Interactive comment on “ The use of remote sensing to quantify wetland loss in the Choke Mountain range , Upper Blue Nile basin , Ethiopia ” by E .

The authors wish to thank the reviewer #1 for the review and critical comments. The critical parts of the paper were revised and the paper is now clearer and more consistent. We agreed with almost all the comments of reviewers and reworked the paper based on the comments and suggestions. We hope to get invited to submit the revised version. We feel the paper has benefited from the comments and appreciate the suggestions. Please see below details of how and where we added the required


Introduction
It is well documented that wetlands play critical roles in the hydrological system (e.g.Ehrenfeld, 2000;Tiner, 2003).Moderating the impact of extreme rainfall events and providing base flow during dry periods is one of the key functions of wetlands.On the one hand, wetlands can moderate flow dynamics and mitigate flooding.On the other hand, they play a regulating role on the water quality by capturing sediment and capturing and converting pollutants.Wetland functions are particularly important in areas with high rainfall variability as they help to sustain smaller discharges during the dry season and, consequently, improve the availability of water.Therefore, wetland conservation has a vital role to play in alleviating water problems at different scales (e.g.Koeln, 1992;Jensen et al., 1993;Reimold, 1994).
Estimates of global wetland area range from 5.3 to 12.8 million km 2 (Mathews and Fung, 1987;Finlayson et al., 1999).About half the global wetland area has been lost as a result of human activities (OECD, 1996).In tropical and subtropical areas conversions of wetlands to alternative land uses have accelerated wetland loss since the 1950s (Moser et al., 1996) and agriculture is considered the principal cause for wetland loss.The degradation and loss of headwater wetlands leads to increased flooding, soil erosion, degradation of water quality, reduced dry season flows, lower groundwater tables and less availability of water and moisture during the dry season.In Ethiopia due to limited research work on wetland resources, the management of wetlands is impaired by the general public and decision makers.Hence many of the wetlands are drying up and disappearing; a good example is the collapse of Lake Alemaya in Eastern Ethiopia (Lemma, 2004;Yohannes, 2005).
E. Teferi et al.: The use of remote sensing to quantify wetland loss in Ethiopia In the Blue Nile basin, wetlands in general are given limited attention and the role of headwater wetlands is especially not well addressed.This seems to be somewhat different for the White Nile basin, where the ecosystem functioning and its services were studied in the Lake Victoria region (Loiselle et al., 2006;Van Dam et al., 2007) and the Sudd wetland in Sudan (Mohamed et al., 2005).It is common that small headwater wetlands are ignored in wetland studies and often the importance of large wetlands is recognized and appreciated.But these seemingly insignificant individual wetlands can collectively play important roles in moderating flows and improving water quality (McKergow et al., 2007).A large number of such small wetlands, ranging from sedge swamps to seasonally flooded grasslands, exist in the Mt.Choke headwater areas.However, very little is known about the spatial distribution, the variability in space and time and their hydrological and ecological functioning.Most previous research on wetland was conducted in southwest Ethiopia with a focus on wetland management and policy implications (Hailu et al., 2003;Wood, 2000), hydrological impacts of wetland cultivation (Dixon, 2002;Dixon and Wood, 2003), socio-economic determinants (Mulugeta, 1999), and gender dimension of wetland use (Wood, 2001).
An effective and efficient management of wetlands requires an exhaustive survey (mapping) of their distribution and determination of whether or not they have changed over time and to what extent (Jensen et al., 1993;Baker et al., 2006).Ground-based survey of wetlands of large, or even small, wetlands is very time consuming.The use of remote sensing techniques offers a cost effective and time saving alternative for delineating wetlands over a large area compared to conventional field mapping methods (Ozesmi and Bauer, 2002;Töyrä and Pietroniro, 2005).Landsat, Satellite Pour l'Observation de la Terre (SPOT), Advanced Very High Resolution Radiometer (AVHRR), Indian Remote Sensing satellites (IRS), radar systems (Ozesmi and Bauer, 2002), Advanced Space-borne Thermal Emission and Reflection Radiometer (ASTER) (Wei et al., 2008;Pantaleoni et al., 2009) and Moderate-resolution Imaging Spectroradiometer (MODIS) (Callan and Mark, 2008) are the most frequently used satellite sensors for wetland detection.However, there is no standard method for computer-based wetland classification (Frazier and Page, 2000).
The aim of this study is to quantify wetland dynamics and estimate wetland loss in the Choke Mountain range.To achieve this objective, a hybrid supervised/unsupervised classification of Landsat imagery acquired in 1986 and 2005 were considered.Before classification it was necessary to create images of surface reflectance that are radiometrically consistent and to ensure interimage comparability between TM and ETM+ images.This was done by applying a combination of cross-calibration and atmospheric correction (Vogelman-DOS3) methods (Paolini et al., 2006).Following this introduction, the definition of wetlands implemented in this paper is given.The remainder of this paper is sub-divided into three sections.The materials and methods are described in the first section, following which there is a section on Result and Discussion.The paper ends with a section on Conclusion.

