Fine-Resolved, Near-Coastal Spatiotemporal Variation of ...

3 downloads 10463 Views 4MB Size Report
This study uses GIS-based modeling of incoming solar radiation to quantify fine-resolved spatiotemporal responses of ..... dication from the ESRI (2012) helpdesk that the diffuse ... this spatial cloud variability, even though local topogra-.
1208

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

VOLUME 52

Fine-Resolved, Near-Coastal Spatiotemporal Variation of Temperature in Response to Insolation NIKKI VERCAUTEREN AND GEORGIA DESTOUNI Department of Physical Geography and Quaternary Geology, and Bert Bolin Center for Climate Research, Stockholm University, Stockholm, Sweden

CARL JOHAN DAHLBERG AND KRISTOFFER HYLANDER Department of Botany, Stockholm University, Stockholm, Sweden (Manuscript received 27 April 2012, in final form 28 November 2012) ABSTRACT This study uses GIS-based modeling of incoming solar radiation to quantify fine-resolved spatiotemporal responses of monthly average temperature, and diurnal temperature variation, at different times and locations within a field study area located on the eastern coast of Sweden. Near-surface temperatures are measured by a network of temperature sensors during the spring and summer of 2011 and then used as the basis for model development and testing. The modeling of finescale spatiotemporal variation considers topography, distance from the sea, and observed variations in atmospheric conditions, accounting for site latitude, elevation, surface orientation, daily and seasonal shifts in sun angle, and effects of shadows from surrounding topography. The authors find a lag time between insolation and subsequent temperature response that follows an exponential decay from coastal to inland locations. They further develop a linear regression model that accounts for this lag time in quantifying fine-resolved spatiotemporal temperature evolution. This model applies in the considered growing season for spatial distribution across the studied near-coastal landscape.

1. Introduction Near-surface temperatures vary at fine spatial and temporal scales in a landscape. Many factors affect temperature at a given location and point in time, such as the large-scale climate, the energy input from the solar radiation, and the local soil energy budget (Geiger 1965). In the context of changing climate, the spatial distribution of temperature is likely to also be affected by the temporal changes in different influencing parameters. There is growing recognition among ecologists of a need for finerscale temperature models, as species distribution might be highly modified by small-scale temperature variations in space and time (Trivedi et al. 2008; Dobrowski 2011). Not least topographic heterogeneity affects the microclimate and thus regulates expansion or restriction of ˚ str€ species in a landscape (S€ oderstr€ om 1981; A om et al. 2007; Virtanen et al. 1998) and influences complex

Corresponding author address: Nikki Vercauteren, Department of Physical Geography and Quaternary Geology, Stockholm University, 106 91 Stockholm, Sweden. E-mail: [email protected] DOI: 10.1175/JAMC-D-12-0115.1 Ó 2013 American Meteorological Society

ecohydrological interactions among snowmelt, rainfall, and vegetative uptake (Broxton et al. 2009; Lyon et al. 2008). Climatic variables and their spatiotemporal dependencies are thus important in determining the distributions of plants and animals (Gaston 2003; Thuiller et al. 2004) even if the relationships can be complex and indirect (Hampe and Jump 2011). It is then important to bear in mind that, not only mean values, but also climatic extremes at different temporal scales, can affect species distributions and their changes. For instance, numerous examples of species distribution models find that summer maximum temperatures (Mclaughlin and Zavaleta 2012) or winter minimum temperatures (Ashcroft et al. 2011) limit species. Areas with similar mean temperature but different diurnal temperature ranges can have different ecological conditions (Thompson et al. 1977; Kirschbaum 2004). Identifying which factors influence the diurnal temperature variability across a landscape can therefore be as important as understanding the spatial variation in mean temperature values. Direct beam solar radiation is among the major factors influencing average temperature (Chung and Yun

MAY 2013

VERCAUTEREN ET AL.

2004) and its diurnal oscillations (Dai et al. 1999; Bristow and Campbell 1984). This radiation, however, is itself controlled by other factors like slope aspect, duration of sunshine hours, and shading by adjacent hillslopes (Pierce et al. 2005). These factors can be modeled using GIS tools as presented by Fu and Rich (1999a,b; 2002). Several investigations of temperature distribution have been carried out in mountain environments, focusing on cold air drainage and other phenomena specific to mountain settings (Lundquist et al. 2008; Dobrowski et al. 2009; Pepin et al. 2009). More general temperature distribution studies consider soil moisture, insolation, and stream location (Fridley 2009). On relatively large spatial scales, effects of the presence of a large lake on temporal temperature responses, and how the lake effect can be handled in downscaling or interpolation have been studied by Holdaway (1996) for Lake Superior. In that paper, the author used a 30-km spatial resolution for interpolation over a wider area (400 km), concluding that the lake presence affected temperature only close to the lakeshore. The spatial repartition of measurement points, however, did not allow for a deeper analysis of lake effects on small-scale temperature variation, that is, on a spatial resolution of around 1 km or less. In this study, we investigate microscale temperature variations in space and time (with spatial resolution of ,1000 m and consider also diurnal temporal variations over the growing season) in a forested landscape in Sweden on the coast of the Baltic Sea. In this landscape, we study in particular how topographical differences and distance to the sea combine to affect spatiotemporal temperature patterns. Temperature measurements of high spatial resolution are labor intensive and not practical in a long-term perspective. Yet availability of such data can be a good basis for developing empirical models, based on more readily available parameters, like meteorological station data and/or modeled products. Examples of this approach can be found in Lundquist et al. (2008), Daly and Conklin (2010), and Holden et al. (2011). Holden shows an interpolation approach building on reanalysis data. As pointed out by Huang et al. (2008), several interpolation methods have been developed to derive spatial temperature coverage from a few weather stations. However, weather stations are usually located in flat open areas and not representative of more complex topography that can largely influence small-scale temperature variations. In this study, we have developed and use a network of temperature measurements in the investigated Swedish coastal and forested landscape to assess the impact of topography and the nearby sea on the spatiotemporal variation of local temperature. The study’s focus is on assessing dominant parameters, which can be relatively

