TCThe CryosphereTCThe Cryosphere1994-0424Copernicus PublicationsGöttingen, Germany10.5194/tc-11-47-2017Satellite microwave assessment of Northern Hemisphere lake ice phenology
from 2002 to 2015DuJinyangjinyang.du@ntsg.umt.eduKimballJohn S.DuguayClaudehttps://orcid.org/0000-0002-1044-5850KimYoungwookWattsJennifer D.Numerical Terradynamic Simulation Group, College of Forestry &
Conservation, University of Montana, Missoula, MT 59812, USADepartment of Geography & Environmental Management and
Interdisciplinary Centre on Climate Change, University of Waterloo,
Waterloo, Ontario N2L 3G1, CanadaJinyang Du (jinyang.du@ntsg.umt.edu)12January2017111476318August20165September20161December201613December2016This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://tc.copernicus.org/articles/11/47/2017/tc-11-47-2017.htmlThe full text article is available as a PDF file from https://tc.copernicus.org/articles/11/47/2017/tc-11-47-2017.pdf
A new automated method enabling consistent satellite assessment of seasonal
lake ice phenology at 5 km resolution was developed for all lake pixels
(water coverage ≥ 90 %) in the Northern Hemisphere using 36.5 GHz
H-polarized brightness temperature (Tb) observations from the
Advanced Microwave Scanning Radiometer for EOS and Advanced Microwave
Scanning Radiometer 2 (AMSR-E/2) sensors. The lake phenology metrics include
seasonal timing and duration of annual ice cover. A moving t test (MTT)
algorithm allows for automated lake ice retrievals with daily temporal
fidelity and 5 km resolution gridding. The resulting ice phenology record
shows strong agreement with available ground-based observations from the
Global Lake and River Ice Phenology Database (95.4 % temporal agreement)
and favorable correlations (R) with alternative ice phenology records from
the Interactive Multisensor Snow and Ice Mapping System (R=0.84 for water
clear of ice (WCI) dates; R=0.41 for complete freeze over (CFO) dates)
and Canadian Ice Service (R=0.86 for WCI dates; R=0.69 for CFO
dates). Analysis of the resulting 12-year (2002–2015) AMSR-E/2 ice record
indicates increasingly shorter ice cover duration for 43 out of 71
(60.6 %) Northern Hemisphere lakes examined, with significant
(p< 0.05) regional trends toward earlier ice melting for only five
lakes. Higher-latitude lakes reveal more widespread and larger trends toward
shorter ice cover duration than lower-latitude lakes, consistent with
enhanced polar warming. This study documents a new satellite-based approach
for rapid assessment and regional monitoring of seasonal ice cover changes
over large lakes, with resulting accuracy suitable for global change studies.
Introduction
Ice phenology describes the seasonal cycle of lake ice cover and encompasses
freeze-up and breakup periods and ice cover duration (Duguay et al., 2015a).
Freeze-up corresponds to the time period between the beginning of ice
formation and the formation of a complete sheet of ice; breakup involves the
time period between the onset of spring melt and the complete disappearance
of ice from the lake surface (Kang et al., 2012). These ice phenology
variables are key metrics sensitive to weather and climate conditions and
influence lake–atmosphere interactions and hydrological and ecological
processes in high-latitude and high-altitude regions (Duguay et al., 2006,
2012, 2015a; Mishra et al., 2011). By insulating lake water from the
overlying atmosphere and minimizing water and atmosphere heat and gas
exchanges, lake ice has a controlling influence on water-column oxygen
concentration, water temperature, and the composition and abundance of
aquatic species (Livingstone, 1997; Bengtsson and Herschy, 2012; Kang et al.,
2012; Wrona et al., 2016). In addition to the impacts on aquatic life, the
formation and disappearance of lake ice also has a significant influence on
the spread of man-made pollutants such as perfluorinated chemicals (PFCs)
(Veillette et al., 2012; Wrona et al., 2016). The extent and duration of lake
ice also affect human activities, including hydroelectric power generation,
navigation and winter transportation, and production and distribution of food
and water (Schröter et al., 2005; Weyhenmeyer et al., 2011). Moreover,
lake ice phenology is closely coupled with atmospheric heat fluxes (Latifovic
and Pouliot, 2007; Park et al., 2016) and sensitive to the alteration of
weather patterns under projected global warming (Magnuson et al., 2000).
Accurate and consistent records of lake ice phenology especially in
data-sparse regions, including much of the pan-Arctic and Qinghai–Tibetan
Plateau regions, provide valuable information for monitoring global change
impacts on high-latitude and high-altitude environments (Magnuson et al.,
2000). Previous studies have documented significantly earlier ice breakup
between the 1950s and 2000s for lakes in Canada (Duguay et al., 2006; Latifovic
and Pouliot, 2007; Prowse et al., 2011) and decreasing ice cover duration of
Eurasia lakes during the last few decades (Vuglinsky and Gronskaya, 2006;
Karetnikov and Naumenko, 2008; Prowse et al., 2011). Shorter ice cover
seasons may promote greater CH4 emissions from northern lakes (Greene et
al., 2014), which could reinforce further climate warming due to the role of
CH4 as a potent greenhouse gas. Despite a general tendency for later
freezing and earlier breakup in the Northern Hemisphere (Magnuson et al.,
2000), various tendencies including earlier ice formation and later ice
breakup over specific lakes and time periods may exist. For example,
observations from satellite altimetry and radiometry over 1992–2004 for Lake
Baikal showed a tendency for colder winters, with earlier ice formation,
later ice breakup, and ice duration increase (Kouraev et al., 2007a, b).
A historic lake ice phenology database has been assembled from long-term
ground-based observations across the northern domain (Magnuson et al., 2000;
Benson et al., 2012); however, the number of monitoring sites is extremely
sparse, with variable observational recording periods and methods, which
limits capabilities for regional assessment and monitoring of environmental
changes. Data acquired from spaceborne optical–thermal infrared (TIR) and
microwave sensors have also been applied for monitoring river and lake ice
phenology at regional and global scales (Chu and Lindenschmidt, 2016). Optical–TIR
remote sensing can provide accurate estimation of land surface temperature
(LST) and classification of land cover types at relatively fine spatial
resolutions (∼ 10s to 100s of meters), while LST (1 km resolution) and
snow cover (500 m resolution) products derived from MODIS (Moderate
Resolution Imaging Spectroradiometer) have been used to infer lake ice
conditions (Nonaka et al., 2007; Hall et al., 2010; Kheyrollah Pour et al.,
2012). Time series of AVHRR (Advanced Very High Resolution Radiometer)
imagery have also been used to classify Canadian lake ice phenology events
with relatively high accuracy (Latifovic and Pouliot, 2007). However,
regional monitoring of lake ice dynamics from satellite optical–TIR sensors
is constrained by signal degradation and data loss stemming from seasonal
reductions in solar illumination at higher latitudes and persistent cloud
cover, smoke, and other atmospheric aerosol contamination (Maslanik and Barry,
1987; Jeffries et al., 2005; Helfrich et al., 2007; Kang et al., 2012).
Satellite microwave remote sensing at lower frequencies
(∼< 89 GHz) is relatively insensitive to solar illumination and
atmosphere constraints, while current microwave radiometers onboard polar
orbiting satellites provide frequent (∼ daily) observations spanning
northern (≥ 45∘ N) land areas. The active and passive
microwave retrievals are also highly sensitive to the large contrast in
surface dielectric properties between open water and ice cover over large
lakes. Despite successful applications using active microwave remote sensing
in lake ice retrieval (Leconte and Klassen, 1991; Nolan et al., 2002; Howell
et al., 2009; Geldsetzer et al., 2010), capabilities for global lake ice
monitoring from satellite radar sensors have been constrained by limited
global coverage and temporal frequency of observations. Alternatively,
spaceborne microwave radiometers have provided brightness temperature
(Tb) observations since 1978 with relatively high temporal
fidelity (∼ 1–2 days) especially at higher (≥ 45∘ N)
latitudes. Frequent microwave radiometer data acquisition and complete time
series of images are valuable to ice phenology studies and also supportive to
improving numerical weather model predictions (Helfrich et al., 2007) and
timely monitoring of lake ice events, including transient ice disturbances
(Jeffries et al., 2005). Despite having relatively coarser spatial resolution
retrievals than optical–TIR sensors, the capability for consistent daily lake
ice monitoring available from passive microwave observations provides added
precision for delineating lake ice phenology trends, which may be much
smaller than year-to-year ice cover variability. The satellite Tb
retrievals are capable of detecting lake ice phenology events coinciding with
large changes in surface emissivity, but the passive microwave retrievals are
constrained by a generally coarser spatial resolution than radar and
optical–TIR sensors. Despite these limitations, ice freeze-up and breakup
events for Great Slave Lake (GSL) were monitored using a threshold-based
method for SSM/I (Special Sensor Microwave Imager) observations at 85 GHz
(Walker and Davey, 1993; Ménard et al., 2002). Recently, H-polarized
AMSR-E (Advanced Microwave Scanning Radiometer for EOS) Tb
observations at 19 GHz were analyzed to determine ice phenology for GSL and
Great Bear Lake (GBL), the two largest lakes in northern Canada (Kang et al.,
2012, 2014). Similar Tb records from SSM/I and SMMR
(Scanning Multichannel Microwave Radiometer) were also used to monitor lake
ice phenology for Nam Co Lake (Ke et al., 2013) and Qinghai Lake (Che et al.,
2009) within the high-elevation Qinghai–Tibetan Plateau. Previous studies
based on satellite passive microwave remote sensing have mainly focused on
one or two lakes using empirical algorithm thresholds developed for specific
study areas. There is a great need to develop a universal lake ice detection
algorithm and establish a consistent ice phenology database covering major
lakes at the global scale for climate impact assessments such as those
published by the Intergovernmental Panel and Climate Change (Stocker et al.,
2013).