Definition of wetlands
Successful identification of wetlands starts with a clear definition of wetlands.Numerous wetland definitions have been developed for various purposes.Most recent ones are definition given by RAMSAR and USGS.The Ramsar convention (Ramsar Convention Secretariat, 2006) defined wetlands as: ". . ..areas of marsh, fen, peatland or water, whether natural or artificial, permanent or temporary, with water that is static or flowing, fresh, brackish or salt, including areas of marine water the depth of which at low tide does not exceed six metres."The United States Geological Survey (USGS) defined wetland as a general term applied to land areas which are seasonally or permanently waterlogged, including lakes, rivers, estuaries, and freshwater marshes; an area of lowlying land submerged or inundated periodically by fresh or saline water (Mac et al., 1998).Taking the RAMSAR and USGS definitions into consideration, the wetlands mapped in this study included water bodies (static or flowing water), seasonal wetlands with high moisture, and seasonal wetlands with low moisture.Riparian vegetations were not included in the wetland classes of this study.

The study area
The study area covers the whole of the Choke Mountain range, the most important source of water for the Blue Nile river system in Ethiopia.The area extends between 10 • to 11 • N and 37 • 30 to 38 • 30 E, the highest peak is located at 10 • 42 N and 37 • 50 E.It is situated in the south of Lake Tana, in the central part of the Amhara National State of Ethiopia (Fig. 1).Elevation extends from 810 m a.s.l. to 4050 m a.s.l.The mean annual rainfall varies between 995 mm a −1 and 1864 mm a −1 based on data from 13 stations for the years 1971-2006.The mean annual temperature ranges from 7.5 • C to 28 • C (BCEOM, 1998a).
The Choke Mountain is the water tower of the region serving as headwater of the upper Blue Nile basin.Many of the tributaries of the Upper Blue Nile originate from this mountain range.A total of 59 rivers, and many springs are identified in the upper catchments of Choke Mountain.The Blue Nile is the largest tributary of the Nile river, the runoff being generated almost all in Ethiopia.Estimates of the mean annual flow range from 45.9 × 10 9 m 3 a −1 to 54 × 10 9 m 3 a −1 (UNESCO, 2004;Conway, 2005).The annual sediment discharge of the basin at the Sudanese border is between 130×10 6 t a −1 and 335×10 6 t a −1 (BCEOM,  ).The major sources of sediment is erosion of the intensively farmed highland areas in the central and eastern parts of the basin (Bewket and Teferi, 2009;UNESCO, 2004;Zeleke, 2000).
The main geological unit in the area is the Tarmaber Gussa formation which represents Oligocene to Miocene basaltic shield volcanoes with minor trachyte and phonolite intrusions (Tefera et al., 1999).The soil units covering the Choke Mountain are Haplic Alisols (deep soils with predominant clay or silt-clay texture), Eutric Leptosols (shallow soils with loam or clay-loam texture) and Eutric Vertisols (deep soils with clayey texture and angular/sub-angular blocky structure) (BCEOM, 1998c).Due to active morphological processes (erosion, land slides etc.) the soil depth can vary between zero and several meters.
There is no longer significant natural forest cover in this mountain range.The major remaining natural habitats are moisture moorland, sparsely covered with giant lobelias (Lobelia spp.; Jibara/Jibbra), lady's mantle (Alchemilla spp.), Guassa grass (Festuca spp.) and other grasses.There is very little natural woody plant cover; heather (Erica spp.; Asta) and Hypericum (Hypericum revolutum; Amijja) are found in patches.Bamboo or Kerkeha (Arundinaria alpina) is found as homestead plantation as well as part of the natural vegetation cover in the area, albeit very sparsely.Korch (Erithrina brucei) is commonly grown as border demarcation plant in the area.Eucalyptus globulus is extensively grown in plantation, and some of the residents have become dependent on it for their livelihoods.

