Arid Lands Newsletter (link)No. 51, May/June 2002
Using geospatial technologies to understand dryland dynamics

Using remote sensing and indigenous knowledge for management of ephemeral surface water

by Eric Patrick

"Two approaches to analyzing and interpreting satellite imagery in order to find suitable areas for water harvesting were compared. Results of this study indicated that visual interpretation by local land users was more fruitful that computer-based analyses of spectral data by an external researcher."

Spatial patterning in dry environments and effects of disturbances on these patterns

(Back to top)
In dry environments lack of rainfall, or a poor distribution thereof, leads to low levels of vegetation for much of the year, such that the soil surface is normally bare at the beginning of the rainy season. As a result, soil crusting often occurs after the initial rains, due to the beating effect of raindrops. These crusted surfaces almost always cause reduced infiltration rates, generating runoff which changes the geomorphology of the surrounding areas. In short, crusting is an important control on the (re)distribution of fluxes of moisture and nutrients, concentrating these scarce resources in sinks where they move up the trophic chain, for example in the form of grass, thereby playing an essential role in creating and maintaining productivity.

The spatial manifestation of this phenomenon is a characteristic pattern of relatively fertile patches interspersed amongst relatively barren, highly crusted supply zones from which water and nutrients are collected and concentrated in the fertile patches during flash floods. As this is an efficient form of making scarce resources available, and indeed is the basis logic of the ecosystem, it is important to determine the degree of disturbance both of the obviously valuable fertile resources but also of the apparently worthless runoff/nutrient source areas around them.

A number of studies have been carried out which attempt to assess the effect of surface disturbance in arid ecosystems. Belnap and Gillette (1997) determined friction threshold velocities for soil crusts in different stages of recovery from a disturbance. Particles on the surface of crusts that had been relatively undisturbed for at least 20 years were found to have significantly higher friction threshold velocities than those that had been disturbed 1, 5, or 10 years previously. This implies

a) that a surface disturbance can have an impact over at least a 20-year period; and

b) that aeolian erosion will occur at much lower wind velocities for disturbed surfaces, with consequences for ecosystem productivity through selective loss of the more fertile fine particles, as well as possibly causing respiratory problems for desert dwellers.

James et al. (1999) studied the provision of watering points in the Australian arid zone. They reviewed changes in vegetation in response to grazing around artificial sources of water, finding that the development of a zone of extreme degradation (up to 0.5 km) where soil crust is broken, resulting in a high rate of erosion once this protective cover is gone. A similar methodology was employed by Bosch and Kellner (1991), laying out transects both on time series imagery and on the ground moving outward from areas of concentrated populations of animals.

The potential of remote sensing for quantifying and monitoring landscape patterns

(Back to top)
There are many successful examples of the use of remote sensing as a key data source in the study of landscape ecology (Frohn 1998), so long as calibration can be achieved to ground conditions. Indeed, it may be possible to use range exclosures as a baseline for 'virgin' conditions, as has been attempted in Kuwait (Omar 1991) for comparison to areas outside as well as subsequent monitoring of changes. Principles employed in the Rangeland Soil Condition Assessment Manual developed for Australia (Tongway 1994) could be adapted for other areas in order to rate range degradation, as the low rainfall in areas such as the Jordanian desert means that an assessment of the soil surface conditions is a more reliable indicator of range condition than is biomass, which may not be present at the time of the survey and may be of highly variable quality.

Water harvesting for managing ephemeral surface water resources

(Back to top)
Water harvesting is a term which is commonly understood to mean the collection and therefore concentration of runoff from a larger area than the cropping area. Issues of definition and of typology of water harvesting systems, which would serve as a basis for a methodology for assessing suitability of a particular area of the landscape for water harvesting, are beyond the scope of this study. However, a basic distinction relevant here is that between micro and macro water harvesting. The former refers to 'within field' and the latter to external catchment runoff collecting areas. Water harvesting mimics, takes advantage of, manipulates and improves upon natural patterns of rainfall redistribution which occurs on crusted surfaces with low vegetation cover (at least at the end of the dry season). Profiting from the fact that rainfall which does not immediately infiltrate becomes runoff which eventually infiltrates in a sink area some distance from the source area, water harvesting collects runoff from a catchment area normally many times larger than the receiving area, thereby multiplying the effective rainfall at the sink. This serves to mitigate both the spatial and the temporal variability characteristic of rainfall in dry areas, by storing water in the soil profile or some form of reservoir. The water thus harvested can be used for grass or cereal or tree production, or indeed for livestock or domestic water if stored in a reservoir or if water can be pumped from the soil.

