Interactive comment on “ The Sodankylä in-situ soil moisture observation network : an example application to Earth Observation data product evaluation ” by J . Ikonen

During the last decade there has been considerable development in remote sensing techniques relating to soil moisture retrievals over large areas. Within the framework of the European Space Agency’s (ESA) Climate Change Initiative (CCI) a new soil moisture product has been generated, merging different satellite-based surface soil moisture based products. Such remotely sensed data need to be validated by means of in situ observations in different climatic regions. In that context, a comprehensive, distributed network of in situ measurement stations gathering information on soil moisture, as well as soil temperature, has been set up in recent years at the Finnish Meteorological Institute’s (FMI) Sodankylä Arctic research station. The network forms a calibration and validation (CAL–VAL) reference site and is used as a tool to evaluate the validity of satellite retrievals of soil properties. In this paper we present the Sodankylä CAL–VAL reference site soil moisture observation network, its instrumentation as well as its areal representativeness over the study area and the region in general as a whole. As an example of data utilization, comparisons of spatially weighted average top-layer soil moisture observations between the years 2012 and 2014 against ESA CCI soil moisture data product estimates are presented and discussed. The comparisons were made against a single ESA CCI data product pixel encapsulating most of the Sodankylä CAL–VAL network sites. Comparisons are made with daily averaged and running weekly averaged soil moisture data as well as through application of an exponential soil moisture filter. The overall achieved correlation between the ESA CCI data product and in situ observations varies considerably (from 0.479 to 0.637) depending on the applied comparison perspective. Similarly, depending on the comparison perspective used, inter-annual correlation comparison results exhibit even more pronounced variation, ranging from 0.166 to 0.840.


Introduction
Soil moisture fluctuation plays an important part in water, energy, and carbon cycles.Both latent and sensible heat fluxes, governing the air temperature and humidity boundary layer over land, are affected by variations in soil moisture.For these reasons it plays an important role in the climate system (Legates et al., 2011).Global-scale and high temporal resolution soil moisture observations are urgently needed to monitor the ongoing changes in the climate.Soil moisture is estimated by either in situ measurements, remote sensing techniques (Kerr et al., 2010), or by Earth system models, e.g.ECHAM6 combined with Jena Scheme of Biosphere-Atmosphere Coupling in Hamburg (JSBACH) (Roeckner et al., 2003;Reick et al., 2013).Though several soil moisture networks have been developed all over the world, with most of them included in the International Soil Moisture Network's (ISMN) database (Dorigo et al., 2011), there still exists a lack of high-density soil moisture data in many regions.Therefore, proper initiation of soil moisture in numerical models is often difficult and regional to global-scale projections of soil moisture remain relatively uncertain compared to other variables of the water cycle (Legates et al., 2011;Stocker et al., 2013).
During the last decade there has been considerable development in remote sensing techniques of soil moisture enabling large-scale soil moisture observations (Dorigo et al., 2015).In 2009, the European Space Agency (ESA) launched the Soil Moisture and Ocean Salinity (SMOS) satellite (Kerr et al., 2010), the first space mission dedicated to soil moisture observations.In 2015 NASA's Soil Moisture Active and Passive (SMAP) mission was launched with the same objective (Entekhabi et al., 2010(Entekhabi et al., , 2014)).These advancements have placed soil moisture into one of the 50 essential climate variables collected by the global climate observing system to help the work of the United Nations Framework Convention on Climate Change (UNFCCC) and the Intergovernmental Panel on Climate Change (IPCC) (Dorigo et al., 2015).In that context, the ESA Climate Change Initiative (CCI) soil moisture product has recently been developed, merging different satellite-based and a land surface model surface soil moisture products (Liu et al., 2011(Liu et al., , 2012;;Wagner et al., 2012).Remotely sensed data, such as the ESA CCI soil moisture data product, need to be validated by means of in situ observations in different climatic regions.The multidisciplinary research centre of Sodankylä, in northern Finland, is considered as one of the most complete terrestrial environment monitoring stations in the Arctic region, collecting data from the sub-surface to the upper-atmosphere with continuous observations dating back to 1908.It also hosts a network of automatic soil state observation stations, thus forming a calibration and validation (CAL-VAL) reference site for Earth observation (EO) soil moisture data products.
The objective of this paper is to (1) present the Sodankylä soil moisture observation network established as part of the CAL-VAL reference site, its instrumentation, and areal representativeness, as well as (2) to utilize the data collected by the in situ observation network to evaluate the accuracy of the ESA's CCI soil moisture product over the study site, as an example case study.The procedures for choosing the representative sites for individual soil moisture network stations are discussed, as well as the development of a spatially weighted average estimate of top-layer (approximately 0-10 or 0-15 cm, depending on the soil type) soil moisture over the study area.Comparisons of in situ soil moisture observations against ESA CCI soil moisture (Liu et al., 2011(Liu et al., , 2012;;Wagner et al., 2012) data are presented and discussed through various perspectives.
Ongoing SMOS CAL-VAL activities by means of the Sodankylä in situ soil moisture data set (e.g.algorithm adaptations for organic substrate, soil moisture retrieval under forest, and wetlands) will be the subject of another article planned for the near future.

