The data, with a spatial resolution of 30 m and a revisit period of 16 days, include nine spectral bands ranging from visible light to thermal infrared [31]. Cloud and cloud-shadow pixels were excluded based on quality flags. (2) The MODIS reflectance products used include MODIS Surface Reflectance Daily L2G Global 500 m (MOD09GA) and MODIS Surface Reflectance 8-day L3 Global 250 m (MOD09Q1). MOD09GA provided 607 daily scenes per region across seven spectral bands (visible to shortwave infrared) [32]. MOD09Q1 offered 22 scenes per region with 8-day composites in two bands (e.g., red and near-infrared) [33]. All MODIS data were reprojected to the Landsat coordinate system (WGS1984), and cloud and cloud-shadow pixels were removed according to QA pixel Bitmask. NDVI was calculated from both datasets to extract phenological parameters.

2.2.2. Environmental Data

Mangrove distribution data were sourced from the China_Mangrove_2020 dataset [34], which employs Landsat imagery and multi-source auxiliary data processed with object-oriented classification methods and has an accuracy exceeding 90%. This mangrove dataset filters the effects of tides on mangrove areas. The mangrove areas are not prone to flooding, which minimizes the impact of tides on mangrove NDVI analysis. The temperature and precipitation were obtained from Climatologies at high resolution for the earth’s land surface dataset (CHELSA) (https://chelsa-climate.org/ (accessed on 2 August 2024)). CHELSA provided daily climate data with a spatial resolution of approximately 1 km [35]. The temperature and precipitation values were converted to °C and mm, respectively.

2.3. Methods

We developed an optimized mangrove phenology extraction approach (OMPEA) and implemented it on the GEE platform (Figure 2). This approach enhances the quality and continuity of remote sensing imagery for mangrove phenology extraction. The process involves the following points. (1) Landsat Denoising: Pre-processing Landsat 8 NDVI imagery with MOD09Q1 NDVI imagery using GF–SG algorithm to generate 16-day 30 m denoised Landsat NDVI. (2) MODIS Denoising: Pre-processing MOD09GA NDVI data using the simplified GF–SG algorithm to generate 1-day 30 m denoised MODIS NDVI. (3) Spatiotemporal Fusion: Generating high-resolution time-series data via an ESTARFM-like algorithm to create a 1-day 30 m fused NDVI imagery that contains mangrove phenology dynamics. (4) Phenology Extraction: Extracting phenological parameters (e.g., SoS, EoS, LoS, and PoS) using the MS method and analyzing environmental influences with CCF.

2.3.1. De-Noising Using GF–SG Algorithm

The GF–SG algorithm [21] consists of two main steps. (1) The first step is generating MODIS–Landsat NDVI paired time series to fill gaps in the original Landsat NDVI time series. (2) The second step is applying a weighted S-G filter to reduce noise in the NDVI time series. The GF–SG algorithm can directly utilize information from neighboring pixels to estimate the value of the target pixel, leveraging the high spatial autocorrelation of NDVI data without the need for a predefined land cover classification. This process is fully utilized for Landsat denoising, while only the second step is applied for MODIS denoising.

Landsat Denoising

The 16-day 30 m Landsat 8 NDVI and 8-day 250 m MOD09Q1 NDVI imagery were the inputs to GF–SG algorithm for Landsat denoising. Before this process, MOD09Q1 NDVI imagery was resampled to a 30 m resolution using bicubic interpolation to align spatially with Landsat data. To further reduce cloud contamination, S-G filtering was applied to MOD09Q1 NDVI imagery, resulting in a processed dataset denoted as M i n t e r p o l .
The cloud-free Landsat 8 NDVI imagery, denoted as L s e r i e s , was obtained for the available acquisition dates ( T ). For each date t i in T , the closest temporal match in M i n t e r p o l was selected to construct a time-matched NDVI series, denoted as M s e r i e s . Subsequently, the correlation coefficient between the corresponding images of L s e r i e s t i and M s e r i e s t i was calculated using a 31 × 31 pixel moving window, following the formula
R j ( t i , x j , y j ) = c o r j ( t i , x j , y j ) − c o r m i n ( t i , x j , y j ) c o r m a x ( t i , x j , y j ) − c o r m i n ( t i , x j , y j ) ,
where x , y denotes the location of the central pixel and ( t i , x j , y j ) denotes the j-th neighboring pixel within the moving window at time t i . R j ( t i , x j , y j ) denotes the normalized correlation coefficient; c o r j ( t i , x , y ) denotes the correlation coefficient between L s e r i e s t i , x j , y j and M s e r i e s t i , x j , y j ; and c o r m i n ( t i , x , y ) and c o r m a x ( t i , x , y ) denote minimum and maximum correlation coefficients within the window, respectively. The neighboring pixels with R j t i , x j , y j > 0.8 are retained and form a subset, denoted as M s i m i l a r t i , x j , y j .
Furthermore, M s i m i l a r t i , x j , y j is used to calculate the NDVI of reference pixel M r e f e r e n c e ( t i , x , y ) :
w ( t i , x j , y j ) = R j ( t i , x j , y j ) ∑ j = 1 n R j ( t i , x , y ) ,
M r e f e r e n c e ( t i , x , y ) = ∑ j = 1 n w ( t i , x j , y j ) × M s i m i l a r ( t i , x j , y j ) ,
where w ( t i , x j , y j ) denotes the weighted average of n correlation coefficients greater than 0.8 within the window at time t i . The 30 m resolution reference image, M r e f e r e n c e t i , is the weighted average of M s e r i e s t i that are both spatially adjacent to and have value similar to the cloud-free L s e r i e s t i .
M r e f e r e n c e t i is then used to fill gaps in L s e r i e s t i . To address spectral differences between Landsat 8 and MODIS, a linear correction function is applied to refine M r e f e r e n c e t i , x , y , ensuring improved performance:
M a d j u s t e d ( t i , x , y ) = M r e f e r e n c e ( t i , x , y ) × a ( t i , x , y ) + b ( t i , x , y ) ,
where a ( t i , x , y ) and b ( t i , x , y ) denote the slope and intercept, respectively, calculated using the least squares method based on L s e r i e s t i , x , y and M r e f e r e n c e ( t i , x , y ) at time t i . For cloud-free pixels at t i , x , y , the original value of L s e r i e s t i , x , y is retained. For cloud-affected pixels, the adjusted value M a d j u s t e d t i , x , y at 30 m resolution is used to fill the missing data, resulting in a new, synthesized time series denoted as   N L s e r i e s .
However, the residual noise in the original L s e r i e s may affect the smoothness of   N L s e r i e s , and a weighted S-G filter is applied to smooth it. The weight used in the filter is defined as W i × A d j u s t i . In the NDVI time series, W i is determined by the distance between the NDVI value on the i-th date and the long-term trend curve, while A d j u s t i is based on whether the NDVI value on the i-th date originates from the   L s e r i e s   or   N L s e r i e s .
A d j u s t i ( x , y ) = 1 − s t d i ( x , y ) s t d m a x ,
where s t d i ( x , y ) denotes the standard deviation of neighboring pixels within a 21 × 21 pixel window centered at x , y for the i-th date; s t d m a x ( x , y ) denotes the maximum valu. If the pixel value at x , y on the i-th date originates from   L s e r i e s , then s t d i x , y = 0 and A d j u s t i x , y = 1 . The degree and half-width of the smoothing polynomial used in the weighted S-G filter are set to 6 and 4, respectively.

