Introduction

The most important anthropogenic impacts on climate are the emission of greenhouse gases and changes in land use (Kalnay and Cai 2003). Land-use/land-cover (LULC) change was highlighted as a major climate force (National Research Council 2005). Anthropogenic LULC change can affect emissions of CO2, CH4, biomass burning aerosols and dust aerosols. Land cover change itself can also modify surface energy and moisture budgets through changes in evaporation and fluxes of latent and sensible heat, directly affecting precipitation and temperature (Forster et al. 2007). In some cases, LULC change response to climate, in the form of urbanization, agricultural activity and deforestation, may even exceed the contribution from greenhouse gases (Roger and Pielke 2005; Dirmeyer et al. 2010). Previous work has well documented the impacts of different land type changes on climate, both globally and regionally (Lim et al. 2008; Anantharaj et al. 2010; Fall et al. 2010a; Costa and Pires 2010; Kishtawal et al. 2010; Strengers et al. 2010; Xiao et al. 2010). It has been reported that land use changes due to agriculture lead to decreased surface air temperatures (Mahmood et al. 2006; Roy et al. 2007; Lobell and Bonfils 2008). The estimated urbanization impact differs significantly, and depends strongly on the criteria used in classifying urban and rural areas (Easterling et al. 1996; Hansen et al. 2001).

Owing to the difficulty in separating the influence resulting merely from land cover, the observation minus reanalysis (OMR) method developed by Kalnay and Cai (2003) has been used recently to assess the impact of land use change by taking the difference between observations by surface stations and reanalysis data. The OMR method takes advantage of the insensitivity of reanalysis to land surface types, and removes natural variability due to changes in circulation (since they are also included in the reanalysis), thus isolating surface effects from greenhouse warming by subtracting the reanalysis from the surface observations (Kalnay and Cai 2008). The advantage of the OMR method is that it provides a substantial surface climate change signal arising from different land cover types through separating near-surface warming from global warming.

Several studies have used the OMR method to assess the surface air temperature trend caused by land cover change. Kalnay and Cai (2003) initially assessed the decadal surface warming associated with local land uses over the eastern region of the United States by subtracting the reanalysis from the observations. Regional surface warming identified by OMR has good agreement with the trends obtained by Hansen et al. (2001) using nightlights to classify urban and rural stations (Kalnay et al. 2006). Fall et al. (2010b) applied the OMR method to investigate the impacts of the sensitivity of surface air temperature trends to land cover change over the conterminous US. The impacts of different land cover types on temperature in China and Argentina, as well as globally, and the consequences of temperature associated with land use in the Tibetan Plateau, have been assessed by the OMR method (Zhou et al. 2004b; Frauenfeld et al. 2005; Yang et al. 2009; Hu et al. 2010; Lim et al. 2005; Nuñez et al. 2008). Results confirm that the OMR method is a robust tool with which to capture land cover change patterns on a regional scale. Fall et al. (2010c) applied the OMR method in precipitation in India. The results revealed that there is no well defined rainfall pattern as a function of OMR trends. Currently, no study has examined the influence of land cover on climate in greater detail, especially in relation to regional characteristics, which play an important role in local and regional climates.

The purpose of this study was to examine the impact on surface air temperature and precipitation of land cover change and topography together in Jiangxi province in China. By analyzing OMR trends and incorporating regional topography into regression assessments of surface air temperature and precipitation, we assess the relationship between OMR trends and land vegetation types [derived from satellite-normalized difference vegetation index (NDVI)]. OMR trends and the sensitivity of surface air temperature and precipitation to land cover types are presented. The results of our study reveal the significance of regional land cover change for the climate. The teleconnection effect, in which regions remote from the landscape conversion have altered surface meteorology (e.g., as discussed in Avissar and Werth 2005 and Chase et al. 2000), is not the focus of this study.

Materials and methods

Study site

The location of our study site is shown in Fig. 1. Jiangxi is situated in the basin of Poyang Lake along the middle range of the Yangtze River and has a total area about 170,458 km2. Mountains and hills account for 78 % of Jiangxi (Jiangxi Meteorological Bureau 2010). Jiangxi is dominated by humid subtropical climate, Cfa (Trewartha 1968). The annual mean temperature is 16–19 °C and annual mean precipitation is 1,400–2,400 mm. Lake Poyang—the largest freshwater lake in China—is in the north of Jiangxi. It has a total area of 162,200 km2 (Jiangxi Meteorological Bureau 2010). Jiangxi is a province with short arable land that has experienced significant land cover change during the past two decades. As the population increases, a large area of forest, lake and pasture has become occupied for agricultural and urban use (Department of Land and Resources of Jiangxi Province 2011).

