Surface temperature cooling trends and negative radiative forcing due to land use change toward greenhouse farming in southeastern Spain

Abstract

[1] Greenhouse horticulture has experienced in recent decades a dramatic spatial expansion in the semiarid province of Almeria, in southeastern (SE) Spain, reaching a continuous area of 26,000 ha in 2007, the widest greenhouse area in the world. A significant surface air temperature trend of −0.3°C decade⁻¹ in this area during the period 1983–2006 is first time reported here. This local cooling trend shows no correlation with Spanish regional and global warming trends. Radiative forcing (RF) is widely used to assess and compare the climate change mechanisms. Surface shortwave RF (SWRF) caused through clearing of pasture land for greenhouse farming development in this area is estimated here. We present the first empirical evidences to support the working hypothesis of the development of a localized forcing created by surface albedo change to explain the differences in temperature trends among stations either inside or far from this agricultural land. SWRF was estimated from satellite-retrieved surface albedo data and calculated shortwave outgoing fluxes associated with either uses of land under typical incoming solar radiation. Outgoing fluxes were calculated from Moderate Resolution Imaging Spectroradiometer (MODIS) surface reflectance data. A difference in mean annual surface albedo of +0.09 was measured comparing greenhouses surface to a typical pasture land. Strong negative forcing associated with land use change was estimated all year round, ranging from −5.0 W m⁻² to −34.8 W m⁻², with a mean annual value of −19.8 W m⁻². According to our data of SWRF and local temperatures trends, recent development of greenhouse horticulture in this area may have masked local warming signals associated to greenhouse gases increase.

1. Introduction

[2] Anthropogenic changes to the physical properties of the land surface can perturb the climate by altering the Earth's radiative energy balance, and have been regarded as a cause of regional and even global climate change [Sagan et al., 1979]. Furthermore, land use changes are likely to be among the first drivers of climate change at meso- and local scales. Surface albedo affects the shortwave radiation budget by controlling how much incoming solar radiation is absorbed by the surface. Because of this, changes in surface albedo have been suspected of being the dominant influence of mid- and high-latitude land use change on climate [Betts, 2001]. Small changes in Earth's albedo, even below satellite detection limits, can lead to global temperature changes equivalent to those associated with increase in greenhouse gases [Charlson et al., 2005].

[3] Radiative forcing (RF) is a useful concept to assess the relative influence of different human agents on climate change [Forster et al., 2007]. The difference in outgoing shortwave (SW) fluxes between two land uses has been used as an estimation of observational SWRF due to land use change [Betts, 2000]. The local SWRF due to agriculture development is determined by local albedo changes, which depend on the nature of the preexisting vegetation replaced, but also on the reflectivity of the agricultural land. Employing historical reconstructions of croplands, pasture lands and primitive natural vegetation [Ramankutty and Foley, 1999], several estimations of global RF due to surface albedo changes from preindustrial times have been reported [Hansen et al., 1998; Betts, 2001]. These studies simulate the global shortwave radiation budget with radiative transfer models within general circulation models (GCM). In an assessment of these studies, Forster et al. [2007] concluded a best global RF estimate of −0.2 ± 0.2 W m⁻², due to land use related surface albedo change since preindustrial times. Although the level of scientific understanding was raised by these authors to medium-low compared to very low in the previous IPCC report [Ramaswamy et al., 2001], still many uncertainties rise from these estimates. On global RF estimations relative to preindustrial or preanthropogenic times, the biggest uncertainties depend on the characterization of historical and present-day vegetation and surface albedos. While there is general agreement on the albedo associated with grassland, the data sets reveal important differences between the albedo values associated with forest, shrubland and cropland [Myhre and Myhre, 2003]. Myhre et al. [2005a] estimated a global RF of −0.09 W m⁻² due to anthropogenic vegetation change since preagriculture times to present, improving the representation of current surface albedo by using data from MODIS surface albedo data.

[4] For recent land use changes at regional or local scale these uncertainties can be overcome by observational and satellite data available. Few satellite observational estimations of RF based on anthropogenic surface albedo change at local and regional scales have been reported, and few of them quantify the impact of RF on surface temperature trends. Fishman et al. [1994] reported in a field study on the meso-scale cooling effects of high albedo sandy surfaces, compared to surrounding areas. An average impact on the air temperature of 1–2°C cooler during daytime was associated to a mean difference in albedo of approximately 0.40, but no RF estimations were carried out by these authors. In another study, Nair et al. [2007] estimated observational RF values at the top-of-the-atmosphere (TOA) associated with clearing native vegetation for agricultural land use in southwest Australia, using observations from the Clouds and Earth's Radiant Energy System (CERES). Jin and Roy [2005] reported a positive surface RF associated with fire-induced albedo changes over half continental Australia, based on MODIS surface reflectance data. Myhre et al. [2005b] used Meteosat satellite data to calculate RF for changes in the surface albedo from burnt scars due to biomass burning.