MODIS Denoising

The 1-day 500 m MOD09GA NDVI imagery was input into the GF–SG algorithm for MODIS denoising. The MOD09GA NDVI imagery was also resampled to a 30 m resolution using bicubic interpolation and smoothed using a weighted S-G filter.

2.3.2. ESTARFM-like Algorithm

An ESTARFM-like algorithm, adapted from Nietupski et al. [36] as a GEE version of ESTARFM [37], generated 1-day 30 m fused NDVI imagery. The method assumes stable systematic errors between MODIS and Landsat data constant over time. It employs a 10 × 10 pixel moving window method to identify similar pixels, applying weights to the images based on their spatial and spectral similarities:
L x w / 2 , y w / 2 , t p , b = L x w / 2 , y w / 2 , t 0 , b + ∑ i = 1 N W i × V i × M x i , y i , t p , b − M x i , y i , t 0 , b ,
where t 0 denotes a date with both denoised Landsat and MODIS observations, and t p denotes a date without a denoised Landsat observation. L x w / 2 , y w / 2 , t , b denotes the b -band pixel value of denoised Landsat at the center position of the moving window at time.  N denotes the number of similar pixels. M x i , y i , t , b denotes the b -band pixel value of denoised MODIS for the i -th similar pixel within the moving window at time t .   W i denotes the weight of the i -th similarity pixel. V i denotes the spectral conversion factor of the i -th similarity pixel.
The estimates of Landsat images are further corrected according to the weight of the reflectivity change amplitude at different times in MODIS images. According to Equation (6), the reflectivity of the denoised Landsat at t p can be predicted using the denoised MODIS at t n or t m , denoted as L 1 or L 2 , respectively. Furthermore, by applying the change amplitude of the denoised MODIS reflectivity between t n and t p and between t m and t p as the weight r W , a more accurate reflectivity value for the denoised Landsat image at t p can be obtained. The calculations are as follows:
r W k = 1 / ∑ j = 1 w ∑ i = 1 w M x i , y i , t k , b − ∑ j = 1 w ∑ i = 1 w M x i , y i , t p , b ∑ k = 1,2 1 / ∑ j = 1 w ∑ i = 1 w M x i , y i , t k , b − ∑ j = 1 w ∑ i = 1 w M x i , y i , t p , b ,
L x w / 2 , y w / 2 , t p , b = r W n × L 1 x w / 2 , y w / 2 , t n , b + r W m × L 2 x w / 2 , y w / 2 , t m , b .

2.3.3. The MS Method

The MS method [38] was used to extract phenological parameters (e.g., SoS, EoS) by determining dynamic thresholds for NDVI variations:
u = b m a x − b m i n p + b t ,
where b m a x and b m i n denote the maximum and minimum values of the NDVI time series, respectively; b t denotes NDVI at time t ; and p denotes the amplitude percentage, usually set to 0.5.
The NDVI time series was further converted into a binary sequence b ′ t , where b ′ t = 1 if b t > u , indicating that mangrove is in the growing season; otherwise b ′ t = 0. The proportional difference in growing period was calculated by a moving window with the following formula:
d t 0 = ∑ b b e f o r e ′ / n b e f o r e − ∑ b a f t e r ′ / n a f t e r ,
where n b e f o r e and n a f t e r denote the total number of observations before and after the window (30 days), respectively. b b e f o r e ′ and b a f t e r ′ denote observed values before and after the window (30 days), respectively. When d t 0 is closest to −1, the corresponding date is denoted as SoS; when d t 0 is closest to 1, the corresponding date is denoted as EoS. The LoS is the difference between SoS and EoS, while PoS is determined from the peak of NDVI time series.

2.3.4. Statistical Metrics

To comprehensively evaluate the performance of the OMPEA, the correlation coefficient (R2), root mean square error (RMSE), and Structural Similarity Index Measure (SSIM) were used to measure the difference between the original cloud-free NDVI imagery and the fitted NDVI imagery. SSIM is calculated as follows:
S S I M = ( 2 μ I μ I ′ + c 1 ) ( 2 σ I ′ + c 2 ) ( μ I 2 + μ I ′ 2 + c 1 ) ( σ I 2 + σ I ′ 2 + c 2 ) ,
where I and I ′ denote the original cloud-free and fitted NDVI imagery, respectively. SSIM is calculated using the mean μ , variance σ , and constants c 1 and c 2 from the I and I ′ . Note that c 1 = (Peak × 0.01)2 and c 2 = (Peak × 0.03)2.
In addition, CCF [39] was used to analyze the lagged relationships of NDVI with precipitation and temperature, respectively.
C C F = ∑ N i − N ¯ V i + τ − V ¯ ∑ N i − N ¯ 2 V i + τ − V ¯ 2 ,
where N i denotes the NDVI value of i -th day in the full time series, including the cloud-free original NDVI and filled NDVI. V i denotes environmental variables (temperature and precipitation) of i -th day, respectively; τ denotes the lag time.

3. Results

3.1. Image Denoising Performances

The performance of Landsat Denoising (Step 1) in OMPEA was evaluated across four mangrove regions. The 16-day 30 m various NDVI imagery, generated at 3-month intervals from January 2020 to April 2021, are shown in Figure 3. Compared to Landsat 8 NDVI, the denoised Landsat NDVI generated by OMPEA effectively reduced cloud-obscured pixels, such as QPR on 24 May 2021, BNR on 16 July 2020, and DNR on 8 January 2020. It even successfully reconstructed NDVI for fully cloud-covered dates, including QPR on 24 May 2020, BNR on 22 January 2020, and DNR on 4 April 2020. The number of usable remote sensing images in the four mangrove regions increased significantly—from 19, 14, 9, and 16 to 26, 21, 22, and 49, respectively. Statistical metrics indicated that the denoised Landsat NDVI was highly consistent with Landsat NDVI on same dates. The mean R2 and RMSE were 0.985 and 0.016 for ZNR, 0.967 and 0.031 for QPR, 0.893 and 0.015 for BNR, and 0.957 and 0.020 for DNR, demonstrating the effectiveness of OMPEA for Landsat NDVI reconstruction and image denoising.
Figure 4 compares denoised NDVI (1-day 30 m) generated through MODIS Denoising (Step 2) of OMPEA and 1-day 500 m MOD09GA NDVI imagery, for the same test dates as Figure 3. Significant gaps in 1-day 500 m MOD09GA NDVI across regions were effectively filled by OMPEA, resulting in 318, 331, 415, and 342 additional usable images across the four regions, respectively. While the denoised MODIS NDVI provides higher spatial resolution, it does not fully capture the fine details of mangrove imagery due to its interpolation from MOD09GA NDVI. The R2 for ZNR, QPR, BNR, and DNR are 0.788, 0.900, 0.904, and 0.890, with RMSE of 0.123, 0.129, 0.100, and 0.090, respectively. However, in BNR, the missing patches remained visible, likely due to prolonged image gaps that limited the effectiveness of interpolation based on NDVI temporal correlations in OMPEA.