Fig. 1
figure 1

Location of the study site

Full size image

Materials

For the surface observations, monthly mean temperature and precipitation totals were obtained from the Computational and Information Systems Laboratory (CISL) at the National Center for Atmospheric Research (NCAR) (http://dss.ucar.edu) for 36 available stations over Jiangxi and six contiguous provinces, 1951–1999. Of these, only six stations are located in Jiangxi (shown in Table 1). The spatial distribution of the stations is shown in Fig. 2. All stations have continuous monthly data for the entire 49-year period except for the missing data for November and December in 2000. The homogenized monthly surface instrumental temperature and precipitation dataset was obtained using the Standard Normal Homogeneity Test (SNHT). The Global Historical Climatology Network (GHCN) data was used for comparative analysis, which is available online at (http://lwf.ncdc.noaa.gov/oa/climate/research/ghcn/ghcngrid.html) from the National Oceanic and Atmospheric Administration, National Climatic Data Center (Asheville, NC). Both NCAR and GHCN datasets have continuous records for stations in Jiangxi and its six contiguous provinces for the period 1951–1990. The data from GHCN has been homogenized based on pairwise comparison (Menne and Williams 2009).

Table 1 Stations in Jiangxi used in the analysis of regional trends
Full size table
Fig. 2
figure 2

Map of Jiangxi, showing locations of climate stations used in this study [overlaid on a 90 m digital elevation model (DEM)]

Full size image

Monthly surface air temperature mean and monthly mean of precipitation rate time series from NCEP-NCAR reanalysis (NNR) were used, both covering the period 1951–1999. Advanced very high resolution radiometer (AVHRR) NDVI, 15-day composite product dataset at 1° resolution for the period 1982–2000 was derived from Global Land Cover Facility, Global Inventory Modeling and Mapping Studies (GIMMS) (GIMMS; http://glcf.umiacs.umd.edu/data/gimms/). We used NDVI to represent surface vegetation types. NDVI and its seasonal change were related to the decadal OMR and observation trends. This NDVI dataset has been corrected for calibration, view geometry, volcanic aerosols, and other effects not related to vegetation change. Seasons are defined as the 3-month averages for surface air temperature and NDVI, 3-month sum for precipitation: DJF, MAM, JJA, and SON. Shuttle radar topography mission (SRTM) 90 m digital elevation data supplied by Consortium for Spatial Information (CGIAR-CSI) was used as elevation factor for regression assessment and correction of spatialization.

Methods

Maximum value composite of NDVI

Atmospheric calibration of original 15-day composite AVHRR-NDVI data was done to reduce cloud and other noise. The two calibrated 15-day NDVI images for each month were composited first by taking the maximum value within two 15-day time periods for each pixel to eliminate pixels with cloud cover. Then we averaged the monthly NDVI values for each pixel into annual and seasonal mean NDVI values. The result was an almost cloud free data set of peak NDVI value from 1982 to 2000 covering Jiangxi. The maximum-value composite technique has the advantage of minimizing cloud contamination and off-nadir viewing effects since these factors tend to reduce the NDVI values over green surfaces (Holben 1986; Holben and Fraser 1984). Applying the maximum value compositing process, the highest value record in the time range was retained for each pixel location (Holben 1986). Due to the growing circle of vegetation in Jiangxi, the NDVI mean of growing season (April–October) was used to reflect the general vegetation state for the entire study period.

Land cover types were indicated by NDVI categories. Each pixel on an image was assigned to a NDVI category. The NDVI value is defined as the ratio of the difference to the total reflectance: (near-infrared − red) / (near-infrared + red). Green leaves commonly have larger reflectance in the near-infrared than in the visible range. Clouds, water, and snow have larger reflectance in the visible than in the near-infrared, so that negative values of the vegetation index may correspond to snow or ice cover, whereas the NDVI value is almost zero for bare soils such as deserts. As a result, NDVI values can range from −1.0 to 1.0, with higher values associated with greater density and greenness of vegetation covers (Lim et al. 2008). Based on Wang et al. (2006) and Jiangxi’s vegetation distribution, NDVI images after maximum value composite were assigned as six categories, NDVI value (1) −1.0 to 0; (2) 0–0.2; (3) 0.2–0.45; (4) 0.45–0.6; (5) 0.6–0.75; (6) 0.75–1.0. Different land cover types in response to climate changes were further analyzed.