In this study, we present a new automated method to derive lake ice phenology
using 36.5 GHz H-polarized satellite radiometric Tb measurements
from AMSR-E and AMSR2 (Advanced Microwave Scanning Radiometer 2). The
algorithm is used to produce daily lake ice maps with 5 km gridding. The
resulting AMSR-E/2 Lake Ice Phenology (LIP) record encompasses all 5 km by
5 km lake pixels (water coverage ≥ 90 %) within the Northern
Hemisphere (≥ 0∘ N) and spans more than 12 years of
observations including both AMSR-E (June 2002 to September 2011) and AMSR2
(June 2012 to December 2015) satellite sensor records. Here we present a
detailed methods description and evaluation of the LIP product against other
independent observations and alternative lake ice products. A trend analysis
is also conducted to characterize recent regional LIP changes over the study
period.
Three sets of lakes selected for the LIP analysis in the Northern
Hemisphere mid- and high latitudes (≥ 30∘ N). The purple star
symbols denote lakes used for evaluating LIP retrievals on a per pixel basis
against GLRIPD ground-based observations, including Lake Superior in the USA
and Canada and Lake Oulujarvi, Lake Haukivesi, and Lake Paijanne in Finland.
The red star symbols denote 12 lakes (Great Bear Lake, Great Slave Lake,
Smallwood Lake, Nettiling Lake, Dubawnt Lake, Amadjuak Lake, Wollaston Lake,
Baker Lake, Kasba Lake, Lesser Slave Lake, and Peter Pond Lake in Canada and
Red Lake in USA) used for the lake-wide comparisons between the LIP results
from this study and other regional lake ice phenology records (IMS, CIS).
The 71 lakes selected for assessing LIP trends over the 12-year satellite
record are in bright blue.
MethodsStudy domain and datasetsStudy domain
This study utilizes satellite passive microwave remote sensing to detect lake
ice changes for 5 km lake pixels in the Northern Hemisphere, with a
particular focus on lake ice phenology in the mid- and high latitudes (≥ 30∘ N). The domain (Fig. 1) includes the high northern pan-Arctic
region and high-altitude Qinghai–Tibetan Plateau, which are data-sparse but
strongly sensitive to the Arctic amplification effect (Serreze and Francis,
2006; Woo et al., 2007) and/or elevation-dependent warming (Wang et al.,
2011; Mountain Research Initiative EDW Working Group, 2015). Both regions are
also characterized by cold climate conditions with extensive winter ice
cover. The resulting domain includes three sets of lakes for algorithm
evaluation and lake ice phenology analysis. Among the lakes analyzed, four
are represented in the Global Lake and River Ice Phenology Database (GLRIPD)
(Benson and Magnuson, 2000) and were used to evaluate the LIP estimates on a
per pixel basis against available ground-based observations; the four GLRIPD
lakes evaluated include Lake Superior in the USA and Canada and Lake Oulujarvi,
Lake Haukivesi, and Lake Paijanne in Finland. In addition, 12 North American
lakes (GBL, GSL, Smallwood Lake, Nettiling Lake, Dubawnt Lake, Amadjuak Lake,
Wollaston Lake, Baker Lake, Kasba Lake, Lesser Slave Lake, and Peter Pond
Lake in Canada and Red Lake in USA) that experience annual breakup and
freeze-up events were selected for lake-wide intercomparisons between the
LIP metrics derived from this study and alternative lake ice products from
the Interactive Multisensor Snow and Ice Mapping System (IMS) (Helfrich et
al., 2007; http://www.natice.noaa.gov/ims/) and the Canadian Ice
Service (CIS) (Howell et al., 2009).
Finally, regional LIP trends were assessed over the 12-year (2002–2015)
satellite record for 71 Northern Hemisphere lakes identified in the Global
Lakes and Wetlands Database (GLWD) (Lehner and Döll, 2004). The lakes
selected represent approximately 23 % (297 044 km2) of the
estimated total surface area of large lakes (area ≥ 50 km2)
within the domain (Lehner and Döll, 2004). The 71 lakes were selected on
the basis of having (a) at least one 5 km lake pixel with 100 % water
coverage and located outside of a 5 km land buffer zone and (b) all pixels
representing the lake having at least 20 days with full ice coverage and
20 days with open water. Criterion (a) was used to reduce potential
contamination from adjacent land areas since the native resolutions of
36.5 GHz observations are approximately 14 km × 8 km for AMSR-E
and 12 km × 7 km for AMSR2, respectively (Imaoka et al., 2010).
Criterion (b) emphasizes lakes having extended ice and open water seasons
rather than those with a temporary ice cover or short open water phase. The
20-day minimum duration was set according to the predefined subsample sizes
of our algorithm (Sect. 2.3).
Datasets used for algorithm development
The lake ice detection algorithm developed in this study relies primarily on
36.5 GHz H-polarized Tb retrievals from AMSR-E and AMSR2. The
AMSR-E sensor was operational on the NASA Aqua satellite from June 2002 to
October 2011 and provided twice-daily measurements of global microwave
emissions over land with descending/ascending orbital equatorial crossings at
01:30/13:30 local time and vertically (V) and horizontally (H)
polarized Tb retrievals at six frequencies (6.9, 10.7, 18.7,
23.8, 36.5, 89.0 GHz). For this study, we used the AMSR-E ascending
36.5 GHz orbital swath data at the native footprint resolution of
approximately 14 km × 8 km (Kawanishi et al., 2003). After the
cessation of AMSR-E operations on 4 October 2011, its successor AMSR2 was
launched on 18 May 2012 onboard the sun-synchronous JAXA GCOM-W1 satellite.
AMSR2 is similar to AMSR-E in sensor configuration, including frequencies,
incidence angle, and orbital equatorial crossing time. Major AMSR2
advancements over AMSR-E include an additional frequency at 7.3 GHz designed
for mitigating radio frequency interference (RFI) and a larger (2.0 m
diameter) main reflector for enhanced spatial resolution. The AMSR2 L1R
(version 1.2) resampled ascending swath 36.5 GHz Tb retrievals
at approximately 12 km × 7 km resolution were used for this study.
The uncalibrated AMSR2 Tb retrievals were estimated to be
positively biased against AMSR-E by ∼ 1.3 K (Du et al., 2014).
However, the sensor inconsistency is expected to have minimal impacts on our
algorithm, which relies on Tb time series change signal detection
rather than Tb absolute accuracy.
The Level 1 GLWD (Lehner and Döll,
2004) comprises the 3067 largest lakes (area ≥ 50 km2) and 654
largest reservoirs (storage capacity ≥ 0.5 km3) worldwide and
was used for identifying Northern Hemisphere water bodies and the 71 large
lakes used for the LIP assessment (Fig. 1). We also used the MODIS 250 m
land–water mask (MOD44W) for calculating the proportional water coverage of
5 km resolution pixels within lake areas identified by the GLWD (Carroll et
al., 2009).
Datasets used for algorithm evaluation
Four lake ice phenology databases were used to evaluate the LIP retrievals,
including (a) the GLRIPD (Benson and Magnuson, 2000), (b) the National
Oceanic and Atmospheric Administration (NOAA) IMS 4 km daily snow and ice
product (Helfrich et al., 2007; http://www.natice.noaa.gov/ims/),
(c) the CIS lake-wide ice product (Howell et al., 2009), and (d) MODIS
quick-look images for GBL downloaded from the Geographic Information Network
of Alaska (http://www.gina.alaska.edu).
The GLRIPD contains descriptive ice cover data for 865 lakes and rivers in
the Northern Hemisphere (Benson and Magnuson, 2000). The GLRIPD includes
ground-based (lakeshore) observations that were used for evaluating the
corresponding LIP results for the targeted lakes. The GLRIPD records the
first date when the water body was observed to be completely ice covered and
the date when the last ice breakup was observed before the summer open water
phase for each year of record (Benson and Magnuson, 2000). For evaluating LIP
results representing 5 km lake dominant pixels (water coverage ≥ 90 %), the lake was assumed completely covered with ice for the period
between the first date with complete ice cover and last ice breakup date as
recorded in the GLRIPD, while lakes were classified as open water condition
for other dates within each annual cycle. Only four lakes were selected for
the LIP comparisons due to a predominance of ice observations from smaller
lakes in the GLRIPD database. The temporal coverage of GLRIPD observations
for the four lakes that overlaps with the AMSR-E/2 record extends from 2002
to 2007 for Lake Superior and 2003 to 2007 for Lake Oulujarvi, Lake
Haukivesi, and Lake Paijanne.
