Evaluating the capability of low configuration spectrometers to

retrieve far-red solar-induced chlorophyll fluorescence and its

application for identifying crop growth stages

Lei Pei 1, 2, Yongguang Zhang 1, 2, 3*, Zhaoyin Zhang 1, 2, Yunfei Wu 1, 2, Yuqing Hou 1, 2 and Linsheng Wu
1, 2

1 International Institute for Earth System Sciences, Nanjing University, Jiangsu 210023, Nanjing, China;
2 School of Geography and Ocean Science, Nanjing University, Nanjing, Jiangsu 210023, China;

Downloaded from https://spj.science.org on December 13, 2024

3 Huangshan Park Ecosystem Observation and Research Station, Ministry of Education, China;
* Correspondence: ygz@nju.edu.cn；Tel.: +86-25-89681569

Abstract: The ground-based continuous hyperspectral remote sensing technique presents a unique
opportunity to advance terrestrial ecosystem monitoring. Accurate retrieval of solar-induced
chlorophyll fluorescence (SIF) relies heavily on commercially available spectrometers with high
spectral resolution. However, high spectral resolution (SR) spectrometers are resource-intensive
and challenging to maintain, limiting their widespread deployment at observational sites. Here, we
explored an alternative low-configuration and low-cost spectrometer for SIF retrieval at two sites
cultivated with wheat and rice. The results revealed the strong correlations between irradiance and
radiance measurements in the selected SIF bands from the high- and low-configuration
spectrometers (R2 > 0.90, average RMSE < 8.95 mW m-2 nm-1 sr-1, rRMSE < 7%). Far-red SIF
retrieved from both spectrometers exhibited significant correlations at both half-hourly and daily
averaged scales (R2 > 0.90). Additionally, we utilized simulated data to examine the impact of SR
and signal-to-noise ratio (SNR) on the far-red SIF retrieval from both high- and low-configuration
spectrometers. Strong correlations (R2 > 0.80) were found between the simulated data with SNR
≥ 300 and SR < 2nm. Furthermore, we investigated the relationships between crop gross primary
productivity (GPP) and far-red SIF from both types of spectrometers and their ability to identify
crop growth stages. SIF from both types of spectrometers exhibited high correlations with GPP (R2
> 0.70). The dates of various wheat-specific growth stages observed using the low-configuration
spectrometer were consistent with those identified by the high-configuration spectrometer, with
less than a 3-day differences for wheat. Our study confirms the capabilities of the low-configuration
spectrometers for far-red SIF retrievals and promotes their deployment over large-scale ground
observation networks, potentially advancing the large development of low-cost ground-based SIF
measurements.

Keywords: Solar-induced chlorophyll fluorescence, Spectrometer, Phenology, Crop growth

stages

Journal of Remote Sensing Page 1 of 34

 1. Introduction
Solar-induced chlorophyll fluorescence (SIF) is closely linked to photosynthesis and serves
as an indicator to elucidate the mechanisms and dynamics of plant photosynthesis. SIF has
been retrieved from tower-based systems [1, 2], unmanned aerial vehicles (UAV) [3-5], and
satellite platforms [6-8]. Tower-based ground systems are particularly crucial in providing
continuous ground measurements and serving as validation for UAV- and satellite-based
SIF retrievals. Therefore, ground observation networks can potentially advance our
understanding of terrestrial ecosystems and their responses to environmental changes.

SpecNet [9], EUROSPEC [10], and ChinaSpec [11] are globally recognized spectral
observation networks, and one of their key objectives is to achieve tower-based SIF
retrieval. Numerous studies have focused on developing diverse tower-based SIF automated
observing systems, including HIS [12], MRI [13], FFLUOR [13], FluoSpec [14], FLoX
[15], FluoSpec2 [16], PhotoSpec [17], FAME [18], SIFprism [19], SIFspec [20], and
SIFmotor [21]. These systems share a commonality in utilizing hyperspectral spectrometers
for SIF retrieval, such as QEPro (Ocean Optics, Inc., Dunedin, FL, USA). The QEPro is a
high-sensitivity spectrometer characterized by low stray light performance, with a spectral
resolution (SR) typically less than 1 nm and a high signal-to-noise ratio (SNR) that can

Downloaded from https://spj.science.org on December 13, 2024

reach up to 1000 under optimal conditions. SIF retrieval typically utilizes the filling effect
of the Fraunhofer lines or the terrestrial oxygen absorption bands (O2-A and O2-B), and the
widths of some of these used for SIF retrieval may be far below 1 nm. Therefore, SIF
retrieval requires a very high configuration of the spectrometer [22, 23]. The SR of the
spectrometer affects the intensity of the observed Fraunhofer line filling effects and
ultimately the accuracy of SIF retrieval.

However, these high-configuration spectrometers have certain limitations. First, their high
cost presents a significant barrier to the widespread deployment and study of large-scale
tower-based SIF, thereby hindering the expansion of global spectral observation networks
like ChinaSpec with a large number of observational sites. Secondly, spectrometers with
exceptionally high SR are sensitive to external environmental factors such as temperature
and humidity [24-26], resulting in high maintenance costs [27]. To address these challenges
and advance the field of tower-based SIF research, there is a pressing need for promoting
cost-effective spectrometers that offer both affordability and accurate SIF retrieval
capabilities.

However, affordable spectrometers usually have lower configurations, which also affects
the accuracy of SIF retrieval. Damm et al. [28] modeled and quantified the effect of different
spectrometer configurations on the accuracy of SIF retrieval in the O2-A band. The results
showed that 80% of the retrieval errors were attributed to SR and SNR, while 12% and 7%
could be attributed to spectral sampling interval (SSI) and wavelength calibration offsets,
respectively. Julitta et al. [29] also found that SNR and SR play important roles in SIF
retrieval by comparing the abilities of four spectrometers to retrieve SIF. Other studies have
shown that different spectrometer configurations affect the accuracy of SIF retrieval [30-
33]. Furthermore, different SIF retrieval algorithms have specific requirements for
spectrometer configuration (SR, SNR, etc.). Depending on the spectral bands used for SIF
retrieval, these algorithms can be broadly categorized into two main groups: algorithms
based on terrestrial atmospheric absorption lines (mainly O2-A and O2-B) and algorithms
based on solar atmospheric absorption lines [34]. The O2-A absorption band is relatively
broad and thus requires at least a modest spectrometer setup and fine SR (SR < 5 nm) [22].

Journal of Remote Sensing Page 2 of 34

 In contrast, the other absorption bands require a spectrometer with very high SR (SR << 0.3
nm) to ensure accurate SIF retrieval without atmospheric influence [8], such as, O2-B. This
study focuses only on far-red SIF due to the narrow absorption valley of O2-B and the strong
absorption of red SIF by the vegetation canopy, resulting in poor inversion results for red
SIF [35, 36]. Therefore, the primary challenge in utilizing low-configuration spectrometers
is to explore whether suitable retrieval methods can accurately quantify weak SIF signals.
Various studies have demonstrated the efficacy of low-configuration spectrometers in
accurately retrieving SIF under specific retrieval algorithms. For example, Kim et al. [27]
successfully retrieved SIF using the 3FLD algorithm with a self-developed filter-based
spectrometer. Additionally, experimental results from Belwalkar et al. [31, 37] have also
confirmed that low-configuration spectrometers can successfully retrieve SIF.

Although previous findings have demonstrated the efficacy of low-cost spectrometers in

SIF retrieval, there remains a significant gap in understanding their practical applications in
agricultural monitoring, especially in differentiating crop growth stages under variable field
conditions. Understanding these growth stages can provide valuable insights for tasks such
as irrigation, fertilization, pest management, and improving crop yields [38, 39]. Currently,
information on crop growth status relies either on costly and time-consuming manual efforts

Downloaded from https://spj.science.org on December 13, 2024

to record the timing of important growth events [40, 41] or on satellite remote sensing time-
series data to determine the dates of key phenological transitions in crop growth stages [42,
43]. However, there is a time lag between the dates of phenological transitions detected by
remote sensing and the physiological developmental stages of the crop. In contrast, the
tower-based SIF observation system can continuously monitor the crop canopy at a high
frequency over a long period of time [44, 45]. Additionally, combining the spectral
observation network with the tower-based SIF automated observation system can improve
the observation range and validate the accuracy of crop growth data derived from satellite
SIF. This process will further improve the accuracy of crop growth stage monitoring
information. Therefore, promoting the widespread use of low-configuration spectrometers
can further facilitate the application of SIF in the field of identifying different growth stages
of crops. For example, Liu et al. [43] developed the shape model fitting by the separate
phenological stage method (“SMF-S”) to monitor crop growth stages and validated the
accuracy of the method with winter wheat phenology observation data collected from
ground stations. Similarly, Ganeva et al. [46] achieved simultaneous observations of several
crops, such as wheat, maize, sunflower, and rapeseed, using in-situ data, elucidating distinct
phenological stages and their interactions with environmental factors. This comprehensive
dataset enables a detailed understanding of crop-specific growth patterns and physiological
responses, facilitating targeted agricultural interventions and resource optimization across
the agro-ecosystem. Therefore, it is an urgent need to promote low-configuration
spectrometers to build spectral observation networks for simultaneous detection of different
crops.

Here, we conducted simultaneous spectral measurements at two sites where wheat and rice
were grown using two types of spectrometers: a high-configuration spectrometer with SR =
0.3 nm, SNR = 1000:1 and a low-configuration spectrometer with SR = 1.5nm, SNR =
300:1. We aimed to assess the ability of the relatively low-configuration spectrometer to
retrieve far-red SIF to explore whether the SIF retrieved from a low-configuration
spectrometer was consistent with those retrieved from a high configuration spectrometer in
terms of trends as well as sensitivity. Additionally, we explored the potential application of
the low-configuration spectrometer for field observations, particularly concerning the
relationship between gross primary productivity (GPP) and far-red SIF, as well as the

Journal of Remote Sensing Page 3 of 34

 monitoring of crop growth stages. The specific objectives of our study are as follows: (1) to
compare the spectral measurements and SIF retrieval between the high- and low-
configuration spectrometers; (2) to explore the effect of SR, SNR and sky conditions on the
far-red SIF retrieval accuracy; (3) to assess the ability of far-red SIF retrieved from the low-
configuration spectrometer in relation to GPP and the identification of crop growth stages.

2. Materials and Methods

2.1 Sites
2.1.1 Jurong
This study was conducted at the Jurong Farmland Ecological Station of Nanjing University,
located in Jurong, Jiangsu Province, China (31.8068°N, 119.2173°E). The farmland was
managed under a summer rice cropping system, with an annual precipitation of 1099.1 mm
and an average annual temperature of 15.5°C. Spectral observations were collected from
June 7, 2021 to November 18, 2021, covering the nutritional, reproductive, and maturation
stages of rice. The rice growth cycle was divided into two phases: the nutritional growth

Downloaded from https://spj.science.org on December 13, 2024

stage, and the reproductive growth stage. Additionally, two non-growth stages, pre-planting
and post-harvest, were included.

2.1.2 Shangqiu
The Henan Shangqiu National Field Scientific Observatory for Farmland Ecosystems
(Shangqiu station) is located in the western suburb of Shangqiu city, Henan Province. Its
geographic coordinates are 34.587°N, 115.5753°E, with an altitude of 51.7 meters above
sea level. The cropland at this station was managed under a rotation-system of winter wheat
and corn. The average annual precipitation was 708 mm, and the average annual
temperature was 13.9 °C. Spectral observations for this study were collected from February
2, 2021 to May 31, 2021 with a focus on the winter wheat crop. Winter wheat had three
growth stages: nutritional growth stage, concurrent nutrient and vegetative growth stage and
reproductive growth stage.

