We publish and describe a surface spectral reflectance data record
of seasonal snow (dry, wet, shadowed), forest ground (lichen, moss) and
forest canopy (spruce and pine, branches) constituting the main elements of
the boreal landscape. The reflectances are measured with
spectro(radio)meters covering the wavelengths from visible (VIS) to
short-wave infrared (SWIR) (350 to 2500 nm). In this paper, we describe the
instruments used and how the spectral observations at different scales along
with the concurrent in situ reference data have been collected, processed
and archived. Information on the quality of the data and factors causing
uncertainty are discussed. The main experimental site is located in
the Sodankylä Arctic Space Centre in northern Finland (67.37
High-latitude regions are facing fundamental and rapid changes in climate and hydrology due to raising mean annual temperatures (ACIA, 2005; AMAP, 2017). The climate-warming-induced changes in snow cover appearance, shifting of the vegetation zones and distribution of animal species, have complex impacts on ecosystems and people (Brown and Mote, 2009; Callaghan et al., 2011). Monitoring of the seasonal snow cover of the spatially vast subarctic and boreal zone benefits from remote sensing for various scientific and socio-economic uses, e.g. related to the assessment of carbon balance in the boreal and subarctic forests (Böttcher et al., 2014; Pan et al., 2011; Pulliainen et al., 2017). Remote sensing has developed over decades into an increasingly reliable and cost-effective way to estimate the decadal and annual changes in the Northern Hemisphere terrestrial snow cover (Brown and Robinson, 2011; Choi et al., 2010; Derksen and Brown, 2012; Dietz et al., 2012; Frei et al., 2012; Hori et al., 2017). The development of reliable methods to map snow extent, including the fractional snow cover (FSC), remains a challenging task especially due to the disturbing effect of forest canopy and heterogeneous land cover. Several approaches have been used to estimate the FSC from satellite imagery (Klein et al., 1998; Hall and Riggs, 2007; Dozier et al., 2009; Nolin, 2010; Dietz et al., 2012; Frei et al., 2012; Metsämäki et al., 2015). These methods, such as the semi-empirical-reflectance-model-based method SCAmod (Metsämäki et al., 2005, 2012), used for the detection of snow cover in forested areas, have benefited from accurate reference spectral measurements enabling better characterization of the model parameters (i.e. spectral endmembers). Spectral endmember refers to a “pure” reflectance spectra of a distinct surface type such as a distinct type of snow or tree species.
Field spectroscopy techniques have evolved into a widely used tool to understand the effects of the measured target on the propagation of electromagnetic radiation. This provides observations under more controlled conditions compared to measurements from satellite platforms (Aoki et al., 2000; Bänninger et al., 2008; Horton and Jamieson, 2017; Milton et al., 2009; Painter et al., 2013; Peltoniemi et al., 2005; Pirazzini et al., 2015; Tanikawa et al., 2014). In order to establish improved optical snow mapping methods for forested areas, detailed surveys of satellite scene reflectance contributors are required, as the relatively large satellite footprint may contain both fractional snow and forest cover. Additionally, snow characteristics may vary according to land cover type, e.g. between forests and open areas. Using continuous spectral signatures, i.e. from instruments with very narrow bandwidths, various land cover constituting elements can be spectrally characterized and their contribution to satellite scene reflectance then identified. In boreal landscape, reference spectroscopy measurements are valuable in defining the spectral endmembers of the satellite scene reflectance, namely snow, snow-free terrain after melting and forest cover. These data can be obtained from controlled-condition laboratory spectroradiometer observations, portable field spectroscopy campaigns, mast-borne spectral monitoring and aerial surveys. These approaches provide observations at different scales. Laboratory measurements can generate detailed information on the spectral signature of a trunk, branch or leaf of a single tree, whereas using portable field spectroscopy, several land cover categories or shrub layer vegetation types can be spectrally characterized. Mast-borne monitoring of scene reflectance facilitates time series production and the study of the seasonal behaviour of fractional snow- and forest-covered scene reflectance. Aerial surveys are useful in extending the observations to a larger variety of landscape properties in particular during the melting season and still maintaining the advantage of high spatial resolution.