1209

easily modeled or measured to derive maps of temperature and its temporal variation at microscales. Such dominant parameters are indeed identified here and used within a simple empirical model for creating temperature maps at 50-m resolution (the spatial resolution of the available digital elevation model). The underlying temperature sampling is then located on slopes of different aspect and orientation and confined to forested localities. With a focus on the temperature spatiotemporal variation in the landscape, and the main factors influencing it, we use data from this sampling network to build on the methodology used by Huang et al. (2008) for a mountain ecosystem and develop it further for use on the present coastal study site, accounting also for the influence of the sea presence. Specifically, the study examines mean monthly temperatures and diurnal temperature ranges for 4 months in the main growing season (June–September), addressing and providing answers to the following two main questions: 1) How does local temperature vary in space and time across the landscape, especially in relation to topography and distance from the sea? 2) What is the time lag between insolation and temperature response, and does it vary in relation to the distance from the sea?

2. Material and methods a. Study location and instrumentation The study area is located in Sweden on the High Coast (H€ oga Kusten) of the Baltic Sea, in the municipalities of Kramfors and H€ arn€ osand (Fig. 1). Because of its rugged terrain all the way out to the coast, and inland deep river valleys, the area is well suited for the present study, since sampling could be done at a wide range of elevations, irrespective of distance to the coast. Instrumentation consists of a total of 63 Maxim 1922L iButton temperature sensors (Hubbart et al. 2005) that are placed in a 2500-m2 area ranging from latitude 62840 to 63810 N and from longitude 178140 to 188330 E. The locations of temperature measurements are chosen to represent prevailing differences in slope orientation and elevation. The digital elevation model of the area and the location and orientation of the temperature sensors can be seen in Fig. 1. Elevation ranges from 0 to 470 m above sea level. The area is characterized by a wide forest cover, ensuring natural shielding for the temperature sensors (Lundquist and Huggett 2008) and a land cover homogeneity across the sampling sites. We collected then near-surface temperature at approximately a height of 1 m. The 1-m height (see also

1210

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

VOLUME 52

FIG. 1. Location and DEM (50-m resolution) of the forested study site.

Fridley 2009) was chosen instead of more commonly used, larger heights—for example, 2 m (e.g., Mahrt 2006; Blandford and Humes 2008; Minder et al. 2010). The reasons for this choice were practical and to get a closer estimate of the air temperature experienced by the herbal layer and ground-living bryophytes. The latter reason was important because a transplant experiment with ground-living bryophytes was conducted in parallel at the same site. Temperature was recorded every half hour between the end of May and the beginning of October 2011. To minimize radiation heating effects on the temperature measurements (Huwald et al. 2009), the sensors were covered by a white shield (not ventilated). The shield consisted of two plastic cups of small size to minimize effects of stagnant air. Validation of the dataloggers has further been performed by Maxim integrated, certifying that the loggers meet the accuracy specifications. Validations were done with reference instrumentation in accordance with the National Institute of Standards and Technology (NIST) (International Certificate of Validation, Maxim integrated). The accuracy of the loggers is 60.58C from 2108 to 1658C. The clock drift follows a hump-shaped curve with a peak slightly over 1 min (month)21 at around 1278C. The lowest values are around

24.3 min (month)21 at 2408C and 23.7 min (month)21 at 1858C (data sheet for temperature logger, Maxim).

b. Solar radiation modeling The major mechanism controlling temperature variation is the energy balance at the surface. The sun warms the surface of the earth by the incoming solar radiation, and this energy is then partly used to warm the air and control its temperature, while some is used for evapotranspiration and some is stored in the ground. The amount of energy used to warm the air depends largely on land cover type (Oke 1987). Furthermore, the amount of solar radiation available at each location in a landscape depends on sky obstruction and surface orientation. The basis of the present analysis is to decompose the effect of topographic variables on local air temperature and its spatial distribution. For any given altitude, prevalence of major surface water bodies is further expected to slow the temporal response of temperature to the incoming solar radiation. We here derive an explanatory model for monthly average temperature and diurnal temperature amplitude, which includes such delay effects, using solar radiation, elevation, and distance from the Baltic Sea as inputs. The incoming solar radiation is modeled with the Solar Radiation Tool of ArcGIS Spatial Analyst [Dubayah and

MAY 2013

VERCAUTEREN ET AL.

