Dense time series of imagery from Landsat 8 and Sentinel 2 are creating exciting new opportunities to monitor, map, and characterize temporal dynamics in land surface properties with unprecedented spatial detail and quality. By combining imagery from the Landsat 8 Operational Land Imager and the MultiSpectral Instrument on-board Sentinel 2A and 2B, the remote sensing community now has access to moderate spatial resolution imagery with repeat periods more than three times more frequent than those available prior to the launch of Sentinel 2A. At the same time, the large data volumes and high-dimensionality of blended time series from Landsat 8 and Sentinel 2 introduce substantial new challenges for users who wish to use these data. Land surface phenology (LSP) products, which synthesize the timing of phenophase transitions and quantify the nature and magnitude of seasonality in remotely sensed land surface conditions, provide a simple and intuitive way to reduce data volumes and redundancy, while also furnishing rich feature sets that are useful for a wide range of applications including ecosystem and agro-ecosystem modeling, monitoring the response of terrestrial ecosystems to climate variability and extreme events, crop-type discrimination, and land cover, land use, and land cover change mapping. To support the need for operational LSP data sets, here we describe a continental-scale land surface phenology product based on Harmonized Landsat-Sentinel 2 (HLS) data from NASA. This product includes science data sets that: (1) identify the timing of phenophase transitions, (2) provide the user community with reduced dimensionality image data sets that capture the primary modes of multispectral and multi-temporal variability and minimize temporal correlation in image time series, and (3) identify phenological anomalies in near real-time. In this way, the LSP data sets provided by this product go well beyond what current coarse spatial resolution LSP products provide, and are designed to support a wide and diverse user community. We describe the algorithm we have developed for this effort, LSP data sets for North America based on three years of HLS data, and present results from assessment efforts comparing HLS-based LSP retrievals with observations of phenology derived from a network of ground-based webcam imagery.

Landsat-8 and Sentinel-2 provide high resolution, cloud-free, near-nadir imagery, which provides a valuable resource for the examination of surface heterogeneity, vegetation status, and biomass extent. However, these near nadir sensors cannot provide adequate multi-angle observations to accurately capture the true effects of surface anisotropy (i.e., the Bidirectional Reflectance Distribution Function, BRDF) which allow the production of realistic bihemispherical albedo quantities of the land surface. Therefore, to acquire BRDF information we rely on wide swath, multi-angle, multi-spectral imagers, such as MODIS and VIIRS, which use a semi-empirical BRDF model (the RossThickLiSparseReciprocal (RTLSR) Lucht et al., 2000) and the provide products at 0.5 and 1.0 km resolutions. With a well-established protocol, we are able to combine the BRDF information from MODIS and VIIRS sensors with Landsat (and Sentinel-2) surface reflectance products to generate high resolution albedo. Here we use the newly released Harmonized Landsat and Sentinel-2 (HLS) surface reflectance product (Claverie et al., 2018) to generate a 30m albedo product. The use of the HLS dataset provides continuity within the surface reflectance inputs ensuring comparability and consistency across space and time. The continuity of data between VIIRS and MODIS has been previously presented (Li, et. al., 2018), and will be examined further here.

We use this high spatial and temporal resolution dataset to examine the surface energy implications of boreal fire across the high northern latitudes. As these regions continue to warm, shifts in vegetation structure, composition and distribution have been widely observed. Coupled with these shifts has been an increase in both the number and intensity of disturbances. North American boreal forests have, over the past five decades, been subject to increases in fire severity, the total burned area and frequency of large, naturally ignited fires (Kasischke and Turetsky, 2006; Turetsky et al., 2010). This shift in fire frequency and severity extends northward into the arctic tundra, which has previously been relatively free of fire since the early Holocene (Higuera et al., 2008). The drivers of these changes are diverse and include the likely increase in the number of lightning strikes (Veraverbeke et al., 2017), lengthening of the fire/growing season (Kasischke and Turetsky, 2006; Soja et al., 2007; Turetsky et al., 2010), and shifts in hydrological processes and general vegetation health.