[5] The province of Almeria in southeastern Spain (Figure 1) has experienced from the 1970s a rapid development of greenhouse horticulture (GH). From the mid-1980s, the land covered by greenhouses doubled, reaching an area of almost 26,000 ha in 2007 [Sanjuan, 2007] (Figure 2). The coastal plain called Campo de Dalias accounts for 70% of total greenhouse area in the province and has become a continuous greenhouse-covered surface of 18,300 ha in 2007. Before this farming development, semiarid pasture communities had been replacing most previous natural vegetation of semiarid shrubland (Rhamno angustifolii–Mayteneto europaei sigmetum association) [Rivas-Martinez, 1987]. Favorable climatic conditions, due to high insolation (around 3000 hours a⁻¹) and mild temperatures during the growing season (October–May), with absence of frost, have been the basis for the success of greenhouse horticulture in this province, which still continues at an average net growth rate of 500 ha a⁻¹ [Sanjuan, 2007].

Figure 1

Study site. (left) SE Spanish INM Meteorological Stations used as temperature series controls: Murcia (MU), Granada (GR), Malaga (MA) and Almeria (AL) airports. Province of Almeria is shown in grey color. (right) Greenhouse farmland (GH) Campo de Dalias (in white). Agroclimatic Experimental Stations: Mojonera (MOJ) and Palmerillas (PAL).

Figure 2

Growth of land area dedicated to greenhouse farming from 1985 to 2007 in the study site (province of Almeria and coastal flatland Campo de Dalias) (Data from Sanjuan, 2007).

[6] Spatial patterns of temperature change for the Spanish region have been recently described using high quality controlled and homogenized series to elaborate the Spanish Daily Adjusted Temperature Series (SDATS) [Brunet et al., 2006], the 22 most reliable, longest, continuous, homogenized and quality-controlled surface air temperature time series in Spain. Brunet et al. [2007] selected these records to generate the Spanish regional temperature series (STS), composed of the regional mean, maximum and minimum temperatures time series developed from the 22 daily adjusted records. A general and highly significant warming has been observed for the 1901–2005 and 1973–2005 periods, with STS annual mean temperature trends of +0.13 and +0.48°C decade⁻¹, respectively. Three climate stations surrounded by greenhouse development were selected in the province of Almeria as representative of GH farming area. To minimize possible impacts due to subtle circulation changes caused by agricultural land on temperature trends of nearby pasture area selected to determine SWRF, three reliable stations far enough from this area (at least 120 Km away), and with long enough available time series of temperature, were selected as temperature series controls.

[7] Using an empirical approach, here we assess the hypothesis of a causal connection between SWRF associated to land use change toward greenhouse farming, and detected trends in local surface air temperature series of the last twenty years. Our assumption is that change in surface albedo because of rapid GH development has created a response in near surface air temperature that is measurable by weather stations that monitored the farm area throughout the last decades, showing a significant cooling effect. In the null hypothesis of no differential forcing, the long-term climate trends should be very similar between greenhouse and control stations, where no localized forcing is assumed to occur.

[8] MODIS surface albedo data were used to estimate the seasonal change in shortwave radiation budget between both land uses. To minimize differences in outgoing shortwave radiation (OSR) related to other factors than albedo change, such as spatial variability of atmospheric turbidity due to altitude, water vapor, ozone, or aerosols, sectors of well-conserved shrubland adjacent to greenhouse land were selected as representative of semiarid pasture surfaces for albedo and radiation measurements.

2. Data and Area of Study

[9] The study area is known as Campo de Dalias, the widest greenhouse area in the world, and is located on the coast of the Almeria province (SE Spain) (Figure 1). It is a coastal plain with a relatively gentle relief, limited by the Mediterranean at the south and the Sierra de Gador range (above 2000 m high) at the north, and occupies an area of around 33,000 ha. Its climate is Mediterranean, with mild winters and low annual precipitation: average annual temperature and rainfall are 18.8°C and 220 mm, respectively. Detailed description of the study area is provided elsewhere [Castilla and Hernandez, 2005; Pulido-Bosch et al., 2000; Fernandez et al., 2007]. Most greenhouses are called parral type, consisting of low-cost structures covered with a flat layer of plastic (polyethylene) and without heating equipment. This type of greenhouse is considered the archetype of the Mediterranean greenhouse agrosystem, characterized by low technological and energy inputs [Baille, 2002]. During summer months, natural ventilation is not sufficient to extract excess of energy from the greenhouses, so farmers usually whiten the roofs by whitewashing (painting with slaked lime) to reduce incident radiation and avoid excess heating and humidity of growing crops inside. At the end of August, this slaked lime layer is removed to allow enough solar radiation inside for winter and spring crops.