Sodankylä region soils and landscape
The Sodankylä region represents a typical northern boreal forest/taiga environment; 71 % of the surrounding landscape, within an 80 km radius is forested.While open and forested bogs cover 18 % of the area.The landscape is relatively flat with moderate hills reaching 505 m a.m.s.l.(metres above mean sea level), while the lowest point within the area is at a height of 91 m a.m.s.l.
Soils within the Sodankylä region have only been formed fairly recently, after the last, Weichselian glaciation, period (Sippola and Yli-Halla, 2005;Yli-Halla and Mokma, 2002).Owing to rather weak development due to weatheringresistant felsic parent material and a cool climate, the soils of the Sodankylä region have been classified according to national standards and mapped, primarily by GTK (Geological Survey of Finland) according to grain size and content of organic matter.Little attention has been paid to pedogenic classification.Grain size and organic-matter-based classification serves practical soil-related activities in Finland well; however, this makes it difficult to present soil data from Finland in an international context.During the last few years, soil classes based on the Finnish classification system have been related to the Food and Agriculture Organization of the United Nations' (FAO) World Reference Base (WRB) pedogenic classification systems by MTT (Agrifood Research Finland).We have combined national GTK Quaternary deposits map data with northern Finland MTT pedogenic soil map units, Corine 2006 land cover information, and in situ field campaigns to produce a new detailed pedogenic soil map of the Sodankylä region.
Glacial till (moraine) is the most widespread deposit in the region and is primarily covered by sparse and relatively dense pine forests.Peatlands, formed in depressions in glacial till and moist semi-organic soils with mixed forests (mostly Pine, Spruce, and Birch) are also widespread.podzolization is the primary pedogenic process within the region, although the podzolization process is clearly fairly week in parts of the region.As is the case in other parts of Finland, soils developed on sandy and loamy glacial till are often weakly podzolized and marginally meet the requirements of Podzols of the WRB system (Yli-Halla and Mokma, 2002).
Weakly podzolized, Haplic Podzols, covering approximately 55 % of the area, with varying grain sizes are present through-out the region, except for the central riverbed plain (see Fig. 3).The general distribution of organic soils (Histosols), covers 18 % of the area and clearly follows a northwest to south-east diagonal (see Fig. 4).The northern portion of the area contains larger and more extensive areas of organic and semi-organic (Umbric Gleysol) soils.Semiorganic Umbric Gleysols soils cover 14 % of the Sodankylä area (see Fig. 4).Fine sandy-loam soils, determined in our classification as Haplic Arenosols are found predominantly within the immediate proximity of large rivers, particularly in places where the river channel has changed its shape, exposing old channel bed materials.Haplic Arenosols only cover an area of 1.6 %.Terminal moraines at the edges of the greatest extent of past glaciers have formed deposits of silty parent material for forming Eutric Regosols.As can be observed The measurement principle in Decagon 5TE is based on a capacitance technique in which the sensor supplies a 70 MHz oscillating wave to the sensor prongs that change according to the dielectric constant of the material (Vaz et al., 2013;Decagon Devices Inc., 2014).The CS655 measurement is based on the so-called transmission line oscillation (TLO) technique in which the sensor sends electromagnetic pulses along its two stainless steel rods at a frequency of 175 MHz.The sensor then measures the time period the signal takes