The effect of weathering on the surface moisture conditions of Norway spruce under outdoor exposure

ABSTRACT Understanding wood surface moisture variations is fundamental to the modelling of moisture content. Exposure to sunlight, fluctuating temperature and rain leads to superficial deterioration and reduced surface hydrophobicity. Here, the effect of weathering was studied through monitoring the surface and global moisture content of axially matched pre-weathered and planed sets of specimens exposed outdoors over 18 months. The difference in surface conditions was characterised by analysing the rain-induced peaks. The results suggest that, following precipitation, weathered surfaces remain wet over a longer time period. No effect was observed through the global moisture content. After one year of exposure, the difference between pre-weathered and planed surfaces was small to insignificant. In service life modelling, it is therefore unnecessary to consider the unweathered state and simulations should target the behaviour of weathered wood. Numerical simulations were able to capture the general behaviour of the surface and global moisture content, but discrepancies over individual peaks were observed. Finally, the study demonstrates how peak analysis and signal analysis can be used for isolating subtle differences between time-series of surface moisture content. The same techniques can be used in future studies to characterise other factors influencing surface conditions, such as wood species and detailing.


Introduction
Wood products employed in exposed environments become subject to a variety of biotic and abiotic deterioration agents. Fungal decay impairs the structural integrity of the material and thereby defines an end of service once the wood product requires replacement or repair. The onset and rate of fungal decay can be described as a function of the wood moisture content and temperature as well as the inherent resistance of the material (Brischke et al. 2006, Brischke andMeyer-Veltrup 2016). Methods for service life design of wood products rely on these dependencies to assess service life performance from hygrothermal conditions (Viitanen et al. 2010, Saito et al. 2012, Isaksson et al. 2013. Wooddiscolouring fungi, photochemical deterioration, and moisture-induced cracks are generally less harmful than fungal decay, but can facilitate the process through affecting the moisture conditions (Žlahtič-Zupanc et al. 2018, Osawa et al. 2019, Sjökvist et al. 2019. Weathering is one of the most frequently observed degradation mechanisms in exposed environments. Wood weathering typically comprises a cascade of events that follow photochemical degradation and leads to discolouration, increased surface roughness and checking (Evans et al. 2008, Žlahtič andHumar 2016a). Ultra-violet (UV), and to a lesser extent, visible light, causes photochemical damage to most compounds in wood, with lignin and extractives being more susceptible (Feist 1989). Damage to the cell wall constituents and leaching of the decomposition products often result in exposing other constituents that, in turn, become subject to degradation from solar radiation (Feist 1989).
The region affected by this cascade of mechanisms is superficial, ranging between 0.25 and 1.0 mm from the exposed surface (Derbyshire and Miller 1981, Wang and Lin 1991, Kataoka et al. 2007), but has been recorded to depths of up to 2.5 mm (Browne and Simonson 1957). Wetting and drying of the affected region tend to also increase the potential of moisture-induced check formation (Zabel and Morrell 2012), which can extend to several millimetres (Evans et al. 2003). The kinetics of the weathering process depends on the climate conditions and surface microclimate during exposure, the most important factors being solar radiation and water (Oberhofnerová et al. 2017). However, the damaged superficial layer largely develops within a few months of exposure in sunny, warm climates (Feist 1989, Evans 2013. Weathering affects wood-water interactions and thereby the service life of wood. The photochemical decay leads to the wood surface becoming more hydrophilic (Feist 1989, Žlahtič andHumar 2016a). The hydrophilic nature of the surface, in conjunction with surface checks, tends to increase moisture uptake (Žlahtič and Humar 2016b, Žlahtič-Zupanc et al. 2018, Keržič and Humar 2021 and facilitate biological growth through increased duration of elevated moisture conditions (Niklewski et al. 2018). These effects have so far not been considered in the modelling of moisture conditions for performance-based design.
The objective of this study was to investigate the effect of weathering on Norway spruce boards subject to natural outdoor exposure and its implications for modelling moisture conditions in the context of service life design of wood. Five sets of specimens (s2-s6) were exposed outdoors at different starting points over a period of 12-18 months. Each set consisted of a pre-weathered reference and two axially matched specimens with their original surface recovered through planing. Surface moisture conditions were monitoring continuously while weight was recorded a few times per month. The characteristic rain-induced peaks in the surface measurements were detected, segmented and analysed to study the effect of weathering and its variation over time. An existing numerical model for analysis of moisture transport in wood, which has frequently been used for decay prediction, was then tested against the dataset. The aim was to understand whether existing models better describe wood having weathered or planed surfaces. Developing a numerical model to explicitly consider the moisture dynamics of the damaged surface layer is outside the scope of the present study.