OMR method

Surface climate change trends are analyzed using OMR method suggested by Kalnay and Cai (2003). No surface data were included in the data assimilation of NNR. Due to the insensitivity of the reanalysis to surface processes, surface effects caused by regional vegetation type changes could be separated from greenhouse warming. In the reanalysis (a statistical combination of 6-h forecasts and observation), surface observations of temperature, moisture and wind over land were not used (Kistler et al. 2000). The reanalysis, combined with its model parameterizations of surface processes, creates its own estimate of surface fields from the upper air observations. The surface parameters in a reanalysis have less dependence on local characteristics than the actual surface observations. As a result, the reanalysis excludes surface effects even though it should show climate change to the extent that they influence the observations above the surface (Fall et al. 2010b; Kistler et al. 2000).

As described by Kalnay and Cai (2003), the OMR decadal trend is obtained by taking the average of two decadal mean differences (the 1990s minus the 1980s and the 1970s minus the 1960s). The trends were computed as changes in decadal averages in order to reduce random errors. Only two decadal trends, the decade 1990–1999 minus 1980–1989 and 1970–1979 minus 1960–1969 were used. The trend changes between the decade 1960–1969 minus 1950–1959 and 1980–1989 minus 1970–1979 were excluded due to the scheduling changes during 1958 and changes in the observing systems starting in 1979. Observation times of the reanalysis for the first decade (1948–1957) were done 3 h earlier than the current synoptic times. This implies less reliable information extraction during the pre-1958 era. Furthermore, introduction of the satellite observing system starting in 1979 resulted in a significant change in climatology data. Before 1979, climatology was dominated more by model climatology in data-sparse areas, leading to the generation of spurious trends (Kistler et al. 2000). These two major changes affect trends in the NNR. The two decadal changes of the 1990s minus 1980s (20 years with satellite data) and the 1970s minus 1960s (20 years essentially without satellite data) were used. Thus, we obtain decadal trends from two independent and largely homogeneous 20-year periods.

Results

Comparison between the temperature trends from GHCN and from our analysis on adjusted data

Figure 3 compares the time trends in homogenized annual mean temperature at six stations over Jiangxi (1951–1990) obtained from GHCN and from our analysis on adjusted data. Temperature from both analyses shows a similar increasing trend but a slight difference in amplitude. Both are not statistically significant, P > 0.05. The temperature from GHCN (0.48 °C/100a) was 0.30 °C/100a warmer than that from our analysis (0.18 °C/100a). The temperature from GHCN analysis in 1951–1954, 1986 and 1989, was slightly cooler than that from our analysis. The differences can be explained by different station use and different methods for homogeneity test in both analyses. Nevertheless, the overall temperature trend was almost identical from both datasets but there was a slight difference in the increasing amplitude.

Fig. 3
figure 3

Surface air temperature anomaly relative to 1961–1990 averaged over Jiangxi (1951–1990), annual mean for Global Historical Climatology Network (GHCN) (triangle-dashed line), annual mean for adjusted data (dotted line), 5-year running average for GHCN (crossed-solid line) and 5-year running average for adjusted data (solid line). P > 0.05

Full size image

Spatial pattern of surface air temperature changes based on local linear trends (1951–1990) is shown in Fig. 4, from the top to the bottom: annual mean, winter, spring, summer and autumn mean temperature, respectively. Our analysis (Fig. 4a) indicates more heterogeneity warming than GHCN (Fig. 4b). No noticeable warming center was observed from either analysis due to the heterogeneity. Warming in the west and east for both analyses was in similar amplitude. GHCN reveals a different pattern of high warming in the south and low in the north. Concerning seasonal pattern, both analyses show a similar spatial distribution (winter warming vs summer cooling; autumn warming vs spring cooling) but different amplitudes (strong spring cooling from GHCN). For both analyses, strong and heterogeneous winter warming is presented and summer cooling concentrates on the north along Poyang Lake watershed. Spring cooling is noticeable in the south, Wuyi Mountain. However, our analysis indicates strong warming autumn in the west and weak in the east but inversely from GHCN. The temperature computed from different datasets over the same region can thus show different trends.

