Results and Discussion
Temporal-Spatial Tendencies of NDVI Variations in China
China Nationwide
A countrywide PCA was undertaken in order to detect the overall tendencies of temporal-spatial variations of NDVI in China. Using 216 monthly Maximum value composite NDVI images from 1982 through 1999, an overall spatial pattern has been identified through component 1, which is coincident with the vegetation distribution of China (Hou, 1982). That is, in the eastern part of China, NDVI presents higher values, particularly in the southwest and northeast where natural vegetation is dense and comparatively less influenced by human activity. In the west, NDVI is much lower, where infertile deserts and high elevation plateaus are dominant. Further, the temporal curve of component 2 shows a periodic change of NDVI, which is typically influenced by the change of seasons, with a peak in July and August and a trough in December and January within each year. Based on the China Vegetation Distribution Map (Hou, 1982), it is clear that component 2 is showing natural deciduous vegetation in China. That is, in north China, especially in the northeastern area, the seasonal change of NDVI strongly corresponds to the one-peaked seasonal fluctuation characterized by the temporal curve of component 2, yet southward, the trend breaks gradually. On the other hand, there are regions uncorrelated with this seasonal fluctuation, which reveals the regional diversity of NDVI change patterns in China. As a result, different regions command different factor composites that conduct differential temporal-spatial distributions of NDVI. Component 3 shows a different temporal and spatial pattern. It has been identified to indicate a cropping pattern, based on the temporal and spatial analysis. This is an area of wheat and rice agriculture (Hou, 1982). Therefore, on the countrywide scale, PCA is able to show the overall photosynthesis activity in China as well as showing major phenological trends involving both natural deciduous vegetation and agriculture vegetation.
Sub-regions
As is well known, component 1 generally indicates overall NDVI distribution. Therefore, the illumination below is chiefly focused on later components.
Northeast China
The temporal curve of component 2 shows a highly regular fluctuation in each year, with a peak in December and January and a trough in July and August. The spatial dimension of component 2 shows that nearly the entire region is strongly negatively correlated to the tendency of the temporal curve, which reveals that this region's NDVI variation indeed follows such a uni-peaked progression, but in the reverse direction as shown in the curve. That is, the maximum NDVI value occurs in the period of July to August instead of December to January. This pattern clearly corresponds to the pattern addressed by the temporal curve of component 2 for the entire China above. Compared to the China Vegetation Distribution Map (Hou, 1982), the pattern can be explained by the characteristics of climate, natural vegetation, and agricultural cropping structure of this region. In this region the winter is frigid, but summer's average temperature is generally over 24°C. Its adjacency to the Pacific Ocean makes the yearly average precipitation reach 700-800 mm (Liu, 1985). The natural vegetation type is forest, composed of masses of conifers and broad-leaf deciduous trees.
The Daxing'an and Changbei-shan Mountains in this area are in China's largest natural forest zone. It can be concluded that it is the specific characteristics of climate and vegetation that make this region's NDVI obviously follow the uni-peaked change pattern within one year. In summer, NDVI goes up to its maximum value along with the luxuriant growth of vegetation and increase of rainfall; in winter, NDVI declines to its harsh winter environment. On the other hand, this region's agricultural cropping structure in large part contributes to the explanation of the uni-peaked pattern as well. Restricted by temperature as well as precipitation conditions, crops in this area can only be harvested once per year. Crops are seeded in late spring, and harvested in early autumn (Liu, 1985). In late July to August, crops enter the most luxuriant growth period, which directly leads to the maximum NDVI value. It is the temporal-spatial coincidence of variations of natural vegetation and crops along with seasonal change that makes the variation of NDVI of this area concerted with the repeated uni-peaked fluctuation rhythm over years.
Yellow River Valley
The temporal curve of principal component 2 of this area exhibits a two-year period in which there are two big peaks and one small peak. The two big peaks occur in August to September, separately in each year, and the small peak grows from November of the previous year and reaches its crest in late February to March of the following year. Spatially, Weihe basin, Henan province, Anhui province, and north Jiangxu province, all located in the middle-lower reach of the Yellow River, show significant positive correlation with the temporal curve. On the other hand Shanxi , Inner Mongolia, Ninxia, and Shanxi provinces at the upstream exhibit a negative correlation. By reference to China Vegetation Distribution Map (Hou, 1982), it is found that the former is subject to the agricultural zone with three harvests every two years. Summer and fall months ranging from May to November are the growth periods of wheat and rice, and the stage spanning November of the previous year to May of the following year is that of winter wheat (Liu, 1985). The upstream area is principally covered by semi-arid grasslands. It can be concluded that the PCA methodology can clearly pull out agricultural zones under different phenological conditions.