Sample preparation
Material was obtained from three different glulam beams made of Norway spruce (Picea abies) which had been stored outdoors for a period of two years, during which the weather and wood moisture content were monitored. The top lamellas were first separated by longitudinal cuts. These lamellas were then cut at their finger-joints, resulting in five different boards. Each board was then cut into a set of several smaller specimens, each having a length of 250 mm. Three axially matched specimens from each set without major imperfections (e.g. large cracks) were then selected while the rest were discarded. Two out of the three specimens were planed from the weathered face to a depth of about 5 mm and, turned over and planed from the cut face to a final thickness of 25 mm. The third specimen was planed from the cut face to the same total thickness, thus preserving the original weathered surface. The specimens were then conditioned in a room with a controlled climate (70% relative humidity, 20°C) until equilibrium wood moisture content was reached. Finally, the specimens were equipped with electrodes and sealed on all short sides as well as the back face using a silicone-based adhesive. The specimens were then put back into the climate room until they were mounted on the outdoor rack.

Experimental setup
Five sets of specimens, each set consisting of two planed and one pre-weathered specimen, were subjected to outdoor exposure over a period of 12-18 months. After 18 months, the specimens were covered and kept outdoors for 6 more months.
Specimens were mounted on a rack at an angle of about 30 degrees with one face being exposed to rain. Weather data and wood moisture content were monitored over time. The experiment was started on the 23rd of March 2020 when the first set of specimens (s3) was mounted on the rack. The remaining four sets were delayed by 7 (s4), 21 (s6), 56 (s5) and 212 (s2) days, respectively. Figure 1 shows the test rack after the last set of specimens had been mounted.

Weather data
The variation of weather variables is shown in Figure 2 as daily values, including precipitation, relative humidity, air temperature, wind direction, wind speed as well as globaland diffuse radiation. Radiation was captured with a temporal resolution of 60 min and the remaining variables were captured with a resolution of 10 min. Precipitation was measured by the tipping-bucket-technique, recording every instance of 0.2 mm accumulated water. In addition to the main station located approximately 100 m from the experimental site, precipitation was recorded by two additional stations located onsite and about 500 m from the site, respectively. These stations shared the same settings as the main station (0.2 mm tipping bucket and 10 min resolution) and were used for cross-validation and to substitute any missing data from the main station.
On each occasion when a new set of specimens were added to the rack, the entire rack was sprayed with water for 15 min. The same procedure was then repeated 24 h later. The motivation for this manual intervention was twofold. First, it provided a set of points in time when the entire rack was subjected to a known duration of wetting, as opposed to the measured accumulation of rain. Second, the spraying ensured that the new sets of specimens were exposed to precipitation at least two times prior to the development of notable surface deterioration, thus defining the initial behaviour. Figure 1. The test rack after the last set of specimens was mounted on the rack. The contrast has been increased to highlight subtle differences in colour between the pre-weathered specimens (dark grey) and the planed specimens (lighter grey) and the pair of new specimens (yellowish).

Liquid absorption test of smaller specimens
Smaller specimens (L×W×T=150×45×20 mm) were subjected to a 24-hour floating test, where one face is submerged in water. The test was conducted to document the initial differences in terms of liquid water absorption between the weathered and planed specimens. Twelve smaller pairs of specimens without cracks or knots used here were obtained from the vertical side of the same glulam beams. All sides except for a side-grain (150 × 45 mm) face were sealed using silicone-based adhesive. After conditioning in 65% relative humidity and recording the initial weight, the specimens were put with the unsealed face submerged in distilled water (20°C). The weight of the specimens was then recorded after 5, 20, 60, 120, 240, 480 and 1440 min.