Data used and image preprocessing
The study utilized 27 topographic maps of the area at a scale of 1:50 000 and 1:250 000 dated 1984, geological maps at the scale of 1:250 000, and Landsat images of years 1985/6 (TM) and GLS 2005 (ETM+).Wet and dry season images of each year were acquired.Scenes were required to be of the same phenological cycle (dry or wet season) and have little or no cloud cover.The characteristics of the image data are shown in Table 1.All Landsat images were accessed free of charge from US Geological Survey (USGS) Center for Earth Resources Observation and Science (EROS) via http://glovis.usgs.gov/.All scenes supplied by the EROS Data Center had already been georeferenced to the Universal Transverse Mercator (UTM) map projection (Zone 37), WGS 84 datum and ellipsoid.The GLS 2005 Landsat 7 ETM+ image was processed to Standard Terrain Correction (Level 1T-precision and terrain correction) in the USGS-EROS using the Level 1 Product Generation System (LPGS) (Williams, 2009).The Landsat 5 TM (1986) standard product is processed using the National Land Archive Production System (NLAPS).Several preprocessing methods were implemented before the classification and the change detection.These include geometric correction, radiometric correction, atmospheric correction, topographic normalization and temporal normalization.The importance of each preprocessing steps has been discussed extensively in the literature (e.g., Hill and Sturm, 1991;Han et al., 2007;Vicente-Serrano et al., 2008).The specific preprocessing steps followed in this research are briefly described below.

Geometric correction
Re-projection to the local level projection system was made (UTM, map projection; Clarke, 1880, Spheroid, and Adindan Datum).In some parts of the study area, there were significant discrepancies between the imageries and the underlying GIS base layers, which were extracted from high resolution SPOT-5 imagery.The misaligned scenes were georectified to the underlying base layers by using control points.
In this study, first the GLS 2005 Landsat 7 ETM+ was geometrically rectified using 25 control points taken from the topographic map at 1:50 000 and 11 GPS points taken in the field.The other images were then registered to the ETM+ image with the same projection.An RMS error of less than 0.5 was achieved.The nearest neighbor resampling method was used to avoid altering the original pixel values of the image data.Because of the high incidence of varied topography in the study area, ortho-correction was found to be necessary and hence undertaken to further enhance the image geometry by accounting for the significant spatial distortion caused by relief displacement.

Radiometric calibration
The absolute radiometric correction method used in this paper involved a combination of cross-calibration method developed by Vogelman et al. (2001), with an atmospheric correction algorithm based on COST model (Chavez, 1996), denoted DOS3 by Song et al. (2001).Paolini et al. (2006) called this combination of methods "Vogelman-DOS3".

Cross-calibration
When using both Landsat TM and ETM+ images in studies that require radiometric consistency between images, special attention has to be paid to the differences in sensors response.Previous studies have demonstrated significant differences in the radiometric response between Landsat ETM+ and Landsat TM spectral bands (Teillet et al., 2001;Vogelmann et al., 2001).To reduce these differences and ensure image intercomparability, a cross-calibration was needed (Paolini et al., 2006;Vicente-Serrano et al., 2008).In the cross-calibration procedure, the Landsat TM images DN (DN5) were first converted to the Landsat ETM+ DN (DN7) images using the equation developed by Vogelmann et al. (2001).After this conversion, the L5 TM DN is treated as L7 ETM+ DN, and L7 ETM+ gain, offset and ESUN values are used respectively for radiance and Top-Of-Atmosphere (TOA) reflectance.