Rich 1995; Environmental Systems Research Institute (ESRI)]. The tool accounts for the effect of elevation, surface orientation, sun–Earth geometry, sky obstruction caused by surrounding topography, and atmospheric conditions. It then calculates insolation for each location in a landscape. The inputs are the digital elevation model (DEM) for the study area, and estimates of atmospheric parameters affected by cloud cover. The latter are the atmospheric transmissivity (the fraction of total radiation that passes through the atmosphere) and diffuse proportion (the proportion of diffuse radiation as a fraction of total radiation). The national weather service, the Swedish Meteorological and Hydrological Institute (SMHI), provides monthly data on percentage of cloud cover over the study area, which we can then convert to empirical estimates of atmospheric transmissivity and diffuse proportion. The SMHI dataset is interpolated from a sparse network of point measurements (around 40 stations in the country) and is available on the SMHI website (http:// www.smhi.se/klimatdata/meteorologi/moln). From this dataset, we extract monthly cloud cover across our study area. We further use the Food and Agriculture Organi˚ ngstr€ zation A om–Prescott model to estimate the atmospheric transmissivity t (Yang et al. 2006) and a similar expression for the diffuse proportion d: t 5 0:25 1 0:75s/S and d 5 0:7 2 0:5s/S ,

(1)

where s is the number of sunlight hours and S is the maximal amount of sunlight hours. In our case we approximate the ratio s/S by the monthly percentage of cloud cover. The formulation for d is based on the indication from the ESRI (2012) helpdesk that the diffuse proportion has an inverse relationship to the cloud cover and on the typical values indicated for clear-sky and cloudy conditions. From Eq. (1) and the percentage of cloud cover for each month, we calculate the solar radiation for the months of April–September 2011. Since there are no further data available to investigate cloud cover variability on smaller spatial scales, we cannot further resolve this spatial cloud variability, even though local topography could influence cloudiness and some sites could experience systematically higher or lower cloud cover than others.

3. Results a. Monthly temperature 1) LAPSE RATE AND INSOLATION EFFECTS We first observe the effect of elevation, which ranges between 0 and 400 m in our dataset, on monthly averaged

1211

FIG. 2. Average lapse rates computed for the listed months. The filled circles represent points located inland (.30 km from the coast) whereas the open circles represent points located near the coast. The dry adiabatic lapse rate is shown as a reference. The inserted figure shows the elevation of sampling sites vs the distance to the sea.

temperature. The observed temperature lapse rate ranges between 228 and 258C km21 (Fig. 2; June: 22.38C km21, July: 23.58C km21, August: 24.88C km21, and September: 25.08C km21). These values are obtained through linear regression on measured temperature variations with elevation (represented by the slopes of the lines in Fig. 2). This yields a similar value range to that found elsewhere (Huang et al. 2008; Simoni et al. 2011; Minder et al. 2010; Blandford and Humes 2008). The dry adiabatic lapse rate is 29.88C km21and is also shown for reference in Fig. 2. This study’s sampling setup further ensures that multicollinearity is not present between elevation and distance to sea. In other words, the spatial repartition of the elevation of the measurement points is independent of their distance to the coast (see the inset plot in Fig. 2). Figure 2 also shows how the lapse rate each month relates to two different classes of measurement points, characterized by different distances to the coast. No systematic difference in lapse rate can then be seen between the two classes of sampling points, near to or far from the coast, except perhaps a small such difference being apparent in September. Further statistical tests revealed that estimated lapse rates fell in similar confidence intervals independent on the distance class and coefficient of determination r 2 values were too low to justify different treatment between the two cases (ranging between 4% and 57% depending on the cases).

1212

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

VOLUME 52

FIG. 3. Monthly (30 day) averaged air temperature vs monthly averaged solar radiation of (left) the same month and (right) a month before. The black points show an example for one sampling point. The black line is a linear fit through the values for this particular location. The r 2 coefficients for the regression lines for this particular site location are 13% for the no-lag scenario and 99% for the lagged scenario. The average lagged r2 for all sites is 90%, with standard deviation of 5%.

Furthermore, because the available dataset is limited in time to just the growing season of a single year, this study cannot from these data alone draw any general conclusions regarding seasonal variability of the lapse rate [as can be found for example in Huang et al. (2008)]. In the following, a constant lapse rate of 24.08C km21 will therefore be used for all considered growing season months. Because of the above-discussed absence of spatial correlation between elevation and distance to the sea, this use of a constant lapse rate will have minimal influence on the spatial temperature model results presented below in the paper. We further investigate the relationship between monthly solar radiation and monthly temperature (adjusted with the above-discussed constant lapse rate). Figure 3 shows that the relationship between the monthly averaged solar radiation and the monthly averaged temperature lagged by 1 month is nearly linear, as also found by Huang et al. (2008). The improved relationship between monthly mean solar radiation R and monthly mean temperature T when a lag of 1 month is considered is illustrated well by isolating and emphasizing results for specific one sampling point (black dots in Fig. 3). Resulting multisite statistics of the parameters of the regression equation T 5 aR 1 b between monthly R and monthly T, with a 1-month time lag between them at each site, are 0.09 (0.01, 0.07) and 1.86 (1.43, 0.77) for the mean [standard deviation, coefficient of variation (CV)] of a and b, respectively. The resulting average r2

is 0.90, with standard deviation 0.05 and CV 0.05. The correlation between monthly R and T, and the regression parameters a and b for the best-fit regression equation, vary thus between different spatial locations (between 84% and nearly 100%), and one essential reason for this variation is the influence of the sea, which is discussed and quantified further below.