The NOAA IMS daily snow and ice product provides snow and ice cover extent
information derived from ground observations and an extensive variety of
satellite observations, including AVHRR, GOES (Geostationary Operational
Environmental Satellite), SSM/I, and AMSU (Advanced Microwave Sounding Unit)
(Helfrich et al., 2007). The CIS lake ice product estimates lake ice cover
fraction in tenths (0: open water; 10: complete ice cover) for nearly 140
lakes across Canada and the northern USA from visual interpretation of
1.1 km resolution NOAA AVHRR and 100 m resolution RADARSAT ScanSAR imagery
(Howell et al., 2009), and MODIS (250–500 m) and Visible/Infrared Imaging
Radiometer Suite (VIIRS) I-band (375 m) observations. The CIS product
provides a single lake-wide value per lake on a weekly basis. Both the 4 km
IMS grid products (year 2004–2015) and CIS data (year 2002–2015) were used
for lake-wide comparisons against the resulting LIP retrievals.
MODIS quick-look images in true-color composites (bands 1, 4, 3 in RGB) were
selected for qualitative visual comparisons with the LIP results. The MODIS
images were acquired over the breakup season (2012–2013) with clear-sky
conditions on 22, 27 June and 8 July 2013 and extensive cloud cover on
5 July 2013. The quick-look products were provided at 250 m resolution in
Albers equal-area conic projection.
In addition, ERA-Interim (Dee et al., 2011) quarter-degree reanalysis surface
air temperature (SAT) data was analyzed for evaluating LIP trends over the 71
Northern Hemisphere lakes selected (Fig. 1). ERA-Interim is a global
atmospheric model data reanalysis produced by the European Centre for
Medium-Range Weather Forecasts, and the data assimilation system used to
produce ERA-Interim is based on a 2006 release of the IFS (Cy31r2) including
a four-dimensional variational analysis (4D-Var) with a 12 h analysis window
(Dee et al., 2011). Daily average SAT over the spring (MAM) and fall (SON)
seasons of years 2002 to 2015 was extracted for the quarter-degree grids
encompassing the lake centers.
Data processing
To derive the LIP estimates, AMSR-E/2 36.5 GHz orbital swath Tb
data were spatially resampled to a 5 km resolution polar EASE-Grid
(version 2) format using an inverse distance squared weighting method
(Ashcroft and Wentz, 2000; Brodzik et al., 2012, 2014). It is worth noting
that the Tb spatial gridding is posted at 5 km resolution while
the original 36.5 GHz AMSR-E/2 observations have coarser native sensor
footprints (∼ 12 km for AMSR-E and 9 km for AMSR2). The finer grid
spacing is intended to facilitate product comparisons and analyses with other
alternative lake products derived at similar resolutions, including the NOAA
IMS 4 km daily snow and ice product and a land surface fractional open water
cover dataset derived from AMSR-E/2 at 5 km resolution (Du et al., 2016).
Before carrying out the Tb gridding process, an additional
altitude correction was made to the AMSR-E data by considering the actual
surface of the Earth instead of that of an ideal Earth ellipsoid. The same
altitude correction was used for the AMSR2 L1R data (Maeda et al., 2016).
According to Maeda et al. (2016), an altitude of 3000 m leads to about 4 km
displacement of AMSR2 Tb geolocation. Thus the altitude
correction is a necessary prerequisite to ensure reliable analysis of
AMSR-E/2 lake ice phenology retrievals at higher elevations, including the
Qinghai–Tibetan Plateau.
The finer-resolution MOD44W static open water maps were aggregated to the
same 5 km resolution polar EASE-grid 2.0 projection format as LIP and used
with the GLWD to identify dominant lake pixels (water coverage ≥ 90 %) where the AMSR-E/2 lake ice detection was made. The 250 m
resolution MODIS quick-look images were re-projected to the EASE-grid 2.0
projection for visual comparisons with the LIP results.
Algorithm theoretical basis
Accurate modeling of satellite observed microwave emissions from lakes is
complex and requires good understanding of microwave scattering and emitting
mechanisms from atmosphere and lake elements. Microwave emissions from a
non-scattering atmosphere are governed by both air temperature and atmosphere
optical thickness, which is approximately the sum of the optical thickness of
oxygen, cloud liquid water, and atmospheric water vapor (Wang and Tedesco,
2007; Du et al., 2015). Microwave emissions from a lake with an upper layer
that may consist of water, ice, and snow are determined by a number of
factors; these factors include lake surface roughness, water dielectric
properties mainly affected by water salinity and temperature, ice thickness
and dielectric properties, and snow cover dielectric properties mainly controlled
by snow density and wetness, snow particle size, and stratification of snow
and ice layers (Du et al., 2010; Lemmetyinen et al., 2010, 2011). Despite the
complexity of the lake emission problem, sharp changes in satellite microwave
Tb observations at multiple frequencies are evident during the
transitions between lake freeze-up and breakup periods. For example,
previous studies showed low Tb measurements (< 150 K) from
H-polarized 37 GHz SMMR data over low-emissivity open water regions of the
Great Lakes and Gulf of Mexico, contrasting with much higher Tb
values (> 215 K) over western Lake Superior under frozen conditions due
to the high emissivity of lake ice (Ferraro et al., 1986). Similarly, the
H-polarized emissivity at 35 GHz and 50∘ incidence angle is
approximately 0.356 for a calm and unfrozen lake at 0–8 ∘C and is
well below the emissivity (> 0.610) of different types of snow and ice
(Mätzler, 1994). These studies suggest a very large Ka-band
Tb difference (> 60 K) between a lake at 0∘ with no
ice and 100 % ice coverage. The timing of ice formation and disappearance
can therefore be determined by the large characteristic Tb
changes indicated from satellite passive microwave observations (Walker and
Davey, 1993; Che et al., 2009; Kang et al., 2012).
Algorithm development
For identifying freeze-up and breakup events, a moving t test method (MTT)
was introduced to detect abrupt temporal changes in the H-polarized 36.5 GHz
Tb observations from AMSR-E and AMSR2. Selection of 36.5 GHz
Tb observations from other available AMSR-E/2 frequencies
represents a compromise between finer spatial resolutions gained from higher
frequencies and less sensitivity to potential atmosphere contamination
available from lower Tb frequency observations. Moreover,
H-polarization Tb retrievals were used instead of V-polarization
data due to their reported higher sensitivity to lake freeze-up/breakup
signals (Kang et al., 2012). The detailed lake ice detection method used in
this study is described below.
Step 1: detection of abrupt changing point
The MTT method was initially developed for detecting abrupt climate changes
by examining whether the difference between the mean values of two subsamples
is statistically significant (Jiang and You, 1996; Xiao and Li, 2007). As
detailed in the literature (Xiao and Li, 2007), for a time series with n
elements, a t test is made at each point xk for evaluating the
difference of the two subsets xk1 and xk2 (n1≤k≤n-n2; n1, n2 are the subsample sizes) before and after
xk. The t statistic is defined as
t=xk2‾-xk1‾sk1n1+1n2,
where sk=n1sk12+n2sk22n1+n2-2,
xk2‾ and xk1‾ are the mean values, and
sk12 and sk22 are the variances for the two subsets,
respectively. Given a significance level α, xk is determined as
an abrupt changing point if t≥tα. In this study,
we define α as 0.005 and temporal subsample sizes as n1=n2=20 days. The 20-day requirement is set for excluding potentially
dynamic Tb changes caused by short-term weather events such as
storms.
Step 2: determining reference Tb values for lake
ice conditions
For a group of detected changing points sequenced from p to q, the mean
Tb values xp1‾ and xq2‾ as defined
in Eq. (1) are representative of the satellite observations over the stable
stages before and after the changing period, respectively. For a lake
experiencing a complete annual freeze-up/breakup cycle, at least two groups
of seasonal changing points can be defined. Besides the sharp Tb
increases as lake water freezes, the melting of dry snow overlying lake ice
to wet snow can induce further increases in the observed Tb since
microwave emissions from wet snow are close to that of a blackbody (Ulaby et
al., 1986). Therefore, assuming xp1‾ is always smaller than
xq2‾, we define the lowest xp1‾ of all changing
groups as the reference Tb for lake water and xq2‾
from the same group as the reference Tb for lake ice. Lake ice
conditions for a given date i can thus be determined as
Ice dominant ifTbi≥TbthresholdWater dominant ifTbi<Tbthreshold,
where Tbi is the Tb for date i,
Tbthreshold=(xp1‾+xq2‾)/2.0, and
xp1‾-xq2‾ is required to be larger
than 30 K since liquid–ice phase changes of lake water can lead to large
Tb changes exceeding 60 K as introduced in Sect. 2.2.
Step 3: deriving lake ice status
Based on Eq. (2), lake ice status is first derived for each point i in the
Tb time series where n1≤i≤n-n2 using a
temporally smoothed Tbi defined as the mean Tb
within the range[i-n1/2,i+n2/2]. The use of a smoothed
Tbi minimizes the impact of high temporal frequency events in
the time series while emphasizing lower frequency lake ice-covered and
ice-free signals. Thus for point j, whose temporally adjacent points have
different lake ice status, the refined lake ice detection is carried out
using Eq. (2) for each observed Tb value within the range
[j-n1/2,j+n2/2]. For running the algorithm, missing daily
Tb retrievals were obtained through temporal linear interpolation
of adjacent successful Tb retrievals acquired from the same
ascending orbits (Kim et al., 2012). However, only the lake ice detection
results corresponding to the actual satellite observations were output for
further analysis. The above lake ice detection process was carried out for
each Tb time series from AMSR-E and AMSR2 separately because of
the 7-month gap (4 October 2011–18 May 2012) in the observation records
between the two sensors.