Conversion to at-sensor spectral radiance
At-sensor spectral radiance was computed for the crosscalibrated Landsat 5-TM (1986 images) quantized calibrated pixel values in DNs and Landsat 7-ETM+ quantized calibrated pixel values in DNs using sensor calibration parameters published by Chander et al. (2009) and in image header file.

Conversion to Top of Atmosphere (TOA) reflectance
To correct for illumination variations (sun angle and Earth-Sun distance) within and between scenes, TOA reflectance for each band was calculated.This conversion is a very important step in the calculation of an accurate Normalized Difference Vegetation Index (NDVI) and other vegetation indices.Some of the parameters for the conversion are available in the image header files, while the exo-atmospheric irradiance values for Landsat 7 are published by Chander et al. (2009).

Atmospheric correction
TOA reflectance value does not take into account the signal attenuation by the atmosphere, which strongly affects the intercomparability of the satellite images taken on different dates.But atmospheric correction methods account for one or more of the distorting effects of the atmosphere and thereby convert the brightness values of each pixel to actual reflectances as they would have been measured on the ground.
Several atmospheric correction models have been developed to eliminate atmospheric effects to retrieve correct physical parameters of the earth's surface (e.g.Surface reflectance), including SMAC (Rahman and Dedieu, 1994), 6S (Vermote et al., 1997), MODTRAN (Berk et al., 1999), ATCOR (Richter, 1996), and COST (Chavez, 1996).Despite the variety of available techniques, a fully image-based technique developed by Chavez (1996) known as the COST model that derives its input parameters from the image itself, has been proved to provide a reasonable atmospheric correction (Mahiny and Turner, 2007;Berberoglu and Akin, 2009).The COST atmospheric model was used to convert the atsensor spectral radiance to reflectance at the surface of the Earth.

Topographic correction
In mountainous regions such as Upper Blue Nile Basin, topographic effects that result from the differences in illumination due to the angle of the sun and the angle of terrain produce different reflectances for the same cover type.Hence, the classification process for rugged terrain is seriously affected by topography and needs careful treatment of the data (Hodgson and Shelley, 1994;Hale and Rock, 2003).
To reduce the topographic effects methods such as spectral-band Ratioing (Holben and Justice, 1980), application of a Lambertian model (Cosine Correction) (Smith et al., 1980) and application of a non-Lambertian (Minnaert Correction and C-correction) (Meyer et al., 1993) have been developed.The C-correction (Teillet et al., 1982) is chosen for this study to correct for uneven reflectance patterns, as Meyer et al. (1993) and Riaño et al. (2003) showed it to be the most effective method of reducing the topographic effects for Landsat imagery (McDonald et al., 2000).The cosine of the incident solar angle was calculated using the equation of Holben and Justice (1980) and Smith et al. (1980).The variables used in the equation were derived from a digital elevation www.hydrol-earth-syst-sci.net/14/2415/2010/ Hydrol.Earth Syst.Sci., 14, 2415-2428, 2010 and cosine of the incident solar angle (the independent variable).The derived C-correction coefficients (Table 2) were then used as parameters to normalize both images for topographic effects.
Besides visual analysis, the evaluation of topographic correction was based on the computation of R 2 -values for the linear regression between corrected data and cosine of the incident solar angle.Since the topographic impact varies with wavelength and is most prevalent in the near-infrared spectral region, a reduction in the R 2 -value for the near-infrared band was observed (0.214 to 0.008 for the 1986 image and 0.291 to 0.0342 for the 2005 image).This indicates the effectiveness of the non-Lambertian C-correction.

Temporal normalization
Vicente-Serrano et al. ( 2008) recommended an additional step (temporal normalization) to completely remove nonsurface noise and improve temporal homogeneity of satellite imagery.Therefore 9 reservoir sites and 5 bare sites were used to normalize the 2005 ETM+ image (slave) with respect to 1986 TM image (reference).The resulting normalization equation is shown in Table 3.

Image classification
A hybrid supervised/unsupervised classification approach was used to classify the images.This approach involved: (1) unsupervised classification using ISODATA (Iterative Self-Organizing Data Analysis) to determine the spectral classes into which the image resolved; (2) using the spectral clusters ground truth (reference data) were collected to associate the spectral classes with the cover types observed at the ground for the 2005 image and for the 1986 image reference data were collected from the 1984 Topographic map (1:50 000 scale); and (3) classification of the entire image using the maximum likelihood algorithm.A hierarchical class grouping was adopted to label and identify the classes (Thenkabail et al., 2005) in both the wet and dry season images.First, the unsupervised ISODATA clustering was used.25 initial classes were found and then after a rigorous identification and labeling process the original 25 classes were progressively reduced to 17, 7 and 4 classes by using ground truth data.Ground truth data on spatial location, land cover types, soil moisture status, and topographic characteristics were collected from selected sample sites during the period from June 2009 to March 2010.Stratified random sampling design was adopted for the selection of 342 point sample sites.Stratification was based on the accessibility of the sites.Spatial locations were obtained using a GPS (Global Positioning System).
To assist within class identification process in the unsupervised classes, the at-sensor reflectance based NDVI (Normalized Difference Vegetation Index) values were calculated.Wetlands with covers of barren lands and/or with sparse vegetation have lower NDVI values as a result of soil moisture that is relatively higher than the surrounding areas.When wetlands have vegetation or crops the NDVI will vary depending on vegetation density and vigor.

Post-classification processing and assessment
A variety of change detection techniques have been developed (Yuan et al., 1998;Lu et al., 2004).Each of them has its own merit and no single approach is optimal and applicable to all cases.Yuan et al. (1998) divide the methods for change detection and classification into pre-classification and postclassification techniques.In this paper post-classification comparison change detection approach which compares two independently produced classified land use/cover maps from images of two different dates (Jensen, 2005) was used.Other studies (Mas, 1999;Civco et al., 2002) demonstrated that it was the most accurate measure of change.The principal advantage of post-classification is that the two dates of imagery are separately classified; hence it does not require data normalization between two dates (e.g.Singh, 1989).The other advantage of post-classification is to provide information about the nature of change (including trajectories of change) (Song et al., 2001;Coppin et al., 2004).
The accuracy of the classification results was assessed by computing the confusion matrix (error matrix) which compares the classification result with ground truth information.17 classes were identified using the hybrid supervised/unsupervised classification.Classes of different spectral patterns but the same information classes were selectively combined in the classified images and finally reduced to 7 classes.Majority analysis was also performed to change spurious pixels within a large single class to that Hydrol.Earth Syst.Sci., 14, 2415Sci., 14, -2428Sci., 14, , 2010 www.hydrol-earth-syst-sci.net/14/2415/2010/

Results and discussion
The study area was classified into seven classes: three wetland classes and four non-wetland classes.The description for each class is given in Table 4.Although riparian vegetation cover defines one type of wetland, because of the limitation of the classification method used it was classified as woody vegetation.
A Confusion Matrix was computed to assess the accuracy of the classification by comparing the classification results with ground truth information.From the Confusion Matrix report 94.1% and 93.5% overall accuracy for the 1986 and 2005 imageries were attained, respectively.The overall accuracy was calculated by summing the number of pixels classified correctly and dividing by the total number of pixels.An overall accuracy level of 85% was adopted as representing the cutoff point between acceptable and unacceptable results according to Anderson et al. (1976).The Confusion Matrix also reports the Kappa Coefficient (k) which is another measure of the accuracy of the classification (Cohen, 1960).The k values for the 1986 and 2005 image classification were 0.908 and 0.913, respectively.The values can range from −1 to +1.However, since there should be a positive correlation between the remotely sensed classification and the reference data, positive k values indicate the goodness of the classification.Landis and Koch (1977) character-ized the possible ranges for k into three groupings: a value greater than 0.80 (i.e., 80%) represents strong agreement; a value between 0.40 and 0.80 (i.e., 40-80%) represents moderate agreement; and a value below 0.40 (i.e., 40%) represents poor agreement.
The accuracies for both the 1986 and 2005 image classification were considered good enough to apply postclassification change detection analysis.Basically this analysis focuses on the initial state classification changes (1986 image).For each initial state class (1986), the postclassification comparison technique identifies the classes into which those pixels changed in the final state image (2005).

Space-time dynamics of non-wetland land cover types
Table 5 lists the initial state classes in the columns, the final state classes in the rows and the unchanged areas in the diagonal cells.For each initial state class (i.e., each column), the table indicates how these pixels were classified in the final state image.A total of 891 km 2 was classified as woody vegetation in the initial state image (1986), while 2965 km 2 was classified as woody vegetation in the final state image (2005) (Table 5).2074 km 2 woody vegetation cover has increased over the 20 years studied.This is mainly attributed to the increase in Eucalyptus plantation in the area (e.g.Raba Forest and Wildlife Reserve).After significant deforestation in the past the scarcity of wood for fuel and other uses has forced the local people to plant trees (Bewket, 2003).About 46% (1351 km 2 ) of the existing woody vegetation has come from OW = open water, SWH = seasonal wetlands with high moisture, SWL = seasonal wetlands with low moisture, AF = afro alpine herbaceous vegetation, WV = woody vegetation, CL = cultivated land, BL = bare land cultivated land and 30% (873.2 km 2 ) from bare land categories.Since the productive capacity of the the cultivated land is being threatened by the loss of nutrients through erosion, farmers in the area are looking for another option because they could not cope with declining crop yields.Every year more and more agricultural land is being converted into Eucalyptus forest plantation which ensures income security for local residents.This trend in land-use and landcover change is altering the soil (Bewket and Stroosnijder, 2003) and hydrologic (Bewket and Sterk, 2005) characteristics of upland watersheds of the Choke Mountain range.This may also influence the livelihoods of the population living in downstream areas by changing critical watershed functions (e.g.availability of water during the dry season).

Space-time dynamics of wetlands
In the initial state (1986) image, 946.3 km 2 of land was classified as seasonal wetlands with low moisture, but only 339.5 km 2 of land was classified as seasonal wetlands with low moisture in the final state image.This means a total of 607 km 2 of seasonal wetlands with low moisture were lost and/or converted to another land cover/use type during the period 1986 to 2005 (Table 5).Out of the 946.3 km 2 of seasonal wetlands with low moisture class in the initial state image, only 111.5 km 2 (12%) remain unchanged and 835 km 2 (88%) of the initial class has been converted and this makes the class most dynamic (unstable) of all observed land cover types.About 422 km 2 (∼51%), 275.3 km 2 (∼33%), 64 km 2 (8%), and 59.3 km 2 (7%) of the previous seasonal wetlands with low moisture cover were converted to cultivated land, seasonal wetland with high moisture, bare land, and woody vegetation respectively.Every year 21 km 2 of seasonal wetlands with low moisture land cover types are converted to cultivated land.This indicates that the experience of wetland cultivation which is common in south-wet Ethiopia (Dixon and Wood, 2003;Dixon, 2002) is also practiced in Choke Mountain range.With the increase in population pressure and shortage of land, government considers wetland cultivation as a remedy to address the food needs (Wood, 1996;World Bank, 2001).As a result, seasonal wetlands with low moisture have come under pressure.
Again based on Table 5, 2040.9 km 2 of land was classified as seasonal wetlands with high moisture in the initial state image, and 2999.6 km 2 of land was classified as seasonal wetlands with high moisture in the final state image.A total gain of 959 km 2 in the seasonal wetlands with high moisture land cover was observed (Fig. 3).Out of 2999.6 km 2 seasonal wetland with high moisture class, 54% (1621.7 km 2 ) have come from cultivated land, 9% (275.3km 2 ) have come from seasonal wetlands with low moisture class and 25% (748.7 km 2 ) remain unchanged.The contribution of cultivated land to the increase in the seasonal wetland with high moisture class is significant (54%).Since the gain is mostly from classes initially classified as cultivated land, the increase in moist farmlands (irrigated land) caused the increase in the seasonal wetland with high moisture class.This is related to the expansion of small scale irrigation (man-made Mountain are now being converted to bare land.This was also substantiated by field observation.In contrast, Bahirdar pond (Fig. 5) at Yenebirna village in Chemoga watershed showed a gradual increase through time (Table 6).To analyze this situation 10 satellite images (5 on dry season and 5 on wet season) were taken and the size of Bahirdar pond was determined.The observed smallest size (4.2 ha) could be because of the 1986 drought occurred throughout Ethiopia.This indicated that the condition of wetlands of the area is dependent on climatic factors beyond human influences.The increasing trend of rainfall of the area around the pond (Fig. 6) associated with increasing runoff and sediment inflow from the surrounding watershed could be one of