[10] Long-term temperature time series (1950–2006) were obtained from meteorological stations located in Figure 1. All stations selected are outside urban locations, whether in airports or rural experimental stations, so urban heat island effects can be neglected. Two agroclimatic experimental stations in the province of Almeria were selected as GH representative sites: Mojonera (MOJ) at 36°47′N, 2°42′W (Institute for Research and Training in Agriculture and Fisheries IFAPA, Junta de Andalucia), and Palmerillas (PAL) at 36°48′N, 2°43′W (Las Palmerillas–Cajamar Foundation Research Station). These two stations are located inside the coastal plain Campo de Dalias. Almeria airport station (AL) is located by the sea 20 Km east from this main greenhouse plain. Greenhouse facilities have more recently spread at the east of this station, but it is not completely surrounded by them as MOJ and PAL. Granada (GR), Malaga (MA) and Murcia (MU) airports stations, (120 km, 180 km and 170 km away from AL station, respectively), were selected as control stations around the GH area, assuming negligible influence of greenhouse forcing in their climatic data. Records from GR and MA series have been recently used to elaborate the SDATS [Brunet et al., 2006]. No inhomogeneity breakpoints were found by these authors in GR and MA series for the periods considered here. GR is the only inland station and exhibits greater continentality, as it is located 600 m and separated from the sea influence by Sierra Nevada (3400 m high); MA, MU, AL, MOJ and PAL stations lay next to the coast and have a milder Mediterranean climate. The stations in southeastern Spain (MU, AL, MOJ and PAL) are characterized by a more arid climate than GR and MA. MU and AL are first order stations of the Spanish Meteorological Office (INM), and MOJ and PAL are agroclimatic stations included in the cooperative network of the INM. Raw data from all stations have been subjected to different quality controls, mostly gross error checks, internal consistency, temporal and spatial coherence. Data homogenization of series not included in the SDATS was addressed based on available metadata, according to WMO Guidance on Metadata and Homogeneity [Aguilar et al., 2003]. RClimDex software has been used for quality control and RHtest (with 5 years test window) was carried to detect inhomogeneities in stations at the province of Almeria. RHtest is based on a two-phase regression model with a linear trend for the entire series and identifies step changes in station temperature time series [Wang, 2003]. Additionally, no statistical differences were found between MOJ and PAL time series (Kolmogorov–Smirnov test, P > 0.05). Both GH stations lay just 1.8 Km from each other and are highly correlated, as shown by Pearson product–moment correlation (P < 0.05) obtained from their first difference series.

[11] Mean annual values for each record were obtained from monthly means based on daily maximum and minimum surface air temperatures. Records for 1950–2006 were supplied by the INM for AL, MA, MU and GR. Data from 1972 are available for these stations, but there are only common records available from 1983 for both MOJ and PAL. MOJ time series was obtained from IFAPA (Junta de Andalucia), and PAL series from Cajamar Foundation, and were selected as temperature series representative of the GH area (Campo de Dalias). Trend values are the slopes of the least-squares fit lines of mean annual temperatures versus time, in units of °C decade⁻¹. All trends were tested for statistical significance at the 95% level.

[12] In order to remove the variability common to different data sets, difference time series were generated from every pair of single data series, and the significance of the trends of every difference series was then assessed. This method isolates those differences that may be attributed to differences in data set production methods, canceling out a large fraction of the noise that obscures the underlying linear trends [Wigley et al., 2006].

[13] Time series of surface reflectance at 500 m resolution for the area of study were acquired from the MODIS (Moderate Resolution Imaging Spectrometer) instrument on board of the NASA Terra polar orbiting satellite for the period 2001 to 2005. The Surface Reflectance product (MOD09A1) provides surface spectral reflectance estimates for bands 1–7 corresponding to 8 days composites removing atmospheric scattering and absorption effects [Vermote and Vermeulen, 1999].