Definitions of remotely sensed lake ice phenology variables from
this study in relation to the other lake ice observational datasets used for
the LIP validation assessment on a per pixel basis and for entire lakes.
Per pixel basis Entire lake TerminologyDefinitionTerminologyDefinitionIce-on dateDay of year on which a pixel becomes totally ice coveredComplete freeze over(CFO) dateDay of year when all pixels become totally ice coveredIce-off dateDay of year on which a pixel becomes totally ice freeWater clear of ice(WCI) dateDay of year when all pixels become totally ice freeIce cover duration of entire lake (ICDe)number of days between CFO and WCIAssociated dataset: LIP Associated datasets: LIP, IMS, CIS
The above MTT algorithm was applied to all 5 km pixels with dominant (≥ 90 %) open water coverage within the Northern Hemisphere domain on a
daily basis to generate the AMSR-E/2 LIP dataset describing lake ice
conditions. The dominant (≥ 90 %) open water coverage criterion is
set to include lake pixels while reducing potential contamination from
adjacent land areas.
Evaluation of lake ice phenology retrievals
The resulting LIP retrievals were evaluated against other available lake ice
databases, including the GLRIPD, IMS, and CIS products. The remotely sensed
lake ice phenology variable definitions from this study are summarized in
Table 1 relative to the other lake ice observational data records used for
LIP validation on a per pixel basis and for entire lakes (Kang et al., 2012;
Duguay et al., 2015a).
For comparing with the GLRIPD ground-based records, the LIP ice-on and
ice-off dates were extracted for the 5 km pixel closest to the GLRIPD
observation site. The pixel representing Lake Superior has 100 % water
coverage (lat/long: 46.78∘ N/-90.45∘ W). For the pixels
representing Lake Oulujarvi (lat/long: 64.3∘ N/27.3∘ E),
Lake Haukivesi (lat/long: 62.07∘ N/28.57∘ E), and Lake
Paijanne (lat/long: 61.19∘ N/25.55∘ E), the water coverage
is 100, 91.4, and 95.7 %, respectively.
The LIP-derived annual CFO (complete freeze over) and WCI (water clear of ice) dates for the 12 selected lakes were also
compared with alternative IMS and CIS ice products. Different from the
dominant open water coverage (≥ 90 %) requirement set for
generating the LIP database, only lake pixels with complete (100 %) open
water coverage and outside of the 5 km land buffer zone were considered in
the IMS and LIP comparisons; this same criterion was set for the lake-wide
comparisons to minimize potential contamination from adjacent land areas
since the native 36.5 GHz AMSR-E/2 footprint ranges from approximately
9 to 12 km. Considering possible retrieval uncertainties, the CFO–WCI dates
derived from the LIP and IMS datasets were slightly adjusted from the
definitions in Table 1 and were determined as the dates when most (99.5 %
for this study) lake pixels were identified as ice/water. The CIS CFO dates
were determined when the reported lake ice fraction was 9, followed by changes
from 9 to 10, and WCI dates were derived when the reported lake ice fraction
was 1 followed by changes from 1 to 0. The derived CIS CFO–WCI dates are
comparable with corresponding LIP results, excluding waters adjacent to land
such as part of the eastern arm of the GSL where a high concentration of islands
exists and ice formation/melting timing was found to be different from other
GSL areas (Howell et al., 2009; Kang et al., 2012).
Analysis of lake ice phenology changes
Based on the LIP database covering the AMSR-E (June 2002–October 2011) and
AMSR2 (June 2012–December 2015) observation periods, we selected 71 lakes in
the Northern Hemisphere from 250 of the world's largest lakes (including both
natural and artificial lakes), as described in the GLWD, to analyze potential
lake ice phenology trends, including WCI date, CFO date, and annual ICDe. In
order to assess the pattern of recent Northern Hemisphere lake ice phenology
changes, a temporal trend analysis was performed on the 12-year LIP record
for each of the 71 lakes. The assumption of independent observations was
first determined using a correlogram (Noguchi et al., 2011). For ice
phenology time series without significant autocorrelation detected, the
magnitude and significance of temporal trends were tested using the
non-parametric Mann–Kendall and Sen's methods (Sen, 1968; Duguay et al.,
2006). Alternatively, for a time series with persistent serial correlation,
additional prewhitening approaches were applied (Zhang et al., 2000). For
evaluating LIP-derived lake phenology, a similar temporal trend analysis was
also carried out on the ERA-Interim daily average SAT over the lakes for the
spring and fall seasons from 2002 to 2015.
Comparison of lake ice status for Lake Superior, USA and
Canada (a),
Lake Oulujarvi, Finland (b), Lake Haukivesi, Finland (c), and Lake
Paijanne, Finland (d), derived from the Global Lake and River Ice Phenology
Database (GLRIPD) (blue dots) and AMSR Lake Ice Phenology retrievals (LIP)
(red dots). The AMSR-E/2 36.5 GHz H-polarized daily Tb retrievals
used in the LIP algorithm are also plotted for reference (black line). The
blue/red dots represent GLRIPD/LIP-derived ice conditions as indicated by
their y axis positions for the dates described by their x axis coordinates.
Summary of the comparison results for water clear of ice (WCI) and
complete freeze over (CFO) dates derived from AMSR Lake Ice Phenology (LIP),
Canadian Ice Service (CIS) datasets for the period 2002–2015, and the
NOAA/IMS (IMS) dataset for the period 2004–2015. RLIP,CIS/IMS denotes
the correlation coefficient between the LIP and CIS/IMS datasets;
DLIP,CIS/IMS is the average difference (unit: day) in WCI or CFO dates
calculated by LIP minus CIS/IMS.
Statistics of WCI date comparisons Statistics of CFO date comparisons Lake nameRLIP,CISRLIP,IMSDLIP,CISDLIP,IMSRLIP,CISRLIP,IMSDLIP,CISDLIP,IMSGreat Bear0.900.94-2-30.900.5403Great Slave0.850.91-6-50.720.63-23Smallwood0.660.62-6-40.70-0.08-41Nettiling0.910.84-9-40.920.33-210Dubawnt0.920.79-810.340.78-9-5Amadjuak0.890.80-7-40.870.27-26Wollaston0.950.87-500.66-0.42-79Baker0.800.70-10-10.600.26-70Kasba0.960.77-530.190.61-71Lesser Slave0.820.92-12-50.720.75-11-3Red Lake0.860.91-420.770.72-10-9Peter Pond0.850.95-7-10.880.54-50Average0.860.84-7-20.690.41-61ResultsLIP comparisons with GLRIPD lake observations
The lake ice status derived from the LIP and GLRIPD records are plotted for
the selected large lake validation sites (Fig. 2), including Lake
Superior (a), Lake Oulujarvi (b), Lake Haukivesi (c), and Lake Paijanne (d),
along with the daily ascending AMSR-E/2 Tb retrievals. The LIP
results show generally strong agreement with the GLRIPD site observations of
lake ice conditions for the four lakes examined, with overall retrieval
accuracy of 95.4 %. The lake ice/water retrieval error at the beginning
of the record for Lake Haukivesi (Fig. 2c) may be caused by partial melting
of lake ice in January and February 2003 that resulted in low AMSR-E/2
Tb observations. While the AMSR-E/2 Tb observations
show dynamic daily fluctuations due to changing water and atmosphere
properties (Sect. 2.2), lake freeze-up and breakup events constitute the
dominant factors affecting seasonal Tb changes. The effects of
higher temporal frequency Tb variations are minimized in the LIP
algorithm by the predefined 20-day subsample sizes (Sect. 2.3), which
represent a compromise between the algorithm's capability in capturing
shorter-term lake ice formation or melt events, and potentially degraded lake
ice/water seasonal retrieval accuracy.
Comparisons between MODIS quick-look images (left column) and
AMSR-E/2 LIP results (right column) for Great Bear Lake (GBL) on 22 June (a),
27 June (b), 5 July (c), and 8 July 2013 (d). The images are in the EASE-GRID
version 2 polar projection format, consistent with the underlying AMSR-E/2
gridded Tb dataset used for the LIP classification.
LIP comparisons against MODIS imagery, IMS and CIS products
An example visual comparison between the LIP results and MODIS quick-look
imagery (Fig. 3) shows the 2013 spring ice breakup process over the GBL. In
this example, both datasets show a general onset of lake ice breakup on
22 June (Fig. 3a), similar spatial ice distribution patterns on 27 June
(Fig. 3b), and ice-free conditions on 8 July (Fig. 3d). Despite extensive
cloud presence in the MODIS image for 5 July, both MODIS and LIP show
remaining ice cover on the western edge of GBL (Fig. 3c). A few remaining
pixels along the GBL coast line were identified as ice covered in the LIP
results for 8 July (Fig. 3d); the apparent LIP retrieval error is attributed
to land contamination, while the affected pixels are within the 5 km land
buffer zone (Sect. 2.4) and excluded in the final LIP product.
