Digital image processing for large scale irrigation management and monitoring

Digital image processing for large scale irrigation management and monitoring

Janet E. Nichol
Department of Geography, National University of Singapore

Introduction
Accurate monitoring of the social and environmental impacts of large dams in often still only a minor concern in total project design (Adams, 1985: Barrow, 1987: Nichol, 1987). One of the frequently cited explanations for this shortcoming is a lack of data coupled with the difficulty of obtaining it. Project implementation is often the main concern of governments, funding agencies and contractors. Post-project appraisals are therefore often subjective and lack empirical evidence (Barrow, 1987, p. 134).

Te types of data required for impact assessment and management are not easily obtained by conventional methods since they demand observations at regular intervals over long periods, often covering catchments thousands of square kilometers in extent. These types of data include soil erosion, sediment yield, and changes in the water regime and vegetation cover. Increased human activity in an area frequently follows dam construction, and these impacts are likely to be observed throughout the upstream as well as downstream catchments.

Hydrologic impacts of dam building are usually long term, requiring observation for a subsequent period of several years. Land downstream previously subject, to annual flooding may take several years to dry out completely. Thus associated changes in vegetative intensity may not e apparent immediately. On the other hand geomorphic changes such as gull eying and soil erosion tent to occur most dramatically in the year immediately following dam construction.

Since changes in the type and amount of vegetation are intimately associated with all the major environmental factors surrounding water resource development schemes, vegetation monitoring is seen as a key factor in impact assessment.

This study describes technique of using digital satellite data fro measuring changes in vegetation intensity in a reservoir catchment, and for obtaining data on water surface area for input into hydrological models.

Compared with other data sources, satellite data is relatively cheap, once image processing equipment has been obtained. Since the levels of detail increasingly ressemble those obtainable from air photos, land cover of large areas can be rapidly and cheaply monitored. The data is digital and quantifiable, thus objective comparisons can be made between different time periods if other factors are constant.

The study area
The study was carried out in Kano State of Northern Nigeria, where numerous earth filled dams were constructed in the late 1970s and early 1980s. due to its accessibility and proximity to Kano City, a case study was carried out of the Jakara Catchment, (figure 1) from the edge of Kano city to the river's disappearance in the sediments of the Chad Formation, approximately 40 Kilometers tot eh north-east. This was accomplished using a 512 x 512 pixel extract from LANDSAT MSS path 202, Row 52 covering can area o 1,600 square kilometers, for which two dates were available : 10.1.76 (Plate 1), and 19.11.1978.

The Jakara catchment, an area of 559 sq. kms is located within the Sudan Savanna zone of northern Nigeria, a region of semi-arid climate with an average of 800mm rainfall in the May to September we season. Thus non-irrigated cultivation is limited tot eh wet season except in the seasonally flooded fadama (flood plain) areas. The fadama areas which comprise less than 10% of the total land area in northern Nigeria therefore constitute a very important resource whose continued productivity is vital to the local economy.

In 1976 Jakara dam was completed, approximately 20 kilometers downstream from Kano City, and the valley was first flooded during the 1977 wet season. This resulted in the loss of well over 1000 hectares of farmland due to indunation. (The water Resources and Engineering Construction Authority gives 1959ha. projected water surface area estimate (water resources and Engineering and Construction Authority , 19800, much of it fertile fadama (floodplain) as well as much fadama land further downstream due to the controlled water regime (Nichol. Op. cit). The scheme, costing well over US$ 3 million at construction, was originally conceived mainly for the purposes of irrigation, with the aim of irrigating 2150 hectors, though no irrigation works have been commenced to date.

Figure 1 Study Area

Plate 1. Vegetation enhanced false colour composite January 1976 (pre-dam).
Kano city in bottom left

Data collection
The two dates of imagery for which CCTs were available encompassed three wet seasons and three dry seasons, but only two wet seasons and one dry season following dam construction. A further disadvantage was the seven week difference in state of dry season between the two images. Thus the November 1978 image in early dry season would expectedly be more vegetated that that of January 1986.

However, since one aim of the study is to demonstrate a methodology, for detecting change, identification of the expected seasonal changes by the techniques used confirms the potential of the method. Image processing was carried out on the Iconoclast Image Processing System at Aston University. U.K.