3.2. Spatiotemporal Fusion Performance Based on Gap Filling

Figure 5 presents fused NDVI imagery (1-day 30 m) generated through Spatiotemporal Fusion (Step 3) of OMPEA. The fused NDVI imagery has smaller boundaries compared to the input imagery (Figure 3 and Figure 4) and exhibits missing-value patches in QPR, BNR, and DNR, with ZNR being the exception. These missing values arise from gaps in the input data, which ESTARFM-like algorithm propagates to corresponding locations and neighboring pixels during its moving window calculations. Despite these gaps, the fused NDVI imagery demonstrates high consistency with images from adjacent dates.
Figure 6 presents pixel values (e.g., fused NDVI and denoised Landsat 8 NDVI) that align closely along the diagonal, and the marginal histograms of the horizontal and vertical axes show similar distributions. The red columns represent the frequency distribution of Denoised Landsat 8 NDVI values, while the blue columns show the frequency distribution of Fused NDVI values. The lines indicate the fitted curves based on the histogram distributions. R2 for ZNR, QPR, BNR, and DNR are 0.991, 0.836, 0.998, and 0.941, with RMSE of 0.012, 0.253, 0.007, and 0.054, respectively. QPR exhibited lower performance due to temporal imbalances in the original data, while the results for the other mangrove regions were excellent. The study area is primarily confined to mangroves, covering a small area and consisting of only a single NDVI band. As a result, the SSIM accuracy for each region is as high as 0.999, making it difficult for SSIM to distinctly differentiate the fitting effect. Specifically, the temporal imbalance occurred because Landsat data in QPR were entirely missing between 18 February and 24 May 2020, but were otherwise consistently distributed throughout the rest of the study period. These findings confirm that OMPEA effectively integrates the spatiotemporal resolution advantages of multi-source data, producing frequent, high-resolution NDVI imagery suitable for phenology analysis.

3.3. Comparison of Reconstructed NDVI Time Series

Figure 7 compares four types of NDVI time series across mangrove regions, namely Landsat 8 NDVI, MOD09GA NDVI, the denoised Landsat 8 NDVI, and the denoised MODIS NDVI. MOD09GA NDVI is unsuitable for constructing reliable time series due to numerous missing or abnormal values, often concentrated between 0.0 and 0.1. Landsat 8 NDVI has fewer missing values but still lacks the temporal resolution necessary to extract precise phenological parameters. Landsat denoising of OMPEA addresses these gaps, creating smooth denoised Landsat NDVI data. However, its limited temporal variability makes it difficult to capture detailed phenological parameters over a growth cycle.
The spatiotemporal fusion in OMPEA enables fused NDVI to effectively capture detailed fluctuations in mangrove growth. As shown in Figure 8, fused NDVI trends reveal a general decrease or low values from January to April, an increase starting in April, a peak in late summer, and a gradual decline thereafter. Significant regional variations are evident: fused NDVI values ranged from 0.32 to 0.40 in ZNR, compared to 0.22 to 0.35 in QPR, BNR, and DNR. The PoS of fused NDVI occurred on day of year (DoY) 360 in ZNR, 196 in QPR, 416 in BNR, and 336 in DNR. QPR exhibited the greatest NDVI value increase before PoS, while BNR showed the most substantial decrease after PoS. Seasonal responses varied, with ZNR and QPR showing strong reactions to seasonal changes, peaking in summer and autumn. In contrast, DNR displayed a smoother seasonal trend, while BNR experienced a marked decline after PoS, potentially linked to environmental or human disturbances. These findings highlight the capacity of OMPEA to facilitate the detection of both temporal and spatial variations in mangrove phenology.

3.4. Phenological Parameters and Lagged Effects

Figure 9 shows the distribution of phenological parameters across ZNR, QPR, BNR, and DNR, revealing regional differences likely influenced by climate and geographical location. SoS occurs on DoY 201 ± 15 in ZNR, 175 ± 10 in QPR, 140 ± 12 in BNR, and 212 ± 23 in DNR, generally advancing from north to south, except for DNR. This pattern may be attributed to earlier spring and higher temperatures in southern regions, which promote plant growth. EoS occurs on DoY 429 ± 6 in ZNR, 380 ± 12 in QPR, 426 ± 5 in BNR, and 409 ± 41 in DNR. DNR shows the greatest variation, likely reflecting the diverse mangrove species with differing phenological traits. PoS is observed on DoY 363 ± 18 in ZNR, 192 ± 12 in QPR, 408 ± 14 in BNR, and 330 ± 14 in DNR. QPR peaks in summer, while BNR peaks in winter, suggesting that regional temperature and precipitation significantly influence peak growth timing. LoS spans DoY 219 ± 13 in ZNR, 210 ± 16 in QPR, 265 ± 10 in BNR, and 191 ± 9 in DNR. BNR exhibits the longest LoS, indicating favorable climate conditions for mangrove growth.
Typically, higher latitudes are associated with delayed SoS. However, the SoS in the low-latitude DNR occurs later than in the higher-latitude ZNR. To explore this variation, the lagged effects of temperature and precipitation on mangrove phenology were analyzed using CCF analysis. As shown in Table 1 and Figure 10, temperature increases have significant positive lagged effects on NDVI across all regions, with CCF values (p < 0.05) of 0.6276, 0.7791, 0.6576, and 0.5190, and corresponding lag times of 92, 55, 83, and 99 days, respectively. In contrast, precipitation shows weaker and less consistent lagged effects, with CCF values (p < 0.05) of 0.1396, 0.2327, 0.2057, and 0.2613, and lag times of 89, 100, 65, and 65 days, respectively. For example, in DNR, low temperatures and limited precipitation from January to May may delay SoS until July. In QPR, rapid cooling and reduced precipitation from October to December may cause EoS to occur earlier, in January. Overall, temperature exerts a strong influence on mangrove phenology, while precipitation plays a secondary role. These findings highlight the complex interplay between climatic factors and regional phenological variations.