The LIP products were compared against similar lake ice phenology metrics
from the IMS and CIS datasets for the 12 North American study lakes. For GBL
and GSL, the LIP products agreed well with the CIS records; temporal
correlations (R value) of 0.90 and 0.85 were observed for the WCI dates for
GBL (Fig. 4a) and GSL (Fig. 4b), respectively, while correlations of 0.90 and
0.72 were determined for GBL (Fig. 4c) and GSL (Fig. 4d) CFO dates. The LIP
results were also strongly correlated with the IMS record on derived WCI
dates for both lakes (R=0.94 for GBL and R=0.91 for GSL) (Fig. 4a,
b). However, lower correspondence was found between LIP and IMS CFO dates,
with respective correlations of 0.54 and 0.63 for GBL and GSL (Fig. 4c, d).
These results indicate that the LIP-derived lake ice phenology variables are
generally consistent with the IMS and CIS records for the two lakes examined,
with generally higher (lower) correspondence for WCI (CFO) dates. For GBL,
the LIP estimated CFO dates occur similar to the CIS records (0-day
difference) and later than the IMS records by about 3 days; the LIP WCI dates
occur earlier than the CIS and IMS records by about 2 and 3 days,
respectively. For GSL, the LIP record also shows earlier (later) CFO than the
CIS (IMS) records by about 2 (3) days and earlier WCI dates than both CIS
and IMS by about 6 and 5 days, respectively. The intercomparisons between
CIS and IMS show average 5- and 0-day differences in respective CFO and WCI
dates for their overlapping period from 2004 to 2015 for GBL; corresponding
differences for GSL are 5 and 1 day, respectively. The differences between the LIP and CIS/IMS metrics are of similar magnitude as the differences
between the CIS and IMS metrics.
Comparisons of water clear of ice (WCI) dates for (a) Great Bear
Lake and (b) Great Slave Lake and complete freeze over (CFO) dates for
(c) Great Bear Lake and (d) Great Slave Lake, derived from the AMSR-E/2 Lake Ice
Phenology (LIP) dataset developed in this study. The LIP results are
compared against similar metrics derived from the NOAA/IMS (IMS) and
Canadian Ice Service (CIS). Missing LIP data from 2011 to 2012 denote the
period between the end of AMSR-E operations and the start of the AMSR2
record.
The LIP comparison results for all 12 study lakes are summarized in Table 2.
Similar to the comparisons for GBL and GSL, the LIP results are strongly
correlated with both CIS and IMS records for WCI dates, with respective
average temporal correlations (R value) of 0.86 and 0.84. For CFO dates,
the average correlation between LIP and CIS results are also strong (R=0.69) while only moderate correlation (R=0.41) was found between the
LIP and IMS results. The LIP estimated CFO dates tend to occur earlier than
the CIS record by about 6 days and later than the IMS record by about 1 day.
The LIP record also shows earlier WCI dates than the CIS and IMS records by
about 7 and 2 days, respectively.
Analysis of LIP lake ice phenology changes
The magnitude and direction of LIP trends were calculated for the 71 Northern
Hemisphere study lakes for the 12-year AMSR-E/2 record. Among all 71 lakes,
43 (60.6 %) show declining trends in ICDe, indicating an increasingly
shorter ice cover season, while the other lakes show either increasing or
minimal change in annual ice cover (Fig. 5). However, no observed ice trends
are statistically significant (p≥0.05). The lack of significant trends
is attributed to large yearly variability (±13.3 day) in average ICDe and
a relatively short (12-year) LIP observation record. The changing trends also
demonstrate a latitudinal pattern, as 81.0% of the lakes (17 out of 21) at
higher latitudes (> 60∘ N) show declining ICDe trends while only
45.0 % (9 out of 20) of lower-latitude (< 50∘ N) lakes show a
similar trend.
The observed changes in ICDe are the net result of changes in fall CFO and
spring WCI dates. A tendency toward earlier WCI dates was found for 40 lakes,
including 5 lakes (Lake Vygozero, Lake Barun-Torey, Lake Segozero,
Novosibirsk Reservoir in Russia and Lake Teshekpuk in USA) with significant
LIP trends. However, no lakes showed significant trends toward later spring
breakup. Similar to the ICDe analysis, most high-latitude lakes (81.0 %
of lakes above 60∘ N) show earlier spring thaw trends, while only
45.0 % of lower-latitude (< 50∘ N) lakes show similar trends.
A tendency toward delayed CFO was found for 35 of the 71 lakes (49.3 %)
examined, but no trends are statistically significant (p≥0.05). Lake
Bosten in China was the only lake with a significant trend toward earlier
freeze-up. There was no clear relationship between changes in lake CFO dates
with latitude.
Similar analysis of quarter-degree ERA-Interim SAT over the study lakes
indicates a much stronger warming trend in spring
(0.073 ∘C year-1) than fall (0.023 ∘C year-1).
Moreover, similar to the latitudinal pattern shown in the LIP-based analysis,
the SAT increase in the spring is positively correlated with latitude (R=0.33; p=0.005) indicating greater warming during the study period at
higher latitudes, while no SAT correlation with latitude is found for the
fall (R∼0.0).
Changing trends of (a) ice cover duration (ICDe), (b) water clear
of ice (WCI) dates, and (c) complete freeze over (CFO) dates of 71 lakes for
the period 2002–2015. Lake changing trends are shown by bar symbols whose
heights are proportional to the trend magnitudes; the significant trend
lakes are marked by yellow stars, while purple triangles denote lakes where
no trend was detected (rate of change is ∼ 0.0 day year-1).
Discussion
We developed a new satellite approach for regionally consistent
classification of ice phenology for large lakes in the Northern Hemisphere
from the AMSR-E/2 sensor record. We used similar 36.5 GHz H-polarization
daily brightness temperature retrievals from AMSR-E and AMSR2 sensor records
with 5 km posted spatial resolution. The resulting LIP record documents the
timing and duration of seasonal lake phenology events of 5 km lake pixels
(water coverage ≥ 90 %) over the 12-year AMSR-E/2 record. The LIP
results showed strong agreement with GLRIPD site observations from four
lakes, with agreement ranging from 92.4 to 98.7 %. Differences between
the LIP and GLRIPD results can be attributed to several factors. First, each
database has a different definition of lake ice conditions; lake ice coverage
determined by satellite microwave sensors is dependent on ice thickness,
which may vary from the ice detection approach used by on-site observers or
observed from optical sensors. According to the literature (Hall et al.,
1981), lake ice thickness and Tb are linearly related for
multiple frequencies (from 5 to 37 GHz). The reported maximum microwave
penetration depths of fresh lake ice at 37 GHz range from 0.70 to 1.4 m,
depending on ice temperatures (Chang et al., 1997; Surdyk, 2002; Kang et al.,
2014). This implies that the formation of thin ice, resulting in relatively
small Tb increases, may not be detectable using the defined
Tb thresholds in the LIP algorithm. Differences between the LIP
and GLRIPD results may also reflect spatial inconsistencies in lake
observation area between the ground-based lakeshore observations and the
coarser satellite footprint. Thus, the lake area observed on-site may not
completely overlap with the AMSR-E/2 lake pixel used for the LIP
classification. The GLRIPD also does not provide explicit descriptions of
lake ice status for the period between the first date when the water body was
completely ice covered and the date when the last ice breakup occurred;
thus, short-term events such as temporary ice melting or formation may not be
recorded in the GLRIPD. For example, though identified as ice covered in the
GLRIPD, Lake Oulujarvi was more likely to have thawed on 9 January 2007 since
the low Tb (178.7 K) observation is more characteristic of open
water emissions (Fig. 2b). The satellite Tb observations at
36.5 GHz are also affected by other factors than surface freeze-up/breakup
transitions, including changes in atmosphere water vapor and cloud liquid
water (e.g., Sect. 2.2). For example, the LIP detected ice-on conditions for
Lake Haukivesi Finland in mid-summer (30 July 2004) (Fig. 2c), which is
likely incorrect and may be due to increased atmosphere water vapor
concentrations under warm summer conditions, resulting in a large
Tb increase similar to a seasonal freeze-up event.
In the lake-based comparisons for the 12 lakes examined, including GSL and
GBL, the LIP results show strong correspondence with the CIS product for both
CFO and WCI dates and similar high correlations with the IMS results for WCI
dates; however, the LIP WCI (CFO) dates differ by approximately 7 (6) days
from the CIS and 2 (1) days from the IMS. These differences are attributed to
the different sensor spatial/temporal resolutions and retrieval methods
associated with the different products. As described in Sect. 2.1.3, the CIS
product is derived for individual lakes from visual interpretation of imagery
from optical and SAR sensors and has a ±1-week accuracy due to the
weekly product reporting. Both CIS and IMS products rely partially on
observations from optical sensors such as AVHRR and their accuracy is
influenced by adverse weather conditions, including the presence of cloud
cover. IMS-derived lake ice products have been widely used in monitoring
global climate change (Duguay et al., 2013, 2014, 2015b); however, the IMS
detected freeze onset was found to be too early for some lakes in northern
Québec, presumably due to misclassification by inclusion of coarse-resolution
satellite passive microwave observations during periods of prolonged cloud
cover (Brown and Duguay, 2012; Brown et al., 2014); this may cause the low
correlations between LIP and IMS CFO dates, as well as a delayed LIP CFO bias
relative to IMS.
In addition, the relatively coarse spatial resolution of AMSR-E/2
observations limits capabilities for resolving lake ice conditions of finer-scale water bodies. Spaceborne and airborne optical–TIR and radar sensors
are capable of improved delineation of smaller lakes and rivers (Chu et al.,
2016), but at the expense of degraded temporal fidelity for regional and
global applications.