3. Methodology

[14] Two surface types were selected for the determination of OSR fluxes based on MODIS surface albedo data. The whole sector Campo de Dalias was selected as representative of greenhouse surface for OSR determination. Currently, almost 70% of this coastal plain is covered by greenhouses. The rest of the surface in this plain has been intensely anthropized and it was not possible to find parcels at MODIS resolution to represent the primitive natural vegetation or even the last pasture cover that was cleared from the 1970s for GH development. In order to represent the original pasture use of this coastal plain before the greenhouse development from the 1970s, an adjacent parcel called “Las Amoladeras” was selected (centered at 36°49′, 2°15′), after preliminary screening of other parcels. This is a protected and a well conserved coastal plain that lays 30 Km from the main GH area and was chosen because its current land use and vegetation are very similar to the replaced pastures of Campo de Dalias.

[15] Instantaneous surface broadband albedo (α_si), representative of an 8 day period, was obtained through the equation proposed by Liang [2001]:

where ρ is the reflectivity in the MODIS bands indicated by the subscript. We used the instantaneous surface albedo at the time of satellite overpass (α_si) as an estimate of the daily surface albedo (α_s), as Jacob and Olioso [2005] demonstrated that using instantaneous values at times close to the satellite overpass in place of the daily mean albedo value did not produce significant differences (P < 0.01). Outliers uncorrelated with the time series and with albedo values greater than 0.5, corresponding to cloudy days, where identified and removed.

[16] In order to assess remote sensing albedo estimates, we used available field estimates of albedo obtained from the two surface types. Field measurements over the GH plastic surface were acquired on 19 June 2005, using a GER-2600 (SpectraVista) hand-held radiometer (0.35–2.51 μm) covering 0.5 m². Instantaneous directional albedos (nadir) were estimated at midday (12:00 h) from the ratio of outgoing/incoming radiance integral. Pasture surface albedos have been measured in Almería in a station close to our study site (Latitude: 37°8′N, 2°22′W) between November and December 1997 and in spring 1998 [Domingo et al., 2000], using an albedometer placed 0.5 m above canopy (CM11; Kipp and Zonen, Delft, The Netherlands). Data from both types of cover were then compared with MODIS samples for the same date of spectral measurements acquisition.

[17] Incoming shortwave radiation at the surface (ISR) corresponding to 8 d averages was calculated daily using the solar insolation model POTRAD, which calculates the potential amount of radiation on a surface as a function of elevation, latitude and longitude, solar geometry, slope and aspect of a given site, and takes into account the influence of the surrounding topography and atmospheric transmissivity [Van Dam, 2000]. A transmissivity value of 0.6 was used. Detailed information about clouds distribution, thickness or cloud type was not available for the study area, so clouds were not taken into account by the software. To assess the incoming shortwave radiation at surface estimated with POTRAD model, and the influence of cloud cover in the estimation, we used daily means (8-d averages) for incoming shortwave radiation (Wm⁻²) measured with a pyranometer (CM 6B/7B Kipp & Zonen, Delft, BV), in a field experimental station located 40 km distant from the study site (Latitude: 37°8′N, 2°22′W) between 2000–2005. These data were then correlated with expected insolation modeled with POTRAD for the same site and period.

[18] The outgoing shortwave radiation OSR was estimated by equation (2) from the incoming shortwave radiation ISR and surface albedo α (equation (1)). Atmospheric scattering and absorption of solar radiation of reflected light at the surface were not taken into account in OSR calculations:

[19] Daily shortwave radiative forcing (SWRF) at the surface due to albedo change were calculated as the difference in the daily outgoing shortwave radiation fluxes (OSR) between the two different land use surfaces, with the formula [Betts, 2001]:

[20] Finally, seasonal variability curves for albedo, OSR and SWRF represent a synthetic year generated for every date from the average of the 5 records available for the period 2001–2005.

4. Results

4.1 Surface Air Temperature Trends

[21] Remarkable warming signals have been detected for all time series available from 1972, where a clear trend change occurs after a cooling period during the 50′s and 60′s (Figure 3). Considering the period 1972–2006, control stations GR, MA, and MU have comparable and significant warming trends of around +0.5°C decade⁻¹ (P < 0.05). AL trend however is significantly lower, with +0.37°C decade⁻¹ (Table 1). From the mid 80’s, a clear divergence is shown between temperatures registered at control stations and the two stations in the greenhouse area: MOJ and PAL. While control series maintain high warming trends (around +0.4°C decade⁻¹), MOJ and PAL show significant cooling trends (P < 0.01) of −0.29 ± 0.12 and −0.32 ± 0.11°C decade⁻¹, respectively. No significant differences between these two GH time series were detected (P > 0.05), showing similar temporal evolution. AL series shows no significant trend from 1983 (P > 0.1), and annual mean temperatures seem to have stabilized during this period.

Figure 3

Annual mean surface air temperature anomalies time series for the six stations considered. A moving average of 2 terms has been applied to smooth the data. Anomalies are related to the reference period 1961–1990. MOJ and PAL related to AL average value.