4. Discussion

Our study introduces an OMPEA, an optimized mangrove phenology extraction by combining temporal smoothing and spatiotemporal interpolation to generate a 1-day 30 m fused NDVI imagery. The method addresses challenges in data availability caused by cloud cover and Satellite camera angle. For example, in BNR, less than half of the expected annual Landsat and MODIS images were usable due to cloud cover in 2020. While previous studies, such as that by Abowarda et al. [40], suggested combining cloud-free Landsat and MODIS data, significant disparities between the NDVI values from these sources made their direct integration into our study impractical. Compared to Landsat and MODIS, combining Landsat and Sentinel-2 data offers more humanized data with a temporal resolution of 3–5 days and a spatial resolution of 10–30 m. However, Sentinel-2 data may not be adequate for studying long-term changes in mangrove phenology, as it has been available only since 2015. In contrast, the MODIS and Landsat series provide continuous datasets spanning from 2000 to the present, making them more suitable for long-term research applications. The OMPEA effectively harmonized these MODIS and Landsat, producing NDVI outputs that aligned closely with observed seasonal mangrove growth dynamics, as validated through visual inspection and correlation analysis. The method’s robustness across diverse mangrove scenarios highlights its potential to support large-scale and long-term conservation efforts.
However, some limitations were observed in OMPEA’s performance. The method appears sensitive to both the time range and data quality of the input images, with variations in the start and end dates of the imagery impacting its effectiveness. To assess these factors, we compared synthetic results generated with different time ranges and associated cloud cover levels (Figure 11). Figure 11a,c show results generated using a continuous time range (16 December 2019 to 21 May 2021) with an average cloud cover of 54.64%, while Figure 11b,d were generated using three segmented periods, each with varying levels of cloud cover. The results indicate that higher cloud cover, even over a similar number of remote sensing imagery, leads to more significant gaps in imagery, particularly in Landsat denoising of OMPEA. Although the full-time range in Figure 11a did not have the highest average cloud cover, it had larger and more persistent missing values due to the extended time range. These incomplete results reduce the quality of the final fusion outputs in Spatiotemporal Fusion of OMPEA (Figure 11c,d). These findings highlight that both the time range and average cloud cover impact the performance of OMPEA. High cloud cover during shorter periods prevents the generation of complete images, while extended time ranges may hinder the method’s ability to fill missing values, even under similar cloud cover conditions.
In addition, relying solely on OMPEA-generated data for validation also has inherent limitations. While excluding certain high-quality pixels and filling them using the proposed method could provide a more robust evaluation of the model’s performance, the window-based calculations in the OMPEA framework introduce challenges. Excluding a single good pixel can impact the accuracy of multiple neighboring pixels, complicating the validation process. Directly validating the fitting results against actual data, without excluding pixels, has been widely utilized in related studies and demonstrated reliability in assessing model performance [21]. Future research will integrate datasets with repeated temporal coverage and comparable sensor characteristics, such as Landsat 8 and Landsat 9, or high-resolution data, such as 1 m drone imagery, to enable a more comprehensive validation of the model’s applicability and accuracy. It is also important to focus on exploring more advanced cloud detection and removal techniques to further improve the reliability of data fusion.
Incomplete fitted outputs due to missing values are a common challenge in vegetation phenology studies [41,42]. Strategies such as temporally consistent data [37] or adaptive moving window approaches [24] have been proposed, but their applicability is often constrained in mangrove regions with persistent cloud cover. The Spatiotemporal Shape-Matching Model (SSMM) can generate high spatiotemporal resolution time series using historical Landsat time series and limited cloud-free observations, which may be more suitable for monitoring the phenology of mangroves [43]. Ground-based calibration is a standard practice to ensure remote sensing data accuracy, but no official phenological parameters were available for the mangrove regions analyzed in our study. Validation was instead conducted by comparing results with nearby regions. For instance, phenological metrics from the Jiulong River estuary [44] were consistent with ZNR from our study, indicating reliability under similar environmental conditions [45]. Notably, this study found no significant variation in EoS across latitudes, unlike prior studies that reported delayed SoS and EoS at higher latitudes [44]. This discrepancy may stem from differences in data sources, vegetation indices, and phenology extraction methods [10]. Future research should explore strategies for combining optimal extraction methods with suitable vegetation indices to enhance accuracy and reliability.
Phenological differences are not only evident across latitudes but also among different mangrove species. Different mangrove species exhibit variations in SoS, EoS, and other growth dynamics during the growing season [46]. However, because the NDVI signal from mangroves typically reflects a composite of multiple species, directly quantifying these phenological differences remains challenging in this study. To accurately assess interspecies phenological variations, high-resolution remote sensing data or ground-based species classification is necessary. Due to spatial resolution limitations, many mangrove phenology studies, including ours, have focused on the overall phenological characteristics of mixed mangrove communities rather than on species-specific analysis [44,47]. While SoS and EoS are commonly used to identify the onset and cessation of mangrove growth, they cannot capture specific physiological processes, such as flowering, which often occurs during the peak or late stages of the growing season and sometimes multiple times a year [48]. This study focused on optimizing methods for extracting mangrove phenology, with future research expected to explore interspecies phenological differences using high-resolution imagery and advanced classification techniques.
Pastor-Guzman et al. [8] identified precipitation’s lagged effects on mangrove phenology but did not specify its duration. Our study investigated lagged effects of temperature and precipitation on NDVI by CCF analysis. Results confirmed that temperature and precipitation significantly influence mangrove growth, with a 2–3 month lag in NDVI response to temperature and PoS occurring 2–3 months after peak precipitation. For example, drought and low temperatures delayed the SoS in DNR, while similar conditions caused the EoS to occur earlier in QPR. These findings underscore the critical role of climatic anomalies in shaping mangrove phenology, aligning with Chamberlain et al. [20], who noted that prolonged low precipitation led to a reduction in NDVI, delayed SoS, and an earlier EoS.
Beyond climate, other environmental factors such as water, salinity, and typhoons also affect mangrove phenology. Water under mangrove may complicate phenology extraction by influencing NDVI values. High reflectivity in the near-infrared band, especially at mangrove edges, may lower NDVI, particularly during spectral mixing with water at high levels [49]. Additionally, tidal variations lead to inconsistencies in water levels, affecting NDVI across the year, as mangroves in intertidal zones may become submerged during high tides [50]. However, the impact of water depth on mangrove NDVI is generally minimal [51]. Our study used a dataset that excluded tidal influences, but future work could minimize the effects of water by enhancing spatial resolution to reduce mixed pixels and analyzing NDVI under varying tidal conditions. Typhoons also impact seawater salinity through salt intrusion, freshwater dilution, and temperature redistribution, which can affect mangrove growth [52]. Despite mangroves’ self-regulation mechanisms to adapt to salinity changes [53], typhoons may lower NDVI due to leaf and branch damage [54]. While some mangroves recover within a season, others experience slower or continued decline [55]. In 2020, several typhoons impacted regions near our study area, such as Nuri in Guangxi, Sinlaku in Guangdong, Hagupit in Fujian, Higos in the Pearl River Estuary, and Linfa in Hainan, but their effects were limited due to distance. However, the landfall points of these typhoons were approximately 100–200 km away from our study area, resulting in a relatively limited impact. While many studies focus on individual factors, the complex interactions among these variables require further exploration [56] to enhance our understanding and conservation strategies.