to propagate to the end of the rods and back.As the water content in soil increases the propagation velocity of the signal decreases because of increasing dielectric permittivity (Vaz et al., 2013;Campbell Scientific Ltd., 2015).Both sensors have calibration equations that convert the raw EM data to bulk dielectric conductivity (Decagon Devices Inc., 2014;Campbell Scientific Ltd., 2015).The accuracy of the Decagon 5TE soil moisture sensors in mineral soil is reported to be ±0.030cm 3 cm −3 by the manufacturer, while the accuracy of the CS655 sensors in mineral soil is reported to be ±0.025cm 3 cm −3 by the manufacturer.In Vaz et al. (2013) the corresponding accuracies in mineral soil have, however, been found to be ±0.040 and ±0.129 cm 3 cm −3 .The accu-racy of the Decagon 5TE soil moisture sensors in organic soil (> 30 % organic) has been assessed in (Bircher et al., 2016).
In their study an accuracy of ±0.070 cm 3 cm −3 is reported.
The relationship between dielectric permittivity and volumetric water content in mineral soils by Topp et al. (1980) is used to convert sensor output to soil moisture.Based on comparisons with volumetric soil moisture estimates from gravimetric sampling, the Topp equation yields satisfactory results for mineral soils with low organic matter content.Considerable effort has been placed on calibration of the Decagon 5TE soil moisture sensors measuring the volumetric water content in organic layers within the Sodankylä area.As a result of these efforts, a new calibration function has been derived and recently applied to soil moisture measurements in organic layers (Bircher et al., 2016).Equal efforts for the more recently installed Campbell Scientific CS655 sensors are underway and planned for the summer of 2016.
Each station has one vertical measuring profile and two additional horizontal measuring points.The vertical profiles have five sensors placed close to the station at the following depths: −80, −40, −20, −10, −5 cm in mineral soils, and −40, −30, −20, −10, −5 cm in organic bog type soils.The two additional horizontal measuring points have been installed approximately 10 m from the station in opposing directions, in order to catch small-scale variations in soil moisture of the uppermost layer.Both of these measuring points have two sensors at depths of −10 and −5 cm.
Currently eight soil moisture stations have been installed around the Sodankylä CAL-VAL site.See Table 1 for a list of the stations and their soil/land cover types, sensor types, and different measurement calibrations types used.Figures 1 and 2 depict the locations of the CAL-VAL sites.
The options and requirements governing in situ observation site locations have evolved over time from 2011 onwards.In the beginning the aim was to only represent the immediate surroundings of the Finnish Meteorological Institute (FMI) Arctic Research Center.The initial site installations in 2011 reflect this approach and provide relatively good local coverage.During the summer of 2013 an additional  in situ station (UG forest 2) was installed in close proximity to the UG forest 1 station for UG forest 1 station data verification purposes.As such, this in situ site is not very significant when considering its location and does not contribute to increasing the areal representativeness of the in situ observation network.Following wider area field campaigns in 2014 it was acknowledged that the in situ network should be expanded to cover a more diverse range of soil and land cover types.Therefore, two new (HA forest 2 and HP-F forest 1) in situ sites were established.Further new in situ observation sites will be installed during 2016 providing improved areal representativeness of the soil moisture observation network as a whole.This will include the relocation of the UG forest 2 station as mentioned above.

