Estimation of depth, extent and volume in flooding in the Hadejia-Nguru Wetlands of Nigeria, using remote sensing

This study aims to develop models predicting the depth, extent and volume of flooding in the Hadejia-Nguru wetlands of Nigeria from satellite sensor images. In subregions representative of the entire wetlands over 1,500 rigorously coordinated and geocoded depths were observed, simultaneously and near-simultaneously, with Landsat-5 satellite overpass of 2 September 1990. Over 1,000 of the measured depths were reduced to water levels of 2 September 1990 and accurately calibrated to corresponding pixels on the Landsat-TM image of the same date. Depth-radiance power-curve relationships were established using regression analysis based partly on relationships from ground radiometry. An operational model for flood prediction was successfully developed. Area, volume and average depth of flooding predicted from Landsat-TM satellite sensor data were, respectively, 1186km2, 560.92 million m3 and 0.66m on 26 September 1986; and 910km2, 430.79 million m3 and 0.66m on 2 September 1990. Mean water-depths were predicted in open waters from the Landsat-TM imagery with a confidence interval of between 9-20% at depths of 0.10- 6.20m; and in inundated vegetation to 12-14% at 0.25-0.75m depths (and 30-50% at depths less than 0.25m or greater than 0.75m). Applying the developed depth-radiance equations to NOAA-11 AVHRR satellite sensor data of 24 August 1990, they overestimated flood extent by 6% (52 km2) and underestimated volume by 18% (78.8 106 m3). Simulated Meteosat satellite data overestimated flood extent by over 6% (55 km2) and underestimated volume by 37% (160.2 106 m3). Using imagery from 1986, 1987, 1990 and 1991 the frequency of flooding was found to vary spatially in about 50% of the wetlands every 4 or 6 years, with only about 5.6% (242.1 km2) and 2.2% (94.5 km2) of the region flooded 3 times and 4 times, respectively. This means that conventional techniques alone cannot be used to adequately monitor flooding in the wetlands. Operational problems encountered with using satellite sensor images were examined. A significant problem is the low probability of acquiring a cloud-free scene from Landsat-5 satellite (eg a probability of 1 in 32 days at 0.50 average cloud-free cover) in the wet season which is the peak flood period. Also, there is little or no correlation between satellite and ground radiometry data due to their different field- of-views. Vegetation and turbidity also poses problems in deriving relationships between water-depth and satellite pixel radiance. However, the depth-radiance prediction models developed in this research are operational and (statistically) more accurate than some similar models recently evaluated.