To fully benefit from the increasing amount of spectral reflectance data records available in various archives and libraries, it is essential to ensure that the data are of consistent quality and accompanied with information on the sources of uncertainties, such as the variations in the incoming radiation during the measurements or unideal characteristics of the measurement setup or the used reference calibration target. However, in the case of field measurements in the natural environment, it is difficult to provide quantitative uncertainty information due to the lack of repeatability of the exact same conditions on different occasions (even though single measurements may include several spectral acquisitions). In addition, there are limiting factors due to the ambiguous use of reflectance terminology, measurement geometry description and variable measurement protocols (Milton et al., 2009; Schaepman-Strub et al., 2006). To control the limiting factors the provision of metadata and sufficient documentation on the measurement conditions and target characteristics is essential (Rasaiah et al., 2014, 2015). To produce high-quality spectral information, some guidelines for successful measurements and factors influencing the measurement output have been reported by both the instrument manufacturers and the individual scientific projects (Goetz, 2012; Pfitzner et al., 2011). Depending on the application, different levels of quality can be acceptable, but in general common protocols and standardized terminology are required for successful data sharing, fusion of different data sources and for the data comparison (Dor et al., 2015; Milton et al., 2009). Often laborious and time-consuming experimental field work, data review and quality check may limit the resources to compile a thoroughly described metadata (Kokaly et al., 2017; Rasaiah et al., 2015). A set of metadata parameters critical to field spectroscopy have been presented by Rasaiah et al. (2014). They include viewing geometry, location, general target and sampling properties, illumination, instrument properties, reference standards, calibration, hyperspectral signal properties, atmospheric conditions, and general project details. The requirements aim for such metadata and documentation that the user is able to assess the quality level of the spectra and account for the likely variations from one data record to another.
We collected a surface spectral reflectance data record including main
components (model spectral endmembers) of boreal seasonally snow covered
landscape during spring. The collection consists of laboratory, portable
field, mast-borne and selected airborne campaign reflectance observations of
snow and vegetation representing a subarctic site in the northern boreal forest
zone. The site is located at the Sodankylä Arctic Space Centre in
northern Finland (67.37
In this section, we describe the Finnish Meteorological Institute's Sodankylä Arctic Space Centre (FMI-ARC) experimental site characteristics for data collection. Besides the four different spatial-scale datasets (described in the following Sects. 2.2 and 3) from the FMI-ARC area, data are also presented from one aerial survey over Saariselkä, around 120 km north of Sodankylä. Saariselkä is a fell (Arctic hill) region that has a timberline at an altitude of 400 m a.s.l. The treeless altitudes represent fell tundra (Virtanen et al., 2016). Also, some individual field spectral samples from a boreal forest area in Nuuksio, Espoo, southern Finland, are included in the collection (Fig. 1). These additional data were collected to capture more observations from late melting conditions of the snowpack.
The location of the Sodankylä Arctic Space Centre (FMI-ARC), where
most of the data records have been measured. In addition, one aerial survey
was conducted in the Saariselkä fell region, north from the FMI-ARC, and
some individual field spectra were measured in Nuuksio, Espoo, southern
Finland. Distribution of boreal, temperate conifer, temperate broadleaf and
mixed forests, and tundra by the Nature Conservancy (Olson and Dinerstein,
2002,
The Sodankylä station, situated above the Arctic Circle in northern
Finland (67.37
The Sodankylä region is a globally representative example of the boreal forest biome, which encompasses the largest continuous land ecosystem on the planet (ACIA, 2005). Seasonal snow cover is a characteristic feature of the boreal forest zone affecting strongly the functions of the ecosystem, water cycle, and surface–atmosphere interaction. Sodankylä has a subarctic climate due to the warming effect of the Gulf Stream (Kangas et al., 2016). Characteristics for Sodankylä are extreme seasonal temperature variations as well as long and cold winters with a snow season from October until May. The Sodankylä area represents taiga snow, and from 1981 until 2010, the maximum snow depth of approximately 80 cm occurred in late March (Pirinen et al., 2012). The changing seasonal snow cover affects the boreal forest carbon uptake and storage and the hydrological cycle, which are also important features of the boreal ecology in the Sodankylä region (Pan et al., 2011). The landscape around the Sodankylä station is relatively flat, with isolated fells reaching up to 500 m. The landscape consists of sparse-pine-dominated coniferous forests and open areas on mineral soil as well as open peat bogs (Leppänen et al., 2016).
All spectral observations here, regardless of their measurement scale,
correspond to a typical polar-orbiting satellite measurement in the high-latitude spring season with respect to their Sun or illumination source
(calibrated lamp) zenith angle and close-to-nadir instrument viewing angle.
Figure 2 illustrates the radiance (unit of measure: W m
General concept of a satellite- or ground-based remote sensing
measurement of reflected radiance.
In the laboratory, the Spectralon radiance was measured both before and after each pine/spruce branch sample, and the sample radiances were converted to absolute reflectance by dividing with the Spectralon radiance and multiplying with the reference panel calibration data (from the manufacturer). In the case of snow measurements in the laboratory, the Spectralon radiance was measured at the beginning and at the end of the measurements of samples of the specific snow type, and all the Spectralon radiances for each measurement day were then averaged. In field measurements, the Spectralon was measured before each (snow or lichen/moss) sample and repeated when necessary, e.g. if the illumination conditions changed during one measurement event.