5. Conclusions

This study introduces the OMPEA, a novel approach for extracting mangrove phenological parameters by integrating multi-source data (e.g., Landsat and MODIS) with denoising interpolation and spatiotemporal fusion algorithms. The OMPEA enables the generation of 1-day 30 m fused NDVI data, an achievement not possible with individual methods. This makes it particularly valuable for capturing mangrove growth dynamics in data-limited scenarios, providing critical insights for large-scale, periodic mangrove monitoring. Implemented on the GEE platform, the method was applied to four representative mangrove regions in China (i.e., ZNR, QPR, BNR, and DNR), yielding phenological parameter estimates with acceptable performance.
The OMPEA effectively reduces missing values in remote sensing data and significantly increases the availability of 16-day 30 m and 1-day 500 m NDVI imagery while maintaining high spatiotemporal consistency with the original sources (e.g., Landsat 8, MODIS). The generated 1-day 30 m fused NDVI imagery captures the seasonal dynamics of mangrove growth, with NDVI values declining or remaining low from January to April, increasing after April, peaking in late summer, and gradually decreasing thereafter. The study also found that mangrove phenology is closely linked to geographic location, with SoS occurring earlier in southern regions compared to northern ones. However, no consistent geographic patterns were observed for EoS or PoS. Additionally, significant lagged effects of temperature on NDVI were observed, with a 2–3 month delay, while precipitation also showed a positive correlation with NDVI.
OMPEA holds great promise for advancing mangrove ecosystem monitoring, refining ecological restoration strategies, and supporting global mangrove conservation efforts. While the method cannot fully eliminate data gaps due to cloud contamination, the integration of multi-source data and advanced spatiotemporal techniques offers significant improvements. Future studies could further refine mangrove phenology extraction by exploring the optimal combination of datasets, algorithms, and parameters, or by integrating near-surface remote sensing observations.

Author Contributions

Conceptualization, Y.H. and J.L.; methodology, Y.H.; investigation, G.H.; data curation, Y.H., R.Z. and B.C.; writing—original draft preparation, Y.H.; writing—review and editing, Y.H., R.Z., J.L., X.Q., B.C. and Z.H.; visualization, Y.H., R.Z. and K.C.; supervision, J.L.; funding acquisition, J.L., X.Q. and Z.H. All authors have read and agreed to the published version of the manuscript.

Funding

This research was supported by the National Natural Science Foundation of China (42202333); the Key Project of Scientific and Technological Innovation of Fujian Province (2021G02007); Fujian Provincial Science and Technology Innovation Project (Southeast Ecological Restoration [2021], No. 4 KY-090000-04-2021-013); Science and Technology Innovation Project of Fujian Agriculture and Forestry University (KFB23044A and KFB23150); Forestry Technology Research Project of Fujian Province (2024FKJ17).

Data Availability Statement

The data used in the study are archived in GEE and source codes of OMPEA are freely accessible at https://github.com/hongyu0516/GEE-Mangrove-phenology.git (accessed on 1 December 2024).