Creation of an area-representative in situ soil moisture average
Based on land cover and soil type information, an arearepresentative average of the in situ observations within the Sodankylä area have been derived for comparison with EO data-based soil moisture estimates, in this case the ESA CCI soil moisture product.One ESA CCI soil moisture pixel, covering an area of 26 km (north-south) by 10 km (west-east), over the Sodankylä CAL-VAL site, was selected.In situ soil moisture weights were defined based on percentages of composite classes of prevailing soil type and land cover information within the area.Soil moisture from the individual network stations was then multiplied with these weights in order to create a spatially weighted average value, representative of the ESA CCI pixel.Table 2 gives an overview over the applied weights.The applied weights were redistributed when new stations were added to the network.
In practice, in reference to our comparison study, only one new station was added; UG forest 2. In order to verify both the representativeness of the in situ observation sites and the accuracy of our soil type classification, field measurement campaigns were conducted between 24 and 25 June 2014 and on 12 October 2015.Field soil moisture measurements were made at 228 points on all major soil types within and below the organic litter layer (O and A horizons) representing depths of approximately 5 and 10 cm on average.The measurements were conducted with hand held ML2X theta probes.Pictures of the top soil profile and surrounding area were also taken.As expected, the field measurements of soil moisture in each soil type exhibit a large degree of local variability.Despite this variability a clear trend in average soil moisture for each soil type can be observed, with heavier (Eutric Regosol and Umbric Gleysol) soils generally holding more water compared to lighter (Haplic Podzol and Haplic Arenosol) soils.The field campaign results show that the Sodankylä CAL-VAL soil moisture network automatic in situ observations correspond fairly well with the average of field campaign soil moisture measurements in each representative soil type (see Fig. 6).Moreover, the field campaigns were also able to distinguish differences in Haplic Podzol water holding capacity, with coarse Haplic Podzol being able to hold on average slightly less water than fine Haplic Podzol (see Fig. 6).
As a result of the field campaigns, as well as the new soil type classification and mapping effort, the need to expand the    we chose the HA forest 1 site on Haplic Arenosol as the surrogate.This site tends to hold more water than the HA open 1 site and although generally not as moist as the HP-F forest 1 site, it, out of all the options, most closely resembles that of the HP-F forest 1 site.Figure 7 illustrates the impact of the new station (HP-F forest 1) installation on spatially weighted average top soil (0-10 cm depth) moisture in comparison to spatially weighted average soil moisture without the new sta- tion between 1 June and 30 September 2015.As expected using the HA forest 1 as the surrogate for Haplic Podzol soils (fine and coarse) results in generally lower soil moisture levels.Figure 7 also illustrates the difference in soil moisture observations between the surrogate; HA forest 1 site, and the actual HP-F forest 1 site independently.Although the difference in the two is quite pronounced, the overall weighted effect is less notable.