2.2 Data
2.2.1 Spectral data
In this study, a tower-based SIF automated observing system was utilized to collect spectral
data. The tower at the Jurong site had a height of 10 m, while the tower at the Shangqiu site
had a height of 5 m. Both sites employed an automated ground-based continuous
observation system known as AS-SpecFOM system, which is an automated SIF observing
system equipped with two types of spectrometers: the QEPro spectrometer, known for its
high SR and SNR utilized for SIF retrieval, and the FLAME (Ocean Optics, Dunedin, USA)
spectrometer, which has relatively lower SR and SNR and is used for retrieving vegetation
indices. Detailed profiles of the two types of spectrometers are presented in Table 1. Both
the Jurong and Shangqiu sites utilized hemispherical observation for irradiance and conical
(bare fiber) observation for radiance. This combination is collectively referred to as
hemispherical-conical observation, and the almost simultaneous observation of solar
incident radiation and reflected radiation from the ground objects was achieved by switching
optical fibers through the optical path switch. And the fiber placements were both vertically
up and vertically down. Since the two types of spectrometers were placed at the same height
and the optical fibers were placed extremely close to each other, the two types of

Journal of Remote Sensing Page 4 of 34

 spectrometers had nearly overlapped observational footprints (Fig. S1). A schematic
diagram of the observation system is shown in Fig. 1. For a detailed description of the tower-
based SIF observation system are refer to [47]. Note that in our experimental design, both
the high-precision QEPro and the more cost-effective FLAME spectrometers were housed
in a temperature-controlled box to eliminate variability due to environmental conditions.
This setup allowed for a controlled comparison of their performance. However, it is
important to note that this does not imply that low-cost spectrometers like FLAME require
the same level of environmental control in typical usage scenarios. However, the long-term
maintenance costs and environmental sensitivity of low-cost spectrometers under less
controlled conditions were not investigated in this experiment.

Downloaded from https://spj.science.org on December 13, 2024

Fig. 1. Tower-based SIF automated observation system. (a)Schematic diagram of the field
installation of the vegetation canopy AS-SpecFOM system; (b) structural diagram of the spectral
observation system. CC-3: Cosine correction; Shutter: Electronic changeover switches; TEC:
Temperature control system.

Spectral data were collected throughout the entire growth season of rice and wheat, with
daily observations starting at 6:00 a.m. and ending at 6:00 p.m. The observation method
involved a sequence of steps: measurement of the DN value of the downward solar
irradiance, followed by the dark current value of the downward solar irradiance, then the
DN value of the upward radiance of the feature, and finally the dark current value of the
upward radiance of the feature. Each of the two types of spectrometers had a switch to
control the collection of dark current, while the two types of spectrometers were collecting
data in parallel without interfering with each other. Fig 2 illustrated the spectral data for a
specific day in the Jurong and Shangqiu.

Table 1. Information about the QEPro and FLAME spectrometers at the Jurong and Shangqiu
sites.
Observation
Jurong Shangqiu
station

Journal of Remote Sensing Page 5 of 34

 Spectrometer QEPro FLAME QEPro FLAME
Spectral range
730~780 400~1000 650~800 400~1000
(nm)
Spectral
resolution 0.17 1.5 0.3 1.5
(nm)
Spectral
sampling 0.074 0.325 0.145 0.325
interval (nm)
Signal-to-noise
1000:1 300:1 1000:1 300:1
ratio
Observation
Hemispherical-conical observation
method

Downloaded from https://spj.science.org on December 13, 2024

Fig. 2. Examples of spectral data observed by the high- and low-configuration spectrometers at the
Jurong (a, c) and Shangqiu (b, d) sites. (a) denotes the high-configuration spectrometer at 730-780
nm at the Jurong site, (b) denotes the high-configuration spectrometer at 650-800 nm at the
Shangqiu site, and (c) and (d) denotes the low-configuration spectrometers at 400-1000 nm at the
Jurong and Shangqiu sites, respectively. The Jurong data were collected on July 26, 2021, and the
Shangqiu data were collected on April 5, 2021.

2.2.2 Eddy covariance flux measurements

The EC flux system was used to continuously measure CO2, H2O and energy fluxes between
the crop canopy and the atmosphere. A threedimensional sonic anemometer (CSAT-3;
Campbell Science, Logan, UT, USA) and an open-path infrared gas analyzer (LI-7500A;

Journal of Remote Sensing Page 6 of 34

 LICOR, Lincoln, NE, USA) were installed at a height of 4.5 m above the ground. Moreover,
the incoming PAR was measured using an LI-190SL quantum sensor (LI-COR Inc., Lincoln,
NE, USA) above the canopy. We used eddy covariance data processing software, EddyPro
(version 6.1.0, LI-COR Inc., Lincoln, NE, USA), to process the raw flux data. Half hourly
fluxes were computed by processing the high frequency data (10 Hz) using EddyPro.
Daytime ecosystem respiration was estimated using models based on nighttime air
temperature and NEE. Following the approach of Reichstein et al. [48], GPP was estimated
by partitioning the daytime net ecosystem CO2 exchange into gross primary productivity
(GPP) and ecosystem respiration, and averaging the GPP at half-hourly intervals.

2.3 Data processing

2.3.1 Data quality control
Prior to data analysis, quality control (refer to Table 2) were implemented to address missing
data and anomalies caused by weather conditions (e.g., cloudy or rainy days). Additionally,
time alignment was conducted due to the slightly different sampling frequencies of the high-
and low-configuration spectrometers resulting from the use of auto-integration.
Consequently, the data acquisition times for the two types of spectrometers varied.

Downloaded from https://spj.science.org on December 13, 2024

Therefore, the data preprocessing also involved the following steps:

(1) Data acquisition time matching: The data acquisition time of the low-
configuration spectrometers was used as the basis for matching the data of the high-
configuration spectrometers with the corresponding the acquisition time, ensuring the error
of the acquisition time was within 1 minute.
(2) Delete Nulls: Filters out and deletes the null values in the two sets of time-
matched data.
(3) Outlier removal: Data that lies outside three times the standard deviation are
removed.
For more details on data quality control see the article by Cogliati et al. [49].

Table 2. Spectral data preprocessing

Quality control standards Formulas Thresholds
Spectral saturation Max (DN) <60000
(Low)
Spectral saturation Max (DN) <200000
(High)
Reflectance L*𝜋 / E <1
Signal-to-noise ratio DN / DC >30

2.3.2 SIF retrieval

This experiment utilized the 3FLD algorithm for far-red SIF retrieval. The 3FLD algorithm
assumes that SIF and reflectance spectra vary linearly around the absorption bands and
utilizes a weighted average of one band to the left and right of the absorption line to calculate
the incident solar irradiance and the upward canopy radiance.

Journal of Remote Sensing Page 7 of 34

 {𝐿𝐼
𝑜𝑢𝑡 = 𝜔𝑙𝑒𝑓𝑡 × 𝐼𝑙𝑒𝑓𝑡 + 𝜔𝑟𝑖𝑔ℎ𝑡 × 𝐼𝑟𝑖𝑔ℎ𝑡
𝑜𝑢𝑡 = 𝜔𝑙𝑒𝑓𝑡 × 𝐿𝑙𝑒𝑓𝑡 + 𝜔𝑟𝑖𝑔ℎ𝑡 × 𝐿𝑟𝑖𝑔ℎ𝑡
(1)

𝜆𝑟𝑖𝑔ℎ𝑡−𝜆𝑖𝑛 𝜆𝑖𝑛−𝜆𝑙𝑒𝑓𝑡
𝜔𝑙𝑒𝑓𝑡 = , 𝜔𝑟𝑖𝑔ℎ𝑡 = (2)
𝜆𝑟𝑖𝑔ℎ𝑡−𝜆𝑙𝑒𝑓𝑡 𝜆𝑟𝑖𝑔ℎ𝑡−𝜆𝑙𝑒𝑓𝑡

where 𝜆𝑖𝑛、𝜆𝑙𝑒𝑓𝑡 and 𝜆𝑟𝑖𝑔ℎ𝑡 are the wavelengths of the inner, left, and right bands of the
absorption line; 𝐼𝑜𝑢𝑡、 𝐼𝑙𝑒𝑓𝑡 and 𝐼𝑟𝑖𝑔ℎ𝑡 are the irradiance of the inner, left, and right
incident solar irradiance; 𝐿𝑜𝑢𝑡、𝐿𝑙𝑒𝑓𝑡 and 𝐿𝑟𝑖𝑔ℎ𝑡 are the radiance of the inner, left, and
right canopy reflected radiance, respectively. The SIF can be calculated by substituting
Equations (1) and (2) into Equation (3) below:
𝐼𝑜𝑢𝑡 × 𝐿𝑖𝑛−𝐼𝑖𝑛 × 𝐿𝑜𝑢𝑡
𝑆𝐼𝐹𝑖𝑛 = (3)
𝐼𝑜𝑢𝑡−𝐼𝑖𝑛

Here, 𝐼𝑖𝑛、𝐿𝑖𝑛 are the incident solar irradiance and the upward canopy radiance within the
absorption line.

Downloaded from https://spj.science.org on December 13, 2024

Therefore, three bands of irradiance and radiance are needed to calculate the SIF in this
experiment: the absorbing "left shoulder" bands (Left), the absorbing inner bands (In), and
the absorbing "right shoulder" bands (Right). For the selection of wavebands, we selected
in a range of bands to form different combinations of [Left, In, and Right] wavebands,
analyzed the correlation of these combinations, and selected the combination with the best
correlation as the wavebands required by the 3FLD algorithm. The selected bands were
shown in Table 3:

Table 3. 3FLD selected bands. The "Left", "In", and "Right" bands indicate the "left shoulder",
"inner", and "right shoulder" of the absorption bands.
Sites QEPro FLAME
Left In Right Left In Right
755- 760- 772- 750- 758.4- 772-
Jurong
756nm 761nm 773nm 754nm 762nm 775nm
Shangq 757- 760- 770- 754- 759.3- 772-
iu 758nm 761nm 771nm 758nm 762nm 775nm

To enhance the authenticity of the experiments, in addition to using the 3FLD algorithm to
retrieve far-red SIF, we also used the spectral fitting method (SFM) algorithm for the
retrieval of far-red SIF, which assisted in validating the experimental results of 3FLD.
Detailed explanation of the SFM algorithm was given in Meroni et al. [51]. Additionally,
since we focused only on far-red SIF, any term "SIF" that appeared below represents far-
red SIF only.

2.3.3 Analysis on different temporal and spectral resolutions

This experiment involved exploring the correlation between the SIF retrieved from the high-
and low-configuration spectrometers at each of the two experimental sites, Jurong and
Shangqiu, at different temporal resolutions: instantaneous scale (sampling frequency of the
spectrometer as instantaneous scale), 15-minute averaged time scale, 30-minute averaged
Journal of Remote Sensing Page 8 of 34
 time scale, and daily averaged time scale. The correlation between the high- and low-
configuration spectrometers was quantified by the coefficient of determination (R2).
Additionally, we resampled the spectral data observed by the high- configuration
spectrometers to make their spectral resolution consistent with that of low-configuration
spectrometers, and then correlated the resampled high-configuration spectrometers data
with that of low-configuration spectrometers.

2.3.4 Absorption depth

In addition to the SIF retrieved by the 3FLD method, the depth at the O2-A absorption
feature captured by each spectrometer at each site is calculated. The formula follows Liu et
al. [32]:
𝐼𝑜𝑢𝑡−𝐼𝑖𝑛
𝐷𝑒𝑝𝑡ℎ = = 1−𝑘 (4)
𝐼𝑜𝑢𝑡

𝐼𝑖𝑛 and 𝐼𝑜𝑢𝑡 indicate the irradiance inside and outside the absorption band, respectively.
𝑘 represents 𝐼𝑖𝑛/𝐿𝑜𝑢𝑡.

Downloaded from https://spj.science.org on December 13, 2024

2.3.5 Determination of clear and cloudy days
Clear and cloudy days are distinguished by the clear-sky index (CI) [52], which is based on
the ratio of irradiance Rin measured at ground level to the maximum irradiance at the top of
the atmosphere (R0). R0 is calculated from the solar zenith angle (SZA):

CI = Rin/R0 (5)

𝑅0 = 𝑆0 × (1 + 0.033 × 𝑐𝑜𝑠(360 × 𝐷𝑂𝑌/365))/𝑐𝑜𝑠(𝑆𝑍𝐴) (6)

where S0 is the solar constant (1367 W m-2). A commonly used threshold value of 0.5 is
used to distinguish between clear days (daily mean CI > 0.5) and cloudy days (daily mean
CI ≤ 0.5). In this study, a threshold of 0.5 is used to distinguish between clear and cloudy
days.

