Flood monitoring in Hainan Island based on HJ-CCD data

Abstract: Flood monitoring is one of the main tasks of agricultural monitoring. For surface observation, it consumes many manpower, resources as well as delayed information, all of those can be overcome by use of remote sensing monitoring flood disaster. The spatial and temporal resolution of HJ-1A/1B-CCD data is 30 m and 2 d. HJ-1A/1B-CCD data are desired remote sensing data of flood monitoring. We chose Hainan Island as research area and selected 400 training samples. Of them, 100 samples were from surface waters and wetlands, respectively, and 50 samples were respectively for forest lands, cultivated lands, roads and settlement places. In this paper, we analyzed and compared three water indexes: normalized difference water index (NDWI), normalized difference water index based on blue light (NDWI-B) and combined index of NDVI and NIR for water body identification (CIWI) by using the division degree (DD). Result showed that when NDWI-B was used to extract pure water, the division degrees of pure water to forest land, cultivated land, road and habitation were all between NDWI and CIWI. The division degrees of NDWI-B in these aspects were obviously higher than NDWI and CIWI when extracting wetland, which meant that when applying HJ-1A/1B CCD data to extract lack branch, small water bodies and wetland, the NDWI-B was better than NDWI and CIWI. The comprehensive discriminations of NDWI-B model were 31.30% and 28.13%, respectively. Thus, NDWI-B model was the optimal model when carrying out flood monitoring in Hainan Island. Based on the error matrix theory in combination with the actual measurement of sample data, the precision of water (pure water and wetland) in Hainan Island extracted by the NDWI-B was verified. Result showed that the user accuracy of water and non-water were 90.29% and 92.78%, respectively, and the overall accuracy was 91.50%. Through repeated comparison between the water index value and the surface feature type in the sampling point, when using NDWI-B model to extract water, the water scope of threshold value -0.015 was the largest, the broken figure spot was relatively less, and it was easy to exclude road and habitation. As such, the water identification's threshold value of NDWI-B was ensured, which was -0.015. Remote sensing images with better data qualities during the monitoring period (Sep. 26th, Oct. 6th, Oct. 12th, Oct. 20th, 2010, time series data) were chosen and the flood disaster conditions in Hainan Island from Sep. 25th, 2010 to Oct. 25th, 2010 were monitored by the NDWI-B. Result showed that the disastrous situation on Oct. 12th was the severest. The inundated area of the whole island reached 120.22 km2, the ceiling value during the monitoring period. Newly-emerged water bodies, mainly spreading over villages, farmlands and roads, were in all cities and counties except Dongfang, Changjiang and Ledong. Regionally, Wenchang, Qionghai, Haikou and Ding'an in the east were severely flood affected areas; Dongfang, Changjiang and Ledong in the west were less affected. The most affected land-use type by the flood in Hainan Island was the paddy field, whose inundated areas were 54.31 km2, 61.46 km2, and 42.58 km2 on Oct. 6th, Oct. 12th and Oct. 20th, respectively. The secondly affected land-use type was the dry land. Its inundated areas were 22.56 km2, 29.59 km2 and 18.36 km2 on Oct. 6th, Oct. 12th and Oct. 20th, respectively; the inundated areas accounted for 73.13%, 75.73% and 71.37%, respectively in the whole inundated area. Besides, taking Sanjiang farm in Meilan District, Haikou City as an example, we calculated the drowned water depth in the farm at October 6th, 2010. The result showed the drowned water depth of some fishpond and cultivated lands in Sanjiang farm was more than three meters. The flooding and water logging disaster situation was quite serious. Study showed that when recognizing the pure water the distinction degree of NDWI-B was a little lower than that of the CIWI, but high when recognizing the wetland. Besides advantages of larger land and water distinction degree and higher extracting precision of water area, the NDWI-B can also identify the water and wetland in small area. It was a relatively ideal model in monitoring flood disasters in Hainan Island. This paper provided references for the water resource management, the dynamic monitoring of flood disaster and disasters prevention as well as damages reduction in Hainan Island.