Acknowledgments

The authors thank anonymous reviewers for their constructive comments on an earlier version of the paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Biswas, P.L.; Biswas, S.R. Mangrove Forests: Ecology, Management, and Threats, in: Encyclopedia of the UN Sustainable Development Goals; Springer International Publishing: Cham, Switzerland, 2020; pp. 627–640. [Google Scholar] [CrossRef]
  2. Wang, Y.-S.; Gu, J.-D. Ecological responses, adaptation and mechanisms of mangrove wetland ecosystem to global climate change and anthropogenic activities. Int. Biodeterior. Biodegrad. 2021, 162, 105248. [Google Scholar] [CrossRef]
  3. Zhang, R.; Jia, M.; Wang, Z.; Zhou, Y.; Mao, D.; Ren, C.; Zhao, C.; Liu, X. Tracking annual dynamics of mangrove forests in mangrove National Nature Reserves of China based on time series Sentinel-2 imagery during 2016–2020. Int. J. Appl. Earth Obs. Geoinf. 2022, 112, 102918. [Google Scholar] [CrossRef]
  4. Coupland, G.T.; Paling, E.I.; McGuinness, K.A. Vegetative and reproductive phenologies of four mangrove species from northern australia. Aust. J. Bot. 2005, 53, 109. [Google Scholar] [CrossRef]
  5. Wang, L.; Jia, M.; Yin, D.; Tian, J. A review of remote sensing for mangrove forests: 1956–2018. Remote Sens. Environ. 2019, 231, 111223. [Google Scholar] [CrossRef]
  6. Grossman, J.J. Phenological physiology: Seasonal patterns of plant stress tolerance in a changing climate. New Phytol. 2022, 237, 1508–1524. [Google Scholar] [CrossRef] [PubMed]
  7. You, Q.; Deng, W.; Tang, X.; Liu, Y.; Lei, P.; Chen, J.; You, H. Monitoring of mangrove dynamic change in Beibu Gulf of Guangxi based on reconstructed time series images. Sci. Total. Environ. 2024, 917, 170395. [Google Scholar] [CrossRef]
  8. Pastor-Guzman, J.; Dash, J.; Atkinson, P.M. Remote sensing of mangrove forest phenology and its environmental drivers. Remote Sens. Environ. 2018, 205, 71–84. [Google Scholar] [CrossRef]
  9. Gong, Z.; Ge, W.; Guo, J.; Liu, J. Satellite remote sensing of vegetation phenology: Progress, challenges, and opportunities. ISPRS. J. Photogramm. Remote Sens. 2024, 217, 149–164. [Google Scholar] [CrossRef]
  10. Younes, N.; Joyce, K.E.; Maier, S.W. All models of satellite-derived phenology are wrong, but some are useful: A case study from northern Australia. Int. J. Appl. Earth Obs. Geoinf. 2021, 97, 102285. [Google Scholar] [CrossRef]
  11. Berlanga-Robles, C.A.; Ruiz-Luna, A.; Villanueva, M.R.N. Seasonal trend analysis (STA) of MODIS vegetation index time series for the mangrove canopy of the Teacapan-Agua Brava lagoon system, Mexico. GIScience Remote Sens. 2018, 56, 338–361. [Google Scholar] [CrossRef]
  12. Songsom, V.; Koedsin, W.; Ritchie, R.J.; Huete, A. Mangrove Phenology and Environmental Drivers Derived from Remote Sensing in Southern Thailand. Remote Sens. 2019, 11, 955. [Google Scholar] [CrossRef]
  13. Younes, N.; Northfield, T.D.; Joyce, K.E.; Maier, S.W.; Duke, N.C.; Lymburner, L. A Novel Approach to Modelling Mangrove Phenology from Satellite Images: A Case Study from Northern Australia. Remote Sens. 2020, 12, 4008. [Google Scholar] [CrossRef]
  14. Zhang, X.; Friedl, M.A.; Schaaf, C.B. Sensitivity of vegetation phenology detection to the temporal resolution of satellite data. Int. J. Remote Sens. 2009, 30, 2061–2074. [Google Scholar] [CrossRef]
  15. Klosterman, S.T.; Hufkens, K.; Gray, J.M.; Melaas, E.; Sonnentag, O.; Lavine, I.; Mitchell, L.; Norman, R.; Friedl, M.A.; Richardson, A.D. Evaluating remote sensing of deciduous forest phenology at multiple spatial scales using PhenoCam imagery. Biogeosciences 2014, 11, 4305–4320. [Google Scholar] [CrossRef]
  16. Hong, Y.; Que, X.; Wang, Z.; Ma, X.; Wang, H.; Salati, S.; Liu, J. Mangrove extraction from super-resolution images generated by deep learning models. Ecol. Indic. 2024, 159, 111714. [Google Scholar] [CrossRef]
  17. Cao, R.; Chen, Y.; Shen, M.; Chen, J.; Zhou, J.; Wang, C.; Yang, W. A simple method to improve the quality of NDVI time-series data by integrating spatiotemporal information with the Savitzky-Golay filter. Remote Sens. Environ. 2018, 217, 244–257. [Google Scholar] [CrossRef]
  18. Wagenseil, H.; Samimi, C. Assessing spatio-temporal variations in plant phenology using Fourier analysis on NDVI time series: Results from a dry savannah environment in Namibia. Int. J. Remote Sens. 2006, 27, 3455–3471. [Google Scholar] [CrossRef]
  19. Yang, N.; Luo, N.; Huang, N.; Wu, N.; Sun, N. Weighted Double-Logistic Function Fitting Method for Reconstructing the High-Quality Sentinel-2 NDVI Time Series Data Set. Remote Sens. 2019, 11, 2342. [Google Scholar] [CrossRef]
  20. Chamberlain, D.A.; Phinn, S.R.; Possingham, H.P. Mangrove Forest Cover and Phenology with Landsat Dense Time Series in Central Queensland, Australia. Remote Sens. 2021, 13, 3032. [Google Scholar] [CrossRef]
  21. Chen, Y.; Cao, R.; Chen, J.; Liu, L.; Matsushita, B. A practical approach to reconstruct high-quality Landsat NDVI time-series data by gap filling and the Savitzky-Golay filter. ISPRS. J. Photogramm. Remote Sens. 2021, 180, 174–190. [Google Scholar] [CrossRef]
  22. Gao, F.; Masek, J.; Schwaller, M.; Hall, F. On the blending of the Landsat and MODIS surface reflectance: Predicting daily Landsat surface reflectance. IEEE Trans. Geosci. Remote Sens. 2006, 44, 2207–2218. [Google Scholar] [CrossRef]
  23. Zhu, X.; Helmer, E.H.; Gao, F.; Liu, D.; Chen, J.; Lefsky, M.A. A flexible spatiotemporal method for fusing satellite images with different resolutions. Remote Sens. Environ. 2016, 172, 165–177. [Google Scholar] [CrossRef]
  24. Liu, M.; Liu, X.; Dong, X.; Zhao, B.; Zou, X.; Wu, L.; Wei, H. An Improved Spatiotemporal Data Fusion Method Using Surface Heterogeneity Information Based on ESTARFM. Remote Sens. 2020, 12, 3673. [Google Scholar] [CrossRef]
  25. Ma, Y.; Jiang, G.; Huang, J.; Shen, Y.; Guan, H.; Dong, Y.; Li, J.; Hu, C. Evaluating the Ability of the Sentinel-1 Cross-Polarization Ratio to Detect Spring Maize Phenology Using Adaptive Dynamic Threshold. Remote Sens. 2024, 16, 826. [Google Scholar] [CrossRef]
  26. Wang, Q.; Atkinson, P.M. Spatio-temporal fusion for daily Sentinel-2 images. Remote Sens. Environ. 2018, 204, 31–42. [Google Scholar] [CrossRef]
  27. Chen, R.; Zhang, R.; Zhao, C.; Wang, Z.; Jia, M. High-Resolution Mapping of Mangrove Species Height in Fujian Zhangjiangkou National Mangrove Nature Reserve Combined GF-2, GF-3, and UAV-LiDAR. Remote Sens. 2023, 15, 5645. [Google Scholar] [CrossRef]