In the northern latitudes, the most significant forcing within fire scars is caused by increasing albedo due to snow cover and phenological events of vegetation regrowth (Rogers et al., 2013). These changes can last many decades and have a substantial cooling effect (negative surface forcing) resulting from consistent snow cover and moderate insolation during February through April in northern latitudes (Randerson et al., 2006; Rogers et al., 2013). We use these higher resolution albedo products to examine trends in landscape heterogeneity post fire, with a focus on early spring albedo in the years immediately following a wildfire. In investigating the impact of fire on albedo recovery trajectories we will focus on selected fires in Alaska and Canada, including select fire scars in Saskatchewan, Canada which burned in 2015. The fires span the boreal shield and boreal plains ecoregions with primary fuel sources of black spruce and jack pine. These lower latitude sites provide sufficient 3 or 4 season albedo values, as sun angle limitations at Landsat and Sentinel-2 overpass times are less restrictive than at higher latitudes. We use both pre and post-fire HLS, MODIS and VIIRS derived albedo to quantify changes in the early spring blue sky albedo and examine implications on landscape fragmentation and surface energy.


Claverie, M., Ju, J., Masek, J. G., Dungan, J. L., Vermote, E. F., Roger, J.-C., Skakun, S. V., & Justice, C. (2018). The Harmonized Landsat and Sentinel-2 surface reflectance data set. Remote Sensing of Environment, 219, 145-161.
Kasischke, E.S., Turetsky, M.R., 2006. Recent changes in the fire regime across the North American boreal region—Spatial and temporal patterns of burning across Canada and Alaska. Geophys. Res. Lett. 33
Li, Z., Erb, A., Sun, Q., Liu, Y., Shuai, Y., Wang, Z., Boucher, P., Schaaf, C., 2018. Preliminary assessment of 20-m surface albedo retrievals from sentinel-2A surface reflectance and MODIS/VIIRS surface anisotropy measures. Remote Sensing of Environment. 217, 352-365,
Lucht, W., Schaaf, C.B.C.B.C.B., Strahler, a. H.A.H., 2000. An Algorithm for the retrieval of albedo from space using semiempirical BRDF models. IEEE Trans. Geosci. Remote Sens.
Higuera, P.E., Brubaker, L.B., Anderson, P.M., Brown, T.A., Kennedy, A.T., Hu, F.S., 2008. Frequent fires in ancient shrub tundra: Implications of paleorecords for arctic environmental change. PLoS One 3, 1–7.
Randerson, J.T., Liu, H., Flanner, M.G., Chambers, S.D., Jin, Y., Hess, P.G., Pfister, G., Mack, M.C., Treseder, K.K., Welp, L.R., Chapin, F.S., Harden, J.W., Goulden, M.L., Lyons, E., Neff, J.C., Schuur, E. a G., Zender, C.S., 2006. The impact of boreal forest fire on climate warming. Science (80-. ). 314, 1130–2.
Rogers, B.M., Randerson, J.T., Bonan, G.B., 2013. High-latitude cooling associated with landscape changes from North American boreal forest fires. Biogeosciences 10, 699–718.
Turetsky, M.R., Kane, E.S., Harden, J.W., Ottmar, R.D., Manies, K.L., Hoy, E., Kasischke, E.S., 2010. Recent acceleration of biomass burning and carbon losses in Alaskan forests and peatlands. Nat. Geosci. 4, 27.
Veraverbeke, S., Rogers, B.M., Goulden, M.L., Jandt, R.R., Miller, C.E., Wiggins, E.B., Randerson, J.T., 2017. Lightning as a major driver of recent large fire years in North American boreal forests. Nat. Clim. Chang. 7, 529–534.

Deforestation and forest degradations are known as critical problems. In this context, it is crucial to monitor timely and accurately forest disturbances. Remote sensing appears to be a good solution to monitoring forest in a large scale, but the existing early warning systems are mainly based on optical images (Hansen and al., 2016; Diniz and al., 2015). Nevertheless, the persistent cloud cover in tropical areas leads to delayed disturbances detection when using optical observations. Moreover, in most studies, the emphasis is given exclusively on alerting deforestation. In order to allow an adapted response to the situation, users need to know the driver.

