关键词: PL影像, 火烧迹地, 归一化植被指数, 火烧程度

Abstract:

Objective: This paper aims to study the feasibility of forest fire image information extraction and data analysis by only using one high-resolution remote sensing image (Planet Labs, PL), so as to provide reliable remote sensing data source and extraction methods for forest fire assessment. Method: In this study, the burned area of the "5·2" big forest fire in Bilahe in 2017 was used as the study area, three phases of PL images including before the fire, the year after the fire and one year after the fire were used as the data source, and the burned area was extracted by using ROIs. The variation characteristics of NDVI before and after fire interference were analyzed. Combined with the ground survey data, the difference normalized vegetation index (dNDVI) was used to divide the fire severity level, and the threshold was verified based on the classification standard of Luo Dekun fire damage levels. The damage status of vegetation in the burned area was evaluated to obtain the spatial distribution pattern of the degree of fire. Result: Fire disturbance led to a sharp decrease in NDVI values, and the NDVI value slightly increased one year after fire, indicating that the ability of vegetation restoration was very limited. The 3 m high spatial resolution of PL Remote Sensing Image made its RGB image highly saturated and every land class clear, and directly classified the land cover types. The training sample separation was above 1.91, the land cover types could be divided into 4 types: forest, herbaceous swamp, road and river. The classification accuracy was 98.05% and the Kappa_Coefficient value was 0.95. The fire severity was classified into 4 levels: unburned area, light burned area, moderate burned area and severe burned area. The overall classification accuracy was 91.55%, and the Kappa_Coefficient was 0.91. The total study area in Bilahe was 10 711.18 hm², and the burned area was 10 130.31 hm², accounting for 94.58% of the total fire area. Most burned area was light burned one with 5 700.78 hm², accounting for 53.22% of the total fire area., The moderate burned area was secondary, with an area of 3 035.12 hm², accounting for 28.34%. The damaged forest area affected by moderate burned was the largest, up to 6 167.48 hm², accounting for 60.88% of the total burned area. The secondary damaged forest area was in the light burned region, reached to 1 846.93 hm², accounting for 29.95%. The damaged forest area in the severe burned swamp region was the least, accounting for only 22.60%. The damage of herbaceous swamp was less, with a burned area of 3 962.86 hm², of which 97.25% of burned swamp area was concentrated in the light burned region. Conclusion: Compared with the traditional research methods, the research method used in this paper can achieve more accurate results in the analysis of forest fire in Bilahe, with higher classification accuracy, more reliable results and more efficient data processing. At the same time, PL remote sensing has ultra-high frequency time resolution, which covers the whole world once a day, and satisfies the requirements of covering different research areas to a great extent.

Key words: PL images, burned area, NDVI, fire severity