Yangtze River Valley
From the PCA results of Yangtze River Valley, it is concluded that the NDVI change pattern in Yangtze River Valley is significantly different from that in Northeast China and that in the Yellow River Valley. In Yangtze River Valley, both natural vegetation and corps are broadly distributed within the area. Roughly, croplands are principally found on the north side of the Yangtze River, whereas natural sub-tropical evergreen broad-leaved forests are dominant on the south side. By reference to the China Vegetation Distribution Map (Hou, 1982), the spatial dimension of component 2 characterizes the geographic distribution of natural vegetation of this area. As opposed to component 2, component 3 spatially presents almost a reverse pattern. The temporal curve of the component 3 exhibits typically two peaks per year, one of which occurs in March to April, and another in August to September. Spatially, north Hunan province around Lake Dongting, Hubei province, Anhui province, Jiangsu province, and Zhejiang province are significantly correlated to the curve. The southern part, however, generally exhibits negative or non- correlation. Referring to the vegetation distribution map (Hou, 1982), the spatial pattern is very highly coincident with the spatial distribution of rice, which, seeded in late January to February and late June to July, respectively, and harvested in May to early June and November to early December correspondingly, is the overwhelming crop type in this area (Huang, 1994 and Liu, 1985). For this, it can be concluded that component 3 represents the part of NDVI change caused by the growth of rice. In addition, the reasons that the southern part of the region presents negative or non-correlation stem from overwhelming sub-tropical natural vegetation there as well as sparse crops with three harvests in one year (Hou, 1982).
Identification of Natural and Human Factors
The temporal-spatial change pattern of NDVI is not only controlled by large-scale factors, e.g., seasonal shifts and regional agricultural cropping structures as addressed above, but also significantly influenced by some small-scale or short-term factors, especially in smaller regions or during some specific periods. In order to identify how these small-scale natural and artificial factors influence the change of NDVI, the author further applied PCA to Lake Dongting, Shanghai, and Pearl River Delta (Guangzhou Area) separately.
Flood Hazards in Lake Dongting
Floods in this area generally take place in May to July. The author therefore extracted the data series of this stage for PCA. However, the resulting components did not reveal significant NDVI change due to the influence of floods. Again, the author attempted to analyze the data series of August and September separately. As a result, the flood accidents can be clearly detected in the first component of the series in both August and September. The resulting temporal curve of the August series shows that the NDVI value of 1997 was much lower than that of any other year, and the same was true of the September series. Why could the series of May to July not identify the flood accidents, but the series for either August or September could? This involves regional climate conditions and the growth pattern of crops. During the period of late May to July, summer crops enter their harvest stage. The croplands are either uncultivated temporarily or just sparsely covered by new rice seedlings. The result is that the NDVI value of this stage is normally much lower. Hence it is difficult to distinguish the NDVI change of this stage because the NDVI of water approximates that of bare lands. Furthermore, cloud cover easily influences the imagery of this stage as well. In August and September, fall crops are normally entering their most luxuriant growth season, and their NDVI should normally approach its maximum. When floods occur, crops will generally be subject to fatal devastation. As a result, the NDVI of this stage in a flood year is much lower than its normal value. Therefore, it is concluded that August and September are the most appropriate months in which to check the NDVI change due to the influence of flooding. Further, the author attempted to delineate the possible flooded extent according to pixels' spatial correlations with the flood accident curve. The method used is to reclassify the spatial characteristic map of component 1 of the September series. Thus, possible flooded areas during 1995 through 1999 may be distinguished. As shown in Map 14, pixels in red represent the highly possibly flooded area, which are at the top 50% of positive index, showing strongly significant correlations to the curve. Pixels in yellow characterize the slightly possibly flooded area, which fall into the lower 50% of positive index, showing weak correlations to the curve. Pixels in green address the impossibly flooded area, which present no or negative correlations to the curve.