Image processing techniques

Vegetation Index
The Normalized Difference Vegetation Index (NDVI)effectively exploits the large difference in radiance values of green vegetation between the red and infra-red bands (MSS bands 5 and 7) by rationing, thus : -

NDVI - MSS 7-5 / MSS 7+5
This effectively creates a new band which can be displayed as a vegetation index image with high values representing vigorous vegetation areas as light tone, while less vegetated areas appear darker.
Change detection
Digital change detection is difficult to carry out accurately. The results are not as accurate as those produced from the visual interpretation of large scale air photos of different dates and transfer of boundaries to a map. However, it is many times quicker and cheaper. Accuracy depends on the ability to acquire comparable imagery or different dates, and to geometrically register the images to the same geographical reference system.

Two techniques are described here :
1. Image differencing and
2. Image overlay
Image differencing
This involves the subtracting of one band of the imagery of one date from that of another. The subtraction results in positive and negative values in areas of radiance change and zero value in areas of no change. This yields a difference distribution which is Gaussian, where pixels of 0change value are distributed around the mean and change pixels are in the tails of the distribution. A critical factor in the method is in deciding where to put the threshold boundary between change and no pixels. Often, one standard deviation from the mean is selected then tested empirically.
Image overlay
This method, described by Howarth and Boasson (1983) involves displaying the MSS band 5 image of one date in blue on the monitor and the same band of a different date in red, areas of no change appear grey, and areas of change appear with differing intensities of blue or depending on the direction and degree of change.

Detection of change in vegetative intensity
Correlation for atmospheric differences between the two dates of imagery was based on examination of the pixel radiance values for a part of the scene least likely to be affected by temporal change. A densely built-up area of Kano City was chosen since seasonal and longer term differences in land cover here might be expected to be relatively low. Mean pixel readings were a follows :

MSS 5	MSS 7
Jan. 1976/36 Nov. 1978/42	Jan. 1976/50 Nov. 1978/50

The readings suggest that atmospheric scattering at the time of the satellite pass on 10.01.76 had the effect of decreasing radiance values in band 5 by a factor of approximately 6. This was confirmed by observations of the statistics for other land cover types. Thus a correction of adding 6 to every pixel of band 5 of the 1976 image was made, before the vegetation index ratio was carried out. A vegetation index image was then created for each date and the 1978 image geometrically corrected to that of 1976. A root mean square error of 1.22 of a pixel (i.e. 0.2% mean geometric error) was obtained. This was considered acceptable.

The image differencing technique w applied by subtracting the 1978 image from that of 1976 thus producing a single image of vegetative change. This multi temporal vegetation index image can be expressed as : -

MTVI = NDVI (76) - NDVI (78)
Visual inspection of the MTVI image suggested little change in vegetative intensity in the Jakara catmint other than that which can be attributed to expected seasonal change. This is hardly surprising in view of the short time elapsed since dam construction.

In order to quantify and enhance the change, the MTVI image was density sliced to limits of one standard Deviation below and above the mean to give areas of positive and negative change respectively (figure 2 and plate 2).

Area exhibiting light image tone, representing decreased vegetative vigour include the flooded reservoir site, as well s areas on the periphery of Kano city where previously vegetated sites may have made way for urban growth. A small area adjacent to the dam site and downstream flood plain which may have suffered loss of vegetation cover and subsequent soil erosion at construction, and some village resettlement sties also show signs of decreased vegetative intensity. Increased vegetation is much more widespread on the image due to the later image being earlier in the dry season. This is particularly apparent in the low lying land between Kano city and the reservoir site, as well as the downstream floodplain with residual soil moisture. Patches of floating vegetation in the reservoir exhibit no change.

The second method of assessing vegetation change that of Image Overlay, gave more visual information than Image Differencing, with apparently more areas exhibiting change in vegetative intensity. Since two image are involved there is no need for density slicing to obtain a color image; thus more subtle differences are evident. However, the method is purely visual and relies on the availability of a color monitor for display. The results obtained are not quantifiable and threshold limits between change and no change cannot be applied.

The monitor display showed areas of increased reflectance, representing decreased vegetation, in red. More areas of decreased vegetation were evident than on the MTVI image including, in addition, a newly eroded gully downstream of the dam, and several of the main tributary valley upstream. Blue areas representing decreased reflectance and indicating increased vegetative intensity included a newly plated forest reserve which was not identified on the MTVI image.