Over the last fifteen years, deforestation and forest degradation have been increasing very quickly in the Guiana Shield, mostly because of gold mining. Mining is the biggest social, sanitary and environmental problem in the Guiana Shield. In fact, large amounts of mercury and cyanide are used for gold mining, with severe impacts on biodiversity, freshwater, forest and human health. As a response, the countries of the Guiana Shield, i.e. Amapá (State of North Brazil), French Guiana, Suriname and Guyana, developed in 2014 a remote sensing system to detect and assess deforestation and forest degradation, using a unified approach (Rahm and al. 2015). However, this system is not fully automated and is based on optical satellite images, which are not adapted to near real-time disturbances detection nor to the need for response (WWF France, 2018).

SAR data present the advantage of being relatively independent of cloud cover, which is the biggest limitation of optical system over tropical areas. Bouvet et al. (2018) presented a reliable indicator to detect deforestation using SAR data, which can be used in near real-time: the Radar Change Ratio. However, Sentinel-1 data, which are the only freely available SAR images with global coverage and fast repeat time (6-12 days), are not well suited to characterize the drivers that lead to forest disturbance. Therefore, optical images could be used in addition to SAR images in order to determine the drivers of disturbances, e.g. to distinguish gold mining from other degradations such as agriculture (using for example the water turbidity).

In this study, we first use the Radar Change Ratio to detect new forest disturbances using Sentinel-1 data time series. Then, a machine learning algorithm is applied to Sentinel-2 data, in order to identify the driver of each disturbance. This demonstration has been conducted over the whole Guiana Shield in a near real-time configuration from 2014 to 2018. Preliminary validation shows a gain of 99 days in average for the disturbance detection with our method. We also discuss deforestation trends in the Guiana Shield over the last four years.

References:
Bouvet, A., Mermoz, S., Ballère, M., Koleck, T., & Le Toan, T. (2018). Use of the SAR Shadowing Effect for Deforestation Detection with Sentinel-1 Time Series. Remote Sensing, 10(8), 1250.
Diniz, C. G., de Almeida Souza, A. A., Santos, D. C., Dias, M. C., da Luz, N. C., de Moraes, D. R. V., Sant’Ana Maia, J., Rodrigues Gomes, A., da Silva Narvaes, I., Valeriano, D. M., Maurano, L. E. P. & Adami, M. (2015). DETER-B: The new Amazon near real-time deforestation detection system. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 8(7), 3619-3628.
Hansen, M. C., Krylov, A., Tyukavina, A., Potapov, P. V., Turubanova, S., Zutta, B., ... & Moore, R. (2016). Humid tropical forest disturbance alerts using Landsat data. Environmental Research Letters, 11(3), 034008.
Rahm, M., Jullian, B., Lauger, A., De Carvalho, R., Vale, L., Totaram, J., ... & Vieira, R. (2014). Monitoring the impact of gold mining on the forest cover and freshwater in the Guiana Shield.
WWF France (2018). Lutte contre l’orpaillage illégal en Guyane : orientation pour une efficacité renforcée.