2.3.6 Relative SIF

To further characterize the ability of low-configuration spectrometers to retrieve SIF, we

introduced the concept of relative SIF (𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒). 𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 eliminated the effect of
sunlight. The specific calculation process was as follows:
𝑆𝐼𝐹
𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 = (7)
𝐸𝑛𝑖𝑟

where 𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒was the relative SIF and 𝐸𝑛𝑖𝑟 was the solar irradiance out to 760 nm.

2.4 SCOPE model simulations

To investigate the impact of the SR and SNR on the correlation of radiance from the high-
and low-configuration spectrometers, we conducted simulations using the SCOPE model.
We simulated the irradiance and radiance at spectral resolutions of 0.3 nm, 0.5 nm, 1.0 nm,
1.5 nm, 2.0 nm, 4.0 nm, and 5.0 nm using the SCOPE model. Subsequently, the far-red SIF
values at the corresponding spectral resolutions were retrieved using the 3FLD algorithm.

Journal of Remote Sensing Page 9 of 34

 The simulated irradiance and radiance, as well as the retrieved SIF, were utilized for several
purposes:

(1) We investigated the impact of the SR on the correlation of SIF between two types
of spectrometers. The retrieved SIF were analyzed for correlation using the following
combinations: [0.3 nm, 0.5 nm], [0.3 nm, 1.0 nm], [0.3 nm, 1.5 nm], [0.3 nm, 2.0 nm], [0.3
nm, 4.0 nm], and [0.3 nm, 5.0 nm].
(2) To quantify the impact of the SNR on the correlation of SIF between the two
types of spectrometers, we introduced noise into the simulated data. The added noise
followed a random Gaussian distribution with a mean value of 0 and the specific SNRs of
100:1, 300:1, 500:1, and 1,000:1 in that order (Eq. (5)) [28, 32].
𝐿𝜆
𝐿𝜆 = 𝐿𝜆 + 𝑛𝑜𝑖𝑠𝑒 (8)
𝑆𝑁𝑅

𝐿𝜆 is the noiseless spectral radiance signal simulated by the SCOPE model. In our noise
analysis experiment, we performed correlation analyses using combinations of [0.3 nm,
SNR] and [SR, SNR], with consistent SNR values for each case. The SR values were 0.5

Downloaded from https://spj.science.org on December 13, 2024

nm, 1.0 nm, 1.5 nm, 2.0 nm, 4.0 nm, and 5.0 nm, respectively. We then repeated the process
with varying SNR values of 100:1, 300:1, 500:1, and 1000:1 in that order, while comparing
the results to the analysis of noiseless data. This allowed us to quantify the impact of the
signal-to-noise ratio on the correlation.
(3) Additionally, the [SR, SNR] of the high- and low-configuration spectrometers
used in this study were [0.3 nm, 1000:1] and [1.5 nm, 300:1], respectively. Therefore, we
then analyzed correlation results between [0.3 nm, 1000:1] and [0.5 nm, 300:1], [1.0 nm,
300:1], [1.5 nm, 300:1], [2.0 nm, 300:1], and [3.0 nm, 300:1].

2.5 Crop growth stages identification

The extraction of crop growth stage information was mainly carried out within the
framework of phenology modeling. Smoothed remote sensing time-series data were used to
extract the key phenological transition dates of the crop growth stage. This experiment
primarily utilized two commonly used phenology models to retrieve the specific growth
stage information for summer rice and winter wheat.
2.5.1 Time series phenological preprocessing
Due to varying light conditions and instrumental noise, time-series SIF data obtained from
spectrometer retrieval may contain implausible or spurious observations that obscure the
underlying phenological patterns of crops. Hence, we converted the SIF data into continuous
SIF data at a temporal resolution of 1 day, forming a SIF time series. Subsequently, we
utilized the Savitzky-Golay (SG) algorithm [54] to smooth the SIF time series data with a
temporal resolution of 1 day, effectively eliminating noise while preserving information
about signal variability.

2.5.2 Phenological modeling of time series

After preprocessing the time-series SIF data, the subsequent step involves modeling the
seasonal phenological development trajectory of crops and revealing their temporal
phenological patterns throughout the year. In this study, the Beck model, a double logistic
function [55], was selected, and the smoothed spline function was considered as a simpler

Journal of Remote Sensing Page 10 of 34

 alternative for comparison.Beck extended the double logistic function to model temporal
variations in crop phenology using six parameters:

𝑓(𝑡) = 𝑎𝑏𝑎𝑠𝑒 + (𝑎𝑚𝑎𝑥−𝑎𝑏𝑎𝑠𝑒) × ( (

1 + 𝑒 −𝑚1
1
× (𝑡−𝑚 ))
2
+ (
1
1 + 𝑒 𝑛1 ×
)
(𝑡−𝑛2))−1 (9)

In the equation, 𝑡 denotes the DOY, 𝑓(𝑡) denotes the fitted value at time t. Whereas 𝑎𝑏𝑎𝑠𝑒
denotes the lowest baseline or off-season value, and 𝑎𝑚𝑎𝑥 denotes the maximum value
during the growing season. 𝑚1 is the rate of increase of the curve at the inflection point
𝑚2, and 𝑛1 is the rate of decrease of the curve at the turning point 𝑛2.

The smoothed spline function (hereafter referred to as spline) was utilized to capture the
seasonal phenological trajectories of crops, while also addressing outliers and mitigating the
impact of off-season vegetation cover. This was achieved by modeling seasonal changes in
SIF through fitting segmented polynomials to time segments of the data and connecting
these segments with continuous curves at specific locations, known as nodes [56]. The
spline curve model not only represented seasonal trajectories but also ensured the continuity
of the modeled curves and their derivatives [56].

Downloaded from https://spj.science.org on December 13, 2024

2.5.3 Phenological transition characterization
By employing smooth time-series curves generated by a phenology model, time-series
phenology characterization can estimate key phenological transition dates specific to the
physiological growth stages of a crop. In this study, two representative methods of
phenological characterization, derivative-based (DB) and curvature-based (CU) methods,
were used to retrieve crop phenological transition dates within this framework.

The DB method characterizes crop growth based on the local extremes of the first-order
derivatives of smoothed time-series curves. This approach estimates three phenological
transition dates, namely, the start of the season (SOS), the peak of the season (POS), and
the end of the season (EOS), which correspond to the absolute maximum, zero, and
minimum values of the first-order derivatives of the smoothed curves, respectively [7].

The CU method retrieves crop phenological periods based on the local extremes of the
second derivative of the curvature of the fitted phenological model [55, 57]. This method
identifies four transition points: growing up (GU), mid-transpiration (MT), senescence (SN)
and dormancy (DM) [48]. The GU and MT dates correspond to the times of the two local
maxima in the second derivative curve, while the SN and DM dates correspond to the times
of the two local minima in the second derivative curve.

The correspondence between the dates of climatic transitions retrieved from the above two
methods and the specific growth stages of winter wheat and summer rice are shown in Table
4.

Journal of Remote Sensing Page 11 of 34

 Table 4. Potential optimal harmonization of transition dates for winter wheat and summer rice
phenological characteristics with crop growth stages.
Wheat DB CU Rice DB CU

Regreen (RG) - GU Tillering (TL) - GU

Jointing (JT) SOS GU Jointing (JT) SOS -

Heading (HD) POS MT Heading (HD) POS MT

Panicle
Milk (MK) EOS SN - -
development (PD)
Ripening
EOS DM Flowering (FR) - SN
(RP)
Harvest (HV) - - Maturity (MT) EOS -

Downloaded from https://spj.science.org on December 13, 2024

Harvest (HV) - DM

*-: There are no suitable transition dates for climatic features.

3. Results
3.1 Correlation analysis of the spectral measurements from the high- and low-
configuration spectrometers
We initially conducted correlation analyses between the high- and low-configuration
spectrometers, considering both incident solar radiation and ground-reflected radiation. The
resulting correlations were presented in Fig. 3. We found that R2 between the high- and
low-configuration spectrometers at the Jurong and Shangqiu sites exceeded 0.9 for both
irradiance and radiance at the instantaneous scale, 15-minute averaged scale and 30-minute
averaged scale, respectively. This suggested a strong correlation between the high- and low-
configuration spectrometers in spectral measurements, and the correlation between spectral
measurements from the two gradually increased as the time scale rose. For example, within
the absorption bands of the irradiance at Shangqiu site, the mean value of R2 increased from
0.95 to 0.99 as the time scale rose. However, the results also showed that the correlation
within the absorption bands was lower than that to the left and right of the bands for both
irradiance and reflected radiation, with the most pronounced fluctuation characteristics.

Journal of Remote Sensing Page 12 of 34

 Downloaded from https://spj.science.org on December 13, 2024
Fig. 3. Correlation analysis of irradiance and radiance between the high- and low-configuration
spectrometers at the instantaneous scale, 15-minute averaged scale, and 30-minute averaged scale
at the Jurong (a, b) and Shangqiu (c, d) sites. "Left", "In" and "Right" correspond to the "left
shoulder", "in" and "right shoulder" of the absorption bands required by the 3FLD algorithm. In
the box, the mean is represented by a solid black circle, the median is represented by a solid black
line inside the box, and the 75th and 25th quartiles are represented at the top and bottom of the
box, respectively. The orange boxes represent instantaneous scales, the amber boxes represent 15-
minute averaged scales, and the yellow boxes represent 30-minute averaged scales.

At the daily mean scale, the R2 between the two types of spectrometers also exceeded 0.9
(Table 5), which was consistent with previously discussed results (Fig. 3). Meanwhile, the
average root mean square error (RMSE) of irradiance and radiance of the three bands
selected by the two types of spectrometers at the Jurong site were 8.95 mW m-2 nm-1 sr-1,
3.42 mW m-2 nm-1 sr-1, and the average relative root mean square error (rRMSE) were 5%
and 7%, respectively. At the Shangqiu site, the average RMSE were 3.57 mW m-2 nm-1 sr-
1, 3.63 mW m-2 nm-1 sr-1, and an average rRMSE of 2% and 7%, respectively. Therefore,

compared with high-configuration spectrometers, the low-configuration spectrometers

could also effectively observe variations of irradiance and radiance. The detailed results of
the correlation analysis between the two sites regarding irradiance and radiance were
presented in the Fig. S2 to Fig. S9.

Table 5. Relationship between the high- and low-configuration spectrometers in terms of

irradiance and radiance at the Jurong and Shangqiu sites at the daily averaged scale. "Left", "In"
and "Right" correspond to the "left shoulder", "in" and "right shoulder" of the absorption bands
required by the 3FLD algorithm.

Journal of Remote Sensing Page 13 of 34

 Sites Jurong Shangqiu
Spectra
Band Righ Righ
l data Left In Left In
s t t
R2 0.96 0.90 0.96 0.99 0.99 0.99
Irradia RMSE
65.51 42.00 63.47 51.09 43.45 47.28
nce (mW m-2 nm-1)
rRMSE (%) 8 24 8 10 31 9
R2 0.97 0.96 0.97 1.00 0.99 1.00
RMSE
Radian
(mW m-2 nm-1 sr- 6.04 3.26 6.19 3.77 3.78 3.88
ce 1)

rRMSE (%) 11 25 10 6 19 5

3.2 Correlation analysis of SIF retrieved from high- and low-configuration spectrometers

Downloaded from https://spj.science.org on December 13, 2024

This study utilized the 3FLD retrieval algorithm to explore the correlation between the two
types of spectrometers in the SIF retrieval as presented in Fig. 4 and Fig. 5. Fig. 4 illustrates
the correlation of SIF from the two types of spectrometers at the instantaneous scale, the
15-minute averaged scale, and the 30-minute averaged scale, respectively. The results
showed that the R2 between the SIF from the two types of spectrometers increased with
time. For example, at the Shangqiu site, the R2 fluctuated between 0.8 and 1.0 at the
instantaneous scale, increased to 0.9-1.0 at the 15-minute averaged scale, and ranged
between 0.95 and 1.0 at the 30-minute averaged scale. Notably, the R2 of the two types of
spectrometers with respect to the SIF exceeded 0.9 for both sites at the 30-minute averaged
scale. However, the results of the correlation analysis between the two sites differed,
especially at the instantaneous scale. At the Jurong site, the R2 between the two
spectrometers in terms of the retrieved SIF were greater than 0.5, while at the Shangqiu site,
it was greater than 0.8. The R2 at the Jurong site fluctuated widely, whereas at the Shangqiu
site, it fluctuated between 0.8 and 1.0, within a smaller range of fluctuations.