Mast-based (forest opening and pine forest) target radiances were converted
to reflectance by using a Spectralon radiance measurement obtained before
each target observation (the Spectralon is pushed under the measurement head
automatically). The instrument is taken down from the mast for the cold and
lightless midwinter. During this time, dark laboratory tests are conducted
to reveal any substantial changes in the instrument response or possible
degradation of the Spectralon panel due to impurities or exposure to UV
(ultraviolet) radiation. The changes in the reference panel reflectance are
tested by measuring the Spectralon against a similar panel without exposure
to any external stresses. In most cases, these measurements are executed
before and after cleaning the panel (pressure air or sanding under running
water), with the former status of the Spectralon being valid for mast
measurements before and the latter for measurements after the laboratory tests.
The observed mast scene absolute reflectance values (
Airborne radiances measured in March 2010 were converted to reflectances by vicarious calibration. Airborne radiances were compared with the concurrent mast-borne radiances from the forest site, and calibration coefficients were determined for AISA data by using a least-squares fitting technique. To obtain reflectances the concurrent calibrated mast-borne Spectralon radiances were utilized (Heinilä et al., 2014). In 2011, the airborne reflectance level was obtained by applying a real-time fibre-optic downwelling irradiance sensor (FODIS).
The reflectance quantity of all observations discussed here, corresponding
to the atmospherically corrected estimate of surface reflectance from satellite
data, is
The reflectance data given here are calibrated by Eq. (2) to approximately correspond to the bidirectional reflectance factor (BRF). This is the case since the calibration is carried out by a white reference panel approximating a Lambertian surface, and the incoming irradiance is predominantly or totally originating from one (narrow) direction of the illumination source (calibrated lamp or the Sun). Additionally, some observations are obtained under diffuse illumination conditions (full cloud cover providing close to hemispheric isotropic illumination) but using the same calibration procedure with a white reference panel. However, the actual measurement setup represents a biconical configuration when the data have been collected in clear-sky conditions or with a calibrated lamp (direct irradiance) and a hemispherical–conical geometrical measurement setup when the data have been collected in overcast (diffuse) conditions (Shaepman-Strub et al., 2006). Since the calibration is a comparison against a Lambertian surface, the recorded reflectance can show values above one.
In this section we give a short overview of the measurement setups and platforms. In Sect. 3 below, the conditions and processing steps for the data collection are described in more detail. The four platforms included are the laboratory, portable field, mast-borne and airborne setup (Fig. 3). The instrument utilized for the first three platforms is the FieldSpec Pro JR spectroradiometer by ASD (Boulder, Co, USA). The laboratory and field measurements were carried out with the same FieldSpec Pro JR unit, whereas the mast-borne instrument is a fixed installation. The AisaDUAL airborne imaging spectrometer by Spectral Imaging Ltd. (SPECIM; Oulu, Finland) was used on the airborne platform. The technical specifications and details of the setups of ASD and AisaDUAL are described in Table 1. An overview of the measurement systems and targets along with the digital object identifier (DOI) for each dataset are given in Table 2. The overlap of the different measurements conducted over different platforms is presented in Fig. 4. In brief, spectra of snow, pine branches and spruce branches were measured by using the laboratory setup (Fig. 5), whereas snow-on-ground and snow-free ground spectra were obtained with the portable setup (Fig. 7). In both cases the footprint of a single measurement was small, on the order of 4–20 cm (diameter) depending on the measurement optics and the distance between the sensor and the target. Mast-borne forest and forest opening spectra were collected during winter and spring-melt periods with a footprint size of about 14 m in diameter. Selected spectral bands from the airborne AisaDUAL measurements from snow and snow-free ground surveys were added to the published collection. The investigated targets are somewhat different between the four scales, but they are the components of the same investigated boreal landscape.
AisaDUAL flight lines measured in 2010 and 2011, and the measurement points of the portable field and the mast-borne measurements at the FMI-ARC main site. The CORINE Land Cover 2018 classification by the Copernicus programme and a basemap is shown in the background (© National Land Survey of Finland, Esri Finland December 2018).
Technical details of instruments and different installation platforms.
n/a – not applicable
The data record measurement dates and targets. The total number of separate measurements per target type and the number of spectra averaged (consequent spectral acquisitions collected at 1 s interval) for each individual target measurement are presented. The accompanied reference measurements are described in Sect. 3.5. In the last column the digital object identifier (DOI) for each spectral dataset is given.
Continued.
Snow, pine and spruce branch reflectances were measured in laboratory conditions to define the endmember reflectances used in the optical remote sensing of snow (Metsämäki et al., 2005, 2012). The experiments were carried out with the same ASD FieldSpec Pro Jr spectroradiometer that was also used in the field measurements. Reflectances of pine and spruce branches were measured in April 2012. The laboratory measurements of snow reflectances were conducted in the springs of 2013, 2014 and 2015 (Hannula and Pulliainen, 2019). The snow measurements were done for different snow types. The properties of snow were also measured in situ (Table 3 in Sect. 3.5).
Overlap of the different measurements conducted over different platforms. For the mast-borne platform, the observations first cover a spectral range of 350–2500 nm and a range of 350–1000 nm from 2016 onwards.