Gravimetric measurements
The weight of the specimens, w tot , was recorded approximately three times per month over the first 12 months of the experiment. Specimens were disconnected from the sensor for about 10 min during this process. When possible, measurements were carried out at times when all surfaces appeared visibly dry. In this state, fine differences in moisture content occurring deeper in the specimens are easier to detect, as they are not concealed by the larger fluctuations occurring near the surface. The oven-dry mass of each specimen, w dry , was recorded at the end of the experiment, at which point the moisture content, u, could be calculated from Equation (1).

Measurements of surface moisture
The electrical resistance of the surface layer was measured every 5 min using a commercial system (Omnisense type S-16) for moisture sensing. The system measures the electrical resistance, which is strongly dependent on moisture content, between two electrodes mounted near the surface. Custom electrodes were produced from stainless steel (A 304) threaded rods (Ø = 2 mm) with sharp-pointed ends. Glued shrink tubing was used to insulate the shaft of the electrodes, leaving only 3 mm of the end uninsulated. The electrodes were installed in pre-drilled holes, using a drill bit of size Ø = 2.5 mm to fit with the shrink tubing and Ø = 1.5 mm for the uninsulated end. To avoid damaging the front face, the smaller hole was first drilled from the front face all the way through the rear side. The larger hole was then drilled from the rear side to a depth of 3 mm from the front face, using the first hole to guide the drill bit. The shrink tubing was coated with silicone adhesive before inserting the electrode to seal any residual gap between shrink tubing and wood. The electrodes were inserted through the rear side until penetrating the front face of the specimen, with 3 mm of the threaded part of the rod being in contact with the wood. The exact location of each electrode pair was selected to avoid proximity to knots and other imperfections which may affect the moisture content locally. The electrode configuration has previously been shown to provide accurate records of surface wetness and moisture content (Niklewski et al. 2018). Each electrode pair was then connected to a sensor by wire and a clamptype connection, so that the wires could easily be detached and re-attached to the specimens. The moisture content was calculated from the measured electrical resistance through a calibration curve developed by Hjort (1996), where the resistance is described as function of wood moisture content and temperature. The wood temperature was assumed to be equal to the air temperature. The error stemming from this assumption is small in general, but increases when the surface is either heated by sunlight or subject to radiative cooling. In general, if the wood surface is warmer than the air temperature, then the sensors will overestimate the moisture content and vice versa. Important to note is that the calibration curve employed here was not calibrated for the over-hygroscopic moisture range, where the moisture-dependency on resistivity is far less pronounced (Stamm 1927). Therefore, measurements above approximately 25-30% can be interpreted as the wood moisture content exceeding the hygroscopic range.
A low-resolution digital camera was set up to record the specimens (20 images/hour) over the first three months of the experiment. The resulting images were assessed qualitatively to ensure that the peaks in surface moisture content matched the period when the surfaces appeared wet.

Filtering and interpolation
First, segments of time-series having unreliable or erroneous data were removed. This type of data was mainly caused by either a loss of contact between the sensor and at least one electrode or alternatively by short-circuiting the two electrodes, both of which could be identified and removed programmatically. Other periods were removed manually due to unrealistic fluctuations in the data stemming from, for example, loose contact between sensor and electrode. The remaining data were then interpolated linearly to even 5 min periods (00:00, 00:05 … 23:55) to enable concatenation of the dataset to a common timetable. During interpolation, continuous periods with more than 60 min of missing data were preserved as gaps. Time-series from each pair of planed specimens were then averaged to a single timeseries. On occasions when data was missing in one timeseries, then the corresponding value from the other specimen was used. Sensors of the same pair never failed simultaneously and therefore all time-series were complete. Since each set had only one pre-weathered specimen, any missing data was instead replaced by the average of all other pre-weathered specimens. The difference between the planed specimens within each set, as well as the difference between the pre-weathered specimens of different sets was very small, and therefore the substitution of missing data did not seem to have a meaningful effect on the analysis. The entire procedure resulted in two fully populated time-series of data from each set, i.e. a total of 10 timeseries.