The preceding experimental findings provided qualitative evidence of a strong correlation

between the SIF from the low-configuration spectrometer retrieval and the SIF from the
high-configuration spectrometer retrieval within a specific time scale. To quantitatively
assess the extent to which the ability of the low-configuration spectrometer to retrieve SIF
aligned with the ability of high-configuration spectrometer to retrieve SIF, an analysis of
the correlation between SIF from the two types of spectrometers was conducted at the daily
averaged scale (Fig. 5). The analysis revealed that at both experimental sites, Jurong and
Shangqiu, the R2 values exceeded 0.9 at the daily averaged scale, with R2 values of 0.90 for
Jurong and 0.94 for Shangqiu. Additionally, the correlation analysis at the daily average
scale provided a linear conversion function between the SIF from the low- and high-
configuration spectrometers retrievals. Specifically, at the Shangqiu site, the conversion
function was represented as Y = 2.78*X + 1.43, where X represented the daily SIF results
from the high-configuration spectrometers retrieval and Y represented the daily SIF results
from the low-configuration spectrometer retrieval, and the conversion of SIF between the
high- and low-configuration spectrometers could be roughly realized according to this
formula.

Journal of Remote Sensing Page 14 of 34

 Downloaded from https://spj.science.org on December 13, 2024

Fig. 4. SIF correlation analysis between the high- and low-configuration spectrometers at the
Jurong (a, c, e) and Shangqiu site (b, d, f) at the instantaneous scale, 15-minute averaged scale, and
30-minute averaged scale, respectively The dots represented the R2 of each day.

Journal of Remote Sensing Page 15 of 34

 Fig. 5. Correlation in terms of the daily mean SIF retrieved by the 3FLD algorithm for the high-

Downloaded from https://spj.science.org on December 13, 2024

and low-configuration spectrometers at the Jurong and Shangqiu sites. The dots represent the
daily SIF values and the black line represents the SIF conversion formula for the high- and low-
configuration spectrometers. The SIF from the low-configuration spectrometers retrieval is higher
than that from the high-configuration spectrometers retrieval due to the effect of the difference in
the SR of the instruments.

Fig. 6. Correlation analysis of the 𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 of the two types of spectrometers retrievals for the
Jurong and Shangqiu sites.

Although the SIF retrieved by the low-configuration spectrometer and those retrieved by
the high-configuration spectrometer demonstrated good correlation in the correlation
analysis, the SIF retrieved by the low-configuration spectrometers were more than three
times as large as the SIF retrieved by the high-configuration spectrometers (shown in Fig.
5). Therefore, we introduced a relativity analysis of relative SIF (𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒), and the results
were shown in Fig. 6. The 𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 of the low-configuration spectrometer showed good

Journal of Remote Sensing Page 16 of 34

 correlation with the 𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 of the high-configuration spectrometer (R2 > 0.89) and the
size relationship of the 𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 was within a factor of two.

At the same time, because the difference in the selection of 3FLD bands would have a
certain impact on the results, the SFM algorithm results were also used as an additional
validation to improve the authenticity of the experiment. The SFM algorithm experimental
results (Fig. S10 and Fig. S11) showed a decrease in correlation compared with the 3FLD
results for different time scales, which may be related to the SFM algorithm that requires a
higher configuration of the spectrometer. However, the results at 15-minute averaged, 30-
minute averaged, and daily averaged scales also showed that there were significant
correlations between the retrieved SIF of low- and high-configuration spectrometers,
especially at the Shangqiu site, which had a correlation as high as 0.9 between the daily
averaged SIF. Finally, Fig. 7 demonstrated the SIF retrieval errors (RMSE and rRMSE) for
the two types of spectrometers using 3FLD, with the average RMSE of the two types of
spectrometers at the Jurong site being 0.10 mW m-2 nm-1 sr-1 and 0.40 mW m-2 nm-1 sr-1,
respectively, and the average rRMSE being 7% and 18%, respectively. At the Shangqiu site,
the average RMSE was 0.12 mW m-2 nm-1 sr-1 and 0.34 mW m-2 nm-1 sr-1, respectively, and
the average rRMSE was 5% and 16%, respectively.

Downloaded from https://spj.science.org on December 13, 2024

Fig. 7. 3FLD algorithm retrieval error (RMSE and rRMSE) for the high- and low-configuration
spectrometers at the Jurong (a) and Shangqiu site (b).

The correlation between the SIF from the two types of spectrometers revealed notable
differences, particularly on an instantaneous scale. Specifically, the R2 between the
instantaneous SIF retrieved by the two types of spectrometers at the Jurong site (R2 > 0.5)
were significantly different from those at the Shangqiu site (R2 > 0.8). The R2 values
exhibited a larger fluctuation range (0.5-0.9) at the Jurong site, whereas at the Shangqiu site,
although characterized by fluctuations, the R2 values remained within a smaller range of
0.8-1.0. Upon analyzing the data from the two sites, it was observed that the R2 between the
absorption depth of the high- and low-configuration spectrometers at the Jurong (R2 = 0.71)
was inferior to that at the Shangqiu (R2 = 0.96) (Fig. 8). Further examination of the Shangqiu
data revealed a decrease in the correlation between the absorption depths of the two
spectrometers during the summer months. For instance, the R2 value was 0.96 in the winter
and spring of 2021, decreasing to 0.76 in the summer (Fig. S12). Therefore, we hypothesize
that it may be related to changes in the depth of absorption. It was also worth noting that the
SIF retrieved by the low-configuration spectrometer was greater than the SIF retrieved by
the high-configuration spectrometer by a factor of more than three for both 3FLD and SFM.

Journal of Remote Sensing Page 17 of 34

 3.3 Factors influencing the correlation between the high- and low-configuration
spectrometers
In our investigation of the factors influencing the correlation between the high- and low-
configuration spectrometers, we initially simulated irradiance and radiance using the
SCOPE model. Subsequently, we calculated the SIF using the 3FLD to assess the impact of
SR and SNR on the correlation. Following this, we used the CI to differentiate clear and
cloudy skies, and examined their influence on the correlation.

Downloaded from https://spj.science.org on December 13, 2024

Fig. 8. Relationship of the absorption depths from high- and low-configuration spectrometers at
the Jurong and Shangqiu sites in 2021.

3.3.1 Effect of the SR and SNR on the correlation of SIF between the high- and low-
configuration spectrometers
Fig. 9 presents the results of the correlation analysis between SIF and different SR and SNR.
The first analysis of the effect of SR was shown in Fig. 9 for the "Noiseless" results, where
the R2 values for [0.3 nm, 0.5 nm], [0.3 nm, 1.0 nm], [0.3 nm, 1.5 nm], [0.3 nm, 2.0 nm],
[0.3 nm, 4.0 nm], and [0.3 nm, 5.0 nm] were 1.00, 0.99, 0.98, 0.95, 0.94, and 0.88,
respectively. These results indicated that changes in SR did not have a significant effect on
the correlation in the noiseless condition. Additionally, the effects caused by different SNR
exhibited significant differences. Specifically, when the SNR = 100:1 and 300:1, there was
a notable effect on SIF correlation at different SR. At the SNR = 100:1, all correlation results
yielded an R2 were less than 0.2, indicating a weak correlation. Conversely, at SNR = 300:1,
the R2 for [0.3 nm, 0.5 nm], [0.3 nm, 1 nm], [0.3 nm, 1.5 nm], [0.3 nm, 2 nm], [0.3 nm, 4
nm], and [0.3 nm, 5 nm] was 0.95, 0.90, 0.81, 0.77, 0.58, and 0.44, respectively. These
results suggested that in the case of SNR = 300:1, the SIF data with SR > 2nm did not
accurately reflect the true retrieval results. Furthermore, when SNR was 500:1 and 1000:1,
all the R2 were greater than 0.8, indicating that when SNR ≥ 500:1, the effect of the SNR
gradually weakened. This indicated that as the SNR increased, the correlation results
between the noisy SIF data converged more closely to the correlation results between the
noiseless SIF data.

Journal of Remote Sensing Page 18 of 34

 Fig. 9. (a) Effect of SR and SNR on SIF correlation between high- and low-configuration
spectrometers. Initially, noiseless irradiance and radiance simulated by the model were converted
to SIF using the 3FLD algorithm. Correlations were then analyzed between SIF data with SR=0.3
nm and varying SRs (0.5 nm to 5.0 nm), illustrating the impact of SR variations. Additionally, noise

Downloaded from https://spj.science.org on December 13, 2024

at different SNRs (100:1, 300:1, 500:1, and 1000:1) was added to examine its effect on SIF
correlation. The black dashed line represents the R2 between the SIF data with SRs of 0.3 nm and
1.5 nm. (b) Correlation analysis for combinations of [0.3 nm, 1000:1] with [0.5 nm, 300:1] up to
[5.0 nm, 300:1]. The "★" marks the correlation results between [0.3 nm, 1000:1] and [0.5 nm,
300:1], with an R2 of 0.81.

In addition to analyzing the correlation between SIF under different SR and SNR, we also
resampled the data from the high-configuration spectrometers, and the results are presented
in Fig. 10. The R2 between the resampled SIF from the high-configuration spectrometers
and the low-configuration spectrometers at the Jurong site were all greater than 0.8, while
those at the Shangqiu site were greater than 0.9. This result indicates a strong correlation
between the SIF data with a SNR of 1000:1 and that of the SIF data with a SNR of 300:1 at
SR = 1.5 nm. Furthermore, we conducted a targeted correlation by correlating the SIF data
according to the combinations of [0.3 nm, 1000:1] and [0.5 nm, 300:1], [1.0 nm, 300:1],
[1.5 nm, 300:1], [2.0 nm, 300:1], [4.0 nm, 300:1], [5.0 nm, 300:1]. The results were shown
in Fig. 9(b), where the "★" icon highlights the correlation analysis for [0.3 nm, 1000:1] and
[1.5 nm, 300:1] (R2 = 0.81). According to the model simulation, a significant correlation
exists between SIF with SR = 0.3 nm and SNR = 1000:1 and SIF with SR = 1.5nm, SNR =
300:1.

The spectrometers used to observe the spectral data were the high-configuration
spectrometers (SR = 0.3 nm, SNR = 1000:1) and the low-configuration spectrometers (SR
= 1.5 nm, SNR = 300:1). Therefore, the results of the above analyses indicated that the
differences in SR and SNR between the low- and high-configuration spectrometers used in
this study did not significantly affect the correlation between their retrieved SIFs.

Journal of Remote Sensing Page 19 of 34

 Fig. 10. Resampled irradiance and radiance from high-configuration spectrometers at SR 0.3 nm
to SR 1.5 nm, correlated with low-configuration spectrometer SIF at the instantaneous scale using
the 3FLD algorithm at the Jurong (a) and Shangqiu (b) site.

Downloaded from https://spj.science.org on December 13, 2024

We found that the SIF values retrieved by the low-configuration spectrometer were more
than three times those retrieved by the high-configuration spectrometer. To investigate this
discrepancy, we simulated noise-free spectral data with SR of 0.3 nm and 1.5 nm using the
SCOPE model, and simultaneously added noise to them with SNR of 100, 300, 500, and
1000, respectively. The analysis of these simulated data revealed that the SIF values from
the 1.5 nm data consistently exceeded those from the 0.3 nm data. For instance, as shown
in Fig. 10, the average SIF value at SR = 0.3 nm and SNR = 1000 was 0.88, while at SR =
1.5 nm and SNR = 300, it was 4.58, over three times higher.