Fig. 4
figure 4

Surface air temperature change based on local linear trends (1951–1990) for a our analysis and b the global historical climatology network (GHCN), from the top to the bottom: annual mean, winter, spring, summer and autumn mean temperature, respectively

Full size image

Temperature trends of observation and reanalysis

Monthly surface air temperature anomalies for five decades (1950s–1990s) from the observation and NNR reanalysis relative to the normal period 1961–1990 averaged over Jiangxi are shown in Fig. 5. The correlation between observations and NNR is also presented. There was a good agreement in the inter-annual variability and the long-term trends between observations and NNR in the 1960s, 1970s, 1980s and 1990s, with correlation coefficients r = +0.80, +0.78, +0.79 and +0.76, respectively. Relatively low correlation is indicated in 1950s (r = +0.53), because the observation times of the reanalysis for the first decade were considerably less reliable than in later decades (Kistler et al. 2000). The difference (TOMR) between the observations and NNR is the largest, increasing to 0.05 °C in the 1970s (Fig. 5). As is revealed in many other studies (e.g., Kalnay and Cai 2003; Lim et al. 2008), the NNR has a strong ability to capture surface air temperature variations caused by atmospheric storms, advection of warm/cold air, and variations in the frequency or track of major storms. In contrast to the actual surface observations, no statistically significant difference was found in the NNR estimation of station trends. The NNR trend is not affected by surface property changes. These arguments suggest that we could attribute the differences between monthly or annually averaged surface-temperature trends derived from observations and from the NNR primarily to land use change. Decadal trends can be dominated locally by inter-annual and decadal temperature variability due to anomalies in the circulation rather than to land cover change effects that are excluded by taking the differences between observation and NNR temperatures.

Fig. 5
figure 5

Monthly mean surface air temperature anomalies [observation (dashed red)] and NCEP-NCAR reanalysis (NNR) (solid blue; in °C) over Jiangxi relative to 1961–1990. Five decades (1950s–1990s) are compared. TOMR is observation minus reanalysis temperature. r denotes correlation coefficient between NNR and surface observations

Full size image

Figure 6 shows the decadal temperature trends for the observations, the NNR and the difference between these two trends (OMR) averaged at six stations over Jiangxi. The reanalyses exhibit a weaker warming than observations, as obtained by Kalnay and Cai (2003) and Lim et al. (2005). This feature was found across most of Jiangxi province. Thus the OMR pattern obtained by subtracting the reanalyses from observations shows a positive trend. This is at least partially attributable to land use change. The average warming amplitude of the observations, NNR and OMR was +0.078 °C , +0.051 °C and +0.027 °C per decade respectively (Fig. 6). Therefore, about 35 % of the observation warming could be associated with land surface property changes. The observation, NNR and OMR trends show an overall consistent warming effect. The strong warming concentrated on Northern Jiangxi, both for OMR and observations. In comparison with NNR, the OMR trend indicates a weaker and inverse warming effect. To assess the surface air temperature trend associated with land surface types, we may conclude that the OMR has the advantage of indicating the effect resulting from land cover change on climate.

Fig. 6a–c
figure 6

Decadal trends of monthly mean surface air temperature. The observation minus reanalysis (OMR) trend per decade (in °C) at each grid point was obtained by the average of the ‘1990s minus 1980s’ and ‘1980s minus 1970s’ temperatures. The mean value of the decadal trend is denoted in each panel on the left. a Observation; b NNR and c OMR

Full size image

Precipitation trends of observation and reanalysis

Daily mean precipitation anomalies for five decades (1950s–1990s) from observation and NNR relative to the normal period 1961–1990 over Jiangxi are shown in Fig. 7. The correlation between observations and NNR is also presented. There was a good agreement in the inter-annual variability and the long-term trends between observations and the NNR in the 1970s, 1980s and 1990s, with correlation coefficients of r = +0.66, +0.64 and +0.67, respectively. Poor correlation is indicated in the 1950s and 1960s (r = +0.49 and r = +0.20 respectively). The results may stem from the less reliable observing system during the 1950s. As revealed in Fig. 7, the NNR has good ability to capture precipitation variations during the 1970s, 1980s and 1990s. No statistically significant difference was found between the NNR and actual surface observations. The NNR trend is not influenced by the change of surface information. Therefore the differences between precipitation trends derived from observations and from the NNR could be attributed primarily to land use change.