The ESA CCI soil moisture product
The ESA CCI soil moisture data represent a homogenized and merged product of surface soil moisture with global coverage.The product currently covers the years 1978 to 2014 with daily time steps at a spatial resolution of 0.25 • , spanning the entire period covered by the individual sensors.The Active ESA CCI soil moisture product is produced by merging scatterometer data, derived from Active Mi- crowave Instrument Wind Scatterometer (AMI-WS) (prior to 2007) and Advanced Scatterometer (ASCAT) instruments.The passive product is based on merging derived soil moisture data from Scanning Multichannel Microwave Radiometer (SMMR), the Special Sensor Microwave Imager (SSM/I), the Tropical Rainfall Measuring Mission's (TRMM) Microwave Imager (TMI), Advanced Microwave Scanning Radiometer -Earth Observing System (AMSR-E), and the Advanced Microwave Scanning Radiometer 2 (AMSR2) instruments (Chung et al., 2014b).
The generation of the ESA CCI soil moisture data set involves three steps: (1) merging the original passive microwave soil moisture products into one product, (2) merging the original active microwave soil moisture products into one product, and (3) blending the two merged products into one final data set together with the GLDAS (Global Land Data Assimilation System) model estimates.The input data sets for generating the merged soil moisture product consist of (1) scatterometer-based soil moisture products, (2) radiometer-based soil moisture products, (3) modelled 0-10 cm soil moisture from the Noah land surface model of the GLDAS version 1.In step 3 the active and passive data sets are blended together by re-scaling both to GLDAS-Noah soil moisture data values with a cumulative distribution function (CDF) matching approach.This imposes GLDAS-Noah model-based absolute value ranges on the original EO observations, but does not have an effect on the original EO data dynamics (Chung et al., 2014b).As part of the steps used to create the merged ESA CCI data product, it is stated that the use of GLDAS-Noah model data to impose absolute soil moisture values to the ESA CCI data product renders statistical comparison metrics, such as root mean square difference (RMSD) and bias, somewhat scientifically meaningless.The ESA CCI soil moisture product should in fact be used, and considered as a reference product for computing correlation statistics, not as an absolute soil moisture content estimate (Wagner et al., 2012).
Active and passive data sets are combined into the merged ESA CCI data product based on data availability and their sensitivity to vegetation (i.e.vegetation density).An average vegetation optical depth (VOD) value is obtained from the land parameter retrieval model (LPRM) (Owe et al., 2008; Meesters et al., 2005).The VOD value is used to identify sparsely vegetated areas from densely vegetated areas (e.g.boreal forests).The active soil moisture product is used in areas with high VOD values (i.e. higher than a predefined threshold), whereas the passive product is used in areas with low VOD values, i.e. in semi-arid and arid regions (Chung et al., 2014b;Wagner et al., 2012).Dorigo et al. (2015) explained in depth the effect of vegetation on soil moisture retrievals using both active and passive instruments.For the Sodankylä region only active microwave data are used, since from the perspective of the ESA CCI product merging algorithm, the Sodankylä region falls within a region with high vegetation density, exceeding the predefined threshold value.