Research on links between urbanization and urban heat islands (UHI), their effects, and mitigation strategies have gained significant impetus over the past decade due to a warming planet and overall increase in the number of city dwellers during the 21st century. For example, it is projected that nearly 66% of the world's population will live in cities by 2050. The negative effects of urbanization and related UHI’s are dangerous and widespread and include increases in the intensity and frequency of heat waves (and related heat stress/mortality), increased energy demands and economic cost, poor air quality, and changes in regional climate. The urban landscape represents a complex heterogeneous surface that strongly influences the development of the UHI and cannot be adequately characterized using traditional structural and optical based remote sensing classification techniques (i.e., land use/cover types) since they do not relate to the physical functioning of the surface energy budget. Including thermal infrared (TIR) derived land surface temperature (LST) information provides the critical missing element for quantifying the effect of urbanization on the UHI and helping understand UHI and heat wave interactions. However, current moderate resolution TIR data from Landsat/ASTER (~100m) are unable to resolve fine-scale urban features (<50m), compounded by infrequent observations of at most 16 days in clear skies, making it impossible to effectively model the daily energy balance cycle and provide heat mitigation strategies. As part of a NASA Land Cover Land Use Change (LCLUC) Multi-source Land Imaging (MuSLI) project, we will produce an urban LST product by innovatively combining Landsat-8 TIR and Sentinel-2 optical multispectral imagery with GOES-16/17 high frequency TIR data. The result will be a high spatio-temporal resolution LST product at 30m spatial resolution and in 30 min time-steps optimized for urban environments. We will show a prototype of the urban LST product over the Los Angeles basin during a heatwave event in August 2018 and validate the urban LST product with coincident ECOSTRESS LST data acquired from the ISS between Aug – Sep 2018. The product will be used to investigate and better understand heat wave and UHI interactions and will be extended to four US megacities; Los Angeles, Chicago, Phoenix, and Washington DC. The product will further be incorporated into a heat vulnerability model that can be used in near real time to provide data that would help local officials and decision makers to develop effective adaption and mitigation strategies, issue heat warnings, and help identify communities most vulnerable to the effects of extreme heat events.

Since early 2018, the Landsat 8 Operational Land Imager (OLI)/Sentinel-2 Multi-Spectral Instrument (MSI) Earth observation constellation has been acquiring multispectral visible-to-shortwave infrared (VSWIR) solar-reflected measurements for global landmasses and near shore coastal regions at an approximate temporal frequency of 3 days. This imaging constellation reflects a new observational era for multi-source satellite remote sensing of Earth’s cryosphere at medium resolution spatial scales. Given Landsat 8 OLI’s absolute radiometric calibration, measurement cross-validation between OLI and Sentinel-2A/2B MSI is required to ensure that calibrated land and aquatic products are interoperable across missions. Ongoing efforts to compare Landsat 8 and Sentinel-2A/2B top-of-atmosphere (TOA) measurements for mid-latitude desert pseudo invariant calibration sites (PICS) indicate that OLI and MSI agree within 1-2%. For cryospheric applications, there is pressing need to quantify OLI and MSI measurement performance over high radiance snow and ice surface mediums at latitudes north and south of 45° for both cold and warm season atmospheres. In this communication, we briefly describe our approach for quantifying OLI and MSI measurement performance for snow on land, inland lake ice, and land ice by exploiting coincident-to-near coincident, National Institute of Standards and Technology (NIST) traceable in situ ground and low altitude airborne VSWIR spectrometer measurements acquired during field campaigns to U.S. Midwest Glacial Lakes and Mid-Atlantic regions, and the northwestern Greenland Ice Sheet over the period 2015 to 2018.

Our technical approach to quantify Landsat 8 OLI and Sentinel-2A/2B measurement performance is to use reference full spectrum (0.35 to 2.5 μm) in situ ground and low altitude airborne VSWIR measurements of snow on land, inland lake ice, and land ice, coupled with atmospheric measurements from either an AErosol Robotic NETwork (AERONET) CIMEL or other daily satellite sensor retrievals to initialize the MODerate resolution atmospheric TRANsmission (MODTRAN) Version 6 radiative transfer code. First, we simulate TOA radiance for OLI and MSI relative spectral response functions and compare with observed Level-1 TOA radiance products for a pixel neighborhood (120 m x 120 m footprint on the ground and along low altitude airborne flight lines) during overpass. Second, we then compare reference in situ ground and atmospherically-corrected airborne reflectance with U.S. Geological Survey (USGS) Landsat 8 and European Space Agency (ESA)/Copernicus Sentinel-2 standard Level-2 surface reflectance products within the same pixel neighborhood footprint. We use a difference ratio to quantify measurement uncertainty for Level-1 and Level-2 products and consider MODTRAN TOA simulations and in situ ground and low altitude airborne reflectance as ‘truth’ because they are common references to both OLI and MSI.