Fig. 7
figure 7

Daily mean precipitation anomalies [observation (dashed red)] and NCEP-NCAR reanalysis (NNR) (solid blue; in mm) over Jiangxi relative to 1961–1990. Five decades (1950s–1990s) are compared. TOMR is observation minus reanalysis precipitation. r denotes correlation coefficient between NNR and surface observations

Full size image

The decadal precipitation trends for observations, the NNR and the OMR averaged at six stations over Jiangxi are shown in Fig. 8. The observations exhibit a stronger wetting than the NNR. Thus, the OMR pattern obtained by subtracting reanalyses from observations indicates an overall positive trend. This can be attributable partially to land use change. The wetting amplitude of the observations, NNR and OMR was 0.320 mm, 0.024 mm and 0.296 mm per decade, respectively. In comparison with the observation precipitation, the NNR exhibits an inverse pattern. This may be due to the high variability of topographical gradients. As reported by Zhao et al. (2004), who compared the observation and reanalysis precipitation in China and found more observations than the reanalyses. The spatial pattern of OMR precipitation was similar to that of observations, but with a weaker wetting effect. The feature of more wetting in the north and less wetting in the south for both OMR and observation precipitation was revealed. The decadal trend of observation precipitation is in good agreement with that of the observation temperature. To assess the precipitation change in response to land surface types, the OMR trend of precipitation has the ability to represent the effect of land cover change.

Fig. 8a–c
figure 8

Decadal trend of daily precipitation averaged at stations over Jiangxi. The OMR trend per decade (in mm) at each grid point was obtained by the average of the ‘1990s minus 1980s’ and ‘1980s minus 1970s’. The mean value of the decadal trend is denoted in each panel on the left. a Observation, b NNR, c OMR

Full size image

NDVI trend

Figure 9 indicates the geographical variations of the annual and seasonal mean NDVI in Jiangxi (1982–2000). Surface vegetation patterns with their greenness ranges are shown. Different NDVI greenness ranges are good representatives of different land cover types. The range is from the least sparse vegetation cover, 0 to the highest density vegetation canopy, 1.0. Water bodies with negative NDVI values had nearly no seasonal change. The very sparse vegetation areas with vegetation index (0∼0.2) show a slight seasonal change. The areas characterized by vegetation index (0.2∼0.75) show a noticeable seasonal change (Fig. 9b–e). Evergreen tree cover areas [NDVI (0.75∼1.0)] exhibit a relatively weak seasonal change. An overall noticeable seasonal NDVI change is shown. The findings are similar to those of Wang and Li (2008)

Fig. 9a–e
figure 9

Vegetation index map derived from NDVI. a Annual mean, b winter (DJF), c spring (MAM), d summer (JJA) and e autumn (SON)

Full size image

Surface air temperature and precipitation trends with respect to surface vegetation type changes

Relationship between surface air temperature and surface vegetation types

To examine the surface air temperature with respect to surface vegetation type change in Jiangxi, we related the temperature change estimated by observation, NNR and OMR to land vegetation types. Decadal observation, NNR and OMR trends of temperature at each grid point plotted against the NDVI mean of growing season are shown in Fig. 10. The results show that the decadal observation, NNR and OMR trends were all inversely proportional to the surface vegetation index. The decadal trend in NNR shows no significant relationship with NDVI (r = −0.02). Poor correlations between NDVI and decadal temperature trends of OMR and observations are indicated (r = −0.20 and r = −0.25, respectively). The surface warming response to the surface vegetation cover was poorly represented by OMR. In contrast, Lim et al. (2005, 2008) and Fall et al. (2010b) reported good representative of OMR trends. However, the large positive correlations in the scatterplots were all in the areas with low NDVI values (0–0.4) for OMR and observations, whereas this was not the case for NNR. Consistent results obtained by Kalnay and Cai (2003) pointed out that the NNR should not be sensitive to land use effects, although it will show climate changes to the extent that they affect observations above the surface. However, NNR from modeling experiments reveals that a strong surface warming correlated well with low vegetation index areas (Xue and Shukla 1993; Dai et al. 2004; Hales et al. 2004), which contradicts our findings.