Fig. 11. SCOPE simulation of SR at 0.3 nm and 1.5 nm, with the retrieved SIF after adding noise
at signal-to-noise ratios of 100, 300, 500, and 1000. Each box represents the concentration area of
the data, the "black dot" denotes the average value, and the upper, middle, and lower lines within
each box indicate the maximum, median, and minimum values, respectively.

Journal of Remote Sensing Page 20 of 34

 3.3.2 Effect of sky conditions on the relationships between the high- and low-
configuration spectrometers measurements
Initially, we analyzed the diurnal pattern of SIF under different sky conditions using the
high- and low-configuration spectrometers retrieval at the two sites. Fig. 12 illustrates the
diurnal pattern of SIF on a half-hourly scale for a specific sunny and cloudy day at Shangqiu,
with the black dashed line marking 12:00. The results indicated that the SIF retrieved by the
high- and low-configuration spectrometers exhibited consistent diurnal variations for both
cloudy and clear days, but the SIF retrieved by the high-configuration spectrometers showed
more subtle variations.

Additionally, Fig. 13 shows the R2 box-plot of the SIF retrieved by the high- and low-
configuration spectrometers at the instantaneous scale under different sky conditions. In
general, the findings revealed that the R2 of sunny days at both sites was slightly higher than
that of cloudy days (Fig. 13(a)). The mean R2 of the SIF at the Jurong site was exceeded 0.8
on a sunny day, and ranged between 0.7 and 0.8 on a cloudy day. Similarly, at the Shangqiu
site, the mean value of the SIF correlation was 0.9 on a clear day and ranged between 0.8
and 0.9 on a cloudy day. Fig. 13(b) provides a more intuitive representation of the consistent
correlation trend between SIF from the high- and low-configuration spectrometers under

Downloaded from https://spj.science.org on December 13, 2024

both sunny and cloudy conditions. Combined with the results in Fig. 13(a), these findings
demonstrate that the correlation between SIF from the spectrometers remained largely
unaffected by varying sky conditions, indicating robust correlation strength of SIF between
the high- and low-configuration spectrometers regardless of sky conditions.

Fig. 12. Diurnal pattern of SIF under sunny (a, b) and cloudy (c, d) at the Shangqiu site. The dashed
line in the figure indicates 12:00 p.m. Data were collected on May 22, 2021 for sunny day and on
April 19, 2021 for cloudy day.

Journal of Remote Sensing Page 21 of 34

 Fig. 13. Effect of sunny and cloudy sky on the correlation of two types of spectrometers in SIF
retrieval. (a) Box plot showing SIF correlations at the instantaneous scale under sunny (yellow
boxes) and cloudy (orange boxes) conditions. Black circles indicate means, and solid lines within
the boxes denote medians. Scatter plots of R2 values for SIF correlations at Jurong (b) and
Shangqiu (c) under clear and cloudy skies, with black dots for cloudy days and red dots for sunny
days.

3.4 The relationship between GPP and SIF from the high- and low-configuration
spectrometers

Downloaded from https://spj.science.org on December 13, 2024

To evaluate if low-configuration spectrometer retrieval SIF could estimate the GPP
effectively, we analyzed the SIF-GPP relationship throughout the entire growing season at
two sites, with data processed at daily scale. The results showed substantial linear
correlations between SIF and GPP from both high- and low-configuration spectrometers at
the Jurong and Shangqiu sites, with similar R2 (Fig. 14). Specifically, the R2 between SIF
and GPP for the high- and low-configuration spectrometer observations at the Jurong site
were both 0.73, while at the Shangqiu site, they were 0.85 and 0.79, respectively. This
indicated that the low-configuration spectrometer could accurately analyze the relationship
between SIF and GPP. And the relationship between SFM-based SIF data and GPP was
shown in Fig. S13. Notably, the R2 at the Shangqiu site remained almost unchanged,
whereas at the Jurong site, there was a significant change, primarily due to the lower
spectrometer configuration. Under the SFM algorithm, the R2 for SIF and GPP was 0.67,
which contrasts with an R2 of 0.73 under the 3FLD algorithm.

Journal of Remote Sensing Page 22 of 34

 Downloaded from https://spj.science.org on December 13, 2024
Fig. 14. Relationship between daily SIF using 3FLD and GPP at the Jurong (a, b) and Shangqiu (c,
d) site.

3.5 Capabilities of SIF to monitor crop growth stages

We utilized daily SIF data retrieved from the high- and low-configuration instruments to
monitor crop growth stages. Employing the Beck-DB method, we identified the start of
jointing (JT), heading (HD), ripening (RP) and maturity (MT) for several phenological
transitions (Fig. 15). At the Jurong site, JT, HD, and MT dates monitored by high-
configuration spectrometers were 201, 232, and 280 days, respectively, compared to 195,
223, and 272 days by low-configuration spectrometers. Similarly, at the Shangqiu site, JT,
HD, and RP dates recorded by high-configuration spectrometers were 88, 106, and 138
days, respectively, with low-configuration spectrometers showing dates of 89, 108, and 132.
These results suggest that monitoring accuracy was higher at the Shangqiu site compared to
Jurong. The mean deviation of the phenological transition dates monitored by the low-
configuration spectrometers in Jurong was approximately 7.7 days relative to those recorded
by high-configuration spectrometers, while at Shangqiu, the mean deviation was only 3.0
days.

Journal of Remote Sensing Page 23 of 34

 Downloaded from https://spj.science.org on December 13, 2024
Fig. 15. Retrieval of crop phenological transition dates based on the Beck-DB method. The dates
of phenological transitions from the high- and low-configuration spectrometers retrieval at the
Jurong (a, b) and Shangqiu (c, d) site. The dots represented the daily SIF values smoothed by the
SG filter. The black solid line depicts the curve simulated by the Beck phenological model. The
intersections of the gray dashed line with the black solid line indicate the retrieved dates for
jointing (JT), heading (HD), ripening (RP), and maturity (MT).

Subsequently, we applied the Spline-CU method to identify key phenological transition

dates of crops, including regreen (RG), tillering (TL), heading (HD), milk (MK), ripening
(RP), flowering (FR), harvest (HV), with results displayed in Fig. 16. Fig. 16 (a) and (b)
show the monitoring results at the Jurong site, where the high-configuration spectrometer
recorded TL, HD, FR, and HV on days 184, 217, 256, and 306, respectively, while the low-
configuration spectrometer recorded these stages on days 177, 215, 253, and 300. The mean
deviation between the two configurations was 7.9 days. Fig. 16 (c) and (d) present the results
at the Shangqiu site, with the high-configuration spectrometer capturing RG, HD, MK, and
RP on days 68, 97, 127, and 142, respectively, and the low-configuration recording on days
71, 101, 126, and 142. The mean deviation of phenological transition dates between the two
configurations was 2.0 days.

Journal of Remote Sensing Page 24 of 34

 Downloaded from https://spj.science.org on December 13, 2024
Fig. 16. Retrieval of crop phenological transition dates using the Spline-CU method. The dates of
phenological transitions from the high- and low-configuration spectrometers retrieval at the
Jurong (a, b) and Shangqiu (c, d) site. The dots represent diurnal far-red SIF values smoothed by
the SG filter. The black solid line depicts the curve simulated by the Spline phenological model,
while intersections of the gray dashed line with the black line mark the retrieved dates for
transitions including regreen (RG), tillering (TL), heading (HD), milk (MK), ripening (RP),
flowering (FR), and harvest (HV).

Table 6 . SIF and GPP data from Jurong and Shangqiu sites to monitor crop phenological
transition dates. 𝑆𝐼𝐹ℎ𝑖𝑔ℎ, 𝑆𝐼𝐹𝑙𝑜𝑤 denoted SIF from high- and low-configuration spectrometers,
respectively. Beck-DB and Spline-CU were methods for retrieving phenological transition dates,
respectively. Regreen (RG), jointing (JT), tillering (TL), heading (HD), milk (MK), ripening (RP),
flowering (FR), and harvest (HV) were the recognized phenological transition dates, respectively.

Crop phenological Beck-DB Spline-CU

transition dates 𝑆𝐼𝐹ℎ𝑖𝑔ℎ 𝑆𝐼𝐹𝑙𝑜𝑤 GPP 𝑆𝐼𝐹ℎ𝑖𝑔ℎ 𝑆𝐼𝐹𝑙𝑜𝑤 GPP

JT 88 89 82 RG 68 71 68

HD 106 108 108 HD 97 101 95

Wheat
RP 138 132 138 MK 127 126 121

- - - - HV 142 142 146

JT 201 195 188 TL 184 177 180

HD 232 223 222 HD 217 215 214

Rice
MT 280 272 277 FR 256 253 250

- - - - HV 306 300 308

Journal of Remote Sensing Page 25 of 34

 A direct comparison of the phenological periods captured by both the low-configuration and
high-configuration spectrometers revealed important insights. Additionally, GPP data were
utilized to monitor the phenological periods of the crops and serve as a reference for
evaluating the differences between the two spectrometers. As shown in Table 6, at the
Shangqiu site, the average deviation of the SIF phenological period recorded by the high-
configuration spectrometer compared to the GPP data was less than 3 days. The average
deviation of the phenological period captured by the low-configuration spectrometer
compared to the GPP data was less than 4 days. At the Jurong site, the average deviations
of the phenological periods from the GPP data were less than 6 days for the high-
configuration spectrometer and less than 4 days for the low-configuration spectrometer..

The results of different phenological transition dates of crops captured by daily SIF using
the SFM algorithm are shown in Fig. S14 and Fig. S15. At the Shangqiu site, the seasonal
trends and curve fitting results show little variation regardless of the SIF retrieval or
phenological conversion algorithms used. The mean deviation in the timing of crop growth
stages from different climatic conversion algorithms at Shangqiu under the SFM algorithm
is 4 days and 1.5 days, respectively. The results from the Jurong site show slightly marked
differences, especially in the seasonal variations in SIF retrieved by the low configuration

Downloaded from https://spj.science.org on December 13, 2024

spectrometer. However, the results are also closer between the different retrieval methods.

4. Discussion
4.1 Comparisons between the high- and low-configuration spectrometers
Although previous studies have demonstrated the efficacy of low-configuration
spectrometers in accurately retrieving SIF [27, 31, 37], there was a notable lack of
experimental evidence under continuous sampling conditions. In contrast, our study
revealed that the data obtained from low-configuration spectrometers correlate well with
those from high-configuration spectrometers throughout the crop growth cycle, particularly
at half-hourly and daily averaging scales (R² > 0.9).

Furthermore, the concern regarding the excessively large SIF values retrieved by low-
configuration spectrometers, which could misrepresent vegetation conditions, has been
significantly addressed by the robust correlation with 𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒. 𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 mitigated the
influence of non-physiological effects, while the physiological effects of SIF rendered it
highly sensitive to various physiological activities of vegetation. The good correlation
provided further evidence that the SIF values retrieved by low-configuration spectrometers
in our experiments had the ability to reflect the growth status of vegetation. Additionally,
high SIF values have been reported in previous studies [31, 33, 37], and it has been observed
that spectrometers with low SR and SNR led to greater retrieval errors [25, 30, 59]. The
SCOPE simulation conducted in this experiment (Fig. 10) also corroborated that low-
configuration spectrometers may produce exaggerated SIF values. Moreover, our
experiments did not need to confirm whether the SIF retrieved by the low-configuration
spectrometer was the real SIF or not, but only needed to determine whether the trend and
sensitivity were consistent with those of the high-configuration spectrometer.

4.2 Factors influencing the correlation between high- and low-configuration

spectrometers

The lower the SR and SNR of a spectrometer, the lower the SIF retrieval accuracy [28, 30,
33, 60-62]. Julitta [29] suggested that spectrometers with an SNR greater than 500 were
necessary for effective SIF retrieval. However, our simulations revealed a strong correlation

Journal of Remote Sensing Page 26 of 34

 (R² > 0.8) between the simulated data when SR < 2 nm and SNR>= 300. This finding, in
conjunction with the correlation analysis of previously measured data, demonstrated that
the SIF retrieved by the low-configuration spectrometer utilized in this study could be used
to observe vegetation. Furthermore, the experimental results of resampling corroborated this
conclusion.