Characterization of wetlands
Although Landsat (TM and ETM+) has 16-day repeat cycle, these 16-day images are not available for the period 2001-2009.Therefore, the class characteristics were studied using Moderate Resolution Imaging Spectroradiometer (MODIS), www.hydrol-earth-syst-sci.net/14/2415/2010/ Hydrol.Earth Syst.Sci., 14, 2415-2428, 2010   The NDVI time-series class characteristics follows the rainfall trend (Fig. 7); vegetation begins to gain vigor in beginning of September and remains at high vigor until November while the rainfall reaches maximum during 28 July to 12 August.Previous studies have investigated the lag and cumulative effect of rainfall on vegetation in semi-arid and arid regions of Africa (Martiny et al., 2006;Wagenseil and Samimi, 2006) and humid regions of Africa (Onema and Taigbenu, 2009).However, this lag effect of rainfall can not be observed in open water.

Conclusions
The study demonstrated the use of remote sensing techniques to delineate headwater wetlands from non-wetlands and determine the dynamics over large areas of Choke Mountains with the overall accuracy of 94.1% and 93.5% and Kappa Coefficient of 0.908 and 0.913 for the 1986 and 2005 imageries, respectively.Therefore, ground-based wetland surveying for large area, especially for small wetlands is a very time consuming and ineffective process.The use of Remote Sensing and GIS techniques in wetland mapping reduces cost and enhances the accuracy.Creating radiometrically consistent image data through (Vogelman-DOS3) methods before applying hybrid supervised/unsupervised classification was found to be effective in wetland mapping.However the application of technique to Landsat images was not effective in mapping riparian vegetation as wetland.To undertake a more focused wetland management plan implementation the use of mandatory environmental characteristics (vegetation, soil, and hydrology) for wetland detection is still important.
Four major trajectories of change that were observed: (1) from seasonal wetlands with low moisture to cultivated land, (2) from cultivated land to woody vegetation (plantation), ( 3) from open water to bare land, and (4) from bare land to cultivated land.In general, 607 km 2 of seasonal wetland with low moisture and 22.4 km 2 of open water are lost in the study area over the 20 years considered.This is one indication of future deterioration in wetland condition.This calls for wetland conservation and rehabilitation through incorporating wetlands into watershed management plans so as to make wetlands continue providing their multiple functions which include flood control, stream-flow moderation, groundwater recharge, sediment detention, and pollutant retention.Further research on specific wetland spots is needed for accurate inventories of wetlands and to explore the reasons why and how wetlands are changing.

Fig. 1 .
Fig. 1.The location and topography of the study area. 1998b

Fig. 7 .
Fig. 7.The time-series characteristics of the wetlands and non wetlands (Mean 16-day 4 annual NDVI and rainfall for the period 2001 to 2009).

Table 1 .
Description of Landsat images used.

Table 2 .
C-correction coefficients derived for the non-Lambertian corrected images.

Table 3 .
Image normalization regression models developed for Choke area image.

Table 4 .
Descriptions of the land cover classes identified in the Mt.Choke range.
class.Finally, a detailed tabulation of changes in wetland and non-wetland classes between 1986 and 2005 classification images were compiled by post-classification comparison

Table 6 .
Time-series of pond size.