Fig. 10
figure 10

Scatter plot between decadal NDVI and temperature trend over Jiangxi. a OMR, b Observation, c NNR. Abscissa NDVI, ordinate temperature trend, r correlation coefficient of all the data points

Full size image

Relationship between precipitation and surface vegetation types

Decadal observation, NNR and OMR trends of precipitation were associated with mean NDVI of growing season to illustrate the relationship between precipitation and surface vegetation types in Jiangxi. The decadal observation, NNR and OMR trends at each grid point was scatter-plotted with NDVI and is shown in Fig. 11. The results show that decadal observation, NNR and OMR trends are not significantly correlated with NDVI (r = 0.02, r = −0.01 and r = 0.03 ,respectively). The OMR trend has low correlation with surface greenness. Thus, surface vegetation types response to precipitation was poorly represented by OMR trend. The findings are consistent with those of Fall et al. (2010c), who found no good correlation between OMR precipitation trend and surface vegetation types in India.

Fig. 11
figure 11

As Fig. 10 but for precipitation trends

Full size image

Trends of temperature and precipitation associated with different surface vegetation types

We related observation and OMR trends of temperature and precipitation to surface vegetation types over Jiangxi. The correlation coefficients are presented in Table 2. Both observation and OMR trends of temperature showed a significant negative correlation with land type [NDVI (0–0.2)]. The correlation coefficients were −0.61 and −0.62, respectively. For both temperatures, the correlation decreased with the increase in vegetation greenness. No significant correlation between other land types and corresponding observation and OMR trends of temperature is indicated. The finding is consistent with that of Lim et al. (2008), who showed that the OMR trend decreased with surface vegetation greenness.

Table 2 Correlation coefficients (r) between surface air temperature/precipitation and corresponding surface vegetation types
Full size table

No significant correlation was indicated between both observation and OMR trends of precipitation and different surface vegetation types over Jiangxi (Table 2). For both trends of precipitation, the correlation does not reveal regular change with vegetation greenness.

Surface air temperature/precipitation change as a function of different surface vegetation types incorporating regional topography

Using ordinary least squares, surface air temperature and precipitation (both for OMR and observation) were regressed on each surface vegetation type and corresponding regional topography including elevation, latitude, longitude and slope according to the form:

$$ \mathrm{Surface}\;\mathrm{air}\;\mathrm{temperature}\;\left( {\mathrm{or}\;\mathrm{precipitation}} \right)=\mathrm{a}*\left( {\mathrm{surface}\;\mathrm{vegetation}\;\mathrm{type}} \right)+\mathrm{b}*\left( {\mathrm{elevation}} \right)+\mathrm{c}*\left( {\mathrm{latitude}} \right)+\mathrm{d}*\left( {\mathrm{longitude}} \right)+\mathrm{e}*\left( {\mathrm{slope}} \right)+\mathrm{f} $$

Where: a, b, c, d, e are corresponding regression coefficients of surface vegetation type, elevation, latitude, longitude and slope, and f is constant.

The regression results are indicated in Table 3. They are statistically significant at 99 % level. Both the observation and OMR surface air temperature trends show a dependence on vegetation greenness. The dependence integrating regional topography on each surface vegetation type performed better than that removing regional topography. A promising finding was that 81.1 % of surface warming (OMR temperature: 0.108 °C/decade) over low vegetation cover areas [NDVI (0 –0.2)] was explained by regional surface vegetation type change and elevation, latitude, longitude and slope; 49.1 % of surface warming was attributable to regional land cover change over these areas. Over the areas with the highest vegetation index (0.75– 1.0), land cover change can explain 19.6 % of the surface warming integrating regional elevation, latitude, longitude and slope. The explanation ability was even poorer when removing regional topography. As land cover greenness increased, the dependence of decadal OMR trends of both surface air temperature and precipitation on vegetation cover greenness got worse (Table 3). The response of land cover change to the OMR temperature trend was consistent with Lim et al. (2008) and modeling analysis by Dai et al. (2004); Hales et al. (2004) and Xue and Shukla (1993).

Table 3 Summary for regression models of different surface vegetation types
Full size table

An overall poor response of land cover change to OMR and observation precipitation, both integrating and removing regional topography, is revealed (Table 3). The results were similar to those of Fall et al. (2010c), who also obtained a poor land cover functional response to OMR precipitation in India. As indicated in Table 2, 38.8 % of the surface drying (OMR precipitation: −0.063 mm/decade) over the areas with vegetation index (0 – 0.2) was attributable to regional land cover change and topography; 28.2 % of the surface drying can be explained without the regional topography into regression assessment. The explanation level of the dependence of OMR precipitation on other surface vegetation type change was worse. The worst performance was the areas with vegetation index (0.75 – 1.0).