Comparisons and results
Comparisons of top-layer soil moisture around the Sodankylä CAL-VAL site between the years 2012 and 2014 using ESA CCI soil moisture data against in situ network observations were conducted.The comparisons were made against a single ESA CCI data product pixel encapsulating most of the Sodankylä observation sites.Soil type classificationbased map unit areas within the ESA CCI data pixel were used to derive in situ observation spatially representative weights.The respective weights for each in situ observation site are provided in Table 2. Since ESA CCI data are, at the time of writing, only available prior to 2015 our comparisons are made against in situ weights using the HA forest 1 site as a surrogate for the HP-F forest 1 site as described in Sect.3.
Comparisons have been made against both daily average ESA CCI soil moisture estimates data (if more than one overpass occurred on that day) and against weekly running average (in situ and ESA CCI data) values (see Figs. 10-13).Further, we have experimented with applying an exponential filter proposed by Wagner et al. (1999) to relate ESA CCI soil moisture measurements to the used in situ measurement depths.In our application we have used a simple recursive formulation of this method, which is given as where SM n is the filtered ESA CCI soil moisture estimate and CCI is the unfiltered ESA CCI data product to which the exponential soil moisture filter is applied to, while the gain K n in time t n is given as (in a recursive form) where T is the temporal variation of soil moisture within the root-zone profile (in days), while K n ranges from 0 to 1 (see Fig. 9).Further information on the theoretical basis and application of the exponential filter used in our study can be found in Brocca et al. (2011).Top soil moisture for each in situ site are approximated from 5 cm depths apart from sites on Umbric Gleysol soil (UG forest 1 and UG forest 2 sites), where the organic layer is typically much thicker than in, e.g., Podzol soils.With these in situ sites, top soil moisture is approximated from depths of 10 cm.Soil moisture comparisons are only conducted during snow-free and thawed periods, as the presence of snow and soil frost interfere with EO data-based soil moisture retrievals (Chung et al., 2014a), such as is the case with the ESA CCI data product.Snow-free and thawed conditions are determined by a data quality filter provided as a part of the ESA CCI product itself (Scipal et al., 2005) combined with direct Sodankylä CAL-VAL site snow observations.Figures 7 and 8 show daily and weekly ESA CCI and in situ soil moisture time series as well as corresponding scatter plots for the years 2012, 2013, and 2014, separately.Corresponding statistical metrics can be found in Table 3.The overall correlation between the daily ESA CCI data product and daily spatially weighted average in situ observations is rather low at 0.479.The best values are achieved in 2012 and 2013 at 0.551 and 0.621, respectively.In 2014 the correlation is very low at only 0.186.The overall unbiased root mean square error (RMSE) is generally relatively low at 0.039, with the highest RMSE value (0.049) observed for 2014.All years exhibit a negative bias ranging from 0.0346 to 0.046.
Application of the exponential filter to daily ESA CCI soil moisture estimates significantly improves correlation results against in situ observations.Application of the exponential filter does not have any effect on bias and the effect on RMSE is insignificant.The degree of improvement in correlation is heavily dependent on the assigned T parameter value (tem-  poral variation of soil moisture within the root-zone profile, in days).In deriving our comparison results we have assumed a T parameter value of 3.5 as suggested by Brocca et al. (2011).We have also included comparison results using a T parameter value of 1 in order to demonstrate the sensitivity of this parameter to the filter.The average overall daily correlation between the ESA CCI data product and spatially weighted average in situ observations improves from 0.479 to 0.629 when assigning a T parameter value of 3.5 and to 0.559 when a T parameter value of 1 is assigned.The improvement in correlation between in situ data and ESA CCI estimates in 2012 ranges from 0.551 to 0.833 depending on which T parameter value is used.Similarly, the improvement in correlation between in situ data and ESA CCI estimates in 2013 ranges from 0.562 to 0.782, again depending on which T parameter value is used.In general correlations  are better with a T parameter value of 3.5.Interestingly the application of the exponential filter has very little effect on correlation results for the year 2014, and, in fact, the correlation is lower when a T parameter value of 3.5 is assigned compared to when a value of 1 is assigned, albeit the effect is rather insignificant.
Averaging ESA CCI soil moisture data from daily to weekly estimates significantly improves both correlation and RMSE, but has little effect on bias.The average correlation between the ESA CCI data product and spatially weighted average in situ observations improves from 0.479 to 0.637 and the average RMSE decreases from 0.039 to 0.026.The improvements in correlation are most pronounced in 2012 and 2013, with an improvement from 0.551 to 0.840 and from 0.621 to 0.813, respectively.Similarly RMSE decreases from 0.034 to 0.015 in 2012 and from 0.033 to 0.020 in 2013.
The temporal correlation statistics between spatially weighted daily in situ data and ESA CCI soil moisture anomalies are very low.Anomaly correlations have been estimated by calculating the difference to a moving average and scaling it to the standard deviation, as in Brocca et al. (2011).Three different anomalies were considered: (1) deviation from 7-day average, (2) deviation from 15-day average, and (3) deviation from 31-day average.The corresponding statistical metrics can be found in Table 4.The low anomaly correlations indicate that the ESA CCI soil moisture product is unable to capture rapid changes in top-layer soil moisture.Interestingly the anomaly correlations are in all cases higher for 2014, although the performance of the ESA CCI soil moisture product is clearly lower according to our comparisons against spatially weighted in situ data in 2014.The higher anomaly correlations for 2014 are, however, not significantly higher, and therefore as such do not provide an explanation to the poor comparison statistics for 2014.In fact, the higher anomaly correlations for 2014 are unexpected since the comparison statistics for 2014 against spatially weighted in situ data are significantly poorer compared to 2012 and 2013.
The general in situ observation soil moisture trend during the summer of 2012 is descending, while switching to an ascending trend towards the autumn period.In 2013 the overall in situ soil moisture observation trend is primarily descending, with the exception of early summer and late autumn.These same general trends are also apparent in the ESA CCI data product, and therefore the correlations for these years are higher, as compared to 2014.It appears that in 2014 in situ observations exhibit larger wet-dry variation compared to the years 2012 and 2013, with no clear overall trend.In situ observation variation and lack of a general trend is, however, not observable in the ESA CCI data product and the general trend, as in all other years is descending, resulting in considerably poorer correlation with both the daily and weekly spatially weighted average in situ observations.In conclusion, it is clear that there is better correlation between ESA CCI soil moisture estimates and daily spatially weighted average in situ observations during years exhibiting a long-term trend in the data and less short-term variation.