Soil-loosening disturbances to a natural ecosystem, such as grazing, can become advantageous if the nutrients in that sediment settle in a sink downslope where soil fertility becomes enriched. Such a distribution of nutrient and moisture rich areas interspersed between -- indeed normally harvested from -- larger areas of poorer resources mimics the natural patchiness characteristic of dry areas.

The physical effectiveness of water harvesting has been established beyond a doubt both by its use in semi-arid and arid areas around the world for at least 8,000 years (Barrow 1999), as well as by modern studies of soil water balance and yield improvements vis-à-vis in situ rainfall receiving controls. Water harvesting has always received much greater attention from practitioners than researchers, and as the 'twain often n'er will meet', the amount of documented research into water harvesting is remarkably small compared to both its historical importance and its current potential.

Geomorphology of case study areas and political ecology of water use

(Back to top)
In the case of the Kenyan study area, the lowlands of Baringo district, two principle zones were identified at the macro scale: the Tugen Hills, step-faulted (horst and grabben) walls of the Rift Valley, enclosing a fluvio-lucustrine valley of loamy soils connecting the Rift Valley lakes. Average rainfall is 650 mm/yr but with a bimodal distribution. This means that the area is best suited for grazing, as the dry period between the peak rainfall periods is often too long for food crops to survive without supplemental irrigation. Within the low slope valley area there are many linear features: gullies, often several kilometres long, which are formed as a result of large, channelled volumes of water emerging from the Tugen Hills during large rain events, but which grow at the headcut area upslope due to local runoff within the valley during smaller events. The spatial distribution of perennial vegetation, which comprises mainly various species of Acacia, correlates strongly to these linear features, due to residual soil moisture and relatively low evaporation. Irrigated agriculture exists at three or four schemes scattered around Lake Baringo, making use of the reasonably reliable rivers which flow seasonally into the lake. Thus, natural water harvesting is occurring at a macro scale in this area. Unfortunately, because of the desirability of the land within irrigation reach of these rivers, tribal politics has partly dictated who benefits from the government-subsidized schemes; often relatively new outsiders are the ones to benefit. A more distributed strategy of water harvesting, albeit at a smaller scale, would help mitigate this unsustainable situation.

In Jordan, the study area is the panhandle region comprising the northeastern portion of the desert. Here, fertile 'oases' (in terms of sinks for runoff and nutrients) occur, especially within the basaltic al-Harra geological unit, as small 'floodplains' or marabs and as enclosed basins or qa'a. Both of these are readily apparent from satellite imagery, and understandably are highly coveted by livestock owners because of good grazing. There are also areas in the wadis known as ghudrans, which transmit runoff from the surrounding rock covered hills; the collected runoff sometimes lasts for many months. Rainfall in this area is very low, ranging from 90 to 150 mm/yr. The dominant land use is pastoralism, with some opportunistic sowing of forage crops and with some areas of permanent agriculture where the state has subsidized groundwater extraction. Often the beneficiaries of drilling permits and subsidies are well-connected outsiders, leaving the traditional occupants of the region to rely on government wells in order to water their animals. Rich grazing occurs in the marabs and, depending on the degree of salinity, in the qa'a; however, the richness of these resources has made them a cause of tension, with better-connected individuals, especially sheiks, making de facto claims on them. For these reasons it would be both environmentally and socially appropriate to seek to develop other surface water resources in the area.

