The role of land use and soils in regulating water flow in small headwater catchments of the Andes

Authors


Abstract

[1] Land use changes can have a significant impact on the terrestrial component of the water cycle. This study provides a comparison of three small headwater catchments in the Andean mountains of Colombia with different composition of land use. Several methods were used to quantify differences in the hydrological behavior of these catchments such as flow duration curves, stormflow analysis, and the linear reservoir concept. They were combined with an analysis of the characteristics of soils that contribute to understanding the aggregate catchment hydrological behavior. Andisols, which are soils formed in volcanic areas and with a large capacity to hold water, amplify differences in land use and limit the potential impact of land use management activities (conservation or restoration) on the water regulation function of catchments. Of the three studied catchments, less variability of flows was observed from the catchment with a larger percentage of area in forest, and a slower decrease of flows in the dry season was observed for the catchment with a relatively higher percentage of area in wetlands. Evidence is provided for the infiltration trade-off hypothesis for tropical environments, which states that after forest removal, soil infiltration rates are smaller and the water losses through quick flow are larger than the gains by reduced evapotranspiration; this is compatible with the results of the application of the linear reservoir concept showing a faster release of water for the least forested catchment.

1. Introduction

[2] Headwater streams of the Andes have highly modified flows due to the development and use of water resources and the alteration of the water regulation capacity of soils [Buytaert et al., 2005, 2006; Diaz and Paz, 2006]. In Colombia as in many parts of the tropics, mountains are preferred areas for human habitation because of their moderate climate. Over 66% of the Colombian population is located in areas above 500 m, which corresponds to the Andean region, and makes up only 24% of the country [Etter and van Wyngaarden, 2000]. The large population of the Andes obtains water from small sources: streams, creeks, and small rivers [Instituto de Hidrología, Meteorología y Estudios Ambientales (IDEAM), 2000]; therefore, the pressure on these catchments is high. Precipitation is spatially and temporally variable in the Andean mountains, with some regions having decreased precipitation with elevation [Buytaert et al., 2010]. Globally, more than 50% of mountain areas are essential in supporting downstream regions [Viviroli et al., 2007]. Increasingly, the population depending on water coming from mountain regions is subject to scarcity and to use restriction of the resource. The water discharge from the Andean region of Colombia decreases by 30% during a dry year, and according to the National Institute for Hydrology, Meteorology, and Environmental Studies (IDEAM), the highest pressure on water resources in the country is concentrated in the Andean region. IDEAM estimates that in scenarios of dry years in the near future, over 60% of the Colombian population could be subject to water scarcity [IDEAM, 2001].

[3] Changes in land use have potentially large impacts on water resources, yet quantifying these impacts remains among the more challenging problems in hydrology [Stonestrom et al., 2009]. Quantifying the effects of land use on water availability is critical, particularly where communities are dependent on small water sources that are more vulnerable to interannual fluctuations in precipitation and also to climate change. In tropical environments hydrological data is scarce and is concentrated in large rivers despite the fact that it is the small streams that directly supply water for most human use, particularly in mountain areas. Knowledge about the hydrology of small tropical mountain headwater catchments is still limited.

[4] The Andean landscape has been influenced and transformed by the introduction of cattle by the Spaniards in the 16th century. Large areas of sloping terrain were transformed into extensive grazing lands with negative effects, e.g., landscape homogenization [Etter and van Wyngaarden, 2000], erosion [Murgueitio and Calle, 1999], and loss of the capacity of watersheds to regulate hydrological flows.

[5] This paper is a contribution to the understanding of process hydrology in the Andes by relating land use differences to the hydrological behavior of small headwater catchments of tropical Andean mountains of volcanic soils. It addresses the characteristics of the soil in relation to water movement that limit the potential influence of land use management (conservation or restoration) on the water regulation function of catchments. Through a comparison of three small neighboring catchments of the Andes, it is shown that deforestation in a tropical environment reduces dry season streamflow, which coincides with the infiltration trade-off hypothesis (the reduction in the infiltration of water in the soils is larger than the gains because of a reduced evapotranspiration) [Bruijnzeel, 1988], and that the drainage of wetlands reduces the transit time of water.

[6] In order to isolate the impacts of land use change on water resources, a comparative field measurements study between three small neighboring headwater catchments was proposed as a way to minimize confounding effects. The catchments were chosen on the basis of their strategic importance for the supply of water to a small Andean community of 15,000 people. This community depends on the water flow regulation of the catchments for the provision of water in the dry season and the attenuation of stormflows in the wet season, since there is no water storage infrastructure.

2. Methods

2.1. Study Site

[7] The catchments are located in Filandia (4.67°N, 75.63°W) at 2000–2200 m in elevation, in the coffee-growing region of Colombia, on the western side of the central branch of the Andes and drain to the Cauca River, which flows north to the Atlantic Ocean. The microcatchments are called Barro Blanco (White Mud in Spanish, referring to the ash layer found in parts of the catchment) and Bolillos (the name of an endemic palm tree that grows in the area). The Bolillos catchment comprises two smaller minicatchments called Bolillos 1 and Bolillos 2. For convenience the catchments are referred to as BB, B1, and B2, respectively. Most of the land in the three microcatchments is dedicated to extensive cattle rearing. The decline in coffee prices since the mid 1990s and the increase in cattle and beef prices stimulated the expansion and intensification of cattle grazing in the three catchments, leading to significant land use changes from forests and wetlands to grazing lands.

[8] Central Colombia and the western Andean cordillera experience a bimodal annual precipitation cycle, where rainfall peaks during April–May and October–November, followed by drier periods during December–February and June–August. This is caused by the double passage of the Intertropical Convergence Zones. The seasonal strengthening (September–November) and weakening (February–March) of the Chocó Jet partially explains why the October–November rainy season is more intense than that of April–May over central and western Colombia [Poveda et al., 2006]. Average annual precipitation in the region recorded since 1972 has been 2990 mm. The high precipitation and the topography of the three catchments have been conducive to the formation of wetlands.

[9] The topography has been described as hummocky [Guarín et al., 2004], and the parent material of soils are fluviovolcanic deposits, mostly clays of uniform size. Soils formed on these sediments are classified as Andisols (Acrudoxic Hapludans) [IGAC, 1996]. After Histosols, Andisols are the soils with the highest organic matter content, which results in light soils with very high water-holding capacity. At depths below 1.5 m in the three catchments, there is a volcanic ash layer, which can reach tens of meters in thickness at some locations. The layer is characterized by very low hydraulic conductivity constituting an aquiclude, limiting deep percolation of water and creating a shallow water table.

[10] A land use map was developed using a 2003 QuickBird satellite image with a resolution of 0.6 m combined with field verification using a GPS (Navman). The field mapping was subsequently transferred to ArcGIS (Environmental Systems Research Institute) for spatial analysis (Figure 1 and Table 1).

Figure 1.

Location of study site (4.67°N, 75.63°W) and the three small headwater catchments compared. The detailed map shows the wetlands found in each of the three catchments.

Table 1. Land Use Areas per Catchment
 B1B2BB
Area (ha)PercentageArea (ha)PercentageArea (ha)Percentage
Riparian and natural forest815149271625
Plantation forest28172135
Grasslands4830124693962
Wetlands113146
Roads and buildings112111
Total area15910018010063100

[11] The catchments differ in size and land use. BB is the smallest catchment (0.6 km2) but has the highest proportion of wetland area. B2 is the largest catchment (1.7 km2), with the highest proportion of grasslands. B1 (1.5 km2) is slightly smaller than B2 but has a large proportion of natural and riparian forest (Table 1). Figure 1 shows the concentration of natural and riparian forests in B1 and of grasslands in B2 and BB. It is also evident that wetlands occupy a larger percentage (6%) of the catchment area in BB. As an indicator of the differences in topography, Table 2 shows the altitude of the highest point in each catchment (maximum elevation) and the altitude at the catchment drainage point (minimum elevation). The differences in elevation vary relative to the distance from the drainage point to the most distant point in the catchment. As indicated by the slope, topography does not appear to be a major factor that would significantly affect the shape of the hydrograph.

Table 2. Maximum and Minimum Elevation for Each Catchment
 Maximum Altitude (m)Minimum Altitude (m)DifferenceDifference, Distance to Outflow Point
B12,2111,99921210%
B22,1301,9991317%
BB2,1482,03511313%

2.2. Data Collection

[12] Three data logging rain gauges were installed in the three catchments to measure precipitation and to account for spatial variability (Figure 1). These rain gauges were installed at elevations between 2100 and 2135 m and recorded precipitation from October 2004 until May 2007. The seasons were separated using weekly precipitation. The dry season in 2006 runs from the end of June until the second week of October; the wet season of 2006 goes from the second week of October until the third week of December; and the first or little dry season of 2007 goes from the end of December 2006 until the first week of March 2007. An event is defined as precipitation equal to or larger than 2 mm that ends when there has been no precipitation for the following 2 h. The average annual precipitation in the region since 1972 has been 2990 mm. Precipitation occurs with a unimodal diurnal peak in the afternoon explained as convective precipitation associated with solar thermal forcing favored by the entrance of low level moisture-laden winds onshore from the Caribbean and Pacific which ascend because of orographic lifting [Poveda et al., 2005].

[13] For continuous measurement of water level in the streams, three pressure transducers (Instrumentation Northwest) were used to record data every 15 min from June 2005 until May 2007. Stage-discharge relationships were determined for each site to relate water level in the stream channel with water flow. This was done by taking water flow measurements in a wide range of water levels using a current meter built by OTT. These water level data series were converted into flow measurements with stage-discharge relations using 62, 44, and 50 discharge measurements for catchments B1, B2, and BB, respectively (see Figure 2 for a hydrograph of the three streams). The data series collected for streams were analyzed and were used to build flow duration curves (FDC) and to calculate water yields for different time scales and lag times. Potential evapotranspiration was calculated using the Penman combination equation taken from Dunne and Leopold [1978] with daily data.

Figure 2.

Hydrographs for the three catchments.

[14] Soil moisture retention curves were built on the basis of 24 soil samples corresponding to the three major land use types (grasslands, riparian forests, and plantation forests) and were analyzed for water retention-release capacity using pressure plates [Klute, 1986] at the soils laboratory of the International Center for Tropical Agriculture in Colombia.

[15] Differences in water storage were calculated according to land use and season on the basis of volumetric soil moisture measurements taken using a reflectometry sensor (Campbell Scientific, Inc.) with a resolution of 1% and an accuracy of ±3.0%. Measurements were taken weekly at chosen sites for the major three land use types: grasslands, riparian and natural forests, and plantation forest, from June 2006 until June 2007. In B1 three riparian or natural forest sites were monitored, one plantation forest site was monitored, and three grassland sites were monitored. In B2, three riparian forest and three grassland sites were monitored. In BB, six riparian or natural forests, two plantation forests, and seven grassland sites were monitored. Water infiltration rates were determined for soils under grass using a handheld minidisk infiltrometer (Decagon Devices, Inc.). Measurements were done at 15 sites for grassland soils.

[16] Overland runoff was quantified using a runoff plot connected to a barrel of 0.5 m3 (similar design to an erosion plot). The barrel was buried at the lower end of a 10 m by 10 m square area of pasture delimited by tin sheets inserted 20 cm into the surface. Measurements were taken daily from November 2005 until May 2007.

[17] Subsurface lateral and vertical flow of water in the soil was determined in the field using a vertical trench 2 m long and 1.5 m deep, corresponding to the ash layer at this location. The vertical surface was covered by a permeable fabric, and a perforated hose was placed at the bottom of the trench between the permeable fabric and a plastic sheet that covered the permeable fabric. A 3 m hose was connected to a barrel located at a lower level than the leveled hose to collect and measure the water coming out of the trench. Measurements were taken daily from May 2006 until May 2007.

2.3. Data Analysis

[18] The study was designed as a comparative catchment study, similar to paired catchment studies that are used to compare two catchments with similar characteristics in terms of slope, aspect, soils, area, climate, and vegetation located adjacent or in close proximity to each other [Brown et al., 2005]. The principle of a comparative catchment study is to select, in this case, three catchments as similar as possible (in particular in terms of size, morphology, geology, and climatic forcing) and then to monitor them jointly during a given time period to understand their differences [Andréassian, 2004]. Since the major conversions of land use in the catchments occurred more than 10 years ago, it is assumed that the catchments are stable and the differences in hydrological behavior correspond mainly to differences in land use. For pair catchment experiments, this assumption is avoided by monitoring two catchments and converting one of them during the monitoring period; this allows accounting for other differences such as topography, soils, and vegetation [Buytaert et al., 2007]. The catchments were monitored from May 2005 until May 2007.

[19] In order to compare the hydrological behavior of the three catchments and to relate it to the land use characteristics of the catchments, the analysis and monitoring program was divided into two components: (1) land use differences in the three catchments and characteristics of soils in relation to the water retention release processes for each land use type and (2) hydrological response of the three catchments at various time scales (daily, monthly, and seasonal fluctuations of discharge and storm response). Once the hydrological differences between the three catchments were established for low flows and storm response, they were linked to the previously described traits of each of the catchments. Low flows were analyzed through flow duration curves [Smakhtin, 2001] and water yields, and stormflows were analyzed using hydrographs, time lag differences, and the linear reservoir concept [Dingman, 1994]. The most widely used definition of low flow is any flow that is exceeded for 70%–99% of the time [Smakhtin, 2001], and high or peak flows are considered to be the flows that are exceeded 1%–5% of the time [Brown et al., 2005].

[20] As part of the characterization of soil-water dynamics in the catchments, three processes were quantified: surface and subsurface runoff processes occurring in grasslands and infiltration rates of water in soils under different land use types. The rates of these processes were measured to explain the hydrological processes taking place at the catchment scale.

[21] The response of catchments to the different types of events was analyzed using the time lags between the peak in rain and the peak in flow (TLPF). Events were also analyzed using the linear reservoir model, which was applied to 15 rain events chosen on the basis of the spatial heterogeneity of precipitation in the three catchments [Roa-García, 2009]. The linear reservoir equation describes discharge (Q) as a variable dependent on the water stored in a catchment (S) [Dingman, 1994]:

equation image

where Q is the outflow, S is the stored amount of water, t is the time, k is a rate constant which is the inverse of the time constant T and is an index of the speed at which a reservoir drains, and Qo is the streamflow at the beginning of the interval. T indicates the buffering capacity of a reservoir or the “slowness” of the water release. As such, it is a good means for estimating the water-buffering capacity of the water storage reservoir (S) [Buytaert et al., 2004].

[22] The linear reservoir concept is used to assess the overall retention capacity of the catchments in terms of peak flow. In this model, based on the recession limbs of the drainage hydrographs, every catchment is considered as a series of independent reservoirs, each characterized by mean residence times (T). T is related to the time required for water to travel to the catchment outlet, so it is dependent on watershed size (larger catchments have larger T), soils and geology, slope, and land use [Dingman, 1994]. The linear reservoir concept is a tool to compare the behavior of catchments by splitting events into three reservoirs of discharge, each of them with a particular time of discharge −T. The three reservoirs represent pools of water that flow into the stream in short, medium, and long periods of time. The comparison of these times provides an idea of the regulatory capacity of the soils to hold and discharge water.

[23] Discharge of events was defined and calculated using total stream discharge from the beginning of the rain event until 2 h after the rain had stopped. The 2 h period corresponds to the median time that it takes for the streams to return to base flow conditions. This is used to compare the response of catchments to precipitation.

3. Results

3.1. Soil Moisture

[24] Soil water content is relatively high in the three catchments most of the year. The average percentages of volumetric soil moisture content are shown in Table 3. These data show the variation between the dry and wet seasons for the three types of land uses; there is no statistically significant difference between riparian or natural forest (n = 513) and grasslands (n = 527), but there is a lower soil moisture content under plantation forests (n = 122) in all seasons (α = 0.05).

Table 3. Average Values of Soil Volumetric Moisture Content by Catchment, Land Use Type, and Seasona
 B1B2BB
ValueSDValueSDValueSD
  • a

    SD, standard deviation.

Riparian and natural forest
   Dry 200655%7%45%6%56%7%
   Wet 200667%4%63%8%63%6%
   Little dry 200758%5%57%6%59%3%
   Average60% 55% 59% 
Grassland
   Dry 200658%9%49%8%56%8%
   Wet 200665%7%60%6%63%5%
   Little dry 200755%5%53%4%57%6%
   Average59% 54% 59% 
Plantation forest
   Dry 200647%9%  47%8%
   Wet 200652%6%  56%3%
   Little dry 200746%6%  51%4%
   Average48%   51% 

[25] Despite the low bulk density values of the soil of three land use types (typical Andisols range between 0.6 and 0.8 g cm−3), shown in Table 4, saturated hydraulic conductivity rates (Ksat) for 15 surface grassland sites were of the order of 10−3 cm s−1, similar to the ones found for other Andisols [Nanzyo et al., 1993; Poulenard et al., 2001; Jiménez et al., 2006], which suggests that surface runoff corresponds largely to saturated overland flow.

Table 4. Bulk and Particle Density, Total Porosity, and Available Water by Land Use Type
 nBulk Density (g cm−3)Particle Density (g cm−3)Total Porosity (%)Field Capacity (%)Permanent Wilting Point (%)Available Water, equation imagefcequation imagewp (%)
Riparian and natural forest100.62.37661537
Grassland110.72.07160528
Plantation forest20.72.37064596

[26] Soil moisture retention curves of the three land use types of the study site were compared with the typical curves for clay, loam, and sand (Figure 3). The three soils at the study site show consistent higher water retention at all pressure levels. To further compare the soils, a storage release coefficient was calculated dividing the soil moisture content at field capacity by the soil moisture content at the permanent wilting point [Topp, 2000]. The higher the number, the higher the proportion of water held in the soil that can be released. The values obtained are between 1.3 and 1.5 for the soils in the study site and between 2.1 and 8 for the typical clay-loam-sand soils, indicating the high water retention capacity of Andisols.

Figure 3.

Soil moisture retention curves showing average values for typical soils (asterisk) [Brady and Weil, 2002].

3.2. Overland and Subsurface Flow

[27] Daily overland flows on grassland measurements are shown in Table 5 classified according to runoff amount (more than 5 mm, between 5 and 2 mm, and less than 2 mm); for each of the groups, the median runoff was calculated, as well as the amount of precipitation that fell during the previous day, 2 days, and 3 days; the maximum 15 min intensity of precipitation; and surface runoff as a percentage of the last day of precipitation. Table 5 shows that relatively low daily precipitation (11 mm during the previous 24 h) combined with antecedent wet conditions produce saturated overland runoff, and substantial portions of rain become surface runoff when antecedent precipitation is higher, reflected in runoff of more than 20% of the daily rain.

Table 5. Surface Runoff on Grassland
 Groups by Runoff Amount
≥5 mm5 > X > = 2<2
Number of events, n294456
Median runoff (mm)641
Precipitation, 1 previous day (mm)282011
Precipitation, 2 previous days (mm)463121
Precipitation, 3 previous days (mm)594529
Maximum 15 min intensity (mm)655
Percent runoff over previous day's precipitation23%20%9%

[28] The comparison between precipitation and the data obtained for subsurface runoff in the hillslope plot shows that the average daily precipitation required to produce a subsurface runoff equivalent to 0.1 mm is 15 mm d−1 when preceded by rainy days. On average, precipitation in the previous 2 days of 29 mm and in the previous 3 days of 48 mm is required to produce measurable subsurface flows. The maximum subsurface flow measured between 15 November 2005 and 10 March 2007 occurred on 12 November 2006 and was 7.8 mm, associated with a precipitation of 57 mm corresponding to two events that occurred within 15 h. This indicates that shallow subsurface lateral flow is minimal during periods of no rain.

3.3. Water Yields and Low Flows

[29] The total water yield, defined as the water leaving the catchment as streamflow as a percentage of the annual precipitation, for each of the catchments is given in Figure 4. The higher water yield for B2 in relation to B1 could be attributed to the smaller area of forest cover of B2 and the larger extension of area in grasslands. BB has a smaller water yield in comparison with B2; catchments B2 and BB have similar areas of forests and grasslands and differ primarily in wetland area.

Figure 4.

Annual water yields for the studied catchments. NF, natural forest; PF, plantation forest; G, grasslands; W, wetlands.

[30] These values show the highest water yield in the catchment with the lowest forest coverage (B2) and the lowest water yield in the catchment with the highest area of wetlands (BB). B1 has 24% more area in forests than B2, which could be related to its smaller annual water yield.

[31] Flow duration curves were used for the analysis of low flows. Figure 5 shows the FDC in mm d−1 for the three catchment streams. Considering the definition of low flows (flows exceeded between 70% and 99% of the time), only the flows exceeded more than 90% of the time show significant reductions. B1 and B2 display similar behavior during 60% of the time when the flows are higher, although the large differences in event flows are not apparent on a log scale (Figure 5). When the flows begin to decline with the onset of the dry season, the flow of B1 (forest dominated) decreases first. However, when the dry season advances, B2 (grassland dominated) shows a clear drop in flow. In the dry season B1 presents a significantly reduced flow, though an order of magnitude higher than the flow of B2. B2 has a significantly reduced discharge in the dry season.

Figure 5.

Flow duration curves of daily stream discharge. NF, natural forest; PF, plantation forest; G, grasslands; W, wetlands.

[32] BB has a smaller discharge in the medium range of flows in relation to the other two catchments, but in the dry periods (10% of the time) its flow is higher than the other two streams. BB is the stream that sustains flow the longest in the dry season. It drops below the B1 flow only during the end of the dry season.

[33] Given the large precipitation and fairly high soil moisture content throughout the year, evapotranspiration in the three catchments is likely near the potential value. To differentiate potential ET from evaporation from an open surface, Penman [1967] used ratios of potential ET to open water evaporation and suggested using 0.7 as the ratio for a short grass in the humid tropics. Similarly, Pereira suggested 0.75 as the ratio for young pines in humid Kenya and 0.9 for mature evergreen rain forest in the same area [Pereira, 1967]. This would suggest that in the dry season the forest demand for water could be estimated to be around 28% higher than that of grasslands. This difference is apparent in the evapotranspiration calculation for the catchments and could help explain why the FDC for B1 (forest dominated) shows lower flows per unit area than B2 (grassland dominated) for the flows exceeded between 60% and 90% of the time (Figure 5).

3.4. Stormflows

[34] Rain events were classified in five groups as shown in Figure 6a. This classification was done considering the distribution and differences found between the combinations of total precipitation in the event, intensity in the 30 min interval, and duration. In the intermediate range of event size (between 10 and 20 mm), rainfall intensity was a more influential factor of classification than the duration of the event, and therefore, events in this size range were split by 30 min rainfall intensity. The distribution of the resulting event types is shown in Figure 6b.

Figure 6.

Criteria for the classification of rainfall events and frequency distribution of event types.

[35] Differences in the stream response to events were analyzed on the basis of two indicators: the volume response to individual events (discharge) and the lag times in hydrological response from the peak in precipitation of individual events to the peak in stream discharge. There appears to be no difference in discharge of the three catchments for event types 1, 2, and 3 (dry season), as can be observed in Figure 7. B1 (forest dominated) is the catchment with the lowest discharges for event type 3 in the wet season and for event type 4 (significantly different from B2 and BB, α = 0.05). Catchment B2 shows highest maximum values of discharge (higher variability), and catchment B1 shows the lowest variability (greatest buffering). This suggests the influence of forests on reducing the variability in discharge (B1).

Figure 7.

Discharge for catchments by season and event type. The box plots represent the minimum, lower quartile, median, upper quartile, and maximum of observations.

[36] The difference in time lag to peak between the three catchments is small as can be seen in Table 6, all lags being around 1.8 h and shorter for intense events (types 3 and 4) for all streams in the wet season [Roa-García, 2009]. The relatively large wetland in BB with an area of 0.6 ha consistently showed a lag in time between the peak in rainfall and the peak in wetland outflow [Roa-García, 2009], providing an explanation for the slower response of its host catchment.

Table 6. Time Lag to Peak in Flow for Streams
 Streams
B1 (Forest)aB2 (Grassland)bBB (Grassland and Wetlands)c
Number of EventsTLPF (min)Number of EventsTLPF (min)Number of EventsTLPF (min)
  • a

    Area 159 ha.

  • b

    Area 179 ha.

  • c

    Area 62 ha.

Wet309109303112321117
Dry10913290134106104

[37] Events were also analyzed using the linear reservoir concept. Out of the 15 events considered for this analysis, 11 were events of more than 20 mm in total and duration of 5 h or less and exhibited even spatial distribution (large intense events). Two of the analyzed events had a total precipitation between 10 and 20 mm with maximum 30 min intensity of 10 mm or less and an even spatial distribution (medium intense events); one event had a total precipitation between 10 and 20 mm with maximum 30 min intensity of more than 10 mm and an even spatial distribution (medium and slow event); and one event was a large intense event in two catchments and a medium slow event in the third catchment. Consequently, the analysis has been done predominantly for large intense events. The median time for each of the three theoretical reservoirs to drain out of the catchments is shown in Table 7.

Table 7. Median Indices of Water Release
 B1B2BB
ValueSDaValueSDaValueSDa
  • a

    SD, standard deviation.

T1 (h)9±68±57±6
T2 (h)26±2626±1617±15
T3 (h)137±67117±6657±30
Area (ha)159 179 62 

[38] For each of the events, three different T were calculated by dividing the corresponding hydrograph into three slopes and using the linear reservoir equation to calculate T for each section. T1 (fast), T2 (moderate), and T3 (slow) represent the time in hours that each of the theoretical reservoirs takes to drain out of the catchments. Three slopes were observed to be consistent in the hydrographic behavior for rain events, with an initial fast drainage followed by a slower recession limb and later by a clearly slower limb tending to base flow. Similarly to the case study in Ecuador [Buytaert et al., 2004] the catchments presented here have a low-permeability layer which limits the hydrological processes to saturated overland flow and shallow subsurface flow. As in that case study, three slopes were considered appropriate to represent the main drainage processes.

4. Discussion

4.1. Effects of Grasslands and Forests on Stream Discharge

[39] The relative hydrological behavior of the three catchments in the dry season shown through the FDCs provides evidence of the infiltration trade-off hypothesis. The significantly lower water yield of catchment B2 could be explained by the compaction of soil by grazing in the grasslands relative to the characteristics of soils in riparian or natural forests and the absence of the litter layer and canopy cover. Compaction reduces rainfall infiltration opportunities, making groundwater replenishment insufficient during the rainy season and producing strong declines in dry season flows [Bruijnzeel, 2004]. This is contrary to the conclusions of the majority of studies conducted in temperate areas, where as a result of deforestation, increases of base flow, and associated reduction in ET, are almost consistently observed [Hornbeck et al., 1993; Andréassian, 2004]. Under mature tropical forest typically 80%–95% of incident rainfall (rainfall reaching the canopy) infiltrates into the soil [Bruijnzeel, 2004]. The rate of evapotranspiration from undisturbed tropical forests is around 1400 ± 100 mm yr−1 [Roberts et al., 2005] and close to 950 ± 50 mm yr−1 from grasslands in moderately seasonal conditions [Scott et al., 2005; Grip et al., 2005]. On the basis of these data, replacing forests with grasslands would be expected to produce an increase in annual water yield between 300 and 400 mm yr−1, except if the forest removed was a cloud forest (characterized by persistent or frequent cloud cover at the canopy level) receiving an important amount of precipitation in the form of fog captured by the trees or if the soils are degraded to the extent that more water is lost from the catchment during the rainy season than what is potentially gained by a reduced evapotranspiration [Bruijnzeel et al., 2005]. Rather than this effect as the main driver for the increase in annual flows after forest clearing, which has been found to be the main factor in temperate areas [Bosch and Hewlett, 1982], in the tropics it appears to be the reduced infiltration capacity of the soil. According to studies by Bruijnzeel [1988, 2004], if infiltration rates after forest removal decrease to the extent that the amount of water leaving an area as quick flow exceeds the gain in base flow associated with decreased evapotranspiration, then dry season flows will be reduced.

[40] The use of the linear reservoir concept corroborates the results of the FDCs by showing that relative to size, for catchment B2, the time the water takes to leave the catchment is shorter than for the other two catchments. This indicates a smaller water storage capacity and higher surface runoff rates, related to grasslands that have less capacity to store water than forests.

[41] The comparison of discharge between catchments B1 (forest dominated) and B2 (grassland dominated) shown in Figure 7 indicates that forests reduce the variability of discharge for medium-size events in the wet season, highlighting the importance of forests in stormflow attenuation.

[42] Not only do forests appear to reduce discharge related to medium size events, they also appear to slow the rate of the discharge. The results of the linear reservoir analysis showed that although B2 is a larger catchment than B1, T1 and T3 (attenuation coefficients) are smaller than for B1 (forest dominated), suggesting the impact of the predominance of grasslands in B2, whose soils have a reduced porosity in comparison with natural forest (Table 4). Lumped lag times do not show significant differences between the peak in rain event and the peak in discharge for the dominantly forested catchment (B1) and the dominantly grassland covered catchment (B2), indicating that both catchments have a synchronized peak (the intervals between peak in rain and peak in discharge are similar) but different rates of release.

4.2. Effects of Wetlands on Stream Discharge

[43] Comparative catchment studies involving wetlands are not very common. Burt [1995] compared three headwater catchments located in the Yorkshire region of Britain, one of them dominated by peat deposits. The three catchments compared had very different geology, precipitation patterns, and soil types. The peat-covered catchment located in the uplands had an annual precipitation regime of more than double the other two catchments that were located in the lowlands. By comparison with the two lowland catchments, the peat-dominated catchment generated large quantities of flood runoff since there was little storage capacity available, providing a flashy runoff regime [Burt, 1995]. Assessments of the role of wetlands in headwater catchment hydrology require a closer observation of the paired catchment method in order to isolate the presence of wetlands from climatologic and geologic factors.

[44] The flow duration curves in this study may reflect the differences in catchment flow influenced by wetlands. One major difference between the FDC comparisons of Figure 5 is that BB shows a higher discharge per unit area for the flows in the 90%–95% exceedance probability. This suggests the effect of “large” wetlands with open water surface, with no vegetation and a surface outflow. A wetland in B2 was monitored for surface outflow using a water level recorder and displayed a hydrological behavior characterized by a prolonged surface flow in the dry season and a higher yield than its host catchment during part of the dry season. Out of the total number of wetlands (52) in the catchments only four wetlands with a total area of 0.9 ha have these properties (open water surface with no vegetation and a surface outflow), and of these, three are located in catchment BB with an area of 0.7 ha [Roa-García, 2009]. The hydrological behavior of prolonged surface flow contrasts with catchment response to individual events. The discharge values for large or intense events of catchment BB coincide with other wetland studies that have found wetlands as factors of increased flood peaks [Bullock and Acreman, 2003]; this is compatible with what was observed from the BB catchment wetland, which showed a “bursting” behavior in conditions of high antecedent precipitation [Roa-García, 2009]. This is particularly important in a region where the annual precipitation is concentrated in large events of high intensity; 34% of annual precipitation corresponds to events type 4, as seen in Figure 6.

[45] Even if wetlands appear to produce a “bursting” effect for intense rain events, they increase the residence time of water in the catchment as indicated by the linear reservoir analysis. Results for T attenuation coefficient of the linear reservoir analysis indicate that the higher the values for T, the longer it takes for the water to leave the catchment. It is also evident that BB, with only a third of the size of B2, has values for T that are only half those of BB, indicating a larger water storage capacity per unit area. B2 has a faster speed of drainage considering its size, which helps to explain the characteristics of its flow duration curves during the dry season. These results are compatible with what was found in a residence time study of the same catchments using isotopes [Roa-García and Weiler, 2010]. Lag time differences coincide with the linear reservoir analysis for the comparison between BB (the catchment with a 6% of area in wetlands) and B2. They indicate that despite the smaller size of catchment BB, there is a delay effect of peaks and water release, which could be attributed to the presence of wetlands. BB is also the catchment with the smallest average discharge, although discharge for intense and/or large events is larger than for the other two catchments. A detailed study of wetland and catchment hydrology would be necessary to understand in more detail the effect of 6% of wetland area on catchment BB since this percentage of area is considered small to produce such large differences in discharge, lag times, and annual yields.

[46] Besides discharge, another possible effect of wetlands in the hydrological response of catchment BB is the lag time to peak. Despite BB being almost a third of the size of B1 and B2, the lag times to peak in the wet season are very similar, which could be associated with the presence of relatively large wetlands in BB and the similarity in the drainage network density. This suggests that catchment size is not the only determinant factor of hydraulic lag as suggested by McGlynn and McDonnell [2003].

4.3. Effects of Soils on Stream Discharge

[47] There is likely a relationship between the presence of the volcanic ash layer at a relatively shallow depth (1.5 m or more) and the movement of water through surface and subsurface runoff since the ash layer limits the percolation of water and contributes to saturated conditions in the shallow soil layers. The quantification of runoff according to runoff volume and antecedent conditions of precipitation could contribute to the understanding of the larger stream discharge from catchments with a higher percentage of area in grasslands since surface runoff was observed to occur only on grasslands.