2) INFLUENCE OF THE SEA We here further investigate the response of the 1-month lagged temperature to the monthly solar radiation. Figure 4 shows the variation of monthly insolation during the growing season (from May until September). The maximum insolation occurs in June, followed by a decrease. The monthly mean temperature variations show similar dynamics to that of the insolation with the time lag of approximately 1 month. However, the decrease of monthly temperature after the maximum (in July) is faster for the points located farther from the sea (gray color in the figure). The presence of a large water body is likely to affect the time lag of the response of temperature to solar radiation. Indeed, the heat capacity of water is much larger than that of soil, meaning that more heat can be absorbed into water than into soil for the same temperature change. Thus, the diurnal cycle of radiation that reaches the surface of the sea is almost completely balanced by a corresponding cycle of energy transport and subsequent storage into the sea (e.g., Oke 1987; Stull 1988). The air temperature over water is by this effect slower

MAY 2013

VERCAUTEREN ET AL.

1213

FIG. 4. Monthly evolution of (left) insolation and (right) mean temperature during the growing season. The lines are averaged for all measurement points located close to the sea (black) and far from the sea (gray).

to react to radiation input than that over soil. This will also impact the air temperature in the landscape near the coast, as shown by Holdaway (1996). To quantify the difference in time-lag magnitude among different points in the landscape, which are located at different distances from the sea, we compute the optimal time lag in days between average temperature and average solar radiation. The procedure is the following: We use a running average (over N days) of temperature and of solar radiation at each location. We then select the number of days N that maximizes the correlation between average radiation and lagged average temperature for each location. In other terms, we first compute the averaged series of temperature T and solar radiation R as N

Ti 5

1 N

i12N21

å

k5i1N

Tk

and

N

Ri 5

lag. The time-lag decrease follows an exponential decay (given in Fig. 5) and is less evident after 20 km from the sea. On the basis of the findings above, we further derive a modified (from the above, simple regression equation) linear model to predict average temperature from average insolation at each location. The time period for averaging is then given from the exponential time-lag

i1N21

å

k5i

Rk

(2)

and then compute the correlation coefficient r N between N N the averaged series, that is, r N 5 corr(T , R ). Note also the time lag between the two sums in Eq. (2). Finally we choose N that maximizes r N. In this process, the daily insolation is obtained by using a constant cloud cover for every day of each month, and calculating radiation over 3-day periods. The period of 3 days is chosen because the sun course typically overlaps on such short time intervals, and because this choice saves computation time. Figure 5 shows that the optimal time lag varies with the distance to the sea, so that it is a few days longer for points located by the seashore than for points farther inland. Figure 6 shows the spatial distribution of the time

FIG. 5. Optimal lag time for mean temperature response to mean insolation. The coefficients of the exponential decay curve are T0 5 4.00, a 5 0.09, and C 5 27.39, and D is the distance to the seashore (km).

1214

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

FIG. 6. Spatial distribution of the optimal time lag (days) for the mean temperature response to the mean insolation. The distance classes for the time lag are computed on the basis of a 1-kmresolution coastline.

decay function shown in Fig. 5, with the spatial distribution of the optimal number of days for the averaging shown in Fig. 6. We compute then for each location: N

N

T 5 AR 1 B 1 elevation 3 LR ,

(3)

where N 5 T0 exp(2aD) 1 C as given in Fig. 5, and LR is the lapse rate, taken as 24.08C km21. Note that the time lag between the solar radiation and the temperature is accounted for through the definitions in Eq. (2). On average between all the different sites, the obtained best-fit values for the model coefficients are A 5 2.37 and B 5 3.70, where the temperature is in degrees Celsius and insolation is in kilowatt hours per meter squared, taken as daily average (24 h). The standard deviation of A is

VOLUME 52

0.16 [yielding CV(A) 5 0.07], and the standard deviation of B is 1.36 [yielding CV(B) 5 0.37]. The slope A of the linear model, which determines the delayed temporal response of average temperature T to a change in average incoming solar radiation R, is thus relatively robust among sites. The constant B, which is a correction of average T at each site for other factors than radiation and elevation, varies more among the different sites. The model Eq. (3) expresses that monthly average temperature T varies primarily in response to incoming solar radiation R as shown in Fig. 7a, with variation around the main response component T 5 AR (line in Fig. 7a) being both spatial among different sites in the landscape and temporal from one month in the growing season to the next at each site. The sea contributes to this variation spatially through site distance to the coast, and temporally through the averaging time N and associated lag time quantified by Eq. (2). Furthermore, topography contributes to the spatial variation of T around its main T 5 AR response component as quantified by the lapse rate term in Eq. (3). Finally, the correction term B in Eq. (3) quantifies additional spatiotemporal contributions to T variation around the main T 5 AR response. The spatial variability of the correction term B may depend on different local factors such as local soil moisture, tree cover, or proximity to forest edge. There is, however, also a clear negative correlation between B and the solar radiation R, as shown in Fig. 7b. A best-fit regression between B and R across all observation sites yields B 5 22.3R 1 14 with r2 5 63%. A closer look at values of B for different slope orientations (Fig. 7b) further shows B implies a generally higher negative correction from the primary T 5 AR response for north-facing

FIG. 7. Relation between B from Eq. (3) and average daily solar radiation R, which is also part of Eq. (3). The symbols stand for north- (N), south- (S), east- (E), and west- (W) facing slopes and flat (F) sampling location. (a) The line shows the slope A from Eq. (3), representing the main response component of temperature to radiation. (b) The line shows a best-fit regression between B and R across all observation sites.

