Runoff cannot be directly measured by remote sensing techniques. However, there are two general areas where remote sensing can be used in hydrologic and runoff modeling: (1) determining watershed geometry, drainage network, and other map-type information for distributed hydrologic models and for empirical flood peak, annual runoff or low flow equations; and (2) providing input data such as soil moisture or delineated land use classes that are used to define runoff coefficients.
Remote sensing data can be used to obtain almost any information that is typically obtained from maps or aerial photography. In many regions of the world, remotely sensed data, and particularly Landsat, Thematic Mapper (TM) or Systeme Probatoire, d'Observation de la Terre (SPOT) data, are the only source of good cartographic information. Drainage basin areas and the stream network are easily obtained from good imagery, even in remote regions. There have also been a number of studies to extract quantitative geomorphic information from Landsat imagery [ Haralick et al., 1985]. Topography is a basic need for any hydrologic analysis and modeling. Remote sensing can provide quantitative topographic information of suitable spatial resolution to be extremely valuable for model inputs. for example, stereo SPOT imagery can be used to develop a Digital Elevation Model (DEM) with 10 m horizontal resolution and vertical resolution approaching 5 m in ideal cases [ Case, 1989]. A new technology using interferometric SAR has been used to demonstrate similar horizontal resolutions with approximately 2 m vertical resolution [ Zebker et al, 1992].
Empirical flood formulae are useful for making quick estimates of peak flow when there is very little other information available. Generally these equations are restricted in application to the size range of the basin and the climatic/hydrologic region of the world in which they were developed. Most of the empirical flood formulae relate peak discharge to the drainage area of the basin. See United Nations Flood Control Series No. 7 [ United Nations, 1955]. Landsat data have been used to improve empirical regression equations of various runoff characteristics. For example, Allord and Scarpace [1979] have shown how the addition of Landsat derived land cover data can improve regression equations based on topographic maps alone.
One of the first applications of remote sensing data in hydrologic models used Landsat data to determine both urban and rural land use for estimating runoff coefficients [ Jackson et al, 1976]. Land use is an important characteristic of the runoff process that affects infiltration, erosion, and evapotranspiration. Distributed models, in particular, need specific data on land use and its location within the basin. Most of the work on adapting remote sensing to hydrologic modeling has involved the Soil Conservation Service (SCS) runoff curve number model [ U.S. Department of Agriculture, 1972] for which remote sensing data are used as a substitute for land cover maps obtained by conventional means [ Jackson et al, 1977, Bondelid et al, 1982].
In remote sensing applications, one seldom duplicates detailed land use statistics exactly. For example, a study by the Corps of Engineers [ Rango et al., 1983] estimated that an individual pixel may be incorrectly classified about one-third of the time. However, by aggregating land use over a significant area, the misclassification of land use can be reduced to about two percent which is too small to affect the runoff coefficient or the resulting flood statistics.
Studies have shown [ Jackson et al., 1977] that for planning studies the Landsat approach is cost effective. The authors estimated that the cost benefits were on the order to 2.5 to 1 and can be as high as 6 to 1, in favor of the Landsat approach These benefits increase for larger basins or for multiple basins in the same general hydrological area. Mettel et al, [1994] demonstrated the recomputation of pmf's for the Au Sable River using Hydrologic Engineering Center (HEC-1) and updated and detailed land use data from Landsat TM resulted in 90% cost cuts in upgrading dams and spillways in the basin.
Other types of runoff models that are not based only on land use are beginning to be developed. For example Strubing and Schultz [1983] have developed a runoff regression model that is based on Barrett's [1970] indexing technique. The cloud area and temperature are the satellite variables used to develop a temperature weighted cloud cover index. This index is then transformed linearly to mean monthly runoff. Rott [1986] also developed a daily runoff model using Meteosat data for a cloud index. Recently, Papadaakis et al, [1993] have used a cloud cover index from satellite imagery to estimate monthly area precipitation. A series of non-linear reservoirs then transforms the precipitation into monthly runoff values. This approach was successfully demonstrated for the large (16000 sq km) Tano River Basin in Africa and illustrates the value of remote sensing data when conventional data are not readily available. Ottle et al [1989] have shown how satellite derived surface temperatures can be used to estimate ET and soil moisture in a model that has been modified to use these data. Duchon et al. [1992] have used Landsat to identify uniform land cover areas and GOES data for input insolation for a monthly water balance model.
The pixel format of digital remote sensing data makes it ideal for merging with Geographical Information Systems (GIS). Remote sensing can be incorporated into the system in a variety of ways: as a measure of land use and impervious surfaces, for providing initial conditions for flood forecasting, and for monitoring flooded areas [ Neumann et.al., 1990]. Kite and Kouwen [1992] have shown that a semidistributed model based on Landsat derived subbasins performed better than a lumped model. The GIS allows for the combining of other spatial data forms such as topography, soils maps as hydrologic variables such as rainfall distributions or soil moisture. This approach was demonstrated by Kouwen et al, [1993] where their Grouped Response Unit (GRU) included satellite based land use and lies within a computational element that may be either a sub basin or an area of uniform meteorological forcing. In HYDROTEL, Fortin and Bernier [1991] propose combining SPOT DEM data with satellite derived land use and soils mapping data to define Homogeneous Hydrologic Units (HHU). In a study of the impact of land use change on the Mosel River Basin, Ott et al, [1991], and Schultz, [1993] have defined Hydrologically Similar Units (HSU) by DEM data, soils maps and satellite derived land use. They also used satellite data to determine a Normalized Difference Vegetation Index (NDVI) and a leaf Water Content Index (WCI) which are combined to delineate areas where a subsurface supply of water is available to vegetation. Mauser [1991] has shown how multitemporal SPOT and TM data can be used to derive plant parameters for estimating ET in a GIS based model.
There continues to be speculation about the potential value for soil moisture data as in input variable in hydrologic models, either to establish the initial conditions for simulating storm runoff, or as a descriptor of hydrologic processes. Aircraft data taken during the First International Satellite Land Surface Project (ISLSCP) Field Experiment (FIFE) campaign were used to map the spatial patterns of soil moisture resulting from drainage and ET in a 37.7 ha watershed [ Wang et al, 1989]. These patterns were seen to match the results of a simple slab model and identified the region contributing base flow to the channel [ Engman et al, 1989]. Attempts to use passive microwave measurements in small watershed showed good correlation with the ground data and may yield a reliable technique for calibrating the model [ Wood et al, 1993]. Also, even the relatively low-resolution passive data can improve the water budget calculations of a small basin [ Lin et al., 1994]. Goodrich et al, [1994] studied the prestorm soil moisture at various scales of basin runoff. They concluded that initial values were important but that the resolution of the final remote sensing product was not a limitation.
The area inundated by floods and floodplains can be mapped effectively with remotely sensed data. Satellite data such as those from Landsat and SPOT can be used to define coverage of an entire river basin if weather conditions are favorable. The effects of flooding may be detected for up to two weeks or longer after the passage of a flood; thus it may not be necessary to obtain data exactly during the flood peak. [ American Water Resources Association [1974]). All weather flood mapping is possible with microwave sensors. Radar systems are capable of higher spatial resolution than passive systems under similar situations and should be well suited for this task [ Imhoff et al., 1987]. Koblinsky et al., [1993] discuss how satellite altimetry may be useful in measuring the river levels to estimate discharge or to monitor flooding. Floodplains have been delineated using remotely sensed data to infer the extent of the floodplain from vegetation changes, soils or some other cultural features commonly associated with floodplains [ Rango and Anderson, 1974].