Discussion and conclusions
A comprehensive, distributed network of in situ measurement stations gathering information on soil moisture, as well as soil temperature, has been set up in recent years at the FMI Sodankylä Arctic research station.The network forms a tool to evaluate the validity of satellite retrievals of soil properties.The applicability of point-scale measurements to represent an area covered, especially, by coarse-scale passive microwave sensors is always a subject for debate; however, the employed land-cover sensitive scheme in this study addresses the issue by applying spatial weight factors for mea-surement stations according to their relative representativeness following a new soil type classification.The classification will also be used to assist in the planning for installation of new stations in the near future, with the aim of achieving improved representativeness of the region as a whole.
In our study, soil moisture measurements network data was compared in a land cover sensitive scheme to the blended soil moisture product of the ESA soil moisture CCI initiative over several summer periods.While the overall achieved correlation for the daily ESA CCI was low (0.479), this was largely the result of a single year of observations (2014) with poor correlation metrics.
The ESA CCI data product is derived from AMSR2 and/or ASCAT observations.In the case of the Sodankylä test site and surroundings, only ASCAT observations have been applied.We assume that what is referred to as top-layer soil moisture, can in fact be considered to refer to soil moisture at a depth of only 1 cm or less.However, this has not been explicitly defined in the description of the ESA CCI data product, or definitively proved to be the case in other studies.Nicolai-Shaw et al. (2015) and Dorigo et al. (2015), however, supported our initial assumption that the ESA CCI data product represents soil moisture data at a depth of 0-5 cm.Furthermore, in both of these examples ESA CCI soil moisture is compared to in situ soil moisture observations made at depths of 5 or 10 cm.
Despite our initial assumption on ESA CCI data representativeness in terms of depth, applying an exponential filter to relate ESA CCI data to depths that correspond with the used in situ observation depths can in our view supply valuable additional information.In general, applying an exponential filter to theoretically improve the relation of ESA CCI soil moisture estimates to in situ observation depths significantly improves correlation metrics.There exists, however, a possible drawback in its application related to the need of an additional parameter, T (temporal variation of soil moisture within the root-zone profile, in days).Ideally this parameter needs to be regionalized and should possibly be dynamic, in that the value of the T parameter should ideally change according to soil moisture conditions below the target filtering depth.The uncertainty of the T parameter value comes with potentially undesirable characteristics in that it can be used incorrectly as a way to fit EO-based soil moisture data to in situ observations.In order to reduce uncertainty and the potential for fitting, a physiographic meaning should ideally be attached to the assignment of this value.The issue of applying an exponential filter to adjust EO based soil moisture data and proper T parameter value assignment could perhaps be a topic for an entirely separate study.
Comparisons of weekly ESA CCI estimates to averaged weekly station readings provided improved metrics in terms of correlation and RMSE.As with the application of the exponential soil moisture filter, it can, however, be argued that the smoothing of noise through averaging data over a particular time period can in some cases simply just be used as a way to artificially improve statistical metrics.However, in various operational activities such as terrain trafficability analysis, inflow forecasting for hydropower production and flood forecasting, identifying a longer-term trend in soil moisture data can be useful.Therefore, evaluating the accuracy of EObased soil moisture data with this perspective is warranted, and in some cases even desirable.
It is apparent from the comparison results that the ESA CCI data product has issues with noise between observation steps.Reduction of ESA CCI data noise by smoothing results to a running weekly average improves correlation against spatially weighted average in situ observations considerably.This could possibly point to issues stemming from ESA CCI data acquisition times.In Sodankylä, and in boreal and sub-arctic zones in general, moisture often condenses over night into morning dew on vegetation.This could interfere with active microwave soil moisture retrievals acquired during the early morning hours.In order to investigate this, we created a subset of comparisons where ESA CCI data collected during morning hours were removed.Comparisons were also conducted against hourly in situ observations corresponding to ESA CCI data overpass times.However, neither one of these comparisons resulted in significant changes in correlation, RMSE or bias.Therefore, it is not clear to us if these shortcomings are related to issues with the retrieval algorithm, scaling issues induced by the GLDAS-Noah model or possible issues associated with the LPRM model.The latter, however, seems less likely according to our findings as diurnal effects of vegetation moisture do not appear to be the cause.The effect of conducting comparisons of spatially weighted daily average in situ moisture against ESA CCI soil moisture data, which is intervallic by nature, was also investigated.However, as with the removal of observations conducted during morning hours, this also had little effect on any statistical metrics.
The apparent soil-moisture-scale difference (dry bias) between spatially weighted average in situ observations and the ESA CCI data product is significant but not very meaningful.This is true even when it is likely that our weighted in situ average soil moisture level itself somewhat underestimates overall soil moisture content due to the use of surrogate in situ sites, as detailed in previous sections.
Further studies in to the apparent anomalies in the ESA CCI data product and assumption made in producing the data product should be conducted.For example, the validity of using a single, inter-annually constant, correction variable to counter the effect of vegetation moisture interference in soil moisture retrieval should be assessed.Although our findings do not directly point to an issue related to this, it could still explain at least part of the noise and variability of soil moisture estimates between the daily time steps.Further comparisons and analysis of ESA CCI soil moisture data over larger (boreal) areas and longer time spans against both in situ observations and distributed soil moisture model estimates will be performed in future studies.