[48] According to the surface runoff data, the catchment with the highest percentage of area in grassland (B2) would have a large volume of surface runoff occurring during storm events and flowing into the stream channel, increasing the total volume of water flowing into the stream. Given that saturated hydraulic conductivity was estimated to be of the order of 10−3 cm s−1 (approximately 9 mm per 15 min), the surface runoff is likely dominated by saturated conditions and not by rain intensities in excess of infiltration rates (Hortonian).

[49] The comparison of soil moisture retention curves of the three land use types of the study site with the typical curves for clay, loam, and sand [Brady and Weil, 2002] in Figure 3 shows that despite the large water storage capacity of the soils, they tend to hold water with a high matric potential, reducing its movement. This is a result of the degree of soil development, soil structure, soil texture, parent material, organic matter content, biological activity, and soil management [Cassel and Lal, 1992]. The free draining water or storage capacity between saturation and field capacity of the studied soils does not differ much from that of other soils (around 15%) and constitutes the base flow in streams. But below field capacity (water available to plants) water is held in the soils with the moisture content reduction of only 5% (which coincides with the soil moisture data taken in the field) that does not fluctuate very much between the wet and the dry seasons. A comparison of the three soil moisture curves for the study site (Figure 3) shows that the larger drop in soil moisture content occurs for riparian and natural forests, particularly between saturation and field capacity, which is an indication that these soils have a larger capacity to release water (contribute more to base flows) than those in grassland (statistically significant, α = 0.05).

[50] The relations between water storage and the velocity of water release, which are influenced by land use practices, were estimated using the linear reservoir concept by estimating attenuation coefficients (T). Buytaert et al. [2004] compared two high-elevation headwater catchments (at around 4000 m) in the Andes of Ecuador. The study shows a difference of about 40% in the rate of water release between the cultivated (impacted) and undisturbed catchment [Buytaert et al., 2004]. In the present study, on a per area basis catchment BB presents a consistently slower release of water from the three reservoirs, the difference being greatest for the fast reservoir. B2 shows the smallest attenuation coefficients, which indicates a faster release of water in comparison with the other two catchments (Table 7). These differences between the three catchments indicate the influence of wetlands prolonging the release of water and the effect of grasslands reducing it.

5. Summary and Conclusions

[51] Different methods were used as part of a comparative catchment approach to assess the effect of land use in headwater catchments on stream discharge. The comparative catchment approach minimizes variability related to climate and geologically determined differences in groundwater reserves and deep percolation as well as spatial and year to year variability.

[52] The presence of a volcanic ash layer and the water storage and release characteristics of Andisols influence the hydrological processes by creating slow percolation rates and increased surface runoff during the wet season. The soil moisture curves indicate that the soils under riparian and natural forests have a larger capacity to store and release water than those in grassland. The FDCs illustrate that the catchment with the highest percentage of area in forest (68%) experienced a small reduction of flow during the dry season and maintained a higher flow during the last days of the dry season in comparison with the grassland-dominated catchment. These two findings support the “infiltration trade-off” hypothesis for tropical environments that soils that are subject to compaction (such as highly grazed grasslands) have a reduced rainfall infiltration, which impairs the maintenance of base flows.

[53] Water requires more time to leave the forested catchment, and wetlands have an effect in prolonging the residence time of water, shown by the linear reservoir analysis. Even though catchment BB had higher discharge for large events, the linear reservoir concept showed that on a per unit area basis, water took more time to leave BB in relation to the other catchments. Wetlands also appear to delay the reduction of catchment flows in the dry season, as was illustrated through the FDC. Forests also demonstrated a significant effect in dampening discharge of rain events in water volume released during and immediately after rain events. However, there is no significant difference in lag time from the peak of the precipitation to the peak in discharge between catchments B1 (forest dominated) and B2 (grassland dominated). This means that forests do not produce a delay in peak discharge. Considering the smaller size of catchment BB, its lag time to peak in discharge is larger, suggesting the influence of wetlands in delaying peak discharge. Less clear is the contribution of wetlands to smaller discharge. The three catchments show very similar discharge for small events, but BB shows a higher discharge for large and intense events. The delayed peak and the large discharge suggest that wetlands could “burst” as a consequence of saturated conditions following large events. Although the wetlands contribute to prolonging the transit time of water in the catchments, their current water storage capacity does not compensate for the naturally limited release of water from the catchment soils.

[54] Results of this study highlight the dependency of mountain communities on the ability of the ecosystems in the headwaters to regulate water flows. The transformation of the headwater catchments from forests and wetlands to grasslands is most likely contributing to the reduction in their water regulation capacity, as indicated by this research. The water regulation function of these catchments, while influenced by land use, is overridden by climate regime and soil characteristics and consequently constrains the potential of management to influence dry season flows and stormflow attenuation.

Acknowledgments

[55] This study was funded by the International Development Research Center (IDRC) and the International Center for Tropical Agriculture (CIAT). We thank the Filandia team for data collection and three anonymous reviewers for improving the quality of the manuscript.

Ancillary