Detrending
The seasonal variation in relative humidity and temperature leads to a seasonal variation in moisture content of sheltered wood, with higher moisture content in the winter and lower moisture content in the spring and summer. Subtracting the seasonal variation tends to simplify the peak analysis.
Linear regression was used to find a relationship between the moisture content of sheltered wood and relative humidity and temperature. Data used for model fitting was taken from the period when the specimens were sheltered for six consecutive months at the end of testing as well as a subset of longer sequences with no records of precipitation. The resulting model was then used to calculate a baseline Figure 3. Summary of the signal analysis, showing (a-b) the variation in moisture content together with the baseline, (c-d) peaks (markers) detected by the algorithm, (e-f) removal of peak (red markers) not detected in both time-series and (g-h) removal of peak (red markers) not associated with any recorded rain-event. Note that the time, as given on the x-axis, refers to different windows. The moisture content, u (%), is given on the y-axis. seasonal variation during the entire test period. This baseline was subtracted from the moisture content time-series prior to peak analysis. The baseline and the original data are shown in Figure 3a-b for a reference pre-weathered timeseries (a) and a single set of planed specimens (b), respectively.

Peak analysis
Peak analysis was used as a basis for detecting and isolating distinct time-segments where the moisture content increased rapidly and then decreased gradually. These events were typically associated with wetting by a rain event followed by a period of drying, as the presence of water on the surface tends to greatly reduce the electrical resistance between the electrodes. The nature of these peaks, and their variation over time, were used as a basis for studying the difference between pre-weathered and planed surfaces. The procedure for signal analysis is described in the following paragraph.
The default peak detection algorithm of the MATLAB signal processing toolbox was used to approximate the timing and characteristics of the peaks. Prior to peak detection, the measured data were smoothed using a Savitzky-Golay filter to reduce noise. Peaks with a minimum prominence (difference between peak and valley) of 7% moisture content were then detected in the reference pre-weathered time-series and the average of each set of planed specimens. This procedure resulted in five sets of peaks stemming from the pre-weathered specimens and five sets of peaks stemming from the planed specimens.
The peaks from the axially matched pre-weathered and planed specimens were then cross-referenced with respect to their start time to detect and filter out peaks which were found in only one of the two time-series. An example of this can be seen in Figure 3e-f where only the planed specimens ( Figure 3f) dried out prior to the next rain event, resulting in the algorithm detecting two peaks in one time-series and a single combined peak in the other. In this case, the second peak was discarded from the corresponding dataset. Next, the specific rain event associated with the start of each peak was located in time. The recorded time of this rain event was then defined as a common starting point for both peaks. Peaks not associated with any rain event were discarded. An example of this can be seen in Figure 3h where one peak was induced by large fluctuations in relative humidity and possible surface condensation. Finally, the time-period was split into segments by extracting windows bound by the common start of each peak and the latter of the two ends. In summary, the peak analysis results in a dataset of multiple time-segments, which are useful for characterising the difference in behaviour between pre-weathered and planed specimens and how this difference may change over time.

Numerical model
A numerical model based on Fick's second law of diffusion was used to model the experiment, see Niklewski et al. (2018). The same model has previously been used for estimating the exposure in the context of durability assessment, but has never been compared against surface moisture content. The objective here was not to incorporate the effect of weathering in the numerical model, but rather to add a numerical reference to understand the results better.
According to Fick's second law of diffusion, the change in moisture concentration over time can be described in one dimension according to the following equation: where u (kg/kg) is the moisture content and D (m 2 /s) is the diffusion coefficient. The diffusion coefficient describes the rate of moisture transport and is known to depend on both moisture content and temperature. When subject to ambient changes in relative humidity, the moisture content in the outermost fibres will tend towards the equilibrium moisture content given by the sorption isotherm, at which point the vapour pressure is in equilibrium. The boundary condition accounting for changes in relative humidity is thus given by the following equation: where q(kg/(m 2 s)) is the moisture flux, k p (kg/(m 2 Pa s)) is the mass transfer coefficient and (p vw -p v ) is the difference in vapour pressure between the wood surface, p vw (Pa), and the ambient air, p v (Pa). The vapour pressure, p vw , at the wood surface for a given moisture content given by the sorption isotherm. When subject to wetting, a fixed moisture content, u wet , is assigned to the boundary. The model has been shown to be in good agreement with measurements on fast-grown Norway spruce sapwood when, u wet is set equal to 120%. This value corresponds to a liquid absorption coefficient of approximately 4.0 kg/(m 2 s 0.5 ). When tested under outdoor conditions against Norway spruce with unknown properties, the model has been shown to overestimate the effect of precipitation. This is expected since sapwood content contributes to higher permeability (Metsä-Kortelainen et al. 2006). To accommodate for the fact that the material in this study was on the lower end of the scale (as indicated by the liquid absorption experiment), the model was tested with a value of u wet = 60%, in addition to the default value of 120%. The former corresponds to a liquid absorption coefficient of about 2.0 g/(m 2 s 0.5 ). The geometry was modelled in one dimension with a length equal to the thickness of the specimens (25 mm). The top surface was assumed to be freely exposed to the ambient air and to precipitation. For the sake of simplification, the sealant applied to the bottom face and short sides were assumed vapour resistant, and thus a zero-flux boundary condition was applied. The implications of these assumptions were then tested and discussed further in connection to the results.