MAY 2013

VERCAUTEREN ET AL.

1215

FIG. 8. Maps of temperature in June, July, August, and September as computed by Eq. (3). The color scale shows the modeled temperature, and the circles show the measured temperature. Note that the color scale is different for each month to increase the readability of the figure.

slopes (with commonly lower insolation) than for southfacing slopes (with commonly higher insolation). Physically, the negative B term correlation to R can be explained by regional airflow and mixing of local air temperatures, which should tend to smooth out any primary R-driven T differences between adjacent north- and south-facing slopes. Since south-facing slopes systematically receive more incoming radiation than north-facing slopes, one should expect a preferential mixing airflow that redistributes heat and thus diffuses temperature differences from the typical main response of warmer primary air above south-facing slopes and colder air above north-facing slopes. The net effect is then quantified by the correction term B, which spatially tends to cool areas (slopes) that receive more and warm areas that receive less radiation. Further analysis will investigate if synoptic wind conditions influence the variation of this correction term and thus support the hypothesis of heat redistribution that is contained in this factor. From Eq. (3), monthly averaged temperature maps can be calculated as shown in Fig. 8 for June, July, August, and September in the considered growing

season. Figure 9 further shows a direct comparison between calculated and observed temperature data, with resulting r2 of 84%, root-mean-square error of 1.28C, and maximum absolute error of 3.38C. This relatively good performance of the model is largely due to a good representation of the seasonal cycle of temperature for each individual site, whereas the model performance with regard to spatial variability in a given month is more limited, as quantified by the relatively large spatial variability expressed by CV(B) 5 0.37. The r2 values for each month taken separately decrease to 22%–24% depending on the month, illustrating the spatial limitations of the model. Possible causes for that are discussed later in the paper.

b. Diurnal temperature amplitude Besides mean monthly temperatures, quantities of interest are also the diurnal variations of temperature, for instance in calculations of (growing, melting, and other types of) degree-days (Higley et al. 1986; McMaster and Wilhelm 1997; Hock 2003). With regard to this smallerscale temporal variability, we further investigate here if

1216

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

FIG. 9. Comparison of observed and modeled averaged temperatures (averaged over N days as given in Fig. 5) for all sites and months (June–September).

a similar method to that discussed above for monthly temperature can also be used to quantify the variability and understand the spatial pattern of the diurnal temperature signal. The diurnal temperature range (DTR; maximum minus minimum temperature) is strongly related to the diurnal cycle of net radiation at the surface (Dai et al. 1999; Bristow and Campbell 1984). Indeed the daily maximum temperature is responding to shortwave radiation, whereas the longwave radiation controls the minimum

VOLUME 52

temperature to a large extent. DTR has been shown to be a reliable estimate for surface solar radiation (Makowski et al. 2009). Figure 10 shows an example of the data that were collected for two sites, one located close to the sea and one farther inland. The data are first averaged over 1 month for each time step (with the frequency of 30 min that was used for data storage), and we show a monthly averaged diurnal cycle. This way of averaging the data shows the periodicity of the diurnal signal and dampens the effect of the advection of cold air masses in the monthly amplitude of the temperature signal. The spatial distribution of calculated DTR is shown in Fig. 11. The map shows larger DTR inland than nearer to the coast. A similar effect was also noted by Geerts (2003) with a global perspective, finding a coastal fringe of reduced DTR values in Australia, with an exponential DTR decay occurring over the 50 km closest to the sea. This decay distance is similar to the one observed here for the time lag between solar radiation and temperature response (Fig. 5). Geerts (2003) suggests that the coastal DTR decay is due to the inland evolution of low-level cloudiness, which affects both the shortwave and the longwave radiation, and therefore the energy balance and the minimum and maximum temperature. Our dataset does not include measurement points farther inland than 70 km from the sea. Over this study area, DTR shows a clear dependence on the distance to the coast; however, the scatter does not support an exponential increase of DTR when moving inland. No effect of elevation is observed on this parameter and we derive the following linear regression model for it: DTR 5 E 1 FR 1 G 3 Dist,

FIG. 10. Example of diurnal temperature variation at two localities for the month of June for (left) a site located inland and (right) a site located by the seashore. Each half-hourly point is averaged over the month of June.

(4)

MAY 2013

1217

VERCAUTEREN ET AL.

FIG. 11. Amplitude of the diurnal temperature variation averaged for the month of June based on daily solar radiation averaged over the same period. The color scale shows the modeled daily solar radiation, and the red circles show the measured temperature.

where the overbar stands for monthly averages, and Dist is the distance to the coast. The optimal regression coefficients for each month are shown in Table 1 along with corresponding r2 values. The parameters of this linear regression are relatively stable from month to month. This result is encouraging for the possible applicability of the regression model for calculation of the spatial distribution of monthly DTR. Figure 12 exemplifies a map of DTR calculated for the month of July with Eq. (4).