The pine and spruce branches were collected on the experiment day and placed
inside a fridge until they were measured. Two black boxes were filled with
the branches, one with pine branches and one with spruce branches. A bare
fibre-optic cable was used as a measurement head and spectra were collected at
nadir view angle. The laboratory measurement setup for the spruce spectra
acquisition is illustrated in Fig. 5a. A calibrated tungsten halogen lamp
was used as a light source with a light zenith angle of
Measurement setup
For the snow sample collection, an aluminium sampler of the size of
Examples of the mean spectra measured in the laboratory for pine and spruce branches as well as dry and wet snow types are shown in Fig. 6. The spectra in Fig. 6a and b show the variation in pine/spruce branch reflectance when the sample box was shifted under the measurement head. The standard deviation, defined from the consequent measurement acquisitions collected and averaged for one measurement spectrum, is so small that they are very difficult to distinguish from the plot. In Fig. 6c the standard deviation shows the deviation between different snow samples collected and measured from the same investigated snow type. Although, the experiment setup does not simulate the true conditions accurately, in the laboratory the measurement surroundings can be controlled. By practically removing the effects of changing illumination conditions and diffuse light, it is possible to evaluate the spectral characteristics of the targets. The main findings of the laboratory experiments are the mean and standard deviation of reflectance for dry and wet snow types as well as for boreal pine and spruce branches. These provide reference information to be utilized in the characterization of the spectral endmembers of the remote sensing models.
Reflectance spectra for
The reflectance spectra from snow and the ground underneath the snow cover were measured with the ASD FieldSpec Pro JR, which was also used in the laboratory measurements. Timing of the measurement campaigns was aimed to be both during the cold season and during the melting period, when patches of open ground appear in the snow surface and when the snow properties have higher variation. The measurement targets were characterized with location; landscape characteristics; weather conditions, namely air temperature and cloud cover; and in situ snow properties. Measurements were carried out in the springs of 2010 and 2011 for Sodankylä station and in the spring of 2010 on the grassland site in southern Finland. The data from Sodankylä was collected parallel to airborne campaigns with the AisaDUAL imaging spectrometer.
For field measurements, the spectroradiometer was placed in a polypropylene
case with soft interior padding to protect the instrument during transport.
Additionally, an external battery and a laptop were connected to the
measurement unit. The measurement head was mounted on a camera tripod with
an arm that can extend the measurement head around 40 cm from the centre of
the tripod base (Fig. 7). The tripod was placed with the arm extending towards
the Sun. A second tripod with a bubble level was used to place the
Spectralon panel horizontally under the measurement head for the reference
measurements. At each measurement location the coordinates and general
conditions were logged and the incoming full sky irradiance was measured.
For the reflectance measurements, first the Spectralon reflectance standard
was measured and then the reflectance spectrum of the target, and in
situ measurements of snow were carried out (Table 3). The distance between
the tip of the measurement head and the target (snow surface/ground) was
approximately 45 cm, and the associated full sky irradiance, measured with
the remote cosine receptor (RCR), was also measured from this height. With a 25
Measurement setup for field measurements with the ASD FieldSpec Pro JR spectroradiometer. During the measurement, the operator stepped away and squatted to minimize the effect on the measurements.
Supplementary parameters measured from each snow type condition represented in the portable field measurements and in the snow laboratory experiments.
Measurements were carried out in forests (
In the portable field measurements of reflectance spectrum from snow and the ground underneath the snow cover, the goal was to get better understanding of the variation of the snow reflectance under different snow conditions (e.g. with different snow depths). In Fig. 8 field snow reflectance observations in clear-sky conditions in direct light and in shadow for dry and melting snow and for melting snow with different total snow depths are presented. Observed reflectances drop with increased water content, impurities and larger snow grain size in melting snow. The detection of snow cover in forested areas from optical satellites is also influenced by the shadowing of the ground by trees. The shadows decrease the reflectance considerably. It should be noted that the measurements are of apparent reflectance, i.e. reflectance measured at the Earth-observation instrument and therefore related to the full sky irradiance (Salminen et al., 2009). Depth of the snow pack becomes an important factor when snow cover is at the melting stage and light is passed through to the ground (Salminen et al., 2009). Noise at the water absorption band, characteristic for field measurements, is seen at 1900 nm, where the signal-to-noise ratio is inadequate for meaningful observations. Utilizing field observations, it is possible to study the effect of both the observation geometry and the target properties on the observed reflectance spectra, although controlling the measurement geometry is difficult.