Additionally, illumination conditions significantly influenced SIF retrieval [64, 65]. The
daily SIF data retrieved by the low-configuration spectrometer exhibited a trend consistent
with that of the high-configuration spectrometer across various weather conditions, and the
correlation between the two remained stable regardless of the conditions. This capability
was essential for the broader application of low-configuration spectrometers in tower-based
SIF observing systems, as it not only validated satellite SIF products under clear-sky
conditions but also provided crucial data under cloudy conditions, thereby enhancing the
accuracy of satellite SIF products.

In summary, these findings indicated that the SIF retrieved by the low-configuration
spectrometer in this study demonstrated a consistent trend with that retrieved by the high-
configuration spectrometer.

Downloaded from https://spj.science.org on December 13, 2024

4.3 On the SIF-GPP relationships and crop growth stages identification
This study identified a significant linear relationship between SIF and GPP as retrieved from
low-configuration spectrometers on a seasonal scale (R² > 0.70). Previous research has also
demonstrated this linear relationship between SIF and GPP on a seasonal scale [8, 66-68,
75]. This finding indicated that low-configuration spectrometers can facilitate the
investigation of the SIF-GPP relationship, which was crucial for advancing tower-based SIF
automated observing systems and establishing global spectral observing networks.
Furthermore, existing literature suggested that the SIF-GPP relationship may exhibit
nonlinearity across different observational scales and within various ecosystems [66, 69-
73], with these relationships often diminishing over time [68, 74]. Additionally, most
current studies relied on satellite-derived SIF products with limited temporal and spatial
resolution [78, 79], posing challenges in examining the complex variations of the SIF-GPP
relationship. Therefore, our experimental findings offered a referable solution to address the
limitations of satellite SIF products in comprehensively studying the dynamic SIF-GPP
relationship: Utilizing a global network of spectral observations could yield a deeper
understanding of SIF-GPP interactions at the ecosystem level [11, 80, 81] and provide
essential data for supplementing and validating satellite observations. This enhanced
approach was vital for exploring the spatial variability of SIF-GPP and assessing the
influences of canopy structure, leaf physiology, and environmental factors on this
relationship.

Most research concerning crop growth stages had been conducted at the satellite level, with
comparatively few tower-based studies. Hou et al. [77] highlighted the potential of tower-
based SIF to capture these growth stages. In our study, we successfully retrieved the growth
stages of winter wheat and summer rice using the tower-based SIF observing system. The
positive outcomes of our analysis for winter wheat demonstrated that low-configuration
spectrometers can effectively capture the phenological transition dates of crops despite their
lower SNR and SR. While previous studies [40, 76] had successfully linked phenological
frameworks derived from remotely sensed data time series to specific crop growth stages, a
temporal discrepancy existed between the phenological transition periods and specific
growth stages [40, 82]. This misalignment could introduce errors into satellite-derived SIF

Journal of Remote Sensing Page 27 of 34

 retrieval results. Low-configuration spectrometers may thus facilitate the deployment of
tower-based SIF automated observation systems, providing critical ground truth data for
validating satellite products and advancing crop climate modeling research. While absolute
SIF values from low-configuration spectrometers may not perfectly align with those from
high-configuration systems or satellite measurements, their relative trends and correlations
with GPP were sufficient for specific observational tasks in phenology and plant physiology
studies.

4.4 Prospects and limitations

Although our experiments demonstrated that SIF retrieved by low-configuration
spectrometers exhibited trends and sensitivities consistent with those retrieved by high-
configuration spectrometers, some limitations existed in our study. First, we primarily
focused on the effects of SR and SNR on the correlation of SIF between high- and low-
configuration spectrometers, neglecting other factors such as spectral sampling interval
(SSI), and spectral smoothing (SS). Second, the results obtained from Jurong were relatively
inferior to those from Shangqiu, as illustrated in Fig. 4. This discrepancy may be attributed
to variations in SR, environmental conditions, or other unknown factors. However, we
analyzed only one factor, SR, by resampling the SR of the Jurong high-configuration

Downloaded from https://spj.science.org on December 13, 2024

spectrometer from 0.17 nm to 0.3 nm, followed by an analysis of the correlation on an
instantaneous scale (as shown in Fig. S16). However, the correlation did not significantly
improve. We hypothesized that environmental factors might play a role, as the correlation
of absorption depth decreased in summer (Fig. S12) and the robustness of the SIF retrieval
algorithm relied on absorption depth; shallower absorption depths could adversely affect
retrieval accuracy [58]. Additionally, the results of the SFM search indicated relatively poor
performance for Jurong (Fig. S10).

Finally, regarding the disproportionately high far-red SIF values retrieved by low-
configuration spectrometers in early stages, several factors may contribute. Firstly, the
configuration of the spectrometer significantly impacts SIF retrieval accuracy, as discussed
in earlier sections. Additionally, calibration differences between spectrometers can affect
their baseline readings, with low-configuration spectrometers potentially exhibiting less
effective zero calibrations than high-configuration models, leading to baseline shifts that
manifest as higher SIF values. Sparse early-stage crop growth could also weaken the SIF
signal, exacerbating retrieval errors by low-configuration spectrometers and resulting in
elevated SIF values. At the same time, we also introduced the concept of 𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 in order
to attenuate concerns. The analysis of 𝑆𝐼𝐹𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 had been discussed in Section 4.1 and
would not be redundant here.

5. Conclusion
In this study, we evaluated the performance of low-configuration and cost-effective
spectrometers to retrieve far-red SIF at two agricultural sites cultivating winter wheat and
summer rice, using high-precision, more expensive spectrometers as a reference. The
findings indicated that the spectral data obtained from low-configuration spectrometers
correlated highly with data from high-configuration spectrometers (R2 > 0.90, average
RMSE < 8.95 mW m-2 nm-1 sr-1, rRMSE < 7%). The analysis of winter wheat data was
particularly promising, showing strong correlations in far-red SIF retrievals (R2 > 0.80,
average RMSE < 0.08 mW m-2 nm-1 sr-1) and maintaining a significant linear relationship
with GPP (R2 > 0.70). The monitoring of crop growth phases revealed minimal
discrepancies between the high- and low-configuration spectrometers, with an average

Journal of Remote Sensing Page 28 of 34

 deviation of 1.8 days. Additionally, simulation data indicated that accurate SIF retrieval
requires spectrometers with an SR greater than 2 nm and an SNR above 300. Despite
variable results under different environmental conditions, the correlation of SIF between the
two spectrometer configurations remained robust, even under cloudy conditions. This study
underscores the potential of low-configuration spectrometers in SIF retrieval: although the
SIF retrieved by the low-configuration spectrometer were not true, SIF retrieved by low-
configuration spectrometers had trends and sensitivities that were consistent with those
retrieved by high-configuration spectrometers. Further research is necessary, but the
insights gained are crucial for advancing ground-based SIF research and enhancing global
spectral observation networks.

Acknowledgments

General: We would thank researchers from Shangqiu station of Farmland Irrigation

Research Institute of Chinese Academy of Agricultural Sciences and Jurong Ecological
Experimental Station for their valuable contributions in the field measurements. We would
like to thank AgriSIF Environmental Technology Co., Ltd. located in Nanjing, China for

Downloaded from https://spj.science.org on December 13, 2024

providing equipment support. We greatly appreciate the editors and anonymous reviewers
for their insightful and constructive comments that helped us to improve our manuscript.

Author contributions: Y.Z provided guidance on methodology. L.P conducted the data
analysis. L.P, Y.W and L.W provided data. L.P, Y.W and Y.H designed the program code.
L.P wrote this manuscript. Z.Z, L.W and Y.Z reviewed this manuscript.

Funding: This work was funded by the National Natural Science Foundation of China
(42125105), the National Key Research and Development Program of China
(2022YFF1301900) and the Nanjing University Integrated Research Platform of the
Ministry of Education - Top Talents Program (090414380033).

Competing interests: The authors declare no conflict of interest.

Data Availability:
Data will be made available upon request.

Supplementary Materials:

Fig. S1—Fig. S16.

Journal of Remote Sensing Page 29 of 34

References
1. Kim, J., Y. Ryu, B. Dechant, H. Lee, H.S. Kim, A. Kornfeld, and J.A. Berry, Solar-induced chlorophyll
fluorescence is non-linearly related to canopy photosynthesis in a temperate evergreen needleleaf forest
during the fall transition. Remote Sensing of Environment, 2021. 258: p. 112362.
2. Li, Z., Q. Zhang, J. Li, X. Yang, Y. Wu, Z. Zhang, S. Wang, H. Wang, and Y. Zhang, Solar-induced
chlorophyll fluorescence and its link to canopy photosynthesis in maize from continuous ground
measurements. Remote Sensing of Environment, 2020. 236: p. 111420.
3. Damm, A., L. Guanter, V.C.E. Laurent, M.E. Schaepman, A. Schickling, and U. Rascher, FLD-based
retrieval of sun-induced chlorophyll fluorescence from medium spectral resolution airborne spectroscopy
data. Remote Sensing of Environment, 2014. 147: p. 256-266.
4. Guanter, L., L. Alonso, L. Gómez-Chova, J. Amorós-López, J. Vila, and J.F. Moreno, Estimation of solar
‐induced vegetation fluorescence from space measurements. Geophysical Research Letters, 2007. 34.
5. Zarco-Tejada, P.J., V. González-Dugo, and J.A.J. Berni, Fluorescence, temperature and narrow-band indices
acquired from a UAV platform for water stress detection using a micro-hyperspectral imager and a thermal
camera. Remote Sensing of Environment, 2012. 117: p. 322-337.
6. Frankenberg, C., A. Butz, and G.C. Toon, Disentangling chlorophyll fluorescence from atmospheric
scattering effects in O2 A-band spectra of reflected sun-light. Geophysical Research Letters, 2011. 38(3).
7. Guanter, L., Y. Zhang, M. Jung, J. Joiner, M. Voigt, J.A. Berry, C. Frankenberg, A.R. Huete, P. Zarco-
Tejada, J.-E. Lee, M.S. Moran, G. Ponce-Campos, C. Beer, G. Camps-Valls, N. Buchmann, D. Gianelle, K.

Downloaded from https://spj.science.org on December 13, 2024

Klumpp, A. Cescatti, J.M. Baker, and T.J. Griffis, Global and time-resolved monitoring of crop
photosynthesis with chlorophyll fluorescence. Proceedings of the National Academy of Sciences, 2014.
111(14): p. E1327-E1333.
8. Joiner, J., L. Guanter, R. Lindstrot, M. Voigt, A.P. Vasilkov, E.M. Middleton, K.F. Huemmrich, Y. Yoshida,
and C. Frankenberg, Global monitoring of terrestrial chlorophyll fluorescence from moderate-spectral-
resolution near-infrared satellite measurements: methodology, simulations, and application to GOME-2.
Atmos. Meas. Tech., 2013. 6(10): p. 2803-2823.
9. Guan, B., J. Zhang, W.A. Sethares, R. Kijowski, and F. Liu. Spectral Domain Convolutional Neural Network.
in ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing
(ICASSP). 2021.
10. Porcar-Castell, A., A. Mac Arthur, M. Rossini, L. Eklundh, J. Pacheco-Labrador, K. Anderson, M. Balzarolo,
M.P. Martín, H. Jin, E. Tomelleri, S. Cerasoli, K. Sakowska, A. Hueni, T. Julitta, C.J. Nichol, and L. Vescovo,
EUROSPEC: at the interface between remote-sensing and ecosystem CO₂ flux measurements
in Europe. Biogeosciences, 2015. 12(20): p. 6103-6124.
11. Zhang, Y., Q. Zhang, L. Liu, Y. Zhang, S. Wang, W. Ju, G. Zhou, L. Zhou, J. Tang, X. Zhu, F. Wang, Y.
Huang, Z. Zhang, B. Qiu, X. Zhang, S. Wang, C. Huang, X. Tang, and J. Zhang, ChinaSpec: A Network for
Long-Term Ground-Based Measurements of Solar-Induced Fluorescence in China. Journal of Geophysical
Research: Biogeosciences, 2021. 126(3): p. e2020JG006042.
12. Meroni, M., A. Barducci, S. Cogliati, F. Castagnoli, M. Rossini, L. Busetto, M. Migliavacca, E. Cremonese,
M. Galvagno, R. Colombo, and U.M. di Cella, The hyperspectral irradiometer, a new instrument for long-
term and unattended field spectroscopy measurements. Review of Scientific Instruments, 2011. 82(4): p.
043106.
13. Cogliati, S., M. Rossini, T. Julitta, M. Meroni, A. Schickling, A. Burkart, F. Pinto, U. Rascher, and R.
Colombo, Continuous and long-term measurements of reflectance and sun-induced chlorophyll fluorescence
by using novel automated field spectroscopy systems. Remote Sensing of Environment, 2015. 164: p. 270-
281.
14. Yang, X., J. Tang, J.F. Mustard, J.-E. Lee, M. Rossini, J. Joiner, J.W. Munger, A. Kornfeld, and A.D.
Richardson, Solar-induced chlorophyll fluorescence that correlates with canopy photosynthesis on diurnal
and seasonal scales in a temperate deciduous forest. Geophysical Research Letters, 2015. 42(8): p. 2977-
2987.
15. Campbell, P.K.E., K.F. Huemmrich, E.M. Middleton, L.A. Ward, T. Julitta, C.S.T. Daughtry, A. Burkart,
A.L. Russ, and W.P. Kustas Diurnal and Seasonal Variations in Chlorophyll Fluorescence Associated with
Photosynthesis at Leaf and Canopy Scales. Remote Sensing, 2019. 11, DOI: 10.3390/rs11050488.