As a proxy indicator of climate variability and change (Duguay et al., 2006),
lake ice phenology variables and their changing trends are important for
monitoring and understanding climate change and its feedbacks. As described
above, only 5 of the 71 lakes examined showed statistically significant
trends towards earlier WCI dates while no lake showed a significant later CFO
trend. Earlier ice breakup events are signs of warmer spring conditions,
which promote melting and breakup of lake ice and a lower surface albedo that
absorbs more incoming solar radiation and further intensifies the rate of ice
melt (Mishra et al., 2011). These results are also consistent with previous
studies over Canada that found a general trend toward earlier springs and WCI
dates particularly over western Canada but little change in isotherm and CFO
dates in fall (Duguay et al., 2006). Our results also indicate that lakes
at higher latitudes are more likely to experience trends toward earlier
spring ice breakup and shorter ICDe, which is consistent with enhanced
warming trends at higher latitudes (Solomon et al., 2007; Deutsch et al.,
2008). The above ice phenology trends coincide with regional SAT trends from
ERA-Interim that show an average spring warming rate that is more than triple
that of fall, as well as stronger warming trends for higher latitudes.
Though the LIP lake ice phenology trends are generally consistent with
regional climate warming (Magnuson et al., 2000; Solomon et al., 2007),
further analysis based on a longer period of record is needed for
distinguishing long-term climate trends from large interannual variability
and periodic climate cycles, including the North Atlantic Oscillation (NAO),
El-Niño Southern Oscillation (ENSO), and Pacific Decadal Oscillation
(PDO) (Mishra et al., 2011).
Conclusions
Lake ice phenology is strongly influenced by variations in air temperature,
while consistent long-term records of lake ice changes provide a sensitive
climate change indicator (Magnuson et al., 2000; Weyhenmeyer et al., 2011).
Continuous and accurate monitoring of lake ice dynamics is greatly needed for
studies of global change and for monitoring lake ice impacts on ecosystems
and infrastructure, especially for high-latitude and high-altitude regions.
In this study, we developed a new automated algorithm for consistent daily
retrieval of lake ice conditions over the Northern Hemisphere using similar
36.5 GHz H-polarized Tb observations from AMSR-E and AMSR2
sensor records. The resulting 5 km resolution lake ice phenology record
allows for daily monitoring of lake ice conditions without being
significantly degraded by variations in solar illumination or cloud and
atmosphere contamination effects. In particular, the LIP record distinguished
71 large lakes that satisfied 20-day minimum ICDe and open water season
algorithm thresholds; these lakes represent approximately 23 % of the
total surface area of large lakes (area ≥ 50 km2) within the
Northern Hemisphere domain. Smaller water bodies were excluded from the
lake-wide analysis if the lakes had no pixels with complete (100 %) open
water coverage outside of a 5 km land buffer zone. The relatively coarse
spatial resolution of AMSR-E/2 observations limits capabilities for resolving
finer-scale water bodies, while the conservative lake selection criterion
minimizes potential land contamination effects. The LIP-derived lake ice
conditions were found to be largely consistent with GLRIPD ground-based
observations, with an average agreement of 95.4 % for the four lakes
examined. The LIP record also showed favorable correspondence with other lake
CFO and WCI assessments defined from the CIS and IMS products for 12 large
study lakes. The LIP, CIS, and IMS differences were attributed to the
different data sources and methods used to construct the different products,
including differences in spatial and temporal resolutions of observations,
and distinct nature of optical and microwave remote sensing. Though the
design of the LIP algorithm, including the MTT method, helps to identify lake
breakup/freeze-up events, while minimizing other Tb disturbances
from short-term weather events, atmosphere effects can still lead to
retrieval errors, especially from persistent high atmosphere water vapor
concentrations over high-latitude lakes in the summer. Based on the LIP
record from 2002 to 2015, significant earlier melting of lake ice cover was
detected for 5 of the 71 lakes examined in the Northern Hemisphere, while
lakes at higher latitudes show a more evident warming trend toward earlier
ice breakup and shorter ICDe than those at lower latitudes. As the operations
from AMSR2 and similar sensors continue, the MTT algorithm will allow for
automated retrieval and consistent monitoring of ice conditions for large
Northern Hemisphere lakes into the future.
Data availability
The AMSR-E/2 derived Northern Hemisphere LIP record described in this study is publicly available through the following link
http://files.ntsg.umt.edu/data/AMSRE2_LAKE_ICE_PHEN (Du et al., 2017).
List of Abbreviations and Acronyms
AMSR2Advanced Microwave Scanning Radiometer 2AMSR-EAdvanced Microwave Scanning Radiometer for EOSAMSR-E/2Advanced Microwave Scanning Radiometer for EOS and Advanced Microwave Scanning Radiometer 2AMSUAdvanced Microwave Sounding UnitAVHRRAdvanced Very High Resolution RadiometerCFOcomplete freeze overCISCanadian Ice ServiceERA-Interima global atmospheric model data reanalysis produced by the European Centre for Medium-RangeWeather ForecastsGBLGreat Bear LakeGLRIPDGlobal Lake and River Ice Phenology DatabaseGLWDGlobal Lakes and Wetlands DatabaseGOESGeostationary Operational Environmental SatelliteGSLGreat Slave LakeIMSInteractive Multisensor Snow and Ice Mapping SystemLSTland surface temperatureMODISModerate Resolution Imaging SpectroradiometerMOD44WMODIS 250 m land–water maskMTTmoving t test methodPFCsperfluorinated chemicalsRcorrelation coefficientRFIradio frequency interferenceSATsurface air temperatureSMMRScanning Multichannel Microwave RadiometerSSM/ISpecial Sensor Microwave ImagerTbbrightness temperatureTIRthermal infraredVIIRSVisible/Infrared Imaging Radiometer SuiteWCIwater clear of ice
Acknowledgements
AMSR-E data are produced by Remote Sensing Systems and sponsored by the NASA
Earth Science MEaSUREs DISCOVER Project and the AMSR-E Science Team. Data are
available at http://www.remss.com. AMSR-E data and land cover
classification maps were also provided courtesy of the National Snow and Ice
Data Center (NSIDC). The AMSR2 L1R Tb data used for this study were
provided courtesy of JAXA. The Global Lakes and Wetlands Database is provided
by the World Wildlife organization and created by the Center for
Environmental Systems Research, University of Kassel, Germany. This work was
conducted at the University of Montana with funding from NASA
(NNX15AT74A). Edited by: R.
Brown Reviewed by: two anonymous referees
References
Ashcroft, P. and Wentz, F.: Algorithm Theoretical Basis Document, AMSR level
2A algorithm, Tech. Rep, Remote Sens. Syst, Santa Rosa, CA, 27 pp., 2000.
Bengtsson, L. and Herschy, R. W.: Encyclopedia of lakes and reservoirs,
Monog. Biol., 53, 10–26, 2012.Benson, B. J. and Magnuson, J. J.: Global Lake and River Ice Phenology
Database, Version 1, NSIDC: National Snow and Ice Data Center, Boulder,
Colorado USA, 10.7265/N5W66HP8, 2000 (updated 2012).Benson, B. J., Magnuson, J. J., Jensen, O. P., Card, V. M., Hodgkins, G.,
Korhonen, J., Livingstone, D. M., Stewart, K. M., Weyhenmeyer, G. A., and
Granin, N. G.: Extreme events, trends, and variability in Northern Hemisphere
lake-ice phenology (1855–2005), Climatic Change, 112, 299–323,
10.1007/s10584-011-0212-8, 2012.Brodzik, M. J., Billingsley, B., Haran, T., Raup, B., and Savoie, M. H.:
EASE-Grid 2.0: Incremental but Significant Improvements for Earth-Gridded
Data Sets, ISPRS Int. J. Geo-Inform., 1, 32–45, 10.3390/ijgi1010032,
2012.Brodzik, M. J., Billingsley, B., Haran, T., Raup, B., and Savoie, M. H.:
Correction: Brodzik, M. J. et al.: EASE-Grid 2.0: Incremental but Significant
Improvements for Earth-Gridded Data Sets, ISPRS Int. J. Geo-Inform. 3,
1154–1156, 10.3390/ijgi3031154, 2014.Brown, L. C. and Duguay, C. R.: Modelling lake ice phenology with an
examination of satellite-detected subgrid cell variability, Adv. Meteorol.,
2012, 1–19, 10.1155/2012/529064, 2012.
Brown, L. C., Howell, S. E., Mortin, J., and Derksen, C.: Evaluation of the
Interactive Multisensor Snow and Ice Mapping System (IMS) for monitoring sea
ice phenology, Remote Sens. Environ., 147, 65–78, 2014.
Carroll, M., Townshend, J., DiMiceli, C., Noojipady, P., and Sohlberg, R.: A
New Global Raster Water Mask at 250 Meter Resolution, Int. J. Digital Earth,
2, 291–308, 2009.
Chang, A. T. C., Foster, J. L., Hall, D. K., Goodison, B. E., Walker, A. E.,
Metcalfe, J. R., and Harby, A.: Snow parameters derived from microwave
measurements during the BOREAS winter field campaign, J. Geophys.
Res.-Atmos., 102, 29663–29671, 1997.