Figure 2. Histogram of multitemporal vegetation index image (sub-sampled)

Plate 2. Multitemporal vegetationm index.

Estimation of water surface area
A second aim of the study is to demonstrate the usefulness of satellite data for estimating water surface area for input into hydrological evaporation models which calculate sedimentation rates and water volume. Depth sounding methods are time consuming and unreliable due to unevenness of the bed an uneven distribution of sediment. At medium to low water levels surface area declines proportionally more, relative to gauge height, due to sediment accumulation.

Pixel count
Following geometric correction of the 1978 image to map (0,2% mean error was achieved), the number of pixels representing water was multiplied by the pixel size of 79m x 56m.

Table 1 indicates that MSS band 7 as the ability to clearly separate water from other land cover types due to very low reflectance. Reflectance readings on a scale of 0-255, for 250 pixels representing water gave statistics as follows : -

Table 1

Band 4	Mean 36	SD 7.2	Min. 24	Max. 45
Band 5	Mean 47	SD 15.4	Min. 26	Max. 66
Band 6	Mean 36	SD 6.9	Min. 23	Max. 46
Band 7	Mean 13	SD 1.1	Min. 10	Max 16

Using Band 7 therefore, all pixels falling within a threshold of 10 of the mean value were thought likely to represent water and no other class. A FORTRAN programme was then written to count all pixels falling within these limits. This gave 2510 pixels.

However, aquatic vegetation both submerged and floating had been classified as non-water. These were counted manually using zoom and zoom functions of the cursor. Thus 140 pixels were added to the total giving surface area of 1197 hectares.

The main source of inaccuracy in the method was thought to be the problem of mixed pixels around the water's edge. However the threshold of 10 used in the classification makes it likely that pixels with only a small proportion of land would be classed as water and those with a large proportion of land would be classed as land, thus averaging out the misclassification.

The result obtained was in close agreement with estimates based on the depth are curve supplied by W.R.E.C.A. as follows :-

Observed 1197 hectares (landsat MSS, November 1978)
Expected 1166 hectares (WRECA depth - area curve, (WRECA, 1980)

Summary
The techniques described appear to have identified the major changes in vegetative intensity in the Jakara catchment over the period of dam construction. This assertion is based on the author's knowledge of the area and from the study of air photos, both pre-and post-dam , rather than on empirical field data.

Field radiance measurements obtained at the time of the satellite overpass would enable more précis calibration of change sin actual land cover. With the MTVI values enabling more objective delineation of changes in vegetative vigour. However, due tot eh limited resolution of LANDSAT MSS, precise categories of land use/land cover change may not be identifiable. The methods described are thus more suitable for preliminary investigations over a whole catchment, of areas where vegetation changes are occurring following dam construction. This can assist in selecting areas fro more detailed monitoring in the field using features readily visible on MSS imagery such as villages and medium sized roads.

References

Adams W.M. 1985 the downstream impacts of dam construction: case study from Nigeria, Trans. Inst. Brit. Geog. 292-301.
Barrow C. 1983. The environmental consequences of water resources development in the tropics. Ch. 13 in Natural Re-sources in Tropical Countries O.J. Bee (ed) Singapore Univ. Press.
Barrow C. 1987 Water Resources and Agricultural Development in the Tropics. Longman.
Booth D.J. and R. Oldfield 1988 Estimation of the area of Lake Kariba, Zimbabawe using LANDSAT MSS imagery, Proc. IGARSS 88 Symp., Rem. Sensing Soc. Edinburgh.
Howarth P.J. and E Boasson 1983 LANDSAT digital enhancements for Change detection in urban environments. Remote sensing of environment 13 149-160.
Nichol J.E. 1987 Monitoring the impact of dam construction on fadama cultivation using SPOT and sequential aerial photography. Nat.conf.on desertification and environment Resource Monitoring in Nigeria, Nig.soc. Rem. Sensing.
Pilon P.G. and P.O.Adeniyi 1986 Evaluating the downstream impacts of dam construction on agricultural land use cover using multitemporal LANDSAT MSS data : a study of the Bakalori irrigation project, Nigeria. Proc. 20th int. Symp. On remote Sensing of Env. Nairobi, Kenya.
Water Resources Engineering and Construction Authority 1980. Main Project data for dams and reservoirs. Kano, Nigeria, Nov. 1987.