Figure 1 .
Figure 1.Locations of the Sodankylä study area, FMI's soil moisture CAL-VAL sites, the field campaign measurement sites and the ESA CCI soil moisture data product pixel.

Figure 2 .
Figure 2. Distribution of FMI's soil moisture CAL-VAL sites within the ESA CCI soil moisture data product pixel.The in situ observation sites indicated in yellow (HA forest 2 and HP-F forest 1) where not included in deriving in situ soil moisture comparison results against the ESA CCI data soil moisture product.

Figure 3 .
Figure 3. Distribution of Haplic Podzols within the Sodankylä region (left panel) and typical top soil profiles and vegetation (right panels).

Figure 4 .
Figure 4. Distribution of organic (Histosols) and semi-organic (Umbric Gleysols) soils within the Sodankylä region (left panel) and typical top soil profiles and vegetation (right panels).

Figure 5 .
Figure 5. Distribution of other less dominant soils; exposed bedrock, Leptosols, Eutric Regosols and Haplic Arenosols within the Sodankylä region (left panel) and a typical Arenosol top soil profile and vegetation (right panels).

Figure 6 .
Figure 6.Comparisons of soil moisture data from the June 2014 field campaign measured with hand held Delta-T Theta Probes and from Sodankylä automatic soil moisture network stations, categorized by soil types.The error bars represent the range of field campaign soil moisture measurements, with the blue error bar indicating measurements conducted on 24 June while the red error bar indicates measurements taken on 25 June.As can be observed, the automatic in situ observations on the corresponding days fall well within the average of field campaign measurements for each soil type measured.The only exception to this is UG forest 2 site data, which appears to over estimate soil moisture.Consequentially plans exist to move this site to another location.

Figure 7 .
Figure 7. Illustration of the impact of the new station (HP-F forest 1) installation on spatially weighted average top-layer soil moisture in comparison to average soil moisture without the new station between 1 June 2015 and 30 September 2015.

Figure 8 .
Figure 8.Time series of daily and weekly averaged top soil layer ESA CCI soil moisture data product estimates in comparison against daily and weekly averaged spatially weighted areal average in situ observations for the years 2012, 2013, and 2014.

Figure 9 .
Figure 9.Time series of daily averaged top soil layer ESA CCI soil moisture data product estimates using exponential filters with T parameter values of 1 and 3.5 in comparison against spatially weighted areal average daily in situ observations for the years 2012, 2013, and 2014.

Figure 10 .
Figure 10.Scatter plot of daily averaged top soil layer ESA CCI soil moisture data product estimates using an exponential filter T parameter value of 1 against spatially weighted areal average daily in situ observations for the years 2012, 2013, and 2014.

Figure 11 .
Figure 11.Scatter plot of daily averaged top soil layer ESA CCI soil moisture data product estimates using an exponential filter T parameter value of 3.5 against spatially weighted areal average daily in situ observations for the years 2012, 2013, and 2014.

Figure 12 .
Figure 12.Scatter plot of daily averaged top soil layer ESA CCI soil moisture data product estimate comparison against daily spatially weighted areal average in situ observations for the years 2012, 2013, and 2014.

Figure 13 .
Figure 13.Scatter plot of weekly running average top soil layer ESA CCI soil moisture data product estimate comparison against weekly running average spatially weighted areal average in situ observations for the years 2012, 2013, and 2014.

Table 1 .
(Bircher et al., 2016)entative soil, and land cover types and start of observations as month/year.The sensor types are given as Decagon Devices Inc. 5TE (5TE) and Campbell Scientific Ltd.CS655 (CS655).The last column provides calibration curve information; 1 is factory calibration for mineral soil for 5TE, 2 is custom calibration for organic soil for 5TE(Bircher et al., 2016), and 3 is factory calibration for mineral soil for CS655.

Table 2 .
The respective areal averaging weights used in ESA CCI data comparison/year for each in situ observation site.

Table 3 .
Top soil layer ESA CCI soil moisture data product estimate comparison metrics/year against weighted areal average in situ observations.

Table 4 .
Top soil layer temporal correlation statistics between spatially weighted daily in situ data and ESA CCI soil moisture anomalies.