Table 4 shows the decadal trends of surface air temperature and precipitation in response to surface vegetation types and the number of 1° × 1°grids calculated for each land type. Results show that the OMR temperature trend decreased and the OMR precipitation trend increased with the increase of surface greenness. OMR warming over the areas with low vegetation cover [NDVI (0 – 0.20)] was the strongest, and the areas with NDVI (0.20 –0.45) indicate a second strongest OMR warming. OMR cooling was found over the areas with vegetation index (0.45 –1.0). Very dense vegetation cover signified by evergreen tree cover with less anthropogenic activity exhibited greater cooling than areas with intermediate vegetation cover [NDVI (0.45 – 0.60)]. The assessment reveals that surface warming is larger for areas with low vegetation cover [NDVI (0 –0.20)], anthropogenically developed, but smaller for areas covered with very density vegetation [NDVI (0.75 –1.0)] (Table 4). OMR drying is shown over areas with low vegetation cover [NDVI (0 –0.20)] and water bodies [NDVI (−1.0 –0.2)]. Oher land types indicate OMR wetting, and areas with vegetation index (0.60 –0.75) show the largest wetting.

Table 4 Decadal trends of temperature (°C/decade) and precipitation (mm/decade) response to surface vegetation types and the number of 1° × 1° grids for the calculation of correlation between vegetation types and temperature/precipitation trend
Full size table

Areas with vegetation index (0.6–1) generally are composed of tree cover, which leads to the suppression of surface warming and the increase of surface moisture through the strong transpiration wetting and evaporation cooling from leaves with 0.009 °C/decade and 0.126 °C/decade surface cooling and 0.145 mm/decade and 0.005 mm/decade surface wetting, respectively. This is in agreement with Giambelluca et al. (1997), Lim et al. (2008), Shukla et al. (1990) and Xue and Shukla (1993). Areas with moderate vegetation index (0.45 –0.6) characterized by herbaceous and shrub cover show 0.008 °C/decade surface cooling and 0.114 mm/decade surface wetting (Table 4). The moderate contribution is because the cooling and wetting feedbacks from leaves are weaker than those in trees. Owing to the small amount of evaporation negative feedback in low vegetation cover areas [NDVI (0 – 0.2)], a strong surface warming and drying effect separated from the observations could be explained by regional surface vegetation type change anthropogenically driven under the same amount of radiative forcing. Our findings are in agreement with those of Diffenbaugh (2005), Lim et al. (2005, 2008) and Fall et al. (2010b).

Discussion

We aimed to investigate the impact of regional land cover change on variations of surface air temperature and precipitation. The OMR method (observation minus reanalysis) was used to estimate the impact of changes in land use by computing the difference between the trends of the adjusted surface observations (which reflect all sources of climate forcing, including surface effects) and NCEP/NCAR reanalysis (which contains only the forcing influencing assimilated atmospheric trends). We analyzed the observation and reanalysis trends of decadal temperature and precipitation over Jiangxi and examined the sensitivity of surface air temperature/precipitation to land cover by using OMR trends as a function of surface vegetation types and regional topography. Key findings are presented as follows:

  1. (1)

    The OMR approach is a robust tool with which to examine the temperature and precipitation trends driven by the impact of regional land-cover types because land surface observations are not used in assimilation of the data into a physically consistent atmospheric model; reanalysis data is thus insensitive to local surface properties. This characteristic of the provides us with the possibility of detecting surface climate change associated with regional land cover types by taking the difference between observed and reanalysis climate time series.

  2. (2)

    Our results show a good temporal and spatial consistency between observed and reanalyzed temperature/precipitation trends, and exhibit the ability of reanalysis to capture regional inter-annual variability. Similar findings were obtained by Fall et al. (2010b), Frauenfeld et al. (2005), Kalnay et al. (2006), Lim et al. (2008) and Zhou et al. (2004a).

