Exacerbation of erosion induced by human perturbation in a typical Oceania watershed: Insight from 45 years of hydrological records from the Lanyang-Hsi River, northeastern Taiwan



[1] High precipitation, steep slopes, small basin areas, and frequent flood events can induce high erosion rates on Oceania islands. These natural characteristics make watersheds much more vulnerable to anthropogenic perturbation. We analyzed long-term data sets (1950–1994) of suspended sediment loading from two gauges in Lanyang-Hsi watershed, a typical mountainous river in northeastern Taiwan. Prior to road construction that began in 1957, the annual sediment yield for the downstream gauge near the river mouth was in the range of 730–5400 t km−2 yr−1 with a mean sediment yield of 2800 t km−2 yr−1, which was 18 times higher than the global mean (150 t km−2 yr−1). However, following massive road construction projects, sediment exports increased more than tenfold for the downstream gauge and fortyfold for the upstream gauge at an altitude of 450 m, indicating an exacerbation of erosion induced by human activities. Such conditions of high export lasted for 2–4 years before the sediment yield returned to lower level. From 1957 to 1994 the mean sediment yield for entire watershed went up to 12,800 t km−2 yr−1; ∼75% of that resulted from human perturbation. By comparing the sediment loads from the two gauges we concluded that the extra sediments mainly originated from the upper reach. A huge amount of the sediments apparently had resided in the middle reach; subsequent flood events with enough physical strength may resuspend and sweep the sediments to the sea gradually. Since the Taiwanese rivers represent the extreme conditions in correlation plots for the calculation of sediment export from mountainous rivers, one should exercise great care in distinguishing natural from perturbed conditions when using sediment yield data from Taiwan.

1. Introduction

[2] Land to ocean material fluxes play an important role in global biogeochemical cycles. Recently, more attention has been drawn to human-induced fluxes since human-disturbed drainage systems may affect the transport and processing of sediments, carbon, and nutrients in aquatic systems. This may lead to significant changes in the magnitude and the nature of riverine material exported from terrestrial biomes to the oceans [Meybeck, 1993; Meybeck and Vorosmarty, 1999; Kao and Liu, 2000].

[3] Oceania islands cover <4% of the Earth's surface area but may collectively contribute over 45% (9 Pg yr−1) of fluvial suspended sediment flux to the oceans [Milliman and Meade, 1983] because of high sediment yield, which is defined as sediment production rate per unit area. On the basis of more recent observations, Milliman and Syvitski [1992] revised their estimate of the mean yield for Oceania small rivers to 3000 t km−2 yr−1, which was 3 times larger than a previous estimate (1000 t km−2 yr−1 [Milliman and Meade, 1983]). The mass flux from Oceania islands could be globally important because the exported particulate material would be much more easily transported to the deep seafloor, resulting in substantial carbon burial due to the narrow shelf, frequent flooding [Milliman, 1991], and a small Coriolis parameter [Nittrouer et al., 1995]. Taiwan, a typical Oceania island with many mountainous watersheds, has a very high sediment yield (14,400 t km−2 yr−1 [Milliman, 1991]), which is two orders of magnitude greater than the world mean value (150 t km−2 yr−1). Seven rivers in Taiwan are on the top ten list of the most sediment-yielding rivers [Milliman and Syvitski, 1992]. The abnormally high sediment yield in Taiwan in recent times might result from road construction and intensive agricultural activities [Milliman, 1991; Kao and Liu, 1996; Kao and Liu, 1997] combined with natural conditions, such as steep slopes, high precipitation, and frequent floods.

[4] Long-term historical data are most valuable because they provide information on how the natural system responds to human perturbation. Unfortunately, most Oceania islands have rather limited historical records. By comparison, Taiwan has relatively long hydrological records. Previously, we reported the sediment export for the period of 1970–1991 on the basis of published data of the Taiwan Water Resources Bureau (WRB). Recently, we obtained the entire historical record of two Lanyang-Hsi gauge stations from 1950 to 1994. In this paper, we analyzed these two data sets to evaluate human effects on sediment yield and to explore the associated transportation processes. The additional data not only confirm our previous findings but also provide more details of the behavior of the Lanyang-Hsi watershed in response to human disturbance. New calculations allow us to make more precise estimates of the sediment yield under natural conditions and the cumulative effect of human disturbances. Similar approaches can be used to analyze records from other Oceania watersheds, including many others on Taiwan. Results of these analyses may provide a more accurate database of sediment yield on Oceania islands, which apparently has global significance due to the high volume of sediment production.