The ASD FieldSpec Pro spectroradiometer was installed on a 33 m high mast in the intensive observation area (IOA) of the FMI-ARC for the optical remote sensing validation studies (project NorSEN, Nordkalotten Satellite Evaluation co-operation Network). The mast observations allow the evaluation of at-satellite reflectances in the spatial scale of the satellite image pixels. The dataset covers the spring time periods between 2010 and 2018. In 2010–2012 measurements were collected by an operator at hours 10:00, 12:00 and 14:00 UTC (Coordinated Universal Time) during clear-sky or full-cloud-cover conditions. Additionally, measurements were made more frequently during specific measurement campaigns. The system was automatized during summer 2012, and after that spectral measurements have been collected yearly from February to November every 30 min from 06:00 until 15:00 UTC. As the climatic environment during the snow season is challenging, the measurement pole was fixed over the forest target on 26 August 2013 due to frequent problems with the turning motor. After 29 September 2015 data are only available between 350 and 1000 nm because of breaking up of a non-replaceable part of the instrument (Table 2).
The instrument was placed inside a weather-resistant box for protection. The
ASD standard fibre-optic cable was replaced by a longer 5 m cable by the
manufacturer to enable mounting of the measurement head at the end of a
turning pole. The measurement head is a bare fibre-optic cable with a FOV of 25
The fibre-optic head is tilted 11
Field reflectance measurements of snow. Cloudiness for all
measurements was less than 3 octas. Dry snow
Mast-spectroradiometer measurement areas of
One individual spectrum represents one instant measurement acquisition.
During the automatization process threshold values were set to avoid
collection of poor data. No measurements are executed during rain or snow
events, high winds (gust
The mast-borne reflectance spectra for two measurement areas, sparse pine forest and forest opening during spring 2013 were resampled to correspond to MODIS (Moderate Resolution Imaging Spectroradiometer) band 4 (545–565 nm), which is essential for snow mapping from satellites (Fig. 10). In the resampling the corresponding wavelengths from the mast-borne spectra were chosen, and weighted averages calculated by using the relative spectral response function (RSR) provided by the data provider. The time series describes both the diurnal and within-season changes in reflectance. Measurements for clear-sky and diffuse illumination conditions were separated. The observed values for forest opening are high compared to forest until the end of the snow season. The forest opening scene during the full snow cover is composed of a snow field only, whereas in the pine forest area the reflectance is dominated by the forest canopy. The casting shadows from the surrounding trees increase the reflectance variability especially for the forest opening. Considerable diurnal variation in snow reflectance during the snow cover period is also seen in diffuse illumination conditions (Fig. 10b). With automatic measurements the number of observations can be increased.
Mast-spectroradiometer observations from both forest and forest
opening during the spring of 2013 resampled to correspond to MODIS band 4
(545–565 nm) reflectance in
Two airborne spectral imaging campaigns were organized in Finnish Lapland. The purpose was to investigate the effect of forest canopy on optical remote sensing signals from snow-covered surfaces. The first campaign was organized in Sodankylä on 18 and 21 March 2010 and the second in Sodankylä and in Saariselkä on 5 May 2011. In both campaigns airborne hyperspectral data were acquired with the AisaDUAL imaging spectrometer manufactured by SPECIM. The technical details of AisaDUAL sensors are presented in Table 1. The data record contains 10 m resolution reflectance mosaics of the flight lines for the bands 555, 645, 858.5 and 1640 nm for all measurement days and for both (Sodankylä and Saariselkä) study sites (Table 2).
During the first campaign, in 2010, the ground was covered by a thick
(
During the second campaign, in 2011, the spring snow melting was ongoing and
first snow-free patches had appeared. Snow depth varied between 0 and 30 cm at the Sodankylä site and between 0 and 60 cm at the
Saariselkä site. Additionally, more snow-free pixels were found in
Sodankylä than in Saariselkä. The measurement setup followed the
earlier campaign. The measurements were carried out under direct
illumination (cloud cover
The imaging spectrometer data were radiometrically and geometrically
corrected by using the SPECIM's CaliGeo tool in the ENVI software.
Measurements from Saariselkä were additionally corrected with the
digital elevation model KM10 (Finnish national digital elevation model by
the National Land Survey of Finland) with a pixel size of
As an example, Fig. 12 shows reflectance values on 5 May 2011 observed over different land cover types during partial snow cover along the AisaDUAL flight line. At the very end of the spring melting the observed reflectances are relatively low in all land cover types even with 50 %–60 % snow patchiness.
Airborne spectrometer reflectance at band 555 nm on 5 May 2011 at 10 m resolution. An orthophoto from summer time conditions is shown in the background (© National Land Survey of Finland, December 2018).