Results and discussion
Water absorption of small specimens The results of the floating tests are shown in Figure 4. During absorption, the weathered specimen immediately absorbed about 140 g/m 2 with a subsequent absorption rate similar to the planed specimen. The liquid absorption coefficient of the planed specimen is approximately 1.8 g/(m 2 s 0.5 )which is within, but in the lower end of the range (1.35-4.00 g/ (m 2 s 0.5 )) given by Niemz et al. (2010). The results primarily confirm the initial effect of planing on the surface characteristics and short-term moisture absorption and desorption.
Global moisture content variation Figure 5a shows the variation in global moisture content of the planed specimens during the first 12 months of the experiment measured by weight, and Figure 5b shows the difference between the pre-weathered and planed samples based on the same measurements. In general, it took less than a week for the moisture content of a new set of specimens to reach the level of the already installed specimens. For example, the average moisture content of the existing specimens was about 20% in October when the last set of specimens (s2) was mounted on the frame at an initial moisture content of 13%. After 28 h, the recorded moisture content was already at 18%, and after one week the moisture content of the new set was indistinguishable from the others.
A significant and consistent difference between axially matched pre-weathered and planed specimens could not be established from global measurements. It should be noted here that weight was recorded when the specimen surface was dry. As such, the similarity between pre-weathered and planed surfaces observed from the global moisture content is consistent with the hypothesis of weathering primarily being a surface phenomenon. In contrast, Niklewski et al. (2018) demonstrated a subtle difference between weathered and planed specimens also extending deeper into the wood substrate, which would indicate that the surface effect leads to a subtle increase in global moisture content. The cyclic spray sequence used by Niklewski et al. (2018) was, however, designed specifically to target and to induce this type of effect.
Manual spray cycles Figure 6 shows the specimens' response during and after manual spraying. As aforementioned, manual spraying was applied for 15 min every time a new set of specimens were added to the test (left column of subplots), as well as the day after (right column of subplots). The same procedure was repeated at the very end of the test period (bottom row of subplots).
Figure 6a-f clearly demonstrates the expected effect of weathering at the start of the test and after one and three weeks, respectively. The peak of the pre-weathered specimens was both higher in amplitude and longer in duration. A similar difference was previously reported by Niklewski et al. (2018 ) under laboratory conditions. The difference between sets of planed specimens after three weeks is small, even though two sets of specimens were exposed outdoors for 2 and 3 weeks, respectively. After 56 days, when the fourth set of specimens was added to the test, there was a clear difference between the pre-weathered specimen, the three previous sets of planed specimens, and the new set of specimens. The variation in surface moisture content indicates that the three first sets of specimens had become partially weathered, but not yet fully weathered. No clear difference was observed when the final set of specimens was added to the test after 212 days, which can be explained by the fact that all specimens remained wet for a long period. However, a notable difference was observed in the second spray cycle (213 days).