2. Study Site

[5] The Lanyang-Hsi (Figure 1) watershed is one of the major rivers in Taiwan. It originates at an altitude of 3535 m and runs a total of 70 km. The main channel follows the direction of SW-NE with a mean gradient of 5% with much steeper slope (>20%) for the upper main channel. The annual precipitation in this watershed ranges from 2000 to 5000 mm for the past 50 years, with an average of ∼3000 mm, which is high in comparison with global average but typical for Oceania islands. The denudation rate in Lanyang-Hsi watershed is high, as it is in the rest of Taiwan [Li, 1976]. The basement rock is composed mainly of Tertiary argillite-slate and metasandstone [Ho, 1975]. The lithology and climate conditions are homogeneous in the watershed. There are two gauge stations along the main stem (Figure 1). Gauge 1 (G1) is located at the river mouth but above the tidal zone. Gauge 2 (G2) is located at the upper stream with an altitude of 450 m. The drainage areas above G1 and G2 are 820 and 273 km2, respectively. The Taiwan Provincial Archives' records show that there were two massive road construction events in the study area over the past 50 years. Project 1, from 1957 to 1960, was the construction of the Central Cross-Island Highway, which is a major artery running along the main stem of the Lanyang-Hsi (marked in Figure 1). Project 2 was an extensive construction of paved secondary and tertiary roads within the I-Lan Prefecture during 1975–1980. In 1980, a new bill regarding land use in hillside areas was passed. It allowed farming activities on rather steep slopes to accommodate the livelihood of the increasing population on Taiwan. Subsequently, many vegetable plantations were developed along the river beds and banks of the main channel up to altitudes as high as 1250 m.

Figure 1.

Map of the Lanyang-Hsi drainage area. Two gauge stations are shown as well as main roads, including the Central Cross-Island Highway constructed during 1957–1960.

3. Data Analysis

[6] Sediment load Qs, which is defined as the amount of sediment carried in the river flow per unit time, was calculated from the hydrological data of the two gauge stations in the Lanyang-Hsi watershed provided by the WRB. The record periods were from 1950 to 1994 for G1 and from 1974 to 1994 for G2. The hydrological data include two sets: (1) the discrete data of total suspended sediment concentration (Cs; average 27 samples per year) and corresponding instantaneous water discharge rate (Q) and (2) the records of daily Q, reported hourly during typhoon invasions since 1966. The first set was used to construct the rating curve, which depicts an empirical relationship between Qs and Q [e.g., Campbell and Bauder, 1940; Crawford, 1991]). This relation is usually defined as a power function,

equation image

where Qs = QCs. The logarithmic transformation yields a linear expression,

equation image

where a and b are the intercept and slope of the rating curve and ei is the residual error (the amount by which the observed response differs from the predicted response) corresponding to the ith observation. The least squares logarithmic regression procedure generally results in underestimation because of the overrepresentation given to the deviation's data points below the fitted line [Farr and Clarke, 1984]. Following the recommendation of Cohn [1995], we adopted a nonparametric correction [Duan, 1983] for the transformation bias. The correction term is

equation image

where n is the number of observations in the data set. Through inverse transformation of (3), an unbiased relationship between Qs and Q is then given by

equation image

where c = log β. We performed regression analysis to construct yearly rating curves. For a few years (1958, 1959, and 1990 for G1 and 1977, 1980, 1984, and 1986 for G2) the regression yielded poor correlations owing to undersampling, so we developed new rating curves for these years by using 3 years (1 year on each side) as the interval [Kao and Liu, 2001]. Once we established the rating relationships, we used the second data set to transform daily mean Q into daily Qs by means of rating curve. The annual load was derived by summing up the daily loads. Details of the regression analysis of the data set were reported by [Kao and Liu, 2001].

4. Results and Discussion

[7] In a previous study [Kao and Liu, 1996], we demonstrated enhanced erosion rate after road construction in the 1970s, but the 20 year records were too short for us to establish the baseline for sediment production in the study watershed. The extended records presented in this study include data obtained before project 1, the first major road construction (1957–1960) in modern times in the watershed, revealing the undisturbed condition. Here we present an example of such data obtained in 1956, which is contrasted with the records from 1963 that represent the perturbed condition (Figures 2c and 2d). The measured sediment concentrations in 1963 reached as high as 30,000 mg L−1, whereas the observed maximum in 1956 was <1000 mg L−1. The rating curves of both years show good correlations (Figure 2d), yielding the b value that increased from 1.5 for 1956 to 2.0 for 1963.