With each case of measured spectra of snow and open ground targets described
above, reference in situ measurements and observations of weather
conditions, location characteristics and snow properties were conducted to
help to interpret the changes seen in the measured spectra (Table 2). The
portable field measurements and the accompanied in situ data serve as
reference information also for the airborne measurements (overlapping in
time). The prevailing weather conditions were logged while making the
reference observations. These included measurement of air temperature at 2 m height and observation of cloud cover in octas. For the mast
measurements, cloud cover (in octas, Vaisala CT25K laser ceilometer), air
temperature at 2 m (10 min average) and wind speed in gust at 22 m (10 min
maximum) from an automatic weather station were accompanied with the
measured spectra. Since the measurement automatization in 2012, the
The target scene reflectance inside the satellite footprint, recorded by a remote sensing instrument, is a combination of spectral information of several endmembers which complicates the data interpretation. Thus knowledge of the spectral reflectance characteristics of the target endmember (e.g. snow) as well as the combined effect of several contributing endmembers (e.g. forest and open ground) is needed. Data of the same quantity at several scales allows for the accumulation of understanding from reflective properties of an individual tree branch or snow type to scene reflective properties observed at a mast scale to a scale of an optical satellite footprint of several hundred meters. As the sources of error and uncertainty are variable, data at multiple scales also benefit the recognition and quantification of inaccuracies in the remotely sensed information.
The spectroscopy measurements are affected by manifold factors leading to error and uncertainty in the observations and therefore complicating the understanding of the effects of the measured target on the propagation of electromagnetic radiation. These factors stem partly from the instrument characteristics and partly from prevailing conditions. Spectral and radiometric calibration and stability characterization are required to address the effects of the instrumental uncertainties. The optimal sampling procedure appropriate for the considered application should be chosen and the common measurement protocols and standards followed. The imperfections in the reflectance calibration need to be recognized, and the effect of uncontrolled factors, such as changing illumination conditions, should be minimized and documented (Hueni et al., 2017). The measurements of reflectance properties of snow and snow-free ground targets in different spatial scales have enabled the estimation of the systematic error involved in satellite algorithms for snow retrieval (Salminen et al., 2018). In order to use the subordinate scale, the relevant error sources need to be identified and preferably quantitatively estimated. Here the sources of measurement error and uncertainty of the described datasets are discussed.
In the laboratory conditions the measurements are highly controllable. The
external error sources can therefore be minimized. In such conditions the
precision of the measurements can be estimated based on the repeated
measurements of a reference Spectralon panel. The integrated precision is
determined by (Hannula and Pulliainen, 2019)
Correct calibration is essential to obtain high-quality reflectance data. As
such, the uncertainty at all scales of the data record presented here is
related to the uncertainty in the calibration. In the laboratory, this is mostly
related to the imperfect Lambertian characteristics of the Spectralon panel.
Sandmeier et al. (1998) and Rollin et al. (2000) have shown that Spectralon
panels have anisotropic reflectance characteristics depending on view and
illumination geometry. This causes some systematic (
The measurements in field conditions, including the mast-platform and airborne measurements, are more susceptible to changes in the external conditions. The field measurements are affected by the naturally varying illumination, atmospheric composition, and measurement geometry but also by the possible reflective or obstructive objects in the measurement surroundings. In field measurements the observed target (directional) radiation may change without any changes in the target properties if the distribution of irradiation over the hemisphere is changed (Kriebel, 1976). This is due to the anisotropic reflective properties of natural surfaces. Under clear-sky conditions, changes in the incident irradiance are governed by the changes in Sun zenith angle and the optical depth of the atmosphere (Goetz, 2012; Kriebel, 1976). To minimize these effects the frequency of Spectralon measurements should be adjusted according to the stability of the illumination (see Sect. 2.2) and measured near or at the same location as the target (Goetz, 2012; Mac Arthur and Robinson, 2015). In ideal case the measurements are executed around the local noon if the purpose is not to study the effect of changing illumination conditions, as in the mast-borne measurements. Accordingly, any nearby objects, including the observer, will affect the spectral measurements by blocking part of the diffuse irradiance and on the other hand by reflecting the downwelling (direct and diffuse irradiance) and upwelling (reflected from ground) radiance towards the target (Kimes et al., 1983). If the location of these objects remains the same in relation to the target and the Spectralon, no error is produced, but this is rarely the case in the field. This speaks in favour of fixed installations, such as mast-platform, where at least the measurement setup itself remains unchanged (Hueni et al., 2017). In the portable field measurements, the tripod with the extended arm obscured a part of the diffuse skylight illuminating the target. This effect has not been quantified or corrected in our measurements. The airborne measurements are affected by the adjacency effect in the heterogeneous areas where top of atmosphere (TOA) radiance is decreased over bright pixels and increased over dark pixels (Otterman and Fraser, 1979). This can be reduced by calibrating the TOA radiances using surface radiances from the same target as was done for the AISA radiances in March 2010. The effect of the external factors may become mixed with the reflectance variability caused by the target properties, such as snow characteristics, and thus make conduction of field measurements complex. In field measurements, the uncorrected irradiance levels and other external sources of error together with the BRDF (bidirectional reflectance distribution function) characteristics of the target may compensate for each other, resulting in less variable reflectance (Sandmeier et al., 1998). These interactions are target specific and are typically hard to predict (Sandmeier et al., 1998). Thus, the reflectance observed in laboratory conditions can be more reliably interpreted to be originating from the target's properties.