Difference in peak width
The results from the manual spray cycles suggested that the difference between pre-weathered and planed samples diminished over time as the planed samples became increasingly weathered. Figure 7 compares the distribution of wetting time per time-segment for pre-weathered specimens (grey) and four sets of planed specimens (red, blue, green and purple) for three months at the start of exposure (0-3 m) and  after one year (12-15 m), respectively. Wetting time per timesegment is defined as the cumulative duration when the surface moisture content exceeds 25% and is calculated for each time-segment separately. The fifth set of specimens was excluded because it was only exposed for a total of about 12 months and the first three months did not have enough peaks for a meaningful analysis.
In the first time period it can be noted that the peak widths of the pre-weathered specimens are slightly longer. In the second time period, this difference is small or zero, depending on the set. Interestingly, a Wilcox rank-sum statistical test failed to establish a significant (p < 0.05) difference between the planed and pre-weathered sets of specimens in either period. This indicates that the effect of weathering is relatively subtle in comparison with other confounding factors contributing to the variation of the peak width. Figure 6. The variation in surface moisture during and after 15 min of spraying performed on every occasion that a new set of specimens were added to the test, as well as the day after. The data denoted weathered (mean) is an average of all currently available pre-weathered specimens and weathered (new) is the single new pre-weathered specimen added with delay. Figure 7. Boxplot comparing the peak widths of four sets of planed specimens (red, blue, green and purple) with the corresponding pre-weathered specimen (gray) between 0-3 and 12-15 months of exposure, respectively. The box includes the 25th and 75th percentile of the data, the horizontal line marks the median, the whiskers extend to the maximum and minimum (excluding outliers).
It should be noted here that time-segments are of (1) variable length, (2) are associated with unique weather and (3) may include more than one rain event. In the case of the latter, a single time-segment may also include multiple peaks. The variability in conditions in conjunction with uncertainty stemming from the measurement system lead to considerable variation in the dataset.
Reducing the confounding effects to isolate the effect of weathering is done by studying the difference in wetting time between matched pairs of time-segments, which is possible since peaks were matched during the initial signal processing. Figure 8a shows the monthly difference between planed and pre-weathered specimens in terms of wetting time per time-segment. In addition, Figure 8b-f provides some examples of time-series for qualitative evaluation.
During the first months of testing (Apr-July 2020), the time spent above 25% moisture content after wetting was consistently longer in the case of pre-weathered specimens. In fact, out of the 38 unique rain events which occurred during this period, the first set of specimens recorded 26 events with a positive difference larger than 10 min and only a single event with a negative difference larger than 10 min. Prior to the 28th of June, no sensor recorded a negative difference exceeding 10 min. The results are more scattered from July onward, where longer wetting periods are occasionally recorded in planed sets of specimens. Very few peaks were detected during the cold period (November-February). This is explained by the higher relative humidity leading to elevated moisture levels already prior to rain events. As a result, the rapid increase in moisture content associated with rain events became less pronounced.
A Wilcox ranked sign statistical test was able to establish a significant difference in wetting times between planed and pre-weathered during the first three months of exposure for the first four sets of specimens (p < 0.05). Again, as can be seen in Figure 8, the fifth set (s2) did not exhibit enough peaks during the first three months for a meaningful statistical analysis. Between 12 and 15 months of exposure, the same test was unable to establish a significant difference between the pre-weathered and planed samples of the first two sets. A significant difference (p > 0.05) was, however, detected for the third and fourth sets.
The results confirm the inferences stemming from qualitative analysis of the manual spraying cycles, but provide additional information on how the effect of weathering, or rather the effect of planing, diminishes over time. In general, the data indicates that the effect of planing is significant at the start of the test, but decreases rapidly during the first months. After one year of exposure there was no significant difference between pre-weathered and planed specimens of sets s3 and s4. However, a small significant difference in set s5 and a significant difference in set s6 were observed. The observations are generally in line with the literature, stating that most weathering occurs within the first months of exposure (Feist 1989, Evans 2013). Žlahtič and Humar (2016b) did however report an increase in short-term water uptake between 9 and 18 months of outdoor weathering, implying that the effects of weathering continually developed beyond the first 9 months of exposure. The increased uptake was however in part explained by effects stemming from blue-staining fungi, which would explain the longer development.