Table 1. Decadal Trends and Standard Errors (in °C decade⁻¹) of Mean Annual Surface Air Temperatures for Every Meteorological Station and for the Periods 1972–2006 and 1983–2006^aa Trend values are estimated to be statistically significantly different from zero (at the 95% level).

T time series	MU	GR	MA	AL	MOJ	PAL
1972–2006	0.54 ± 0.07	0.48 ± 0.08	0.48 ± 0.06	0.37 ± 0.06	N/A^cc Non available data.	N/A
1983–2006	0.43 ± 0.12	0.38 ± 0.17	0.40 ± 0.13	0.08^bb Non available at 95% level. ± 0.13	−0.29 ± 0.12	−0.32 ± 0.11

• a Trend values are estimated to be statistically significantly different from zero (at the 95% level).
• b Non available at 95% level.
• c Non available data.

[22] Difference time series generated from every pair of single data series [Wigley et al., 2006], were used to classify the pairs of stations into three time series (1983–2005) groups (Figure 4), indicating the similarities of the series in every group: control, greenhouse and AL. Very significant differences (P < 0.01) were shown only between control (GR, MU and MA) and greenhouse stations (MOJ/PAL). AL showed again an intermediate behavior, with lower but significant trend differences when paired with either control or greenhouse stations. Trend differences were not significant among control stations difference series (P > 0.8), and between MOJ and PAL (P > 0.5), indicating the high degree of homogeneity of their time series.

Figure 4

Trends of the difference time series between stations for the period 1983–2006. Error bars showing 95% confidence intervals. *Nonsignificant (P > 0.1).

4.2. Seasonal Variations in Surface Albedo

[23] Surface albedo values of greenhouse surface were consistently higher than pasture values in all seasons, with annual average values of 0.28 ± 0.05 and 0.19 ± 0.02, respectively (Figure 5). Albedo values exhibited strong seasonality, with a maximum value for GH of 0.35 measured in summer, and a minimum of 0.20 in winter. Pasture surface albedo showed lower seasonal variation, from 0.22 in summer to 0.16 in winter. Biggest differences in albedo between GH and pasture surfaces of approximately 0.15 were measured in summer, and the smallest difference of 0.05 registered in winter, with a mean annual value of 0.09. It is noticeable the asymmetric shape of GH and difference curves, with a step decrease from maximum summer values and a gradual increase from winter minimum value, not reflected in the pasture curve.

Figure 5

Seasonal variations of broadband albedos of greenhouse and pasture surfaces, and for the difference series (GH-P). Average values and for the period 2001–2005. The temporal variations (standard deviations) are plotted as vertical bars.

[24] A mean daily albedo value of 0.4 ± 0.06 was obtained in the field over plastic GH cover with the spectroradiometer at the available date. In order to compare with MODIS daily albedo estimate, we have to take into account that MODIS GH pixels (1 km²) include minor portions of other types of land use (bare soil, roads, etc) with lower reflectivity than plastic surface. Nonetheless, when MODIS GH pixels (n = 45) with the highest proportion of greenhouses cover were selected, albedo estimated for the same date was 0.36 ± 0.04. For pasture land the albedometer provided an average daily albedo value of 0.158 ± 0.002, for available dates [Domingo et al., 2000], comparable to MODIS pasture estimates for the same period.

4.3. Seasonal Variations in Outgoing Shortwave and SW Radiative Forcing (SWRF)

[25] Annual mean incoming SW radiation at greenhouse surface was 195.98 W m⁻² for the 2001–2005 period, with values ranging from 86.97 W m⁻² in winter to 298.11 W m⁻² (data not shown). Seasonal variations of diurnally averaged OSR for greenhouse and pasture areas, as well as the difference time series for the period of study are represented in Figure 6. Strong differences between both land use types were observed, with higher values over greenhouse areas all year around due to higher albedo of plastic cover. Mean OSR annual values were 58.4 W m⁻² for GH and 38.5 W m⁻² for pasture. Maximum fluxes were detected in summer for both surfaces, with 98 W m⁻² and 65 W m⁻² for GH and pasture, respectively. OSR showed the lowest values in winter, with 20 W m⁻² for GH and 13 W m⁻² for pasture. The range of seasonal variation is much wider in GH (78 W m⁻²) than in pasture (51 W m⁻²).

Figure 6

Seasonal variations of daily averaged outgoing shortwave radiation fluxes (OSR) reflected from greenhouse, pasture and the difference series (-SWRF) between surfaces for the 2001–2005 period. The temporal variations (standard deviations) are plotted as vertical bars.