  3. (3)

    OMR trend analyses associated with land cover types using AVHRR-NDVI dataset in Jiangxi show that land surface type with vegetation index (0–0.2) exhibits the largest surface warming and the vegetation index (0.75–1.0) area indicates the strongest cooling trend. We found different positive OMR trends (0.108 °C per decade) in contrast to the findings of Fall et al. (2010b) (0.034 °C per decade) on a regional scale and Lim et al. (2005) (0.3 °C per decade) on a global scale for bare soils. We find a strong surface warming response to land barrenness and weak warming response to land greenness. The results are consistent with those of Dai et al. (2004), Diffenbaugh (2005), Hales et al. (2004) and Lim et al. (2008).

  4. (4)

    OMR precipitation trend is also assessed as a function of surface vegetation types and regional topography. The results show that OMR trend is insensitive to different surface greenness. This may be due to the length of the dry season and the seasonality of precipitation. Fall et al. (2010c) revealed similar findings of precipitation trends as a poor function of land cover change over India. However, NDVI has been shown to be a sensitive indicator of the inter-annual variability of precipitation (Prasad et al. 2005). A positive correlation between the annual mean NDVI and precipitation was indicated in arid regions of Central Asia (Ichii et al. 2002). Moreover, the sensitivity of NDVI to precipitation variability in drier regions was found to be higher than in other wet forest types (Wang et al. 2003). In agreement with our results, NDVI has a relatively better response to the variation of precipitation over barer areas than areas of dense vegetation. Prasad et al. (2005), Schultz and Halpert (1993) and Gómez-Mendoza et al. (2008) pointed to the significance of lag in the correlation of precipitation and vegetation types. The lag response of land cover change to precipitation requires to be considered. Despite the limited relationship indicated, our results provide guidance for attribution of regional precipitation.

Our innovative findings indicate that 81.1 % of the surface warming over the areas with vegetation index (0 – 0.2) was attributed to regional land cover change and topography together, whereas an explanation level of 49.1 % is obtained without regional topography into regression assessment. The contribution capability of land cover change decreases as land cover greenness increases. Regional climate change is a complex process involving the interaction and feedback with LULC change. It is well known that regional climate change is influenced heavily by regional topography. It is worth mentioning that the integral of the entire topography has a negligible effect on LULC change, and thus affects regional climate only indirectly. The topography of hilly landscapes modifies land environment by changing the fluxes of water and energy, increasing the vulnerability of land systems, which could become more accentuated under climate change (drought, increased variability of precipitation). Model simulations show that cropland change is significantly related to a slope × elevation index in the southern Mediterranean climate even under the high emission scenario (Ferrara et al. 2010). Daly et al. (2009) found that a complex temperature landscape composed of steep gradients in temporal variation is controlled largely by the gradients in elevation and topographic position. The introduction of regional topography in the regression estimation allows a more accurate determination of which components forming processes have adapted to ongoing climate change, and to what extent.

It should be noted that the method used to derive NDVI needs further refinement and calibration. On the other hand, the estimate of land-cover influence on temperature and precipitation is limited by station density (six stations over Jiangxi are available). Kalnay and Cai (2003) have found that the correlation between the observation and reanalysis temperature is lower where the density of the meteorological stations is low, thus decreasing the ability of OMR to represent land cover change. Notwithstanding this limitation, the results of this study suggest that land cover change and regional topography together contribute to regional surface air temperature change. To what extent this effect is strongly land-type dependent remains to be determined. Moreover, we further confirm the ability of the OMR method to give a quantitative estimation of surface climate variability with respect to land cover change on a regional scale.

This study provides important considerations in understanding the impacts of land cover change on climate change on a regional scale, especially in a hilly region. Our results also have important implications in the monitoring and modeling processes of regional climate. LULC change should be considered along with greenhouse gas as a forcing factor in regional climate modeling. Furthermore, the application of the OMR method on a regional scale needs to be considered in greater detail. Also the combination and correlation of different sensors and satellites will provide better tools to understand the relationship between LULC change and climate change. People and ecosystems experience the effects of environmental change regionally and not as globally averaged. Besides greenhouse gases and aerosol-driven radiative forcings, the impact of local and regional LULC on climate merits more attention in future regional-scale studies of global change.

Conclusions

OMR trends associated with different land surface types show a strong surface warming response to land barrenness and weak warming response to land greenness; 81.1 % of the surface warming over vegetation index areas (0 – 0.2) is attributed to regional land cover change and topography. The contribution capability of land cover change decreases as land cover greenness increases. OMR precipitation trend has a weak dependence on regional land surface types and topography.