4. Discussion One of the challenges facing climatologists is to match different scales of climate data or model results in a meaningful way (Yang et al. 2011). In particular, climate models provide outputs with coarse resolution (typically 100 km). The subsequent downscaling of climatic trends requires understanding of local geographical influences

on the climatic variables (Tiet€ av€ ainen and Tuomenvirta 2010; Vajda and Ven€ al€ ainen 2003). Few studies have been able to fully document spatial climatic variability across a wide range of scales because of the scarcity of meteorological stations in many topographically complex and relevant areas. Yet such areas typically exhibit important microclimatic features, such as large temperature variation (Mahrt 2006; Simoni et al. 2011; Chen et al. 2007). Our analysis departed from the documented influence of solar radiation on temperature patterns in topographically complex regions (Huang et al. 2008), which was here further elaborated in a coastal area context. The found spatiotemporal temperature responses to solar radiation show a lag of approximately 1 month, which has to be considered in temperature distribution estimations. As the energy budget of a water body differs from that of soil, more energy is stored in the water than in the soil, which thereby leads to slower transfer

TABLE 1. The DTR parameters from Eq. (4) and the coefficients of regression for each month. The 95% confidence interval bounds for each parameter are in parentheses. The E is a constant, F is for radiation effects, and G is for the effect of the distance to the sea.

E (–) F (W h m22) G (km) r 2 (%)

June

July

August

September

21.95 (25.1, 1.2) 1.4 3 1023 (0.9 3 1023, 2.0 3 1023) 3.5. 1022 (2.2 3 1022, 4.8 3 1022) 73.0

20.19 (23.2, 2.8) 1.2.1023 (0.6 3 1023, 1.8 3 1023) 2.6 3 1022 (1.3 3 1022, 3.9 3 1022) 64

0.32 (21.5, 2.1) 1.2.1023 (0.7 3 1023, 1.7 3 1023) 2.5 3 1022 (1.4 3 1022, 3.5 3 1022) 68

1.47 (0.4, 2.6) 1.4 3 1023 (0.5 3 1023, 2.2 3 1023) 1.7 3 1022 (0.9 3 1022, 2.4 3 1022) 62

1218

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

VOLUME 52

FIG. 12. Map of DTR in June as computed by Eq. (4). The color scale shows the modeled temperature, and the circles show the measured temperature. Table 1 gives the r 2 coefficients for the modeled temperature for each month.

of heat to the atmosphere from the former than from the latter. In the present study site, the presence of the Baltic Sea affects mean monthly temperature not only above the sea itself, but also in its proximity, and this inland effect can be taken into account through our proposed modeling framework. The main sea effect is then to regulate the time of response of average temperature to average insolation. The fitted parameters of main model Eq. (3) in the present study apply to the specifically investigated growing season, from June to September, of a single year at the considered Swedish site. The importance of varying solar radiation forcing across different seasons has been discussed by Yang et al. (2011) and Pike et al. (2013) for other sites in northern Scandinavia, and continued data sampling over a longer time period at the present site will be used for further model testing and development over a whole year. In this area, the sea freezes in the winter, which could lead to different spatiotemporal temperature patterns and sea influences in wintertime; but also when the sea freezes or melts, the local temperature is affected. Indeed when the sea is freezing, a large amount of latent heat is released to the atmosphere, which is likely to increase air temperature around the coast. In contrast, when the sea is melting, latent heat is absorbed, thereby likely decreasing air temperature along the coast. Such phase change periods are further also expected to modify the time lag of coastal air temperature response to incoming radiation.

With regard to diurnal temporal variability, local differences in radiation budget affect the daily fluctuations between minimum and maximum temperature. However, the present results show that monthly mean DTR can to a large extent be explained by only the distance to the sea and the insolation. During the investigated growing season, the importance of distance to the sea was found to be larger than that of insolation, and the fitted model parameters were found to be relatively stable among the different months in the season. These findings can and will be further tested based on a forthcoming extended dataset, over a longer time period that covers also the other seasons at the site. With regard to spatial variability, the proposed model Eq. (3) shows that mean monthly temperatures do not only vary in response to incoming radiation and its variation across the landscape. In particular, the quantification of correction term B shows that also other variables, such as soil moisture and canopy properties, but perhaps most importantly local airflows, mixing warm and cold air across the landscape, may considerably modify spatial temperature patterns. The low performance of the model equation within each individual month (with an r2 value around 24%) also points to the important limitation of the spatial performance of the model. Furthermore, cloud cover, which was here assumed to be constant over the whole area, may in fact vary considerably in space because of local topographical features that influence how incoming radiation reaches the surface across the landscape.

MAY 2013

1219

VERCAUTEREN ET AL.

5. Conclusions Readily available GIS tools such as the ArcGIS solar radiation analysis tool can estimate solar radiation using an advanced model, based primarily on a DEM, which is easily available for any location at fine spatial resolution. The present study shows that this insolation model can be used to predict in particular the temporal evolution of mean temperature on monthly scale rather accurately, without requiring too much input data. It can further also be used to study spatial temperature variation across a landscape, even though with less degree of accuracy than for the temporal evolution. A monthly percentage of cloud cover over an area of interest can then be sufficient to estimate the atmospheric parameters of transmissivity and diffuse proportion, which are necessary inputs to the insolation model. Restricting, as was done here, the temperature analysis to the growing season, during which snow cover is not present and thereby does not affect the albedo of the surface, implies that solar radiation and elevation are sufficient to explain most of the temporal evolution of temperature. The present results further show that, for near-coastal sites, a lag of approximately 1 month between incoming solar radiation and subsequent temperature response has to be taken into account and is determined by distance to the sea. This effect can be represented by an exponential decay of the time of temperature reaction from the seashore to more inland locations. Our results quantify the time of reaction of mean monthly temperature to insolation to differ by about 5 days between coastal locations and locations farther than 20 km inland. More generally, the present results also constitute a step toward possible development of a framework for fine-resolved downscaling of temperature results from large-scale climate modeling or historic time series of temperature data. This step has shown difficulties in particular with spatial model representation of fineresolved temperature data. Further research should then be directed to investigating how this spatial representation could be fruitfully developed further for facilitating quantification of small-scale spatiotemporal variability of temperature from model results and/or available historic data that are only available on coarser scales. Acknowledgments. This work has been funded by the strategic research project EkoKlim (a multiscale crossdisciplinary approach to the study of climate change effects on ecosystem services and biodiversity) at Stockholm University. The authors thank Norris Lam, Johannes Forsberg, and Liselott Wilin for their help in the field. The anonymous reviewers provided valuable comments on the paper.