Water harvesting in the case study areas

(Back to top)
Water harvesting has been practiced in Jordan for at least 6,000 years (the oldest site being at the Neolithic settlement of Jawa, within the study area). It was used to maintain outposts and even large towns by the Romans, Byzantines, Ummayad Caliphate, and on a smaller scale by the Bedu. There is also much evidence of water harvesting of one variety or another having been implemented by local people in Kenya. Many projects in Kenya have promoted water harvesting, starting with an attempted transfer of Israeli experience to Turkana, which was not very successful for both technical and socio-economic reasons. In the case study areas a number of development programs have promoted water harvesting; however, none appear to have taken into account indigenous traditions of opportunistic water harvesting (diversion of flow from large gullies, a form of spate irrigation).

Using remote sensing as a tool for understanding surface hydrology and identifying suitable areas for water harvesting

1) Kenyan case study: Computer-based analysis by an external interpreter

(Back to top)
In the Kenyan study there are two distinct patterns of biomass which can be detected from remote sensing. These patterns are potentially useful for reconstructing moisture (and nutrient) fluxes in the system; such fluxes are difficult to observe directly over a large area. As described earlier, one such pattern is that of storied vegetation, mainly trees, lining at high density the linear gully features which are very evident on the imagery. Although they represent important dynamics, their percentage cover of an image is very low, meaning that appropriate algorithms are necessary so as not to miss these features. Unsupervised classifications of multispectral SPOT and Landsat Thematic Mapper (TM) images were found to be good at revealing the overall pattern of land units in the area, but interpretation of these patterns required a familiarity with the area and an understanding of the surface hydrology of this environment. The other pattern, that of the crusted intergully areas used for grazing, is subject to considerable temporal variability, which would require a large number of image dates to analyze properly. Although grass, which has shallow roots, is a sensitive indicator of water, in order to understand the dynamics of the system, imagery with a reasonable resolution, such as SPOT, would have been required for such analysis. This approach proved to be too expensive, and the alternatives, such as Advanced Very High Resolution Radiometer (AVHRR) images, were too coarse-grained (approximately 1 km effective pixels) to provide the necessary resolution.

Analysis of the imagery did not prove to greatly add to the understanding of patterns (at a finer scale) which could be observed from fieldwork; however, it did assist in ranking the relative size of each land unit as well as confirming the functional relationships between them. Supervised classification of the scene based on ground verification data of different types of crusts over the large, relatively homogenous valley bottom was carried out because the crusted surfaces are potentially indicative of movement of sheet (and rill) flow, as well as being source areas of local runoff. This exercise again proved to be disappointing, presumably because the spectral differences between the crust types were too subtle, and varied at too fine a scale, to be distinguished in the imagery. Albedo (total reflection) proved to be the most relevant parameter, rather than any particular wavelength. This is because albedo is in part a function of the microtopography of the surface, which in turn was found to be related to the severity of crusting, which in turn is related to a combination of soil erodibility and the intensity of surface water fluxes across them. As such, albedo could potentially be used to map crust type, but again the distinctions between crusts originating as in this case due to physical degradation are subtle. Crusts of biological and chemical origin are likely to be more readily distinguished on the basis of their spectral characteristics.

Remote sensing might have proved to be more relevant in terms of revealing process from pattern in another landscape; however in this case there was clear evidence for self-similar scaling. In other words, the patterns observed at a fine scale and the patterns observed at a coarse scale are very similar, because both in this case are driven by the movement of water across the surface of the landscape in response to differences in potential energy (i.e. topographic differences). Indeed, a transect down a gradient from the contact zone with the Tugen Hills to Baringo Lake confirmed a decreasing infiltration rate of soil toward the lake, due to the dispersal and sedimentation of fines brought down with the 'flash floods' which are transmitted mainly as channel flow through the gullies and then spread out at the end of the gully.

2) Jordanian case study: Participatory image interpretation

(Back to top)
In response to the disappointing experiences with computer-based analysis of satellite imagery in Kenya, another approach was attempted in the Jordan case study. Images were enhanced through stretching and the use of edge filters (to highlight wadis), amongst other routines, in order to maximize usability for visual interpretation. Key place names were elicited from knowledgeable local people and added to the images using another software. Features digitized from 1:250,000 maps and stored in the form of thematic layers in a GIS were selected on the basis of balancing the need for orientation against the desire to avoid cluttering the image, and were superimposed on the images after rasterizing these vector features in order to make them compatible with the data structure of the pixels. These images were then used in three forms with local people as the principal interpreters. Large paper copies (A0 size, or 841 mm x 1189 mm) were printed off a plotter; somewhat smaller photographs were printed onto photographic-type paper from digital files at a commercial printshop. The files were also kept on the hard drive of a military specification fielduse laptop (dust, water and impact resistant) in order to be viewed 'live', which allows panning and zooming in response to image interpreters' requests. This in turn has the significant advantage of providing interpreters with both an overview and, as required, a maximum resolution image at any point of interest in the landscape.