Initial OLI results for validation sites in U.S. Midwest Glacial Lakes and Mid-Atlantic regions indicate that Landsat 8 standard Level-2 surface reflectance product pixels with snow on land can be up to 30% different than reference in situ ground reflectance. This result is expected in part, because the operational Landsat 8 Land Surface Reflectance Code (LaSRC) is not optimized to compensate for atmospheric effects over snow and ice surface mediums. When Landsat 8 Level-1 TOA radiance is compared with MODTRAN simulated TOA radiance using reference airborne reflectance spectra over the interior northwestern Greenland Ice Sheet, OLI is measuring between 6 and 16% more TOA radiance for flight line pixels. This measurement performance approach is currently being extended to Sentinel-2A/2B acquisitions where in situ ground and airborne VSWIR measurements have been acquired for U.S. Midwest and Mid-Atlantic regions as well as the northwestern Greenland Ice Sheet. The objective of this communication is to establish a first approximation of Landsat 8 and Sentinel-2A/2B measurement uncertainty for snow on land, inland lake ice, and land ice surfaces at Level-1 and Level-2 to inform land cryosphere product development from the Landsat 8/Sentinel-2 Earth observation constellation.

The availability of high resolution data from Sentinel-2 and Landsat 8 (LC8) shows great promise for continuous land surface monitoring. A prerequisite to combine observations from both sensors is that the pre-processing results in surface reflectances that are not biased, and where consistent choices in the pre-processing have been made. Further, combining different sensors requires an understanding of the uncertainty associated with each measurement.

The requirements introduce above are missing from current standard atmospheric correction methods. This contribution exploits a number of recent advances to address the previous requirements. First, global reanalysis data of atmospheric composition is now routinely available from the Copernicus Atmosphere Monitoring System (CAMS). This data can be used to provide a prior distribution of atmospheric constituents such as aerosols, total columnar water vapour and ozone. Second, the MODIS MCD43 product (or any other similar product) provides a global daily estimation of surface BRDF parameters that can be used to predict land surface reflectance at a coarse (>500m) spatial resolution. Third, recent advances in emulation of complex numerical codes, such as atmospheric radiative transfer (RT) models allows for a detailed solution of the atmosphere-land coupled problem, as well as to develop robust and traceable uncertainty quantification schemes.

The SIAC package exploits these advances in the following fashion: the MODIS MCD43 product is used to produce an expectation of surface reflectance at a coarse resolution. After carefully modelling the effective point spread function (PSF) of the MODIS data in terms of the high resolution sensor (S2 or LC8), a mismatch between the expected surface reflectance, propagated through an atmospheric RT model (6S) as a function of atmospheric composition, and the measured top-of-atmosphere reflectance is calculated, taking into account the uncertainty of the measurements, as well as the uncertainty in the MODIS predictions. This mismatch is in effect a cost function which is then augmented with an additional prior constraint from the CAMS data on aerosol optical thickness (AOT), total column water vapour (TCWV) and ozone (TCO3). Finally, a further constraint that encodes spatial smoothness in the atmospheric composition parameters is also added. The combined cost function is then solved by gradient descent methods, making use of Gaussian Process emulators of the 6S model (Gómez-Dans, Lewis, & Disney, 2016).

This contribution introduces the concepts above, and shows the performance of the SIAC approach on both Sentinel-2 and LC8 data, showing comparisons both of atmospheric constituents retrieved by SIAC compared with in situ AERONET measurements, and also comparisons with in situ surface reflectance measurements over the RadCalNet reference sites. AOT and TCWV retrievals are very successful from both S2 and LC8. Linear regression with respect to in situ AERONET measurements shows high correlation of retrieved AOT with in situ (r2>0.9) and TCWV (r2>0.95). Surface reflectance comparisons with RadCalNet are also characterised by a very high correlation (r2>0.98) and very low bias (<0.03 in units of reflectance).