REFERENCES Ashcroft, M. B., K. O. French, and L. A. Chisholm, 2011: An evaluation of environmental factors affecting species distributions. Ecol. Modell., 222, 524–531. ˚ str€ A om, M., M. Dynesius, K. Hylander, and C. Nilsson, 2007: Slope aspect modifies community responses to clear-cutting in boreal forests. Ecology, 88, 749–758. Blandford, T., and K. Humes, 2008: Seasonal and synoptic variations in near-surface air temperature lapse rates in a mountainous basin. J. Appl. Meteor. Climatol., 47, 249–261. Bristow, K., and G. Campbell, 1984: On the relationship between incoming solar radiation and daily maximum and minimum temperature. Agric. For. Meteor., 31, 159–166. Broxton, P. D., P. A. Troch, and S. W. Lyon, 2009: On the role of aspect to quantify water transit times in small mountainous catchments. Water Resour. Res., 45, W08427, doi:10.1029/ 2008WR007438. Chen, D., L. Gong, C.-Y. Xu, and S. Halldin, 2007: A highresolution, gridded dataset for monthly temperature normals (1971–2000) in Sweden. Geogr. Ann., 89A, 249–261. Chung, U., and J. Yun, 2004: Solar irradiance-corrected spatial interpolation of hourly temperature in complex terrain. Agric. For. Meteor., 126, 129–139. Dai, A., K. E. Trenberth, and T. R. Karl, 1999: Effects of clouds, soil moisture, precipitation, and water vapor on diurnal temperature range. J. Climate, 12, 2451–2473. Daly, C., and D. Conklin, 2010: Local atmospheric decoupling in complex topography alters climate change impacts. Int. J. Climate, 30, 1857–1864. Dobrowski, S. Z., 2011: A climatic basis for microrefugia: The influence of terrain on climate. Global Change Biol., 17, 1022– 1035. ——, J. T. Abatzoglou, J. Greenberg, and S. Schladow, 2009: How much influence does landscape-scale physiography have on air temperature in a mountain environment? Agric. For. Meteor., 149, 1751–1758. Dubayah, R., and P. M. Rich, 1995: Topographic solar radiation models for GIS. Int. J. Geogr. Inf. Syst., 9, 405–419. ESRI, cited 2012: Area solar radiation. [Available online at http:// webhelp.esri.com/arcgiSDEsktop/9.3/index.cfm?TopicName5 Area_Solar_Radiation.] Fridley, J. D., 2009: Downscaling climate over complex terrain: High finescale (,1000 m) spatial variation of near-ground temperatures in a montane forested landscape (Great Smoky Mountains). J. Appl. Meteor. Climatol., 48, 1033– 1049. Fu, P., and P. Rich, 1999a: Design and implementation of the solar analyst: An ArcView extension for modeling solar radiation at landscape scales. Proc. 19th Annual ESRI User Conf., San Diego, CA, ESRI. [Available online at http://proceedings.esri. com/library/userconf/proc99/proceed/papers/pap867/p867.htm.] ——, and ——, 1999b: TopoView, Version 1.0 Manual. Helios Environmental Modeling Institute, 52 pp. ——, and ——, 2002: A geometric solar radiation model with applications in agriculture and forestry. Comput. Electron. Agric., 37, 25–35. Gaston, K. J., 2003: The Structure and Dynamics of Geographic Ranges. Oxford University Press, 280 pp. Geerts, B., 2003: Empirical estimation of the monthly-mean daily temperature range. Theor. Appl. Climatol., 74, 145–165. Geiger, R., 1965: The Climate Near the Ground. 4th ed. Harvard University Press, 611 pp.