Knowledgeable local land users, principally elderly men who had been shepherds for many years, were invited to the tent of a host, in order to reproduce the traditional decision-making environment. After making a fire and having tea, the researcher was introduced by the host, a respected individual known in the community, and the objective of the meeting discussed very briefly. This was done intentionally in order to avoid biasing the discussions. The technical objective of the exercise was to determine whether indigenous knowledge, as applied to satellite image interpretation, would be more efficient at identifying suitable areas for water harvesting than computer-based analyses by external agents. Given that the definition of 'suitable' was taken to be based not just on environmental criteria but also on socio-economic criteria, this was considered to be a promising approach. The more general objectives were to a) determine to what degree persons who had (in most cases) never seen a plan view image, or indeed in many cases even a map, could relate to the imagery; b) observe the decision-making process; and c) gain insights into the hierarchy of criteria used to make decisions. Free flow discussions, translated live by a Bedu who had done a PhD abroad and thus could act as an intermediary or cultural bridge, were noted during the meetings.

The results of the general objectives are beyond the scope of this paper; however, the specific objective was satisfied beyond expectation. Within 30 to 45 minutes, informants had oriented themselves to the imagery and started to identify, apparently with great precision, the locations of key wadis and ghudrans, as well as the best locations for artificially impounding wadi flows. A sample of the recommended locations was subsequently visited and the locations were found to be highly suitable for water harvesting based on environmental criteria, as well as on other criteria employed by the informants, such as access for water tankers.


(Back to top)
Two approaches to analyzing and interpreting satellite imagery in order to find suitable areas for water harvesting were compared. Results of this study indicated that visual interpretation by local land users was more fruitful that computer-based analyses of spectral data by an external researcher. The involvement of local land users might be termed 'Participatory Image Interpretation'; it holds great promise for dry areas because the normally low economic value of these areas typicaly results in a deficit of data about them. Furthermore, because of the large extent of these environments and the relatively low population density, it is very expensive to inventory them using traditional survey methods. This limitation is mitigated by employing local land users - to the degree to which they have been mobile around the area. Finally, involving the potential beneficiaries of any resulting development intervention helps ensure that the project design will be suitable on more grounds than just environmental criteria and will thus be more acceptable to local populations.


(Back to top)
Barrow, J. 1999. Alternative irrigation. London: Earthscan.

Belnap, J. and D.A. Gillette. 1997. Disturbance of biological soil crusts: impacts on potential wind erodibility of sandy desert soils in southeastern Utah. Land Degradation and Development 8(4):355-362.

Bosch, O. and K. Kellner. 1991. The use of a degradation gradient for the ecological interpretation of condition assessments in the western grassland biome of southern Africa. Journal of Arid Environments 21: 21-29.

Frohn, R. 1998. Remote sensing for landscape ecology. Baton Rouge: CRC Press.

James, C.D., J. Landsberg and S.R. Morton. 1999. Provision of watering points in the Australian arid zone: A review of effects on biota. Journal of Arid Environments 41(1): 87-121.

Omar, S. 1991. Dynamics of range plants following 10 years of protection in arid rangelands of Kuweit. Journal of Arid Environments 21:99-111.

Tongway, D. 1994. Rangeland soil condition assessment manual. Canberra: CSIRO.

bar denoting end of article text

Author information

(Back to top)
Dr. E. Patrick is a Research Associate in the Geomatics Research Group, Geography Department, Carleton University, Ottawa, Canada. You can reach him for comment by email at:

About the Arid Lands Newsletter

Link to ALN home page Link to index page for back web issues Link to index page for pre-web issue archive Link to ALN #51 table of contents