The measurement scale needs to be taken into account when interpreting the results as the chosen sensor-to-target distance combined with spatial heterogeneity of the target may yield very different outcomes (Milton et al., 2009). This was demonstrated in Fig. 6 where a shift of the pine and spruce sample boxes under the measurement head was followed by a clear change in the target reflectance. Accordingly, a change in Sun azimuth angle over an asymmetric surface (such as forest canopy) without change in the target properties will yield a different reflectance value (Kriebel, 1976). Thus, the representativeness of the dataset has to be judged with respect to the temporal and spatial sampling and the aim of the study. Instrument characteristics may introduce uncertainty. Photodiode detectors utilized in spectroradiometers have temperature-dependent sensitivities (Hueni and Bialek, 2017; Starks et al., 1995). In the mast-borne and in the portable field measurements the spectroradiometer was placed inside an insulated box for protection and to decrease the variability of the ambient temperature. The spectroradiometer utilized in laboratory and portable field measurements has been regularly calibrated by the manufacturer. The mast-borne spectroradiometer has been calibrated on a less regular basis, but the changes of the instrument responsivity have been monitored by yearly laboratory tests to reveal any changes in the instrument behaviour. Some instrument characteristics are hard to determine in detail. ASD spectroradiometer FOVs have shown to differ from the nominal FOV reported by the manufacturer, and the sensor responsivity has shown to be nonuniform within the FOV (Mac Arthur et al., 2012). This complicates the understanding of the relationship between the observation and the target in heterogeneous areas (Hueni et al., 2017). These examples illustrate the complexity of the factors affecting the (field) spectroscopy measurements and highlight the need for comprehensive metadata of the measurement sites to assist the data interpretation.
Comparison of observations collected by different platforms is not always
straightforward. Figure 13 presents surface reflectance spectra observed at
different scales for snow-covered lake ice (a) and forest measurement area
of the mast-borne platform during dry snow conditions (b). For Fig. 13b the
mean of pine branch reflectance measurements, measured in a laboratory, is
also shown. The motivation behind the comparison of measurements collected
at different scales is to understand how the band reflectance value measured
for a coarse-resolution remote sensing image pixel is composed for different
types of heterogenous landscapes. The same motivation behind these studies
may give rise to problems in future data analysis. For a homogeneous area with direct and stable illumination conditions, comparing measurements observed
from a height of 800 m and 45 cm and with a spatial resolution of 10 m and
20 cm may give information, for example, from the success of the atmospheric
correction. In Fig. 13a there are some differences in the snow reflectance
observed at different scales, but the airborne values still fit within the
standard deviation observed on the ground in the portable field measurements.
When more heterogeneous surfaces, such as the forested area in Fig. 13b, are
compared, even small differences in the view angles (nadir for airborne AISA
and 11
In the laboratory experiments the aim was to characterize the variation of the spectral reflectivity of pine and spruce and different snow types (i.e. spectral endmembers) with controlled illumination, a characteristic which cannot be reached in field conditions. Since the pine and spruce sample reflectances at this scale can significantly change depending on the orientation of the target in relation to the measurement head, a number of observations with varying orientations were taken to describe the average variance. Both laboratory and field observations can describe only part of the spatial and temporal variability in the targets' spectral reflectance as only a specific number of measurements at some specific times can be measured. With continuous mast measurements a time series of reflectance spectra of the same target area can be constructed offering data to study the changes in the spectra of a specific land cover type in varying illumination and atmospheric conditions and with seasonally varying target characteristics. In comparison, the airborne data provide the variability between several boreal land cover types. Scaling upward with mast- and airborne data records provides one more link between the remotely sensed and pointwise field observations. The presented data record can be considered representative as it is measured with various temporal and spatial resolutions and has the specific advantage of being coincident in time and from the same locale.
The data record has been utilized in several feasibility studies of satellite snow-covered-area mapping, with most of them focusing on forested areas. The changes in portable field spectra due to snow properties were studied by Salminen et al. (2009) and Niemi et al. (2012). They showed that snow wetness had strong effect on the forward scattering due to the increase in the effective grain size in the optical region (Wiscombe and Warren, 1980). This produced high variability in the reflectance spectrum (Niemi et al., 2012). Wet snow transmits light more efficiently, and therefore, during the spring melting conditions, snow depth starts to play a more significant role in altering the reflectance. The mean snow reflectance can drop from 1.00 to 0.7, when a threshold of 20 cm snow depth is crossed (Salminen et al., 2009).