1220

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

Hampe, A., and A. S. Jump, 2011: Climate relicts: Past, present, future. Annu. Rev. Ecol. Evol. Syst., 42, 313–333. Higley, L. G., L. P. Pedigo, and K. R. Ostlie, 1986: DEGDAY: A program for calculating degree-days, and assumptions behind the degree-day approach. Environ. Entomol., 15, 999–1018. Hock, R., 2003: Temperature index melt modelling in mountain areas. J. Hydrol., 282, 104–115. Holdaway, M., 1996: Spatial modeling and interpolation of monthly temperature using kriging. Climate Res., 6, 215–225. Holden, Z. A., J. T. Abatzoglou, C. H. Luce, and L. S. Baggett, 2011: Empirical downscaling of daily minimum air temperature at very fine resolutions in complex terrain. Agric. For. Meteor., 151, 1066–1073. Huang, S., P. M. Rich, R. L. Crabtree, C. S. Potter, and P. Fu, 2008: Modeling monthly near-surface air temperature from solar radiation and lapse rate: Application over complex terrain in Yellowstone National Park. Phys. Geogr., 29, 158–178. Hubbart, J., T. Link, C. Campbell, and D. Cobos, 2005: Evaluation of a low-cost temperature measurement system for environmental applications. Hydrol. Processes, 19, 1517–1523. Huwald, H., C. W. Higgins, M.-O. Boldi, E. Bou-Zeid, M. Lehning, and M. B. Parlange, 2009: Albedo effect on radiative errors in air temperature measurements. Water Resour. Res., 45, W08431, doi:10.1029/2008WR007600. Kirschbaum, M. U. F., 2004: Direct and indirect climate change effects on photosynthesis and transpiration. Plant Biol., 6, 242–253. Lundquist, J. D., and B. Huggett, 2008: Evergreen trees as inexpensive radiation shields for temperature sensors. Water Resour. Res., 44, W00D04, doi:10.1029/2008WR006979. ——, N. Pepin, and C. Rochford, 2008: Automated algorithm for mapping regions of cold-air pooling in complex terrain. J. Geophys. Res., 113, D22107, doi:10.1029/2008JD009879. Lyon, S. W., P. A. Troch, P. D. Broxton, N. P. Molotch, and P. D. Brooks, 2008: Monitoring the timing of snowmelt and the initiation of streamflow using a distributed network of temperature/light sensors. Ecohydrology, 1, 215–224. Mahrt, L., 2006: Variation of surface air temperature in complex terrain. J. Appl. Meteor. Climatol., 45, 1481–1493. Makowski, K., E. B. Jaeger, M. Chiacchio, M. Wild, T. Ewen, and A. Ohmura, 2009: On the relationship between diurnal temperature range and surface solar radiation in Europe. J. Geophys. Res., 114, D00D07, doi:10.1029/2008JD011104. Mclaughlin, B. C., and E. S. Zavaleta, 2012: Predicting species responses to climate change: Demography and climate microrefugia in California Valley oak (Quercus lobata). Global Change Biol., 18, 2301–2312. McMaster, G. S., and W. W. Wilhelm, 1997: Growing degree-days: One equation, two interpretations. Agric. For. Meteor., 87, 291–300. Minder, J. R., P. W. Mote, and J. D. Lundquist, 2010: Surface temperature lapse rates over complex terrain: Lessons from the

VOLUME 52

Cascade Mountains. J. Geophys. Res., 115, D14122, doi:10.1029/ 2009JD013493. Oke, T., 1987: Boundary Layer Climates. 2nd ed. Routledge, 464 pp. Pepin, N. C., M. K. Schaefer, and L. D. Riddy, 2009: Quantification of the cold-air pool in Kevo Valley, Finnish Lapland. Weather, 64, 60–67. Pierce, K. B., Jr., T. Lookingbill, and D. Urban, 2005: A simple method for estimating Potential Relative Radiation (PRR) for landscape-scale vegetation analysis. Landscape Ecol., 20, 137– 147. Pike, G., N. C. Pepin, and M. Schaefer, 2013: High latitude local scale temperature complexity: The example of Kevo Valley, Finnish Lapland. Int. J. Climatol., doi:10.1002/joc.3573, in press. Simoni, S., and Coauthors, 2011: Hydrologic response of an alpine watershed: Application of a meteorological wireless sensor network to understand streamflow generation. Water Resour. Res., 47, W10524, doi:10.1029/2011WR010730. S€ oderstr€ om, L., 1981: Distribution of bryophytes in spruce forests on hill slopes in central Sweden. Wahlenbergia Scr. Bot. Umensia, 7, 141–153. Stull, R. B., 1988: An Introduction to Boundary Layer Meteorology. 1st ed. Springer, 684 pp. Thompson, K., J. P. Grime, and G. Mason, 1977: Seed germination in response to diurnal fluctuations of temperature. Nature, 267, 147–149. Thuiller, W., M. B. Araujo, and S. Lavorel, 2004: Do we need landcover data to model species distributions in Europe? J. Biogeogr., 31, 353–361. Tiet€av€ainen, H., and H. Tuomenvirta, 2010: Annual and seasonal mean temperatures in Finland during the last 160 years based on gridded temperature data. Int. J. Climatol., 30, 2247– 2256. Trivedi, M. R., P. M. Berry, M. D. Morecroft, and T. P. Dawson, 2008: Spatial scale affects bioclimate model projections of climate change impacts on mountain plants. Global Change Biol., 14, 1089–1103. Vajda, A., and A. Ven€al€ainen, 2003: The influence of natural conditions on the spatial variation of climate in Lapland, northern Finland. Int. J. Climatol., 23, 1011–1022. Virtanen, T., S. Neuvonen, and A. Nikula, 1998: Modelling topoclimatic patterns of egg mortality of Epirrita autumnata (Lepidoptera: Geometridae) with a geographical information system: Predictions for current climate and warmer climate scenarios. J. Appl. Ecol., 35, 311–322. Yang, K., T. Koike, and B. Ye, 2006: Improving estimation of hourly, daily, and monthly solar radiation by importing global data sets. Agric. For. Meteor., 137, 43–55. Yang, Z., E. Hanna, and T. V. Callaghan, 2011: Modelling surfaceair temperature variation over complex terrain around Abisko, Swedish Lapland: Uncertainties of measurements and models at different scales. Geogr. Ann., 93A, 89–112.