Che, T., Li, X., and Jin, R.: Monitoring the frozen duration of Qinghai Lake
using satellite passive microwave remote sensing low frequency data, Chinese
Sci. Bull., 54, 2294–2299, 2009.
Chu, T. and Lindenschmidt, K. E.: Integration of space-borne and air-borne
data in monitoring river ice processes in the Slave River, Canada, Remote
Sens. Environ., 181, 65–81, 2016.
Dee, D. P., Uppala, S. M., Simmons, A. J., Berrisford, P., Poli, P.,
Kobayashi, S., Andrae, U., Balmasewda, M. A., Balsamo, G., Bauer, P.,
Bechtold, P., Beljaars, A. C. M., van de Berg, L., Bidlot, J., Bormann, N.,
Delsol, C., Dragani, R., Fuentes, M., Geer, A. J., Haimberger, L., Healy, S.
B., Hersbach, H., Holm, E. V., Isaksen, L., Kallberg, P., Kohler, M.,
Matricardi, M., McNally, A. P., Monge-Sanz ,B. M., Morcrette, J. J., Park, B.
K., Peubey, C., de Rosnay, P., Tavolato, C., Thepaut, J. N., and Vitart, F.:
The ERA-Interim reanalysis: configuration and performance of the data
assimilation system, Q. J. Roy. Meteor. Soc., 137, 553–597, 2011.
Deutsch, C. A., Tewksbury, J. J., Huey, R. B., Sheldon, K. S., Ghalambor, C.
K., Haak, D. C., and Martin, P. R.: Impacts of climate warming on terrestrial
ectotherms across latitude, P. Natl. Acad. Sci. USA, 105, 6668–6672, 2008.
Du, J., Shi, J., and Rott, H.: Comparison between a multi-scattering and
multi-layer snow scattering model and its parameterized snow backscattering
model, Remote Sens. Environ., 114, 1089–1098, 2010.
Du, J., Kimball, J. S., Shi, J., Jones, L. A., Wu, S., Sun, R., and Yang, H.:
Inter-calibration of satellite passive microwave land observations from
AMSR-E and AMSR2 using overlapping FY3B-MWRI sensor measurements, Remote
Sens., 6, 8594–8616, 2014.
Du, J., Kimball, J. S., and Jones, L. A.: Satellite microwave retrieval of
total precipitable water vapor and surface air temperature over land from
AMSR2, IEEE T. Geosci. Remote Sens., 53, 2520–2531, 2015.
Du, J., Kimball, J. S., Jones, L. A., and Watts, J. D.: Implementation of
satellite based fractional water cover indices in the pan-Arctic region using
AMSR-E and MODIS, Remote Sens. Environ., 184, 469–481, 2016.Du, J., Kimball, J. S., and Duguay, C.: Daily Lake Ice Phenology Data Record
Time Series Derived from AMSR-E and AMSR2, Version 1 (2002–2015), Missoula,
Montana USA, Numerical Terradynamic Simulation Group, University of Montana,
available at: http://files.ntsg.umt.edu/data/AMSRE2_LAKE_ICE_PHEN, last
access: 10 January 2017.
Duguay, C. R., Prowse, T. D., Bonsal, B. R., Brown, R. D., Lacroix, M. P.,
and Ménard, P.: Recent trends in Canadian lake ice cover, Hydrol.
Process., 20, 781–801, 2006.
Duguay, C. R., Brown, L., Kang, K. K., and Kheyrollah Pour, H.: Lake ice, in:
State of the Climate in 2011, B. Am. Meteorol. Soc., 93, S152–S154, 2012.
Duguay, C. R., Brown, L., Kang, K. K., and Kheyrollah Pour, H.: Lake ice, in:
State of the Climate in 2012, B. Am. Meteorol. Soc., 94, S124–S126, 2013.
Duguay, C. R., Brown, L., Kang, K. K., and Kheyrollah Pour, H.: Lake ice, in:
State of the Climate in 2013, B. Am. Meteorol. Soc., 95, S138–S139, 2014.
Duguay, C. R., Bernier, M., Gauthier, Y., and Kouraev, A.: Remote sensing of
lake and river ice, in: Remote Sensing of the Cryosphere, edited by: Tedesco,
M., Wiley-Blackwell (Oxford, UK), 273–306, 2015a.
Duguay, C. R., Brown, L., Kang, K. K., and Kheyrollah Pour, H.: Lake ice, in:
State of the Climate in 2014, B. Am. Meteorol. Soc., 96, S144–S145, 2015b.
Ferraro, R. R., Grody, N. C., and Kogut, J. A.: Classification of geophysical
parameters using passive microwave satellite measurements, IEEE T. Geosci.
Remote Sens., 24, 1008–1013, 1986.
Geldsetzer, T., Sanden, J. V. D., and Brisco, B.: Monitoring lake ice during
spring melt using RADARSAT-2 SAR, Can. J. Remote Sens., 36, S391–S400, 2010.Greene, S., Walter Anthony, K. M., Archer, D., Sepulveda-Jauregui, A., and
Martinez-Cruz, K.: Modeling the impediment of methane ebullition bubbles by
seasonal lake ice, Biogeosciences, 11, 6791–6811,
10.5194/bg-11-6791-2014, 2014.
Hall, D. K., Foster, J. L., Chang, A. T., and Rango, A., Freshwater ice
thickness observations using passive microwave sensors, IEEE T. Geosci.
Remote Sens., 4, 189–193, 1981.
Hall, D. K., Riggs, G. A., Foster, J. L., and Kumar, S. V.: Development and
evaluation of a cloud-gap-filled MODIS daily snow-cover product, Remote Sens.
Environ., 114, 496–503, 2010.
Helfrich, S. R., McNamara, D., Ramsay, B. H., Baldwin, T., and Kasheta, T.:
Enhancements to, and forthcoming developments in the Interactive Multisensor
Snow and Ice Mapping System (IMS), Hydrol. Process., 21, 1576–1586, 2007.Howell, S. E., Brown, L. C., Kang, K. K., and Duguay, C. R.: Variability in
ice phenology on Great Bear Lake and Great Slave Lake, Northwest Territories,
Canada, from SeaWinds/QuikSCAT: 2000–2006, Remote Sens. Environ., 113,
816–834, 10.1016/j.rse.2008.12.007, 2009.
Imaoka, K., Kachi, M., Kasahara, M., Ito, N., Nakagawa, K., and Oki, T.:
Instrument performance and calibration of AMSR-E and AMSR2, ISPRS Archives,
38, 13–18, 2010.
Jeffries, M. O., Morris, K., and Kozlenko, N.: Ice characteristics and
processes, and remote sensing of frozen rivers and lakes in remote sensing,
in: Northern hydrology: Measuring environmental change, edited by: Duguay, C.
R. and Pietroniro, A., Geophysical Monograph 163, American Geophysical Union,
63–90, 2005.
Jiang, J. M. and You, X. T.: Where and when did an abrupt climatic change
occur in China during the last 43 years?, Theor. Appl. Climatol., 55, 33–40,
1996.Kang, K.-K., Duguay, C. R., and Howell, S. E. L.: Estimating ice phenology on
large northern lakes from AMSR-E: algorithm development and application to
Great Bear Lake and Great Slave Lake, Canada, The Cryosphere, 6, 235–254,
10.5194/tc-6-235-2012, 2012.
Kang, K. K., Duguay, C. R., Lemmetyinen, J., and Gel, Y.: Estimation of ice
thickness on large northern lakes from AMSR-E brightness temperature
measurements, Remote Sens. Environ., 150, 1–19, 2014.
Karetnikov, S. G. and Naumenko, M. A.: Recent trends in Lake Ladoga ice
cover, Hydrobiologia, 599, 41–48, 2008.
Kawanishi, T. J., Sezai, T., Ito, Y., Imaoka, K., Takeshima, T., Ishido, Y.,
Shibata, A., Miura, M., Inahata, H., and Spencer, R. W.: The Advanced
Scanning Microwave Radiometer for the EarthObserving System (AMSR-E): NASDA's
contribution to the EOS for global energy andwater cycle studies, IEEE T.
Geosci. Remote Sens., 41, 184–194, 2003.Ke, C. Q., Tao, A. Q., and Jin, X.: Variability in the ice phenology of Nam
Co Lake in central Tibet from scanning multichannel microwave radiometer and
special sensor microwave/imager: 1978 to 2013, J. Appl. Remote Sens., 7,
073477, 10.1117/1.JRS.7.073477, 2013.Kheyrollah Pour, K., Duguay, C. R., Martynov, A., and Brown, L. C.:
Simulation of surface temperature and ice cover of large northern lakes with
1-D models: A comparison with MODIS satellite data and in situ measurements,
Tellus A, 64, 17614, 10.3402/tellusa.v64i0.17614, 2012.
Kim, Y., Kimball, J. S., Zhang, K., and McDonald, K. C.: Satellite detection
of increasing Northern Hemisphere non-frozen seasons from 1979 to 2008:
Implications for regional vegetation growth, Remote Sens. Environ., 121,
472–487, 2012.
Kouraev, A. V., Semovski, S. V., Shimaraev, M. N., Mognard, N. M.,
Légresy, B., and Remy, F.: Observations of Lake Baikal ice from satellite
altimetry and radiometry, Remote Sens. Environ., 108, 240–53, 2007a.
