3. Results and Discussions
For classified images in cool season (Figure 1), the overall accuracy assessed by the existing land use maps in 1995 were 80.64% for only daytime LST and 83.78% for combination of daytme LST (LSTd) and nighttime LST (LSTn), increased 3%. For classified images in hot season (Figure 2), the overall accuracy were 75.35% for only daytime LST and 79.40 % for combination between day and nighttime LST, increased 4%. Using only daytime data of NDVI and LSTd, Built up areas (Ba) and harvested paddy (Ph) cannot be separated. Ph in northeastern part of Thailand was mis-classified as Ba, particularly in hot season. Evergreen forest (Fe) in hot season was mis-classified as orchard (Or) and active paddy (Pa). These misclassifications were be corrected by using a combination of day and nighttime data. Because field crop (Cr) in hot season was in early planting or in harvesting period, thus, it was mis-classified as harvested paddy (Ph) and deciduous forest (Fd). The analysis between LST and two environmental factors, NDVI and elevation showed that LSTd was stronger correlated with NDVI than elevation while LSTn was fairly correlated with both NDVI and elevation (Chada, 2000). Thus, in case of Fd, its LSTd and NDVI were similarly with Cr, Ba, and Ph but its LSTn was remarkably lower than such land cover types because Fd situated on high elevation. Including channel 3 in the classifications gave good accuracy. As shown in Figure 3, the accuracy of three bands of daytime such as NDVI, band 3 and LST (81.61% in cool and 77.87% in hot season) was lower than of two bands of day and nighttime combinations such as NDVI and nighttime band 3 (82.43% in cool and 78.96% in hot), NDVI and nighttime LST (83.62% in cool and 79.85% in hot). These indicated that the accuracy was not improved by number of band but was improved by nighttime data.
The confused analysis of 7 main classes showed in Figure 4. In both cool and hot seasons, the accuracies of built up and forest were obviously increased using the combination between NDVI and LSTn. The best accuracy was obtained from paddy class. From the results, it should be explained that:
- Why accuracy of forest class was improved? Because forest has consistent temperature and homogenous vegetation cover whereas agricultural land has fluctuated temperature and heterogeneous vegetation cover depending on soil water regime, growing stage and land practices. In addition and more noted, the LST estimation is precisely in dense vegetated areas (Kerr et al., 1992). Daytime LST depends on NDVI and relative humidity while nighttime not only depends on NDVI and relative humidity but depends on latitude and elevation as well (Chada, 2000). Thus, forest in high lands where temperature in nighttime decreases can be well classified even though deciduous forest was falling in hot season.
- Why accuracy of built up class was improved? Because temperature difference between day and night times were widest compared with other land uses. Its concrete cover has highly temperature at daytime and has rapidly decreased temperature at nighttime.
- Why accuracy of paddy class was the best? Typical agricultural fields in Thailand, excepting paddy fields, are smaller than each AVHRR pixel, thus, most pixels contain a mixture of land cover classes while paddy classes are less unmixed pixels.
4. Conclusions
The new method by adding nighttime data was effective to classify land cover types by improving classification accuracy, particularly of forest, paddy and built up classes. Nighttime LST can well distinguish built up from harvested land, active paddy from active filed crop or perennial crop, and forest from active paddy and lowland tree. This indicates that nighttime data can be feasibly used to classify topographically related land cover types.
Acknowledgment
The authors wish to express appreciation to Dr. Kiyoshi HONDA, director of ACRoRS, for contributing the AVHRR images, and to Dr. Surat Lertlum, senior researcher of ACRoRS, for his helpful knowledge in NOAA-AVHRR processing. The author sincere thanks to Dr. Supan Karnchanasutham, head of Center for Agricultural Information, for his providing the updated land use map of Thailand.