Numerical simulation
The numerical results have been summarised in Figure 9 together with a subset of experimental data. Figure 9a shows the variation of modelled surface moisture content compared to the average of all five pre-weathered samples and the average of one pair of planed specimens (s4). Figure 9b-d Figure 8. Difference in duration of wetting (u > 25%) between sets of planed wood specimens (s2-s6) and their axially matched weathered references, measured over individual time-segments. Boxes and whiskers mark the 75th percentiles and max/min, respectively, not including outliers. The box is omitted if less than five peaks are detected in a month. The arrowhead markers (in subfigure a) indicate points which reside beyond the axis limits. Subplots b-f show one example of timesegment per specimen, with shaded vertical lines indicating time-steps when precipitation was recorded.
shows a subset of the same data on a different timescale to highlight sets of typical peaks, while Figure 9e shows the cumulative duration of time spent above 25% moisture content. Finally, Figure 9f shows the average moisture content, with measurements based on the gravimetric method.
Qualitatively, Figure 9a shows that the models were able to accurately reproduce the seasonal variation with elevated surface moisture content during the cold months. The timing of distinct periods with frequent peaks, resulting in an overall elevated surface moisture content, were also reproduced.
An overall acceptable agreement between model and measurements is further supported by Figure 9e where the total duration spent above 25% moisture content at the end of the test, as well as the nature of the increase during the period, is captured within the bounds of the two models. However, the figure also shows that the model based on the default value of u wet equal to 120% is more consistent with the measurements compared to the model based on u wet = 60%.
Moreover, Figure 9e shows that the difference between planed specimens and pre-weathered specimens, in terms of cumulative duration spent above 25% moisture content, is negligible. The reason for this is two-fold. First, as discussed in section "Peak analysis", the difference in peak width between the planed samples and the pre-weathered references diminished within a few months of exposure. Second, the difference in peak width between weathered and planed surfaces is small relative to the total time of wetness. Figure 9f shows that the measured global moisture content is captured within the bounds of the model. Contrary to the surface, however, the numerical model with the default value of u wet = 120% generally overestimates the moisture content and the more accurate solution is obtained with a value of u wet = 60%.
It should be reiterated that the model assumes zero interaction between the bottom face of the specimen and the air. Although a very thick layer of surface coating was applied, it is possible that limited water vapour was able to pass. The implication of the assumption was therefore analysed by altering the boundary condition of the bottom face, applying a surface transfer coefficient of k p = 4.0 · 10 −9 kg/(m 2 Pa s) corresponding to silicate paint with low vapour resistance (Fortino et al. 2019). This had a marginal effect near the exposed face, but reduced the average moisture content and therefore improved the performance of the model with a default value of u wet = 120%. To summarise, the assumption regarding the boundary condition of the bottom face did not affect the conclusions.

Conclusions
The following points summarise the main conclusions from this study.
. The effect of weathering on the moisture conditions of Norway spruce was successfully registered through continuous moisture measurements under real climate conditions. . Gravimetric measurements were unable to record a difference in global moisture content between samples of preweathered and planed specimens. . During the first months of exposure, the effect of weathering was shown to have a significant yet subtle effect on the cumulative duration of surface wetness. Figure 9. Summary of the comparison between numerical results (grey area) and surface measurements (red and black lines) and the gravimetric measurements (dots). Subfigure (a) shows the surface conditions over the entire period, (b-d) shows the surface conditions for smaller time windows, (e) shows the cumulative duration when the moisture content exceeded 25% and (f) shows the variation in global moisture content.
. The effect of weathering diminished over time and was either small or insignificant after one year of exposure. Depending on the set of specimens in question, the difference between planed and pre-weathered specimens was either statistically insignificant or considerably reduced after one year of exposure. . Regarding the duration of surface wetness calculated over the entire test period, the difference between pre-weathered and planed specimens was negligible. . Existing numerical models can capture the total time of wetness and its increase over time with reasonable accuracy. However, large errors could be observed upon inspection of specific peaks. . In the context of modelling fungal decay, which can take several years to develop, the initial hydrophobic effect on the surface can be disregarded without introducing major errors. Experiments for characterising moisture dynamics should therefore be based on weathered samples, and similarly, models for moisture prediction should aim to describe weathered wood.
Finally, the study showed how peak analysis and signal analysis are useful for analysing subtle differences between time-series of surface wood moisture content. The same technique can be used to quantify other effects relating to moisture conditions, such as differences between species or detailing. Moreover, while it is more difficult to measure, the surface wood moisture content is a more direct indication of decay compared to other common metrics such as the global moisture content or the moisture content at a certain depth. Afterall, decay generally starts at the surface and is measured at the surface using pick-ratings.