[26] In Figure 6 observational SWRF due to land use change is represented by the difference between GH and pasture OSR averaged values for the 2000–2005 period (plotted here with opposite sign to RF). Mean annual SWRF was −19.8 W m⁻². Minimum forcing remained almost constant (between −5 and −6 W m⁻²) during November and December, gradually increasing since January to a maximum value by the end of July of −34.8 W m⁻². From August, a steep decrease in the forcing is observed, falling by the end of October next to the minimum annual values. This asymmetry in the seasonal difference curve is determined by the asymmetric shape of both the albedo and OSR curves for GH surface (Figures 5 and 6). This curve shape is not depicted by the more symmetric pasture albedo and OSR curves.

5. Discussion

5.1. Surface Air Temperature Trends

[27] Mean annual temperature trends for control time series GR, MA and MU agree with northern hemisphere reported warming trends for the last three decades [Brohan et al., 2006], as well as with regional trends for Spain (STS) and Europe [Klein Tank et al., 2002; Castro et al., 2005; Brunet et al., 2007]. Temperature fluctuations shown in the time series (Figure 3) are attributed to natural processes (such as major volcanic eruptions, ENSO, QBO, etc.), and mean short breakpoints in the long-term warming were detected in all control stations. The most remarkable in all series is the sudden cooling following the eruption of Mt. Pinatubo in 1991. Nevertheless, trend differences between controls and stations in Almeria province (MOJ/PAL/AL) are consistent with our hypothesis that a massive growth of greenhouse horticulture has impacted long-term trends in surface temperatures of these farming areas. AL station is adjacent to high albedo surfaces, and shows a weaker forcing effect than GH stations completely surrounded by agricultural land (MOJ and PAL). The three temperature series in the province of Almería also clearly disagree with the pattern of variability for southeastern and eastern Spain from 1973 to 2005 reported by Brunet et al. [2007], with and annual trend of +0.54°C decade⁻¹ that indicates a strong rise in temperatures and accelerated warming over this subregion.

5.2. Seasonal Variations in Surface Albedo

[28] Mean surface albedo value measured for GH (0.28) is higher than previous estimates of cropland albedo at global scale, ranging from 0.15 to 0.20 [Myhre and Myhre, 2003]. Nevertheless, our estimation was calculated for the whole coastal plain of Campo de Dalias, and there is a wide range of variation depending on the parcel selected (data not shown). Most important variability in GH albedo depend on whether annual whitewashing of the plastic surface had been applied or not. Smaller GH parcels screened (data not shown) where whitewashing had been fully applied, showed the maximum mean annual albedo values of 0.32 ± 0.03, but were not considered representative of the whole GH area selected (Campo de Dalias), that includes small fractions of other types of land uses apart from greenhouses, such as urban cover or abandoned farmland.

[29] The observed asymmetry in the shapes of GH albedo and OSR curves (Figures 5 and 6) is probably attributable to the annual timing of the whitewashing of the plastic covers carried out regularly by farmers. Whitewashing is performed during summer to prevent summer crops from damage caused by insolation excess. After summer, the slaked-lime is washed out to get adequate radiation conditions during the last stage of winter crops (January–February) and the crops change.

[30] Mean annual surface albedo of the pasture area selected (0.19) was inside the range of values for shrubland previously reported at global scale (0.16–0.29) [Myhre and Myhre, 2003]. In preliminary estimations of surface albedo of other pasture parcels screened nearby the GH area, a range of 0.13 to 0.19 was measured depending on surface reflectance, but we selected our representative parcel of study according to higher similarities with past pastures cleared for GH development in the coastal plain. Clearing of pasture land for GH development follows the maximum pattern previously reported for the increase in the surface albedo in the Northern Hemisphere midlatitudes, considering anthropogenic vegetation changes from preagriculture times to present [Myhre and Myhre, 2003], particularly evident in eastern Europe and the eastern United States caused by the conversion of forest to cropland. In our area of study, the measured increase of the annual mean surface albedo of +0.09 is similar to the maximum values calculated at global scale (around +0.10) [Betts, 2001]. Therefore, extensive white plastic cover has a comparable effect on the differences in mean annual albedo as long-lasting snow cover at high latitudes exerts related to previous forest coverage. On the contrary, this increase in GH albedo is opposite to the decrease in the farmland surface albedo of approximately between −0.02 and −0.01 simulated for Mediterranean latitudes in those studies, related to preagriculture times.