Salminen et al. (2009) used their own pointwise portable field spectroradiometer measurements to statistically characterize the variability of boreal ground reflectance and mast-borne time series to study the comparability of pointwise and scene reflectance measurements aiming at optimal band selection and assessment of accuracy when applying the SCAmod method, an algorithm for FSC detection. They concluded that ground reflectance variability can induce errors up to 10 %–12 % in SCAmod estimations and suggested the use of wavelengths 400–480 nm for SCAmod (and other similar) methods for the best detection of snow. The work was continued by Niemi et al. (2012), who utilized the mast-borne observations of a forest and forest opening to investigate the boreal forest scene reflectance behaviour by means of NDSI (normalized difference snow index), NDVI (normalized difference vegetation index) and MODIS bands during the springs of 2010 and 2011. In the forest opening the band indices were well functioning, but at the forest scene they were strongly affected by the illumination geometry. The study of the spectral index behaviour was continued by investigating the linkage between the scene reflectance and the forest canopy characteristics (coverage, tree height) by concurrent use of field, mast-borne, and airborne spectral measurements and LIDAR data (Heinilä et al., 2014, 2019b). Airborne reflectances from snow-covered surfaces were shown to be highly dependent on forest characteristics. In Pulliainen et al. (2014) the mast-borne measurements from 2010 and 2013 and the airborne data record from 2010 were once again utilized to test a zeroth-order radiative transfer approach for snow monitoring from optical remote sensing data. By means of these data records, the spatial and temporal variability of boreal forest reflectance could be investigated and the model validated at several different scales.
In forthcoming research the mast-borne data record will be further utilized to analyse the representativeness of the mast measurements for the larger boreal forest area in FMI-ARC surroundings and to assess the feasibility of the latest optical satellite data provided at higher, 10–30 m spatial resolution. Spectral data at multiple scales offer a possibility to assess the effect of atmospheric correction applied in remote sensing data processing. Meteorological observations as well as manually and automatically measured snow properties from FMI-ARC have also been used to drive and evaluate snow models (Essery et al., 2016; Ménard et al., 2019). Driving models benefits from (hemispherical) albedo measurements, but also reflectance quantities presented here may be of interest for the snow modelling community. Although the collection and analysis of the spectral data record has been driven by the aim to improve optical snow mapping methods, multiple other possibilities for data usage exist. The mast-borne data can serve as a direct validation or cross-reference information for unmanned aerial vehicle (UAV)-borne spectral measurements, and the spectral range (Table 1) is valid for phenology or vegetation spring green-up studies.
The data record is made available through a community in the Zenodo repository
service (
In order to establish new and improved optical snow mapping methods for boreal forested areas, detailed surveys of satellite scene reflectance contributors are required, as the relatively large satellite footprint may contain both fractional snow and forest cover. The spectral reflectance data record described here contains spectral observations of the main components (i.e. spectral endmembers) of a boreal landscape during spring: snow (dry, wet, shadowed), forest ground (moss, lichen) and forest canopy (spruce and pine, branches) corresponding to the atmospherically corrected estimate of surface reflectance from satellite data. The data record contains comparable observations at laboratory, field, mast-borne and airborne scales partially overlapping in time. In addition, the collection includes reference data collected in situ, along with spectral observations.
The main experimental site for data collection in Sodankylä, northern Finland, and the collection and measurement systems for each scale of data record were described in detail and data examples were given. The possible sources of error and uncertainty were discussed and estimated. So far, the data record has been used for various scientific studies, most of them focusing on the improvement of satellite snow cover detection in forested areas. However, the data record at various scales offers numerous other possibilities for data usage such as cross-reference information for UAV-borne spectral measurements or phenology and vegetation spring green-up studies.
HRH was responsible for the planning, coordination and conduction of the snow laboratory measurements; post-processed the mast-borne data for the period 2012–2018; wrote most of Sects. 3.1, 3.3 and 4–6; and was responsible for the overall writing work of the study as well as the submission process. KH was responsible for the coordination and execution of the pine/spruce branch laboratory experiments, post-processed the mast-borne data for the period 2010–2011 and took part in the collection of portable field spectral measurements. She post-processed the airborne data and wrote most of Sect. 3.4. KB took part in the collection of the portable field spectral measurements and post-processed the data, wrote most of Sect. 3.2, and generated the original idea of the article. OPM took part in the collection of the portable field spectral measurements and the writing work of the study. MS wrote most of Sects. 1 and 2. JP was behind the measurement idea of most of the datasets described and acted as the scientific supervisor of the paper.
The authors declare that they have no conflict of interest.
The data record has undergone some preliminary quality check but any further quality control is left for the user as best fits the purpose.
This work has been supported by the Väisälä foundation, Envibase (Ministry of Finance, Finland), SPECIM (Spectral Imaging Ltd.), and the Airborne Imaging Spectroscopy Application and Research on Earth Sciences (AISARES) graduate school of the University of Helsinki.
This research has been supported by the Maj and Tor Nessling Foundation (grant no. 201500276, 201600013 and 201700417), the European Commission Life+ programme (grant no. ENV/FIN/000133), and the Academy of Finland (grant no. CARB-ARC/285630).
This paper was edited by Jens Klump and reviewed by two anonymous referees.