  28. Ma, Y.; Wang, W.; Gao, F.; Yu, C.; Feng, Y.; Gao, L.; Zhou, J.; Shi, H.; Liu, C.; Kong, D.; et al. Acidification and hypoxia in seawater, and pollutant enrichment in the sediments of qi’ao island mangrove wetlands, pearl river estuary, china. Ecol. Indic. 2024, 158, 111589. [Google Scholar] [CrossRef]
  29. Zhang, B.; Zhang, L.; Chen, B.; Deng, L.; Fu, B.; Yan, M.; Ji, C. Assessment of mangrove health based on pressure-state-response framework in Guangxi Beibu Gulf, China. Ecol. Indic. 2024, 167, 112685. [Google Scholar] [CrossRef]
  30. Zhang, Y.; Xin, K.; Sheng, N.; Xie, Z.; Liao, B. The regenerative capacity of eight mangrove species based on propagule traits in Dongzhai Harbor, Hainan Province, China. Glob. Ecol. Conserv. 2021, 30, e01788. [Google Scholar] [CrossRef]
  31. Roy, D.P.; Wulder, M.A.; Loveland, T.R.; Woodcock, C.E.; Allen, R.G.; Anderson, M.C.; Helder, D.; Irons, J.R.; Johnson, D.M.; Kennedy, R.; et al. Landsat-8: Science and product vision for terrestrial global change research. Remote Sens. Environ. 2014, 145, 154–172. [Google Scholar] [CrossRef]
  32. Hao, X.; Huang, G.; Zheng, Z.; Sun, X.; Ji, W.; Zhao, H.; Wang, J.; Li, H.; Wang, X. Development and validation of a new MODIS snow-cover-extent product over China. Hydrol. Earth Syst. Sci. 2021, 26, 1937–1952. [Google Scholar] [CrossRef]
  33. Lu, S.; Jia, L.; Zhang, L.; Wei, Y.; Baig, M.H.A.; Zhai, Z.; Meng, J.; Li, X.; Zhang, G. Lake water surface mapping in the Tibetan Plateau using the MODIS MOD09Q1 product. Remote Sens. Lett. 2016, 8, 224–233. [Google Scholar] [CrossRef]
  34. Jia, M.; Wang, Z.; Zhang, Y.; Mao, D.; Wang, C. Monitoring loss and recovery of mangrove forests during 42 years: The achievements of mangrove conservation in China. Int. J. Appl. Earth Obs. Geoinf. 2018, 73, 535–545. [Google Scholar] [CrossRef]
  35. Karger, D.N.; Conrad, O.; Böhner, J.; Kawohl, T.; Kreft, H.; Soria-Auza, R.W.; Zimmermann, N.E.; Linder, H.P.; Kessler, M. Climatologies at high resolution for the earth’s land surface areas. Sci Data 2017, 4, 170122. [Google Scholar] [CrossRef] [PubMed]
  36. Nietupski, T.C.; Kennedy, R.E.; Temesgen, H.; Kerns, B.K. Spatiotemporal image fusion in Google Earth Engine for annual estimates of land surface phenology in a heterogenous landscape. Int. J. Appl. Earth Obs. Geoinf. 2021, 99, 102323. [Google Scholar] [CrossRef]
  37. Zhu, X.; Chen, J.; Gao, F.; Chen, X.; Masek, J.G. An enhanced spatial and temporal adaptive reflectance fusion model for complex heterogeneous regions. Remote Sens. Environ. 2010, 114, 2610–2623. [Google Scholar] [CrossRef]
  38. Descals, A.; Verger, A.; Yin, G.; Penuelas, J. A Threshold Method for Robust and Fast Estimation of Land-Surface Phenology Using Google Earth Engine. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2021, 14, 601–606. [Google Scholar] [CrossRef]
  39. Andronis, V.; Karathanassi, V.; Tsalapati, V.; Kolokoussis, P.; Miltiadou, M.; Danezis, C. Time Series Analysis of Landsat Data for Investigating the Relationship between Land Surface Temperature and Forest Changes in Paphos Forest, Cyprus. Remote Sens. 2022, 14, 1010. [Google Scholar] [CrossRef]
  40. Abowarda, A.S.; Bai, L.; Zhang, C.; Long, D.; Li, X.; Huang, Q.; Sun, Z. Generating surface soil moisture at 30 m spatial resolution using both data fusion and machine learning toward better water resources management at the field scale. Remote Sens. Environ. 2021, 255, 112301. [Google Scholar] [CrossRef]
  41. Knauer, K.; Gessner, U.; Fensholt, R.; Kuenzer, C. An ESTARFM Fusion Framework for the Generation of Large-Scale Time Series in Cloud-Prone and Heterogeneous Landscapes. Remote Sens. 2016, 8, 425. [Google Scholar] [CrossRef]
  42. Yang, M.; Zhou, Y.; Xie, Y.; Shao, W.; Luo, F. Accuracy Evaluation of Four Spatiotemporal Fusion Methods for Different Time Scales. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2024, 17, 8291–8301. [Google Scholar] [CrossRef]
  43. Zhang, X.; Gao, F.; Wang, J.; Ye, Y. Evaluating a spatiotemporal shape-matching model for the generation of synthetic high spatiotemporal resolution time series of multiple satellite data. Int. J. Appl. Earth Obs. Geoinf. 2021, 104, 102545. [Google Scholar] [CrossRef]
  44. Cao, J.; Xu, X.; Zhuo, L.; Liu, K. Investigating mangrove canopy phenology in coastal areas of China using time series Sentinel-1/2 images. Ecol. Indic. 2023, 154, 110815. [Google Scholar] [CrossRef]
  45. Shahzad, K.; Alatalo, J.M.; Zhu, M.; Cao, L.; Hao, Y.; Dai, J. Geographic conditions impact the relationship between plant phenology and phylogeny. Sci. Total. Environ. 2024, 945, 174083. [Google Scholar] [CrossRef] [PubMed]
  46. Li, H.; Jia, M.; Zhang, R.; Ren, Y.; Wen, X. Incorporating the plant phenological trajectory into mangrove species mapping with dense time series Sentinel-2 imagery and the Google Earth Engine platform. Remote Sens. 2019, 11, 2479. [Google Scholar] [CrossRef]
  47. Huang, K.; Yang, G.; Sun, W.; Fu, B.; Chen, C.; Meng, X.; Feng, T.; Wang, L. The phenology and water level time-series mangrove index for improved mangrove monitoring. Int. J. Appl. Earth Obs. Geoinf. 2024, 134, 104188. [Google Scholar] [CrossRef]
  48. Kamruzzaman, M.; Sharma, S.; Hagihara, A. Vegetative and reproductive phenology of the mangrove Kandelia obovata. Plant Species Biol. 2013, 28, 118–129. [Google Scholar] [CrossRef]
  49. Ruan, L.; Yan, M.; Zhang, L.; Fan, X.; Yang, H. Spatial-temporal NDVI pattern of global mangroves: A growing trend during 2000–2018. Sci. Total. Environ. 2022, 844, 157075. [Google Scholar] [CrossRef] [PubMed]
  50. Zhang, X.; Treitz, P.M.; Chen, D.; Quan, C.; Shi, L.; Li, X. Mapping mangrove forests using multi-tidal remotely-sensed data and a decision-tree-based procedure. Int. J. Appl. Earth Obs. Geoinf. 2017, 62, 201–214. [Google Scholar] [CrossRef]
  51. Younes, N.; Joyce, K.E.; Northfield, T.D.; Maier, S.W. The effects of water depth on estimating Fractional Vegetation Cover in mangrove forests. Int. J. Appl. Earth Obs. Geoinf. 2019, 83, 101924. [Google Scholar] [CrossRef]
  52. Chen, J.; Jiang, C.; Wu, Z.; Long, Y.; Deng, B.; Liu, X. Numerical investigation of fresh and salt water distribution in the Pearl River Estuary during a typhoon using a fully coupled atmosphere-wave-ocean model. Water 2019, 11, 646. [Google Scholar] [CrossRef]