5.3. Seasonal Variations in Outgoing Shortwave and SW Radiative Forcing (SWRF)

[31] The magnitude of localized SWRF due to GH development over the last decades reported here (−19.8 W m⁻²) is considerably greater than most previous computations of simulated global mean RF due to land use change over the past [Hansen et al., 1998; Betts, 2000, 2001; Myhre et al., 2005a]. Although it has been considered −0.2 W m⁻² a good estimation for this global forcing since preindustrial times [Forster et al., 2007], most of these global simulations show a very high degree of spatial variability, with regions showing the strongest local negative RF (−5 W m⁻²) found in the major agricultural areas of North America and Eurasia. In some of these estimations, the increase in surface albedo due to long-lasting winter snow cover in northern deforested latitudes is again responsible for the maximum negative RF estimated values (−10 W m⁻²) [Myhre et al., 2005a]. Mean annual SWRF due to GH development (−19.8 W m⁻²) almost doubles the magnitude of this global maximum, thus reflecting the relative strength of the localized forcing over the last decades in the area of study. Furthermore, GH albedo change maximize in the period with maximum solar radiation (Figure 5) while deforestation at northern latitudes has a maximum albedo change when solar radiation has its minimum (G. Myhre, personal communication, 2008).

[32] Our observational estimation largely offsets the positive forcing of approximately between +2 and +6 W m⁻² assigned to this region in global simulation studies from preagriculture times, and of course, the global forcing relative to preindustrial times exerted by greenhouse gases, estimated at +2.3 W m⁻² [Hansen et al., 1998]. In one of the few related studies on observational estimation of forcing associated with land use change at regional scale to date [Nair et al., 2007] an observational estimate of mean annual SWRF of −7.0 W m⁻² was reported in Southwest Australia, due to land use change for agricultural purposes, considering that half of the whole area of study had been cleared. These authors estimated a maximum SWRF value of −13.9 W m⁻² in case 100% of the land was cleared. Though our SWRF estimate refers to a 70% GH cover, alternative forcing measurements in minor parcels with an almost complete coverage of GH reached the strongest SWRF of −30.2 W m⁻² (data not shown).

[33] As stated before [Myhre and Myhre, 2003], the most important source of uncertainty in the estimation of RF due to land use change is the correct characterization of surface albedo, that depends on both the pasture vegetation and soil, and the reflectance of the plastic surfaces to be compared. In this case, the selection of the parcels representative of preexisting pasture land cover cleared for GH development and the variation in albedo due to the heterogeneity of whitewash surface can extent the range of mean annual SWRF from −19 W m⁻² to a maximum of −37 W m⁻² (data not shown). The degree of vegetation cover of the pastures studied for SWRF estimation, varying from shrubland to grassland, is in great part responsible for this variability. Nevertheless, we considered GH and pasture parcels selected as the most representative of both land uses, although they yielded the weaker negative SWRF of all pairs of parcels compared in preliminary screening.

[34] Another source of uncertainty is the temporal variability of incoming SW radiation observed with instrumental data due to cloudiness or changes in atmospheric conditions. This temporal variability, as represented by the average of standard error for each date during the recorded period, represents a small percentage of the mean annual incoming shortwave radiation at this region (4.75 %). High correlation was found between field measures and POTRAD modeled insolation series for clear sky conditions (Pearson-correlation coefficient = 0.97; Root Mean Square Error = 16.93 Wm⁻²; Mean Annual Error = 3%, expressed as a percentage of the mean of n = 46 dates). Therefore, it can be assumed that the level of uncertainty due to cloudiness in SWRF estimates using the POTRAD model is low.

[35] While there could be a strong correlation between the decrease in temperature records and the increase in SWRF due to GH development, this does not necessarily imply that albedo change is the main causative factor of the observed trends, as reflectivity alterations are not the only effect of land use change on climate. Local climate sensitivity to GH forcing needs further studies to be determined. Related cooling effects of historical land use change toward higher-albedo surfaces have been simulated [Brovkin et al., 1999; Betts, 2001], suggesting that albedo difference is the main driver of temperature change in temperate agricultural regions. Although global mean climate response can be small due to a weak global albedo forcing, the response can be remarkable in some regions, as can be seen when spatial distribution of this forcing is considered [Hansen et al., 2005]. Through these models, annual mean temperatures 1–2 K lower than natural vegetation have been simulated for some agricultural regions, in response to increases in surface albedo of +0.1, comparable to GH development increase reported here. In those cases, the main vegetation change was the conversion of forest to cropland in high latitudes, where snow-covered areas have much higher surface albedo over open land (as cropland) than in forested areas.