Figure 2.

The scatterplots between suspended sediment concentration Cs, sediment load Qs, and water discharge Q for selected years: (a) Q-Cs for G2, (b) Q-Qs for G2, (c) Q-Cs for G1, and (d) Q-Qs for G1. Open and solid symbols represent the years shortly before and after road construction projects, respectively. The regression functions are also shown for rating curves.

[8] The new data set from upstream gauge station G2 also shed new light on how the steeper slope in the upper reach responded to human perturbations. Here we present the contrasting conditions observed at G2 in 1975 and 1980, which were before and after project 2, respectively. Speaking overall, Cs varies over three orders of magnitude, showing a positive relationship with Q in both years (Figure 2a). This indicates the importance of water flow in determining the concentrations of suspended sediment in river waters. At the low end of Q the Cs values from the two records overlap considerably, but at the high end the Cs value increased one order of magnitude for a given Q value after project 2. The huge Cs differences result in distinctive rating relationships so that the b value increased from 2.0 for 1975 to 3.7 for 1980 (Figure 2b). Compared to the example given for G1, the dramatic increase in the b value for the records from G2 indicates the susceptibility of erosion process to human disturbances on steep slopes.

[9] Figure 3 shows all parameter values derived from regression analysis of hydrological data. The three parameter values illustrated in Figure 3 are all exponents (namely, the a, b, and c values) in the expression of Qs in (4). Because of the high variability of the three parameters we plot the 3 year moving averages of the data to depict the temporal trends. The b values vary from 1.2 to 3.7 for G2 and from 1 to 2.7 for G1 (Figures 3a and 3b). The a values are in the range of −1.6 to 1.6 (Figures 3c and 3d), and the c values range from 0.01 to 0.9 (Figure 3e). All b values are >1, except the one for 1950 at G1 (Figure 3b), while the other two parameter values are mostly <1. The rating relationships vary significantly year by year, confirming the suggestion of Walling [1977] that short-term rating curves are more suitable for small watersheds. Because the b value is usually the largest among the three exponents for each year and the base of the second exponential function, Q, is often much larger than the base (10) of the other two exponential functions in the Qs expression, the b value is apparently the dominant factor controlling the magnitude of estimated Qs. Figures 3a and 3b show that the b value increased during the two projects and reached a peak at the end of both road construction periods. After road construction was completed, b values drop to a lower level at both stations, indicating a recovery process. The close match of the timing between increases of b values and road construction events indicates the alteration of the Q-Qs relationship is caused by human disturbances. Accordingly, we suggest that in Lanyang-Hsi watershed the b value could be used as an indicator for the magnitude of watershed erodibility responding to disturbances, such as deforestation, road construction, and planting. Whether this is applicable to other mountainous watersheds warrants further study.

Figure 3.

The temporal variations of regression parameters: (a) slope for G2, (b) slope for G1, (c) intercept for G2, (d) intercept for G1, and (e) the bias correction term (see text) for G1 and G2. Solid and open symbols represent the data for G1 and G2, respectively. Shaded areas are road construction periods. Solid lines in Figures 3a–3d are the 3 year moving averages.

[10] Figure 4 shows the results of sediment load calculation for G2 and G1. As we have stressed the importance of the b value, the base of the exponential function, namely, Q, is important as well. For comparison, we present runoff depth, which is defined as the discharge rate per unit area of the drainage basin in units of mm yr−1. Figure 4a shows that the annual mean runoff depth of G2 tracks that of G1 closely, whereas the maximum daily runoff depths differ considerably, indicating that the torrential rains usually associated with typhoons were not evenly distributed spatially. Although Qs is dependent on Q within each year, we found that the interannual variation of the annual Qs does not follow that of annual runoff depth in general. Apparently, changes in runoff depth were not the main reason for the interannual variation of sediment production. On the other hand, many peaks of sediment loading correspond to high values of maximum daily runoff depth, which represent significant flood conditions. However, not all major floods induce high sediment loading; only those following the two projects have induced exceptionally high sediment loadings.

Figure 4.

The interannual variations of the (a) annual mean runoff depth, (b) maximum daily runoff depth, and (c) annual sediment load for G1 and G2. Solid circles and open circles represent the data for G1 and G2, respectively.