Journal of Remote Sensing Page 30 of 34

16. Yang, X., H. Shi, A. Stovall, K. Guan, G. Miao, Y. Zhang, Y. Zhang, X. Xiao, Y. Ryu, and J.-E. Lee
FluoSpec 2—An Automated Field Spectroscopy System to Monitor Canopy Solar-Induced Fluorescence.
Sensors, 2018. 18, DOI: 10.3390/s18072063.
17. Grossmann, K., C. Frankenberg, T.S. Magney, S.C. Hurlock, U. Seibt, and J. Stutz, PhotoSpec: A new
instrument to measure spatially distributed red and far-red Solar-Induced Chlorophyll Fluorescence. Remote
Sensing of Environment, 2018. 216: p. 311-327.
18. Gu, L., J.D. Wood, C.Y.Y. Chang, Y. Sun, and J.S. Riggs, Advancing Terrestrial Ecosystem Science With
a Novel Automated Measurement System for Sun-Induced Chlorophyll Fluorescence for Integration With
Eddy Covariance Flux Networks. Journal of Geophysical Research: Biogeosciences, 2019. 124(1): p. 127-
146.
19. Zhang, Q., X. Zhang, Z. Li, Y. Wu, and Y. Zhang Comparison of Bi-Hemispherical and Hemispherical-
Conical Configurations for In Situ Measurements of Solar-Induced Chlorophyll Fluorescence. Remote
Sensing, 2019. 11, DOI: 10.3390/rs11222642.
20. Du, S., L. Liu, X. Liu, J. Guo, J. Hu, S. Wang, and Y. Zhang SIFSpec: Measuring Solar-Induced Chlorophyll
Fluorescence Observations for Remote Sensing of Photosynthesis. Sensors, 2019. 19, DOI:
10.3390/s19133009.
21. Zhang, Z., X. Zhang, A. Porcar-Castell, J.M. Chen, W. Ju, L. Wu, Y. Wu, and Y. Zhang, Sun-induced
chlorophyll fluorescence is more strongly related to photosynthesis with hemispherical than nadir
measurements: Evidence from field observations and model simulations. Remote Sensing of Environment,
2022. 279: p. 113118.

Downloaded from https://spj.science.org on December 13, 2024

22. Meroni, M., M. Rossini, L. Guanter, L. Alonso, U. Rascher, R. Colombo, and J. Moreno, Remote sensing of
solar-induced chlorophyll fluorescence: Review of methods and applications. Remote Sensing of
Environment, 2009. 113(10): p. 2037-2051.
23. Zhao, F., W. Ma, J. Zhao, Y. Guo, M. Tariq, and J. Li, Global retrieval of the spectrum of terrestrial
chlorophyll fluorescence: First results with TROPOMI. Remote Sensing of Environment, 2024. 300: p.
113903.
24. Pacheco-Labrador, J. and M.P. Mart í n Characterization of a Field Spectroradiometer for Unattended
Vegetation Monitoring. Key Sensor Models and Impacts on Reflectance. Sensors, 2015. 15, 4154-4175 DOI:
10.3390/s150204154.
25. Kim, J., Y. Ryu, C. Jiang, and Y. Hwang, Continuous observation of vegetation canopy dynamics using an
integrated low-cost, near-surface remote sensing system. Agricultural and Forest Meteorology, 2019. 264:
p. 164-177.
26. Richardson, A.D., B.H. Braswell, D.Y. Hollinger, J.P. Jenkins, and S.V. Ollinger, Near-surface remote
sensing of spatial and temporal variation in canopy phenology. Ecological Applications, 2009. 19(6): p.
1417-1428.
27. Kim, J., Y. Ryu, and B. Dechant, Development of a filter-based near-surface remote sensing system to
retrieve far-red sun-induced chlorophyll fluorescence. Remote Sensing of Environment, 2022. 283: p.
113311.
28. Damm, A., A. Erler, W. Hillen, M. Meroni, M.E. Schaepman, W. Verhoef, and U. Rascher, Modeling the
impact of spectral sensor configurations on the FLD retrieval accuracy of sun-induced chlorophyll
fluorescence. Remote Sensing of Environment, 2011. 115(8): p. 1882-1892.
29. Julitta, T., L.A. Corp, M. Rossini, A. Burkart, S. Cogliati, N. Davies, M. Hom, A. Mac Arthur, E.M.
Middleton, U. Rascher, A. Schickling, and R. Colombo Comparison of Sun-Induced Chlorophyll
Fluorescence Estimates Obtained from Four Portable Field Spectroradiometers. Remote Sensing, 2016. 8,
DOI: 10.3390/rs8020122.
30. Pacheco-Labrador, J., A. Hueni, L. Mihai, K. Sakowska, T. Julitta, J. Kuusk, D. Sporea, L. Alonso, A.
Burkart, M.P. Cendrero-Mateo, H. Aasen, Y. Goulas, and A. Mac Arthur Sun-Induced Chlorophyll
Fluorescence I: Instrumental Considerations for Proximal Spectroradiometers. Remote Sensing, 2019. 11,
DOI: 10.3390/rs11080960.
31. Belwalkar, A., T. Poblete, A. Longmire, A. Hornero, R. Hernandez-Clemente, and P.J. Zarco-Tejada,
Evaluation of SIF retrievals from narrow-band and sub-nanometer airborne hyperspectral imagers flown in
tandem: Modelling and validation in the context of plant phenotyping. Remote Sensing of Environment,
2022. 273: p. 112986.
32. Liu, L., X. Liu, and J. Hu, Effects of spectral resolution and SNR on the vegetation solar-induced
fluorescence retrieval using FLD-based methods at canopy level. European Journal of Remote Sensing, 2015.
48: p. 743 - 762.

Journal of Remote Sensing Page 31 of 34

33. Marrs, J.K., T.S. Jones, D.W. Allen, and L.R. Hutyra, Instrumentation sensitivities for tower-based solar-
induced fluorescence measurements. Remote Sensing of Environment, 2021. 259: p. 112413.
34. Mohammed, G.H., R. Colombo, E.M. Middleton, U. Rascher, C. van der Tol, L. Nedbal, Y. Goulas, O. Pé
rez-Priego, A. Damm, M. Meroni, J. Joiner, S. Cogliati, W. Verhoef, Z. Malenovský, J.-P. Gastellu-
Etchegorry, J.R. Miller, L. Guanter, J. Moreno, I. Moya, J.A. Berry, C. Frankenberg, and P.J. Zarco-Tejada,
Remote sensing of solar-induced chlorophyll fluorescence (SIF) in vegetation: 50 years of progress. Remote
Sensing of Environment, 2019. 231: p. 111177.
35. Wu, L., Y. Zhang, Z. Zhang, X. Zhang, Y. Wu, and J.M. Chen, Deriving photosystem-level red chlorophyll
fluorescence emission by combining leaf chlorophyll content and canopy far-red solar-induced fluorescence:
Possibilities and challenges. Remote Sensing of Environment, 2024. 304: p. 114043.
36. Zhao, F., W. Ma, P. Köhler, X. Ma, H. Sun, W. Verhoef, J. Zhao, Y. Huang, Z. Li, and A.K. Ratul, Retrieval
of Red Solar-Induced Chlorophyll Fluorescence With TROPOMI on the Sentinel-5 Precursor Mission. IEEE
Transactions on Geoscience and Remote Sensing, 2022. 60: p. 1-14.
37. Belwalkar, A., T. Poblete, A. Longmire, A. Hornero, and P.J. Zarco-Tejada. Comparing the Retrieval of
Chlorophyll Fluorescence from Two Airborne Hyperspectral Imagers with Different Spectral Resolutions
for Plant Phenotyping Studies. in 2021 IEEE International Geoscience and Remote Sensing Symposium
IGARSS. 2021.
38. Gao, F. and X. Zhang, Mapping Crop Phenology in Near Real-Time Using Satellite Remote Sensing:
Challenges and Opportunities. Journal of Remote Sensing. 2021.
39. Zhong, L., L. Hu, L. Yu, P. Gong, and G.S. Biging, Automated mapping of soybean and corn using

Downloaded from https://spj.science.org on December 13, 2024

phenology. ISPRS Journal of Photogrammetry and Remote Sensing, 2016. 119: p. 151-164.
40. Gao, F., M.C. Anderson, X. Zhang, Z. Yang, J.G. Alfieri, W.P. Kustas, R. Mueller, D.M. Johnson, and J.H.
Prueger, Toward mapping crop progress at field scales through fusion of Landsat and MODIS imagery.
Remote Sensing of Environment, 2017. 188: p. 9-25.
41. Schnelle, F. and E. Volkert, Internationale phänologische gärten Stationen eines grundnetzes f ü r
internationale phänologische beobachtungen. Agricultural Meteorology, 1964. 1(1): p. 22-29.
42. Diao, C., Remote sensing phenological monitoring framework to characterize corn and soybean
physiological growing stages. Remote Sensing of Environment, 2020. 248: p. 111960.
43. Liu, L., R. Cao, J. Chen, M. Shen, S. Wang, J. Zhou, and B. He, Detecting crop phenology from vegetation
index time-series data by improved shape model fitting in each phenological stage. Remote Sensing of
Environment, 2022. 277: p. 113060.
44. Lawley, V., M. Lewis, K. Clarke, and B. Ostendorf, Site-based and remote sensing methods for monitoring
indicators of vegetation condition: An Australian review. Ecological Indicators, 2016. 60: p. 1273-1283.
45. Yu, L., T. Liu, K. Bu, F. Yan, J. Yang, L. Chang, and S. Zhang, Monitoring the long term vegetation
phenology change in Northeast China from 1982 to 2015. Scientific Reports, 2017. 7(1): p. 14770.
46. Ganeva, D., T. Tallec, A. Brut, E. Prikaziuk, E. Tomelleri, G. Koren, J. Verrelst, K. Berger, L.V. Graf, S.
Belda, Z. Cai, and C.F. Silva, In-situ start and end of growing season dates of major European crop types
from France and Bulgaria at a field level. Data in Brief, 2023. 51: p. 109623.
47. Li, Z., Y. Zhang, Q. Zhang, Y. Wu, X. Zhang, and Z. Zhang, Tower-based automatic observation methods
and systems of solar-induced chlorophyll fluorescence in vegetation canopy. National Remote Sensing
Bulletin, 2021. 25: p. 1152-1168.
48. Reichstein, M., E. Falge, D. Baldocchi, D. Papale, M. Aubinet, P. Berbigier, C. Bernhofer, N. Buchmann,
T. Gilmanov, A. Granier, T. Grünwald, K. Havránková, H. Ilvesniemi, D. Janous, A. Knohl, T. Laurila, A.
Lohila, D. Loustau, G. Matteucci, T. Meyers, F. Miglietta, J.-M. Ourcival, J. Pumpanen, S. Rambal, E.
Rotenberg, M. Sanz, J. Tenhunen, G. Seufert, F. Vaccari, T. Vesala, D. Yakir, and R. Valentini, On the
separation of net ecosystem exchange into assimilation and ecosystem respiration: review and improved
algorithm. Global Change Biology, 2005. 11(9): p. 1424-1439.
49. Cogliati, S., W. Verhoef, S. Kraft, N. Sabater, L. Alonso, J. Vicent, J. Moreno, M. Drusch, and R. Colombo,
Retrieval of sun-induced fluorescence using advanced spectral fitting methods. Remote Sensing of
Environment, 2015. 169: p. 344-357.
50. Maier, S.W., K.P. Günther, and M. Stellmes. Sun-induced fluorescence: A new tool for precision farming.
2003.
51. Meroni, M., L. Busetto, R. Colombo, L. Guanter, J. Moreno, and W. Verhoef, Performance of Spectral
Fitting Methods for vegetation fluorescence quantification. Remote Sensing of Environment, 2010. 114(2):
p. 363-374.