[36] Previous observational studies show a dominant influence of high albedo surfaces on local surface temperature trends [Fishman et al., 1994]. Christy et al. [2006] have reported the darkening and moistening of formerly dry high-albedo surface in semiarid environments as the main cause of increase in surface temperatures caused by farming development in California San Joaquin Valley. In our case, however, the change is an increase in surface albedo, and though the influence of irrigation under a plastic surface still remains undetermined, cooling temperature trends indicate that it probably has little impact compared to negative forcing exerted by increased surface reflectivity.

6. Conclusions

[37] Our results show that, at local and meso-scale, greenhouse farming is very likely the most powerful driver of climate change in the area of study, probably due to the dramatic increase in surface albedo of the highly reflective plastic cover over a widespread agricultural area, which largely offsets positive forcing (+2 W m⁻²) very probably induced by global increase in greenhouse gases [Forster et al., 2007]. The main general implication of these findings is to highlight the importance of human development of high albedo surfaces in the strategies of mitigation and adaptation to global warming at local scale. However control stations records outside the GH area show that little or no effects on surface temperature extend far from the high albedo area, so the forcing caused by greenhouse development seems to be very localized.

[38] Although other climate change agents out of the scope of this work cannot be ruled out, the attribution of the temperature cooling trend in GH area to a negative radiative forcing due to change in surface albedo is strongly supported by our analyses of satellite data. The most likely explanation is that higher surface albedo reduces net incoming shortwave energy, and therefore diminishes the energy absorbed by the surface emitted as longwave radiation, resulting in lower surface skin temperature, as well as a lower transfer of sensible heat over greenhouses with respect to pasture cover. The lower surface temperature, and the decrease in the total available energy at the surface that needs to be dissipated, generates the cooling trend detected in the near-surface air temperatures in this farming area

[39] Further analyses must be undertaken to establish or refute causality between SWRF and temperature trends discussed here. The purpose of this work is to present the first empirical evidences of this causal connection through an estimation of SW budget inside and outside the GH area, but other significant feedbacks to the atmosphere from changes in the cover of terrestrial surface by greenhouse farming still remain to be investigated. Model-simulations and radiative transfer calculations including constraints arising from high albedo GH surfaces at the temporal and spatial scale of this study should be carried out in order to corroborate our results by comparison of expected responses to detected temperature trends. However evaluation of net radiation and soil heat transfer at daily scales, and quantification of the turbulent fluxes (sensible and latent heat), are necessary to fully determine how the surface energy balance is affected by the changes in cover:

[40] Uncertainties in the local energy budget must be reduced with the determination of longwave fluxes, and net radiative forcing due to short plus longwave differences between both land uses has yet to be quantified. Nevertheless, it is important to notice that when comparing RF values from earlier estimates, it has been assumed in this work that net forcing is dominated by the shortwave component and that land use changes do not significantly impact the TOA longwave fluxes [Nair et al., 2007].

[41] The relative influence on local climate of other physical changes must be explored. The land use climate forcing that we have estimated here does not fully represent GH land use effects, as there are other changes in surface properties affecting the surface energy balance that have not been considered. For instance, eco-physiological and aerodynamic changes and alterations of roughness still remain undetermined. The complex role of evapotranspiration associated to this drip-irrigated soil under a plastic cover must be investigated [Fernandez et al., 2007]. Cooling effect of higher albedo could have been enhanced by the increase in latent heat flux derived from irrigation within the greenhouses (released as water vapor by greenhouses ventilation), with respect to previous pasture cover, further reducing sensible heat transfer and surface air temperature. On the contrary, irrigation might also cause a positive forcing by the increase in water vapor in the lower atmosphere [Boucher et al., 2004; Christy et al., 2006].

[42] In order to assess the net influence of greenhouse farming on local climate, the role of annual biomass growth as carbon sink must be quantified and expressed as RF [Betts, 2000]. While forestation at northern latitudes and most conventional crops are generally associated to positive forcing by reduction in surface albedo, greenhouse development in semi-desert surfaces exert a negative forcing that is probably further increased by the forcing caused by carbon sequestration of these highly productive crops.

[43] Even the RF concept might not be the most appropriate concept in our case, so that other alternative metrics [Pielke et al., 2002] could be advisable to estimate and model the net impact on the local climate of GH development.

Acknowledgments

[44] We thank the INM (Spanish National Institute of Meteorology), Cajamar Foundation, and IFAPA (JJAA) for providing climatic data. This work received financial support from several different research projects: PROBASE (CGL2006-11619/HID), funded by the Spanish Ministry of Education and Science; AQUASEM (P06-RNM-01732), funded by the Regional Government of Andalucia; DESIRE (Desertification, mitigation and remediation of land), funded by the European Commission.