Kouraev, A. V., Semovski, S. V., Shimaraev, M. N., Mognard, N. M.,
Légresy, B., and Remy, F.: The ice regime of Lake Baikal from historical
and satellite data: Relationship to air temperature, dynamical, and other
factors, Limnol. Oceanogr., 52, 1268–1286, 2007b.
Latifovic, R. and Pouliot, D.: Analysis of climate change impacts on lake ice
phenology in Canada using the historical satellite data record, Remote Sens.
Environ., 106, 492–507, 2007.
Leconte, R. and Klassen, P. D.: Lake and river ice investigations in northern
Manitoba using airborne SAR imagery, Arctic, 44, 153–163, 1991.
Lehner, B. and Döll, P.: Development and validation of a global database
of lakes, reservoirs and wetlands, J. Hydrol., 296, 1–22, 2004.
Lemmetyinen, J., Pulliainen, J., Rees, A., Kontu, A., Qiu, Y., and Derksen,
C.: Multiple-layer adaptation of HUT snow emission model: Comparison with
experimental data, IEEE T. Geosci. Remote Sens., 48, 2781–2794, 2010.
Lemmetyinen, J., Kontu, A., Kärnä, J. P., Vehviläinen, J.,
Takala, M., and Pulliainen, J.: Correcting for the influence of frozen lakes
in satellite microwave radiometer observations through application of a
microwave emission model, Remote Sens. Environ., 115, 3695–3706, 2011.
Livingstone, D. M.: Break-up dates of Alpine lakes as proxy data for local
and regional mean surface air temperatures, Climatic Change, 37, 407–439,
1997.
Maeda, T., Taniguchi, Y., and Imaoka, K.: GCOM-W1 AMSR2 Level 1R Product:
Dataset of Brightness Temperature Modified Using the Antenna Pattern Matching
Technique, IEEE T. Geosci. Remote Sens., 54, 770–782, 2016.
Magnuson, J. J., Robertson, D. M., Benson, B. J., Wynne, R. H., Livingstone,
D. M., Arai, T., Assel, R. A., Barry, R. G., Card, V., Kuusisto, E., and
Granin, N. G.: Historical trends in lake and river ice cover in the Northern
Hemisphere, Science, 289, 1743–1746, 2000.
Maslanik, J. A. and Barry, R. G.: Lake ice formation and breakup as an
indicator of climate change: Potential for monitoring using remote sensing
techniques, The Influence of Climate Change and Climatic Variability on the
Hydrologic Regime and Water Resources, International Association of
Hydrological Sciences Press, IAHS Publ. No. 168, 153–161, 1987.
Mätzler, C.: Passive microwave signatures of landscapes in winter,
Meteorol. Atmos. Phys., 54, 241–260, 1994.
Ménard, P., Duguay, C. R., Flato, G. M., and Rouse, W. R.: Simulation of
ice phenology on Great Slave Lake, Northwest Territories, Canada, Hydrol.
Process., 16, 3691–3706, 2002.
Mishra, V., Cherkauer, K. A., Bowling, L. C., and Huber, M.: Lake Ice
phenology of small lakes: Impacts of climate variability in the Great Lakes
region, Glob. Planet. Change, 76, 166–185, 2011.
Mountain Research Initiative EDW Working Group: Elevation-dependent warming
in mountain regions of the world, Nat. Clim. Change, 5, 424–430, 2015.
Noguchi, K., Gel, Y. R., and Duguay, C. R.: Bootstrap-based tests for trends
in hydrological time series, with application to ice phenology data, J.
Hydrol., 410, 150–161, 2011.Nolan, M., Liston, G., Prokein, P., Brigham-Grette, J., Sharpton, V. L., and
Huntzinger, R.: Analysis of lake ice dynamics and morphology on Lake
El'gygytgyn, NE Siberia, using synthetic aperture radar (SAR) and Landsat, J.
Geophys. Res.-Atmos., 107, 000934, 10.1029/2001JD000934, 2002.Nonaka, T., Matsunaga, T., and Hoyano, A.: Estimating ice breakup dates on
Eurasian lakes using water temperature trends and threshold surface
temperatures derived from MODIS data, Int. J. Remote Sens., 28, 2163–2179,
10.1080/01431160500391957, 2007.
Park, H., Yoshikawa, Y., Oshima, K., Kim, Y., Ngo-Duc, T., Kimball, J. S.,
and Yang, D.: Quantification of Warming Climate-Induced Changes in
Terrestrial Arctic River Ice Thickness and Phenology, J. Climate, 29,
1733–1754, 2016.
Prowse, T., Alfredsen, K., Beltaos, S., Bonsal, B., Duguay, C., Korhola, A.,
McNamara, J., Pienitz, R., Vincent, W. F., Vuglinsky, V., and Weyhenmeyer, G.
A.: Past and future changes in Arctic lake and river ice, Ambio, 40, 53–62,
2011.
Schröter, D., Cramer, W., Leemans, R., Prentice, I. C., Araújo, M.
B., Arnell, N. W., Bondeau, A., Bugmann, H., Carter, T. R., Gracia, C. A.,
and Anne, C.: Ecosystem service supply and vulnerability to global change in
Europe, Science, 310, 1333–1337, 2005.
Sen, P. K.: Estimates of the Regression Coefficient Based on Kendall's Tau,
J. Am. Stat. Assoc., 63, 1379–1389, 1968.
Serreze, M. C. and Francis, J. A.: The Arctic amplification debate, Climatic
Change, 76, 241–264, 2006.
Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K. B.,
Tignor, M., and Mill, H. L. (Eds.): Climate Change 2007: The Physical Science
Basis, Cambridge University Press, Cambridge, United Kingdom and New York,
NY, USA, 2007.Stocker, T. F., Qin, D., Plattner, G.-K., Tignor, M., Allen, S. K., Boschung,
J., Nauels, A., Xia, Y., Bex, V., and Midgley, P. M.: IPCC, 2013: Climate
Change 2013: The Physical Science Basis. Contribution of Working Group I to
the Fifth Assessment Report of the Intergovernmental Panel on Climate Change,
Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA,
2013.
Surdyk, S.: Using microwave brightness temperature to detect short-term
surface air temperature changes in Antarctica: An analytical approach, Remote
Sens. Environ., 80, 256–271, 2002.
Ulaby, F. T., Moore, R. K., and Fung, A. K.: Microwave Remote Sensing: Active
and Passive, Volume III: from Theory to Applications, Artech House, Norwood,
MA, USA, 1986.
Veillette, J., Muir, D. C., Antoniades, D., Small, J. M., Spencer, C.,
Loewen, T. N., Babaluk, J. A., Reist, J. D., and Vincent, W. F.:
Perfluorinated chemicals in meromictic lakes on the northern coast of
Ellesmere Island, High Arctic Canada, Arctic, 65, 245–256, 2012.
Vuglinsky, V. S. and Gronskaya, T. P.: Changing of rivers and lakes ice
regime within the Russian territory and their possible consequences for
economy, Modern problems of hydrometeorology, 229245, 229–245, 2006.
Walker, A. E. and Davey, M. R.: Observation of Great Slave Lake ice freeze-up
and break-up processes using passive microwave satellite data, in:
Proceedings of the 16th Canadian Symposium on Remote Sensing, June 1993,
Sherbrooke, QC, Canada, 233–238, 1993.
Wang, G., Bai, W., Li, N., and Hu, H.: Climate changes and its impact on
tundra ecosystem in Qinghai-Tibet Plateau, China, Climatic Change, 106,
463–482, 2011.
Wang, J. R. and Tedesco, M.: Identification of atmospheric influences on the
estimation of snow water equivalent from AMSR-E measurements, Remote Sens.
Environ., 111, 398–408, 2007.
Weyhenmeyer, G. A., Livingstone, D. M., Meili, M., Jensen, O., Benson, B.,
and Magnuson, J. J.: Large geographical differences in the sensitivity of
ice-covered lakes and rivers in the Northern Hemisphere to temperature
changes, Glob. Chang. Biol., 17, 268–275, 2011.
Woo, M. K., Modeste, P., Martz, L., Blondin, J., Kochtubajda, B., Tutcho, D.,
Gyakum, J., Takazo, A., Spence, C., Tutcho, J., Di Cenzo, P., Kenny, G.,
Stone, J., Neyelle, I., Baptiste, G., Modeste, M., Kenny, B., and Modeste,
W.: Science meets traditional knowledge: water and climate in the Sahtu
(Great Bear Lake) region, Northwest Territories, Canada, Arctic, 60, 37–46,
2007.
Wrona, F. J., Johansson, M., Culp, J. M., Jenkins, A., Mård, J.,
Myers-Smith, I. H., Prowse, T. D., Vincent, W. F., and Wookey, P. A.:
Transitions in Arctic ecosystems: Ecological implications of a changing
hydrological regime, J. Geophys. Res.-Biogeo., 121, 650–674, 2016.Xiao, D. and Li, J.: Spatial and temporal characteristics of the decadal
abrupt changes of global atmosphere-oceansystem in the 1970s, J. Geophys.
Res.-Atmos., 112, D24S22, 10.1029/2007JD008956, 2007.
Zhang, X., Vincent, L. A., Hogg, W. D., and Niitsoo, A.: Temperature and
precipitation trends in Canada during the 20th century, Atmos.-Ocean, 38,
395–429, 2000.