Journal of Remote Sensing Page 32 of 34

52. Kumar, R. and L. Umanand, Estimation of global radiation using clearness index model for sizing
photovoltaic system. Renewable Energy, 2005. 30(15): p. 2221-2233.
53. van der Tol, C., W. Verhoef, J. Timmermans, A. Verhoef, and Z. Su, An integrated model of soil-canopy
spectral radiances, photosynthesis, fluorescence, temperature and energy balance. Biogeosciences, 2009.
6(12): p. 3109-3129.
54. Luo, J., K. Ying, and J. Bai, Savitzky–Golay smoothing and differentiation filter for even number data.
Signal Processing, 2005. 85(7): p. 1429-1434.
55. Zhang, X., M.A. Friedl, C.B. Schaaf, A.H. Strahler, J.C.F. Hodges, F. Gao, B.C. Reed, and A. Huete,
Monitoring vegetation phenology using MODIS. Remote Sensing of Environment, 2003. 84(3): p. 471-475.
56. Hermance, J.F., R.W. Jacob, B.A. Bradley, and J.F. Mustard, Extracting Phenological Signals From
Multiyear AVHRR NDVI Time Series: Framework for Applying High-Order Annual Splines With
Roughness Damping. IEEE Transactions on Geoscience and Remote Sensing, 2007. 45(10): p. 3264-3276.
57. Klosterman, S.T., K. Hufkens, J.M. Gray, E. Melaas, O. Sonnentag, I. Lavine, L. Mitchell, R. Norman, M.A.
Friedl, and A.D. Richardson, Evaluating remote sensing of deciduous forest phenology at multiple spatial
scales using PhenoCam imagery. Biogeosciences, 2014. 11(16): p. 4305-4320.
58. Cao, R., X. Chen, J. Chen, and W. Yang, An inherent limitation of solar-induced chlorophyll fluorescence
retrieval at the O2-A absorption feature in high-altitude areas. IEEE Geoscience and Remote Sensing Letters,
2013. 10(6): p. 1567-1571.
59. Li, L., C. Dai, Z. Wu, and Y. Wang, Temperature correction method for commercial CCD array
spectrometers used in spectral radiometry measurement. Journal of Physics: Conference Series, 2018. 972(1):

Downloaded from https://spj.science.org on December 13, 2024

p. 012020.
60. Corp, L.A., E.M. Middleton, Y.B. Cheng, P.K.E. Campbell, and K.F. Huemmrich. Impact of Spectral
Resolution on Solar Induced Fluorescence and Reflectance Indices for Monitoring Vegetation. in IGARSS
2008 - 2008 IEEE International Geoscience and Remote Sensing Symposium. 2008.
61. Guanter, L., L. Alonso, L. Gómez-Chova, M. Meroni, R. Preusker, J. Fischer, and J. Moreno, Developments
for vegetation fluorescence retrieval from spaceborne high-resolution spectrometry in the O2-A and O2-B
absorption bands. Journal of Geophysical Research: Atmospheres, 2010. 115(D19).
62. Middleton, E.M., L.A. Corp, and P.K.E. Campbell, Comparison of measurements and FluorMOD
simulations for solar ‐ induced chlorophyll fluorescence and reflectance of a corn crop under nitrogen
treatments. International Journal of Remote Sensing, 2008. 29(17-18): p. 5193-5213.
63. Cendrero-Mateo, M.P., S. Wieneke, A. Damm, L. Alonso, F. Pinto, J. Moreno, L. Guanter, M. Celesti, M.
Rossini, N. Sabater, S. Cogliati, T. Julitta, U. Rascher, Y. Goulas, H. Aasen, J. Pacheco-Labrador, and A.
Mac Arthur Sun-Induced Chlorophyll Fluorescence III: Benchmarking Retrieval Methods and Sensor
Characteristics for Proximal Sensing. Remote Sensing, 2019. 11, DOI: 10.3390/rs11080962.
64. Damm, A., L. Guanter, W. Verhoef, D. Schläpfer, S. Garbari, and M.E. Schaepman, Impact of varying
irradiance on vegetation indices and chlorophyll fluorescence derived from spectroscopy data. Remote
Sensing of Environment, 2015. 156: p. 202-215.
65. Malenovský, Z., L. Homolová, R. Zurita-Milla, P. Lukeš, V. Kaplan, J. Hanuš, J.-P. Gastellu-Etchegorry,
and M.E. Schaepman, Retrieval of spruce leaf chlorophyll content from airborne image data using continuum
removal and radiative transfer. Remote Sensing of Environment, 2013. 131: p. 85-102.
66. Song, Y., L. Wang, and J. Wang, Improved understanding of the spatially-heterogeneous relationship
between satellite solar-induced chlorophyll fluorescence and ecosystem productivity. Ecological Indicators,
2021. 129: p. 107949.
67. Zhang, Y., L. Guanter, J.A. Berry, C. van der Tol, X. Yang, J. Tang, and F. Zhang, Model-based analysis of
the relationship between sun-induced chlorophyll fluorescence and gross primary production for remote
sensing applications. Remote Sensing of Environment, 2016. 187: p. 145-155.
68. Zhang, Y., X. Xiao, C. Jin, J. Dong, S. Zhou, P. Wagle, J. Joiner, L. Guanter, Y. Zhang, G. Zhang, Y. Qin,
J. Wang, and B. Moore, Consistency between sun-induced chlorophyll fluorescence and gross primary
production of vegetation in North America. Remote Sensing of Environment, 2016. 183: p. 154-169.
69. Liu, Z., X. Lu, S. An, M. Heskel, H. Yang, and J. Tang, Advantage of multi-band solar-induced chlorophyll
fluorescence to derive canopy photosynthesis in a temperate forest. Agricultural and Forest Meteorology,
2019. 279: p. 107691.
70. Marrs, J.K., J.S. Reblin, B.A. Logan, D.W. Allen, A.B. Reinmann, D.M. Bombard, D. Tabachnik, and L.R.
Hutyra, Solar-Induced Fluorescence Does Not Track Photosynthetic Carbon Assimilation Following
Induced Stomatal Closure. Geophysical Research Letters, 2020. 47(15): p. e2020GL087956.

Journal of Remote Sensing Page 33 of 34

71. Dechant, B., Y. Ryu, G. Badgley, Y. Zeng, J.A. Berry, Y. Zhang, Y. Goulas, Z. Li, Q. Zhang, M. Kang, J.
Li, and I. Moya, Canopy structure explains the relationship between photosynthesis and sun-induced
chlorophyll fluorescence in crops. Remote Sensing of Environment, 2020. 241: p. 111733.
72. Parazoo, N.C., T. Magney, A. Norton, B. Raczka, C. Bacour, F. Maignan, I. Baker, Y. Zhang, B. Qiu, M.
Shi, N. MacBean, D.R. Bowling, S.P. Burns, P.D. Blanken, J. Stutz, K. Grossmann, and C. Frankenberg,
Wide discrepancies in the magnitude and direction of modeled solar-induced chlorophyll fluorescence in
response to light conditions. Biogeosciences, 2020. 17(13): p. 3733-3755.
73. Zhang, Z., Y. Zhang, A. Porcar-Castell, J. Joiner, L. Guanter, X. Yang, M. Migliavacca, W. Ju, Z. Sun, S.
Chen, D. Martini, Q. Zhang, Z. Li, J. Cleverly, H. Wang, and Y. Goulas, Reduction of structural impacts
and distinction of photosynthetic pathways in a global estimation of GPP from space-borne solar-induced
chlorophyll fluorescence. Remote Sensing of Environment, 2020. 240: p. 111722.
74. Liu, Y., J.M. Chen, L. He, Z. Zhang, R. Wang, C. Rogers, W. Fan, G. de Oliveira, and X. Xie, Non-linearity
between gross primary productivity and far-red solar-induced chlorophyll fluorescence emitted from
canopies of major biomes. Remote Sensing of Environment, 2022. 271: p. 112896.
75. Damm, A., L. Guanter, E. Paul-Limoges, C. van der Tol, A. Hueni, N. Buchmann, W. Eugster, C. Ammann,
and M.E. Schaepman, Far-red sun-induced chlorophyll fluorescence shows ecosystem-specific relationships
to gross primary production: An assessment based on observational and modeling approaches. Remote
Sensing of Environment, 2015. 166: p. 91-105.
76. Bandaru, V., R. Yaramasu, K. Pnvr, J. He, S. Fernando, R. Sahajpal, B.D. Wardlow, A. Suyker, and C.
Justice, PhenoCrop: An integrated satellite-based framework to estimate physiological growth stages of corn
and soybeans. International Journal of Applied Earth Observation and Geoinformation, 2020. 92: p. 102188.

Downloaded from https://spj.science.org on December 13, 2024

77. Hou, Y., Y. Wu, L. Wu, L. Pei, Z. Zhang, D. Ding, G. Wang, Z. Li, and Y. Zhang Identifying Crop Growth
Stages from Solar-Induced Chlorophyll Fluorescence Data in Maize and Winter Wheat from Ground and
Satellite Measurements. Remote Sensing, 2023. 15, DOI: 10.3390/rs15245689.
78. Magney, T.S., M.L. Barnes, and X. Yang, On the Covariation of Chlorophyll Fluorescence and
Photosynthesis Across Scales. Geophysical Research Letters, 2020. 47(23): p. e2020GL091098.
79. Hu, J., L. Liu, H. Yu, L. Guan, and X. Liu, Upscaling GOME-2 SIF from clear-sky instantaneous
observations to all-sky sums leading to an improved SIF–GPP correlation. Agricultural and Forest
Meteorology, 2021. 306: p. 108439.
80. Wu, G., K. Guan, C. Jiang, H. Kimm, G. Miao, X. Yang, C.J. Bernacchi, X. Sun, A.E. Suyker, and C.E.
Moore, Can upscaling ground nadir SIF to eddy covariance footprint improve the relationship between SIF
and GPP in croplands? Agricultural and Forest Meteorology, 2023. 338: p. 109532.
81. Wu, G., C. Jiang, H. Kimm, S. Wang, C. Bernacchi, C.E. Moore, A. Suyker, X. Yang, T. Magney, C.
Frankenberg, Y. Ryu, B. Dechant, and K. Guan, Difference in seasonal peak timing of soybean far-red SIF
and GPP explained by canopy structure and chlorophyll content. Remote Sensing of Environment, 2022.
279: p. 113104.
82. Sakamoto, T., B.D. Wardlow, A.A. Gitelson, S.B. Verma, A.E. Suyker, and T.J. Arkebauer, A Two-Step
Filtering approach for detecting maize and soybean phenology with time-series MODIS data. Remote
Sensing of Environment, 2010. 114(10): p. 2146-2159.

Journal of Remote Sensing Page 34 of 34