Methods

The meteorological data at the stations was interpolated to 10×10 km gridded data set to observe spatial distribution of the climatic change. Several methods were investigated to interpolate the daily station observations. These included surface-fitting procedure thin-plate splines (, ), Thiessen polygon area averaging (Thiessen, 1911), and angular distance weighted (ADW) averaging (New et al., 2000). The thin-plate splines interpolation was found to be unsuitable because there were considerable undershoot and overshoot in the edge of the study area. Thiessen polygon interpolation employ a limited number of data points in the estimation of grid point values. The ADW method was selected for this study.

In estimating each grid point using ADW method, eight nearest stations regardless direction and distance are used to contribute to grid point estimation and form the distance weighting function (Piper and Stewart, 1996). Weights for the eight stations were determined in a two-stage process following New et al. (2000). All stations were first weighted by distance from the gird point (New et al., 2000; Jones et al., 1997):

w_k = (e^-x/x₀)^m (1)

where x is the distance from the grid point of interest, x₀ is the correlation decay distance, m is parameters between 1 and 8. A value of 4 for m was suggested by New et al. (2000). The second component of the distance weight was determined by the directional (angular) isolation of each the eight selected stations:

2#2 = 3#3 (2)

where cos4#4(k, l ) is the angular separation of station k and l, and w_l is the distance weight at station l. The angular-distance weight is then calculated from:

W_k = w_k(1 + 2#2) (3)

Based on the interpolated data, the linear regression model was used to estimate the trend magnitude in each grid point. The regression weight was calculated as:

5#5 = 6#6 (4)

where n is the time series number, t_i is the time number, and y_i is the data value at the time t_i. The statistical significance of the annual trends is evaluated using the Student¡¯s t-test (Haan, 1977). The trend magnitude during the study period was then estimated from the regression weight:

7#7Y = 5#5 ^. T (5)

where T is the span of time during the study period. The relative trend magnitude was represented as:

7#7Y^' = 100 ^. n ^. 7#7Y8#89#9y_i (6)

The trend of precipitation (P), mean relative humidity (U_m), sunshine duration (D_s), mean cloud amount (C_a) and LAI data was presented using relative trend magnitude. The trend of mean temperature (T_m), minimum temperature (T_min), maximum temperature (T_max) and diurnal temperature range (DTR) data was presented using trend magnitude. The relative trend magnitude of reference evapotranspiration (ET) was calculated to watch the change in evaporative demand of the atmosphere (Allen et al., 1998).