[11] The significance of the peak values of runoff in controlling the sediment load illustrates the strong nonlinear response of sediment production to water discharge rate. Similarly, the hourly discharge rates, whose peak values are often considerably higher than the daily mean value, may yield higher sediment loadings if used for Qs calculation. In order to test how significant the differences may be, we did the calculation for those years with hourly data available for flood periods. The average increases in the estimated annual loads are 9 ± 8% for the 27 years of G1 records and 21 ± 12% for the 9 years of G2 records. Because changes of such scales do not alter the temporal trend and hourly data are not available for every year, we decided to make the calculation based on the daily mean discharge rates to be consistent.

[12] Before the start of project 1 in 1957, Qs values calculated for G1 were only 0.6 to 4.4 Mt yr−1 (Figure 4c). The average Qs over the period of 1950 to 1957 was 2.3 Mt yr−1, which gave a sediment yield of 2800 t km−2 yr−1 for the drainage area above G1. For G1, project 1 caused an increase of sediment loading with a maximum reading of 36 Mt yr−1 in 1963; similarly, project 2 induced a maximum loading of 27 Mt yr−1 in 1980. For G2 we also found a low level of sediment load before 1979 ranging from 0.1 to 1.5 Mt yr−1 with a mean of 0.7 Mt yr−1 for 1974–1978. This corresponded to a mean yield of 2600 t km−2 yr−1 for the drainage area above G2, which was close to the mean yield obtained from G1 in the earliest period. At the end of project 2, which marked the beginning of relaxation of land use regulation, the sediment load increased up to 185 Mt yr−1. At both stations the sediment loadings remained high for 2–4 years and then dropped back. The close correlation between peak Qs values and road construction events suggests that the construction and the human activities immediately following road work led to the exacerbation of erosion. During both projects the surges of Qs occurred in the later years of the construction periods, implying a delayed response of erosion, which escalated as the accumulated disturbances reached a certain threshold.

[13] The cumulative curves for the daily Qs values (Figure 5) show step-like increments corresponding to episodic floods. In addition, the curves also show sudden changes in the increasing trend, reflecting changes in erosion conditions due to human perturbations. The linear trend lines in Figure 5 depict the mean accumulating rate or the mean sediment production rate in different periods of time. For G1, project 1 caused a sudden increase of accumulating rate from 2.3 to 21.4 Mt yr−1. Between 1964 and 1979, the rate dropped back to 3.9 Mt yr−1. After project 2, the rate increased to 23.7 Mt yr−1 again and then dropped back, but its values after 1983 remained significantly higher than that before 1957. For G2, strong peaks in 1980 and 1981 make the accumulating rate jump from 0.7 to 150 Mt yr−1, and after 1983, it dropped to 4.7 Mt yr−1, corresponding to a sediment yield of 17,200 t km−2 yr−1, which was much higher than the background level. Apparently, project 2 caused a much more drastic change in the upper watershed such that the Qs remained significantly higher than background level after the initial surge.

Figure 5.

The cumulative sediment loads for (a) G2 and (b) G1. The slopes of the regression lines represent the mean annual sediment loads during the selected periods. The dashed line represents the cumulative background load derived from the mean yield for preevent years (see text). The values of the sediment yield per area are also shown.

[14] If we use the mean accumulating rate in the earliest period of time at each station as the baseline, we may estimate the cumulative sediment production in the study period under undisturbed conditions (see dashed line in Figure 5). During 1950–1974 (encompassing the period of project 1), the cumulative Qs obtained for G1 was 164 Mt, while the background loading accumulated to only 55 Mt. We regarded the difference, 109 Mt, as human-induced loading. Similarly, for the time period of project 2 (1975–1994) the total cumulative loading was 250 Mt for G1 and 555 Mt for G2, and the background loading was 46 Mt for G1 and 15 Mt for G2. The human-induced loading was then 204 Mt for G1 and 540 Mt for G2. The considerably higher loading for G2 during the same period suggests that human-induced loading might originate entirely from the drainage area above G2. This result is consistent with our previous findings [Kao and Liu, 1996] that sediment yield in the tributaries was far lower than the mean of the entire watershed. Furthermore, apparently, a huge amount of the sediments that passed G2 might still reside on the riverbed in the middle reach. The extra sediment loading induced by project 2 was more than twice that by project 1. Also, the effect of project 2 lasted longer, probably related to the relaxing of land use regulation adopted in 1980, which allows growing farming activities in the upland area of the watershed. Up until 1994, the sediment yield did not fully recover to the background condition. In fact, two peak Qs values occurred coincidentally at both stations in 1992 and 1994, >10 years after the end of project 2. The much higher Qs values at G2 again demonstrate that the steep slope in the upper reach makes the drainage area more susceptible to human perturbation. It is not clear why the extraordinarily high Qs occurred in these two years. Figure 4a shows that the annual runoff depths in 1992 and 1994 are significantly higher than those in the 1980s. The unusually high runoff during these two years may be responsible. In addition, the high daily runoff for G1 also occurred in 1992 (Figure 4b).