  53. Wang, W.; Xin, K.; Chen, Y.; Chen, Y.; Jiang, Z.; Sheng, N.; Liao, B.; Xiong, Y. Spatio-temporal variation of water salinity in mangroves revealed by continuous monitoring and its relationship to floristic diversity. Plant Divers. 2024, 46, 134–143. [Google Scholar] [CrossRef] [PubMed]
  54. Perri, S.; Detto, M.; Porporato, A.; Molini, A. Salinity-induced limits to mangrove canopy height. Glob. Ecol. Biogeogr. 2023, 32, 1561–1574. [Google Scholar] [CrossRef]
  55. Yang, X.; Zhu, Z.; Kroeger, K.D.; Qiu, S.; Covington, S.; Conrad, J.R.; Zhu, Z. Tracking mangrove condition changes using dense Landsat time series. Remote Sens. Environ. 2024, 315, 114461. [Google Scholar] [CrossRef]
  56. Asbridge, E.; Lucas, R.; Accad, A.; Dowling, R. Mangrove Response to Environmental Changes Predicted Under Varying Climates: Case Studies from Australia. Curr. For. Rep. 2015, 1, 178–194. [Google Scholar] [CrossRef]
Remotesensing 17 00549 g001
Figure 1. Map of study area: (a) the overall distribution of study area; (b1–b4) Zhangjiangkou National Mangrove Nature Reserve (ZNR) in Fujian Province, Qi’ao Island Provincial Nature Reserve (QPR) in Guangdong Province, Beilun Estuary National Nature Reserve (BNR) in Guangxi Province, and Dongzhaigang National Mangrove Nature Reserve (DNR) in Hainan Province.
Figure 1. Map of study area: (a) the overall distribution of study area; (b1–b4) Zhangjiangkou National Mangrove Nature Reserve (ZNR) in Fujian Province, Qi’ao Island Provincial Nature Reserve (QPR) in Guangdong Province, Beilun Estuary National Nature Reserve (BNR) in Guangxi Province, and Dongzhaigang National Mangrove Nature Reserve (DNR) in Hainan Province.
Remotesensing 17 00549 g001
Remotesensing 17 00549 g002
Figure 2. Workflow of mangrove phenology extraction based on OMPEA.
Figure 2. Workflow of mangrove phenology extraction based on OMPEA.
Remotesensing 17 00549 g002
Remotesensing 17 00549 g003
Figure 3. Landsat 8 NDVI (16-day 30 m) and denoised Landsat NDVI (16-day 30 m) generated by OMPEA. Gray pixel indicates pixel with no data.
Figure 3. Landsat 8 NDVI (16-day 30 m) and denoised Landsat NDVI (16-day 30 m) generated by OMPEA. Gray pixel indicates pixel with no data.
Remotesensing 17 00549 g003
Remotesensing 17 00549 g004
Figure 4. MODIS NDVI (1-day 500 m) and denoised MODIS NDVI (1-day 30 m) generated by OMPEA. Gray pixel indicates pixel with no data.
Figure 4. MODIS NDVI (1-day 500 m) and denoised MODIS NDVI (1-day 30 m) generated by OMPEA. Gray pixel indicates pixel with no data.
Remotesensing 17 00549 g004
Remotesensing 17 00549 g005
Figure 5. The OMPEA-generated fused NDVI imagery. Gray pixel indicates pixel with no data.
Figure 5. The OMPEA-generated fused NDVI imagery. Gray pixel indicates pixel with no data.
Remotesensing 17 00549 g005
Remotesensing 17 00549 g006
Figure 6. Scatter density plots and marginal histograms of fused NDVI and denoised Landsat NDVI.
Figure 6. Scatter density plots and marginal histograms of fused NDVI and denoised Landsat NDVI.
Remotesensing 17 00549 g006
Remotesensing 17 00549 g007
Figure 7. Composite scatter plots and line plots of various NDVI time series.
Figure 7. Composite scatter plots and line plots of various NDVI time series.
Remotesensing 17 00549 g007
Remotesensing 17 00549 g008
Figure 8. Fused NDVI time-series curve and phenological parameters.
Figure 8. Fused NDVI time-series curve and phenological parameters.
Remotesensing 17 00549 g008
Remotesensing 17 00549 g009
Figure 9. Boxplots of mangrove phenological parameters.
Figure 9. Boxplots of mangrove phenological parameters.
Remotesensing 17 00549 g009
Remotesensing 17 00549 g010
Figure 10. The time-series curves for fused NDVI, precipitation, temperature, and their lagged time-series curves with corresponding lag days.
Figure 10. The time-series curves for fused NDVI, precipitation, temperature, and their lagged time-series curves with corresponding lag days.
Remotesensing 17 00549 g010
Remotesensing 17 00549 g011
Figure 11. The OMPEA-generated fused NDVI in QPR from 17 January 2020 to 24 March 2021. (a) Description of denoised Landsat 8 NDVI in a full-time range. (b) Description of denoised Landsat 8 NDVI across three different time ranges, (c,d) is fused NDVI that using (a,b) as inputs, respectively. Gray pixel indicates pixel with no data.
Figure 11. The OMPEA-generated fused NDVI in QPR from 17 January 2020 to 24 March 2021. (a) Description of denoised Landsat 8 NDVI in a full-time range. (b) Description of denoised Landsat 8 NDVI across three different time ranges, (c,d) is fused NDVI that using (a,b) as inputs, respectively. Gray pixel indicates pixel with no data.
Remotesensing 17 00549 g011
Table 1. Correlation and lag days of temperature and precipitation effects on NDVI.
Table 1. Correlation and lag days of temperature and precipitation effects on NDVI.
RegionTemperaturePrecipitation
CCFLag DaysCCFLag Days
ZNR0.6276 *920.1396 *89
QPR0.7791 *550.2327 *100
BNR0.6576 *830.2057 *65
DNR0.5190 *990.2613 *65
Note: * indicates statistically significant cross-correlation at p < 0.05.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Hong, Y.; Zhou, R.; Liu, J.; Que, X.; Chen, B.; Chen, K.; He, Z.; Huang, G. Monitoring Mangrove Phenology Based on Gap Filling and Spatiotemporal Fusion: An Optimized Mangrove Phenology Extraction Approach (OMPEA). Remote Sens. 2025, 17, 549. https://doi.org/10.3390/rs17030549

AMA Style

Hong Y, Zhou R, Liu J, Que X, Chen B, Chen K, He Z, Huang G. Monitoring Mangrove Phenology Based on Gap Filling and Spatiotemporal Fusion: An Optimized Mangrove Phenology Extraction Approach (OMPEA). Remote Sensing. 2025; 17(3):549. https://doi.org/10.3390/rs17030549

Chicago/Turabian Style

Hong, Yu, Runfa Zhou, Jinfu Liu, Xiang Que, Bo Chen, Ke Chen, Zhongsheng He, and Guanmin Huang. 2025. "Monitoring Mangrove Phenology Based on Gap Filling and Spatiotemporal Fusion: An Optimized Mangrove Phenology Extraction Approach (OMPEA)" Remote Sensing 17, no. 3: 549. https://doi.org/10.3390/rs17030549

APA Style

Hong, Y., Zhou, R., Liu, J., Que, X., Chen, B., Chen, K., He, Z., & Huang, G. (2025). Monitoring Mangrove Phenology Based on Gap Filling and Spatiotemporal Fusion: An Optimized Mangrove Phenology Extraction Approach (OMPEA). Remote Sensing, 17(3), 549. https://doi.org/10.3390/rs17030549

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop