Search

The aim of the study was to evaluate, for the first time, the capability of different machine learning (ML) algorithms in classifying along-track lake surface conditions (open water and ice types) across ice seasons (freeze-up, ice growth and break-up periods) from Sentinel-3 A/B synthetic aperture radar altimeter (SRAL) data. To achieve this goal, over 107,500 radar waveforms extracted from 11 large lakes across the Northern Hemisphere and three ice seasons (2018–2021) were manually labelled using complementary satellite data (Sentinel-1 imaging Synthetic Aperture Radar (SAR), Sentinel-2 Multispectral Instrument (MSI) Level 1C, and MODIS Aqua/Terra data) for the training and testing of the ML algorithms in discriminating between open water, young (thin) ice, growing ice and melting ice. The four ML algorithms tested include Random Forest (RF), Gradient Boosting Trees (GBT), K Nearest Neighbor (KNN) and Support Vector Machine (SVM). To characterize the waveforms, seven waveform parameters were derived: Leading Edge Width (LEW), Offset Center of Gravity (OCOG) Width, Pulse Peakiness (PP), backscatter coefficient (Sigma0), late tail to peak power (LTPP), early tail to peak power (ETPP) and the maximum value of the echo power (Max). Accuracies >95% were achieved across all classifiers using a 4-parameter combination (Sigma0, PP, OCOG Width, and LEW). Among all waveform parameters, Sigma0, OCOG width and PP were found to be the most important parameters for discriminating between lake ice types and open water. Despite showing comparable classification performances in the overall classification, RF and KNN are found to be a better fit for global lake ice mapping as both are less sensitive to their internal hyperparameters. Additionally, consistent results (>93.7% accuracy in all classifiers) achieved on the accuracy assessment carried out for each lake (out-of-sample testing) revealed the strength of the classifiers for spatial transferability. Implementation of RF and KNN could be valuable in a pre-or post-processing step for identifying lake surface conditions under which the retrieval of water level and ice thickness may be limited or not possible and, therefore, inform algorithms currently used for the generation of operational or research products. While the research focused on 11 of the largest lakes of the Northern Hemisphere, the classification approach presented herein has potential for application on smaller lakes too since data in SAR mode (∼300 m along-track resolution) are used. • This is the first study using Sentinel-3 SAR altimetry for lake ice classification. • Four different machine learning classifiers (SVM, KNN, RF and GBT) were assessed. • RF and KNN are recommended for classification of ice types along altimeter tracks. • Sigma0, OCOG Width, PP are the best parameters to differentiate lake ice and water. • Over 93.7% accuracy in spatial cross-validation shows broad-scale applicability. [ABSTRACT FROM AUTHOR]

Late-winter lake ice regimes are controlled by water depth relative to maximum ice thickness (MIT). When MIT exceeds maximum water depth, lakes freeze to the bottom with bedfast ice (BI) and when MIT is less than maximum water depth lakes have floating ice (FI). Both airborne radar and space-borne synthetic aperture radar (SAR) imagery (Ku-, X-, C-, and L-band) have been used previously to determine whether lakes have a BI or FI regime in a given year, across a number of years, or across large regions. In this study, we use a combination of ERS-1/2, RADARSAT-2, Envisat, and Sentinel-1 SAR imagery for seven lake-rich regions in Arctic Alaska to analyze lake ice regime extents and dynamics over a 25-year period (1992–2016). Our interactive threshold classification method determines a unique statistic-based intensity threshold for each SAR scene, allowing for the comparison of classification results from C-band SAR data acquired with different polarizations and incidence angles. Additionally, our novel method accommodates declining signal strength in aging extended-mission satellite SAR instruments. Comparison of SAR ice regime classifications with extensive field measurements from six years yielded a 93% accuracy. Significant declines in BI regimes were only observed in the Fish Creek area with 3% of lakes exhibiting transitional ice regimes—lakes that switch from BI to FI during this 25-year period. This analysis suggests that the potential conversion from BI to FI regimes is primarily a function of lake depth distributions in addition to regional differences in climate variability. Remote sensing of lake ice regimes with C-band SAR is a useful tool to monitor the associated thermal impacts on permafrost, since lake ice regimes can be used as a proxy for of sub-lake permafrost thaw, considered by the Global Climate Observing System as an Essential Climate Variable (ECV). Continued winter warming and variable snow conditions in the Arctic are expected and our long-term analysis provides a valuable baseline for predicting where potential future lake ice regimes shifts will be most pronounced. [ABSTRACT FROM AUTHOR]

Lake ice provides important social and economic services to local communities, in addition to being a sensitive indicator of climate change. The reduction of ground observations of freshwater ice has led to an increased reliance on the use of satellite remote sensing data. There is currently interest in the retrieval of lake ice properties (e.g., ice thickness, bubble radius, roughness) using synthetic aperture radar (SAR). Roughness at the ice-water interface is particularly important as it has been identified as the dominant mechanism for increasing SAR backscatter throughout the ice season and must be considered in numerical radiative transfer models. Therefore, this study determines optimal ice-water interface roughness height for two subarctic lakes in northern Canada and models backscatter throughout two ice seasons using the snow microwave radiative transfer (SMRT) model. The two lakes for this study are Noell Lake and Malcolm Ramsay Lake. Field observations of ice thickness, snow depth, snow density, and the Canadian Lake Ice Model (CLIMo) are used to parameterize SMRT. Modelled L, C, and X-band backscatter at different incidence angles is assessed using SAR imagery from multiple satellite missions. Root mean square errors ranged from 0.38 to 1.45 dB for Noell Lake and 0.70 to 2.33 dB for Malcolm Ramsay Lake. Discrepancies between modelled and observed backscatter were found to be connected to the representation of roughness at different interfaces within the ice column and changes that occurred during freeze-melt events. These results provide insight into how changes in ice properties impact backscatter throughout the ice season. SMRT is valuable for modelling backscatter from lake ice during the cold season and could be used to develop retrieval algorithms for estimating ice-water interface roughness. This would allow for the development of other inversion models for retrieval of surface ice conditions and ice thickness. • First application of SMRT for modelling backscatter over a complete ice season. • Multi-frequency backscatter was successfully modelled for two Subarctic lakes. • Variable root mean square height was an improvement over a constant value. • Changes in the ice surface and snowpack result in discrepancies in modelling. • SMRT can be useful for future retrieval of crucial lake ice properties. [ABSTRACT FROM AUTHOR]

Several lake ice phenology studies from satellite data have been undertaken. However, the availability of long-term lake freeze-thaw-cycles, required to understand this proxy for climate variability and change, is scarce for European lakes. Long time series from space observations are limited to few satellite sensors. Data of the Advanced Very High Resolution Radiometer (AVHRR) are used in account of their unique potential as they offer each day global coverage from the early 1980s expectedly until 2022. An automatic two-step extraction was developed, which makes use of near-infrared reflectance values and thermal infrared derived lake surface water temperatures to extract lake ice phenology dates. In contrast to other studies utilizing thermal infrared, the thresholds are derived from the data itself, making it unnecessary to define arbitrary or lake specific thresholds. Two lakes in the Baltic region and a steppe lake on the Austrian–Hungarian border were selected. The later one was used to test the applicability of the approach to another climatic region for the time period 1990 to 2012. A comparison of the extracted event dates with in situ data provided good agreements of about 10 d mean absolute error. The two-step extraction was found to be applicable for European lakes in different climate regions and could fill existing data gaps in future applications. The extension of the time series to the full AVHRR record length (early 1980 until today) with adequate length for trend estimations would be of interest to assess climate variability and change. Furthermore, the two-step extraction itself is not sensor-specific and could be applied to other sensors with equivalent near- and thermal infrared spectral bands. [ABSTRACT FROM AUTHOR]

Abstract: The spatial resolution of passive microwave observations from space is of the order of tens of kilometers with currently available instruments, such as the Special Sensor Microwave/Imager (SSM/I) and Advanced Microwave Scanning Radiometer (AMSR-E). The large field of view of these instruments dictates that the observed brightness temperature can originate from heterogeneous land cover, with different vegetation and surface properties. In this study, we assess the influence of freshwater lakes on the observed brightness temperature of AMSR-E in winter conditions. The study focuses on the geographic region of Finland, where lakes account for 10% of the total terrestrial area. We present a method to mitigate for the influence of lakes through forward modeling of snow covered lakes, as a part of a microwave emission simulation scheme of space-borne observations. We apply a forward model to predict brightness temperatures of snow covered sceneries over several winter seasons, using available data on snow cover, vegetation and lake ice cover to set the forward model input parameters. Comparison of model estimates with space-borne observations shows that the modeling accuracy improves in the majority of examined cases when lakes are accounted for, with respect to the case where lakes are not included in the simulation. Moreover, we present a method for applying the correction to the retrieval of Snow Water Equivalent (SWE) in lake-rich areas, using a numerical inversion method of the forward model. In a comparison to available independent validation data on SWE, also the retrieval accuracy is seen to improve when applying the influence of snow covered lakes in the emission model. [Copyright &y& Elsevier]

Abstract: The algorithms designed to estimate snow water equivalent (SWE) using passive microwave measurements falter in lake-rich high-latitude environments due to the emission properties of ice covered lakes on low frequency measurements. Microwave emission models have been used to simulate brightness temperatures (Tbs) for snowpack characteristics in terrestrial environments but cannot be applied to snow on lakes because of the differing subsurface emissivities and scattering matrices present in ice. This paper examines the performance of a modified version of the Helsinki University of Technology (HUT) snow emission model that incorporates microwave emission from lake ice and sub-ice water. Inputs to the HUT model include measurements collected over brackish and freshwater lakes north of Inuvik, Northwest Territories, Canada in April 2008, consisting of snowpack (depth, density, and snow water equivalent) and lake ice (thickness and ice type). Coincident airborne radiometer measurements at a resolution of 80×100m were used as ground-truth to evaluate the simulations. The results indicate that subsurface media are simulated best when utilizing a modeled effective grain size and a 1mm RMS surface roughness at the ice/water interface compared to using measured grain size and a flat Fresnel reflective surface as input. Simulations at 37GHz (vertical polarization) produce the best results compared to airborne Tbs, with a Root Mean Square Error (RMSE) of 6.2K and 7.9K, as well as Mean Bias Errors (MBEs) of −8.4K and −8.8K for brackish and freshwater sites respectively. Freshwater simulations at 6.9 and 19GHz H exhibited low RMSE (10.53 and 6.15K respectively) and MBE (−5.37 and 8.36K respectively) but did not accurately simulate Tb variability (R=−0.15 and 0.01 respectively). Over brackish water, 6.9GHz simulations had poor agreement with airborne Tbs, while 19GHz V exhibited a low RMSE (6.15K), MBE (−4.52K) and improved relative agreement to airborne measurements (R=0.47). Salinity considerations reduced 6.9GHz errors substantially, with a drop in RMSE from 51.48K and 57.18K for H and V polarizations respectively, to 26.2K and 31.6K, although Tb variability was not well simulated. With best results at 37GHz, HUT simulations exhibit the potential to track Tb evolution, and therefore SWE through the winter season. [Copyright &y& Elsevier]

Impacts of global climate change are expected to result in greater variation in the seasonality of snowpack, lake ice, and vegetation dynamics in southwest Alaska. All have wide-reaching physical and biological ecosystem effects in the region. We used Moderate Resolution Imaging Spectroradiometer (MODIS) calibrated radiance, snow cover extent, and vegetation index products for interpreting interannual variation in the duration and extent of snowpack, lake ice, and vegetation dynamics for southwest Alaska. The approach integrates multiple seasonal metrics across large ecological regions. Throughout the observation period (2001¿2007), snow cover duration was stable within ecoregions, with variable start and end dates. The start of the lake ice season lagged the snow season by 2 to 3 months. Within a given lake, freeze-up dates varied in timing and duration, while break-up dates were more consistent. Vegetation phenology varied less than snow and ice metrics, with start-of-season dates comparatively consistent across years. The start of growing season and snow melt were related to one another as they are both temperature dependent. Higher than average temperatures during the El Niño winter of 2002¿2003 were expressed in anomalous ice and snow season patterns. We are developing a consistent, MODIS-based dataset that will be used to monitor temporal trends of each of these seasonal metrics and to map areas of change for the study area. [Copyright &y& Elsevier]

The temporal evolution of the backscatter coefficient, sigma-nought (σ°) from QuikSCAT was evaluated for monitoring ice phenology on Great Bear Lake (66°N, 121°W) and Great Slave Lake (61°40′N, 114°W), Northwest Territories, Canada. Results indicated that σ° from QuikSCAT can be used to detect melt onset, water clear of ice and freeze onset dates on both lakes. An ice phenology algorithm was then developed to assess the spatiotemporal variability on both lakes from QuikSCAT for the period 2000–2006. Results showed that for Great Slave Lake, the average melt onset date occurred on year day (YD) 123, the average water clear of ice date was on YD164, and the average freeze onset date was on YD330. On Great Bear Lake, the average melt onset date occurred on YD139, the average water clear of ice date was YD191, and the average freeze onset date was YD321. Ice cover remained present for at least five weeks longer on Great Bear Lake than on Great Slave Lake and most of the difference can be explained by earlier ice melt on Great Slave Lake. Spatially, on Great Bear Lake, melt onset took place first in the eastern arm, water clear of ice occurred first in southeastern and western arms, and freeze onset appeared first in the northern arm and along the shorelines. On Great Slave Lake, melt onset began first in the central basin and then progressed to the northern and eastern arms later in the season. The central basin of Great Slave Lake cleared earlier than the periphery due to the discharge from the Slave River. Freeze onset on Great Slave Lake occurred first within the east arm, closely followed by the north and west arms, and then finally in the centre of the main basin. [Copyright &y& Elsevier]

We demonstrate the potential of combining satellite altimetry and radiometry for lake ice studies using the example of Lake Baikal in Siberia. We show the synergy using active and passive microwave observations available from the recent satellite altimetry missions (TOPEX/Poseidon, Jason-1, ENVISAT and Geosat Follow-On), complemented by the SSM/I passive data. We assess the applicability of altimetric and radiometric data for ice/water discrimination, and discuss the drawbacks and benefits of each type of sensor. An ice discrimination method, based on the combined use of the data from the four altimetric missions and SSM/I, is proposed and validated using available in situ observations and MODIS imagery. The method is applied to the entire satellite data set and used to define specific dates of ice events (first appearance of ice, formation of stable ice cover, first appearance of open water, complete disappearance of ice) and associated uncertainties. Using these satellite-derived estimates, we can extend the existing time series of ice events in the Southern Baikal up to 2004 and provide new information on the Middle and Northern Baikal, regions where no recent in situ ice cover observations are available. Our data show that over the last 10–15 years, trends towards earlier ice formation and later ice break-up result in a tendency for longer fast ice duration over the whole Lake Baikal. The methods proposed here have been tested for Lake Baikal, but they are applicable for other lakes and water bodies with seasonal ice cover. [Copyright &y& Elsevier]

Abstract: Variability and trends in lake ice dynamics (i. E. Lake ice phenology) are related to climate conditions. Climate influences the timing of lake ice melt and freeze onset, ice duration, and lake thermal dynamics that feedback to the climate system initiating further change. Phenology records acquired in a consistent manner and over long time periods are required to better understand variability and change in climate conditions and how changes impact lake processes. In this study, we present a new technique for extracting lake ice phenology events from historical satellite records acquired by the series of Advanced Very High Resolution Radiometer (AVHRR) sensors. The technique was used to extend existing in-situ measurements for 36 Canadian lakes and to develop records for 6 lakes in Canada''s far north. Comparison of phenology events obtained from the AVHRR record and in-situ measurements show strong agreement (20 lakes, 180 cases) suggesting, with high confidence especially in the case of break-up dates, the use of these data as a complement to ground observations. Trend analysis performed using the combined in-situ and AVHRR record ∼ 1950–2004 shows earlier break-up (average — 0. 18 days/year) and later freeze-up (average 0. 12 days/year) for the majority of lakes analyzed. Less confidence is given to freeze-up date results due to lower sun elevation during this period making extraction more difficult. Trends for the 20 year record in the far north showed earlier break-up (average 0. 99 days/year) and later freeze-up (average 0. 76 days/year). The established lake ice phenology database from the historical AVHRR image archive for the period from 1985 to 2004 will to a certain degree fill data gaps in the Canadian in-situ observation network. Furthermore, the presented extraction procedure is not sensor specific and will enable continual data update using all available satellite data provided from sensors such as NOAA/AVHRR, MetOp/AVHRR, MODIS, MERIS and SPOT/VGT. [Copyright &y& Elsevier]

Lake ice is an important feature of the physical landscape at northern latitudes. Not only does the presence of lake ice help modulate weather and climate but it also plays an important role in travel between northern communities. Furthermore, the thickness of ice cover can impact overwintering fish habitat and access to fresh water during winter months. Lake ice cover and lake ice thickness are two notable climate proxies identified by the Global Climate Observing System (GCOS) as Essential Climate Variables (ECVs). Changes in ice thickness can impact lakes freezing to bed which in turn affect permafrost thaw while longer periods of open water can impact heat transfer to the atmosphere. With recent climate warming in the Northern Hemisphere, there are general trends towards thinner ice and an increase in the length of the open water season. However, the paucity of in-situ observations in many countries has led to remote sensing playing an ever-increasing role for lake ice monitoring. This paper reviews progress in lake ice research conducted using active microwave remote sensing over the last 50 years and highlights areas where future developments are needed. Analysis of the literature found that the diversity of study areas where this work is taking place has increased, incorporating more lake sites from the mid-latitudes in North America, Europe, and Asia. Additionally, clear connections can be made between the launch of synthetic aperture radar (SAR) satellites and patterns in the use of certain radar frequencies in scientific studies. This review also discusses evolving theories concerning the scattering mechanisms associated with lake ice as well as the current state of lake ice cover and ice thickness retrieval methods. The paper concludes with a suite of recommendations for future research, highlighting work that is needed on: 1) advancing our understanding of the response of active microwave signals to lake ice properties; 2) applying machine learning algorithms to lake ice classification; 3) further exploring the retrieval of ice thickness from imaging SAR and radar altimetry data; and 4) other considerations such as data availability and the topic of snow on lake ice. • Active microwave remote sensing is the most common technology for studying lake ice. • Our understanding of how active data interacts with lake ice properties has changed. • A variety of methods exist for monitoring lake ice cover using SAR data. • Research is needed on interactions between active signals and lake ice properties. • Data availability will help develop algorithms to monitor essential ice properties. [ABSTRACT FROM AUTHOR]

The topic of satellite remote sensing of lake ice has gained considerable attention in recent years. Optical satellite data from the Moderate Resolution Imaging Spectroradiometer (MODIS) allow for the monitoring of lake ice cover (an Essential Climate Variable or ECV), and dates associated with ice phenology (freeze-up, break-up, and ice cover duration) over large areas in an era where ground-based observational networks have nearly vanished in many northern countries. Ice phenology dates as well as dates of maximum and minimum ice cover extent (for lakes that do not form a complete ice cover in winter or do not totally lose their ice cover in summer) are useful for assessing long-term trends and variability in climate, particularly due to their sensitivity to changes in near-surface air temperature. Existing knowledge-driven (threshold-based) retrieval algorithms for lake ice cover mapping that use top-of-atmosphere (TOA) reflectance products do not perform well under lower solar illumination conditions (i.e. large solar zenith angles), resulting in low TOA reflectance. This research assessed the capability of four machine learning classifiers (i.e. multinomial logistic regression, MLR; support vector machine, SVM; random forest, RF; gradient boosting trees, GBT) for mapping lake ice cover, water and cloud cover during both break-up and freeze-up periods using the MODIS/Terra L1B TOA (MOD02) product. The classifiers were trained and validated using samples collected from 17 large lakes across the Northern Hemisphere (Europe and North America); lakes that represent different characteristics with regards to area, latitude, freezing frequency, and ice duration. Following an accuracy assessment using random k-fold cross-validation (k = 100), all machine learning classifiers using a 7-band combination (visible, near-infrared and shortwave-infrared) were found to be able to produce overall classification accuracies above 94%. Both RF and GBT provided overall and class-specific accuracies above 98% and a more visually accurate depiction of lake ice, water and cloud cover. The two tree-based classifiers offered the most robust spatial transferability over the 17 lakes and performed consistently well across ice seasons. However, only RF was relatively insensitive to the choice of the hyperparameters compared to the other three classifiers. The results demonstrate the potential of RF for mapping lake ice cover globally from MODIS TOA reflectance data. • This study assessed the capability of ML classifiers for lake ice mapping from MOIDS. • RF and GBT produced the best performance in terms of classification accuracies. • RF and GBT offered the most robust spatial and temporal transferability. • RF was insensitive to the choice of the hyperparameters compared to other classifiers. • The results show the potential of RF for mapping lake ice cover globally from MODIS. [ABSTRACT FROM AUTHOR]

The freeze/thaw state of permafrost landscapes is an essential variable for monitoring ecological, hydrological and climate processes. Ground surface state can be obtained from satellite data through time series analysis of C-band backscatter from scatterometer and Synthetic Aperture Radar (SAR) observations. Scatterometer data has been used in a variety of studies concerning freeze/thaw retrieval of the land surface. Coarse spatial resolution scatterometer data has great potential for application in this field due to its high temporal resolution (approx. daily observations). In this study, we investigate the influence of sub-grid cell (12.5 km) surface water (ice free and ice covered) on freeze/thaw retrieval based on ASCAT data using a threshold algorithm. We found discrepancies related to the surface water fraction in the detected timing of thawing and freezing of up to 2 days earlier thawing for spring and 3.5 days earlier freezing for autumn for open water fractions of 40% resulting in an overestimation of the frozen season. Results of this study led to the creation of a method for correction of water fraction impact on freeze/thaw data. Additionally, this study demonstrates the applicability of a new approach to freeze/thaw retrieval which has not so far been tested for SAR, specifically Sentinel-1. • Influence of sub-grid cell surface water on freeze/thaw retrieval based on ASCAT data • Discrepancies related to the surface water fraction in the detected timing of thawing and freezing • Found a general overestimation of the frozen season. • Study demonstrates applicability of approach which was so far not tested for SAR. [ABSTRACT FROM AUTHOR]

This study examines links between two sets of parameters obtained from Radarsat-2 Quad-pol data and geophysical information of lake ice and landfast sea ice in the Resolute Bay area, Nunavut, Canada. The first set includes the three orthogonal backscatter coefficients ( σ hh 0 , σ vv 0 , σ vh 0 ) in addition to the total backscatter power (SPAN). The second set includes polarimetric decomposition parameters obtained using Cloude-Pottier and Yamaguchi decomposition methods. Data were collected during the early freezing period from 20 September until 28 December 2017. Meteorological and ice climatological data from the weather station of Resolute Bay were used to facilitate this task. Ice thickness and salinity were obtained from established empirical models. Results show that single-bounce and multiple-bounce scattering mechanisms are active in lake ice during the entire freezing period while multiple-bounce scattering is active in sea ice only during the initial freezing. Backscatter parameters can be used to determine the thickness of lake ice up to any depth, while both backscatter and the tested decomposition parameters are related to the thickness of young sea ice (<30 cm thick). Physical explanations are offered in each case. Spatial maps of ice thickness are produced. Polarimetric decomposition parameters demonstrate erratic behavior throughout the development of lake ice. High entropy (>0.6) is found to be associated with comparable single- and multiple-bounce scattering mechanisms, hence can be used to identify new ice. Orthogonal backscatter coefficients and polarimetric decomposition parameters provide different information, yet their combination offers clues to better identify ice conditions and investigate questions about scattering mechanisms. • Backscatter parameters determine the thickness of lake ice up to any depth. • Backscatter parameters are related to the thickness of young sea ice. • Entropy and Alpha angle are related to the thickness of young sea ice. [ABSTRACT FROM AUTHOR]