[15] Milliman and Syvitski [1992] compiled data from a large number of world rivers, classified them, and obtained regression relationships between sediment loading and watershed area for different types of watersheds. Their results demonstrated the importance of small mountainous rivers in global sediment discharge. Most of Taiwanese rivers are among the highest sediment yielding rivers. As illustrated above, the extreme values of sediment yield are mostly caused by human perturbation, whereas the natural sediment yield is only a fraction of total sediment production. The regression relationship established by Milliman and Syvitski [1992] for mountainous watersheds were probably influenced by data from perturbed systems. The calculated sediment loading was likely biased by signals from anthropogenic disturbances. If the total sediment loading under natural condition is desired, care should be taken to use the sediment yield estimated for undisturbed conditions.

[16] The human-induced erosion observed in the Lanyang-Hsi watershed could be occurring in any Oceania island where land use is changing. Since the increase in sediment yield could be an order of magnitude, the total increase in sediment loading may be significant from a global perspective. The impact on global biogeochemical cycles is worth careful assessment. In terms of carbon cycling, there may be two processes with opposing effects: on one hand, the accelerated erosion exposes more carbon-bearing bed rocks. Once the mineral structure is destroyed, the carbon may be oxidized and released as CO2. Although kerogen in the particulate organic carbon observed in Lanyang-Hsi and other Taiwan rivers show very old apparent ages (10,000–25,000 years [Kao and Liu, 1996; Kao, unpublished data, 2001]), new evidence indicates that newly exposed fossil carbon is metabolizable for microbes (Y. P. Hsieh, personal communication, 2001). On the other hand, the mineral grains may absorb dissolved organics with their freshly exposed mineral surfaces and act as carbon carriers. Keil et al. [1994] suggested that the availability of mineral surface is one of the controlling factors limiting the extent of organic matter preservation in marine sediment. It is also well known that increase in sedimentation rate may increase the burial and preservation of organic carbon. Apparently, accelerated soil erosion on Oceania islands may lead to changes in global biogeochemical cycles, but the magnitude of the impact and the potential consequences are yet to be investigated.

5. Conclusions

[17] By using historical data sets, we differentiate the background sediment loading from human-induced loading. Remarkably high yield periods immediately following two large-scale road construction projects demonstrate the exacerbation of land erosion induced by human activities. From the beginning of major human disturbances in 1957 until 1994, the total sediment load exported from Lanyang-Hsi watershed was 399 Mt with a mean 12,800 t km−2 yr−1, which was at least 4 times the background value (2800 t km−2 yr−1) observed prior to 1957. This indicates that human-induced sediment loading accounts for 75% of the total yield. For drainage area above 450 m in altitude the mean sediment yield was 110,000 t km−2 yr−1, which was ∼40 times the mean value (2600 t km−2 yr−1) observed before 1980, indicating over 95% of the sediment yield from the upper drainage basin was induced by human activity. By comparing the total sediment loads for the two gauges, we concluded that the extra sediment loads mainly originated from the upper reach, which might be related to the relaxation of land use regulation beginning in 1980. Since mountainous watersheds are vulnerable to human perturbation, which may occur commonly in Oceania islands due to economic development, the sediment yield may increase significantly and collectively may lead to nonnegligible changes in the global biogeochemical cycle, which warrants further study.


[18] This study was supported by a grant (NSC 89-2611-M-002-004-op1) from the National Science Council of the Republic of China. This is the contribution 48 of the National Center for Ocean Research. We are grateful for the invaluable data provided by the Water Resources Bureau. We thankfully acknowledge comments from J. M. P. Syvitski (University of Colorado, Institute of Arctic and Alpine Research), I. C. Chang (Tainan Hydraulics Laboratory, NCKU, Taiwan) and anonymous reviewers.