fulbright Taiwan online journal

fulbright Taiwan online journal

Modeling the impact of dam removal on conservation of the Formosan landlocked salmon

           The Chichiawan Stream and its tributaries in central Taiwan are the last refuge of the critically endangered Formosan landlocked salmon Oncorhynchus formosanus.  Over the past few decades, 11 check dams have been constructed in these streams to reduce sediment transport and to prevent the collapse of riverbanks. However, these dams are thought to be a primary factor in the habitat degradation that has led to a decline in salmon abundance. The dams have impacted the salmon by creating reproductive isolation, by reducing the number of accessible large boulders to provide refuge during typhoons, and by preventing salmon from returning upstream after being flushed downstream during typhoons. In addition, the sand and gravel that accumulate due to dam construction can damage salmon eggs. Typhoons, occurring primarily in spring and summer months, are a key factor in salmon population dynamics, and the salmon have adapted to seasonal typhoons in their natural habitat. However, dams have altered their natural habitat, limiting the salmon’s ability to survive typhoons. The salmon abundance began to decline in the 1960s, reaching as low as 200 individuals by 1984. The abundance has increased to over 1,200 in recent years, but the salmon have not completely repopulated to their historic range.

            In recent years, dam removal has become a frequently-used strategy for restoring stream habitats. After the Formosan salmon was classified as critically endangered, dam removal was used as a conservation strategy in Kaoshan Stream, one of the tributaries of Chichiawan Stream. Four check dams were partially removed from Kaoshan Stream from 1999 to 2001, and the salmon population increased, with fluctuation, afterward. We investigated potential outcomes of removing dams in Taoshan West and Chichiawan Streams, where three check dams remain intact.

            During the first half of my year in Taiwan, we devised an age-structured stochastic simulation model to analyze the effect of dam removal on salmon abundance in Kaoshan Stream. The model included demographic stochasticity for fish length, adult survival rates, and the number of juvenile recruits. To account for the effect of typhoons, two parameters (adult survival rates and the number of juvenile recruits) were dependent on the spring discharge, and the model included environmental stochasticity in the forecasting of spring discharge values. This forecasting method also included a climate change effect, based on expectations that the intensity of typhoons will increase with the global temperature.

            In the second phase of research, we made some refinements to the phase 1 model by developing a typhoon score to replace the discharge index. The typhoon score accounts for several typhoon characteristics that may affect the salmon population, but were neglected by the discharge index. In addition, we re-calibrated the model for regions in Taoshan West and Chichiawan Streams to investigate the effects of the dams and dam removal on salmon abundance in these regions. The objectives of this study are: (1) to determine if replacing the discharge index with the typhoon score improves the accuracy of the model; (2) to predict population trends if the dams are not removed, including the effect of climate change; and (3) to determine the impact of each dam removal on salmon abundance, including the combined effect with climate change.

            The salmon habitat is located in the Wuling Basin in Shei-Pa National Park in central Taiwan (Figure 1). Their habitat is comprised of three third-order streams (Chichiawan, Yousheng, and Kaoshan) and two second-order streams (Taoshan West and Taoshan North) on an upstream reach of the Tachia River. The streams are characterized by cool water temperatures and steep channels with high water velocity.

            The model for Kaoshan Stream uses population data from zones 10-13. We also model three regions corresponding to the three remaining dams: the dam 3 region, consisting of zones 1-4, 6, and 7; the dam 4 region, consisting of zones 6, 8, and 9; and the dam 6 region, consisting of zones 8 and 9. The zones on either side of dam 3 make up the dam 3 region in order to model the scenario in which this dam is removed, connecting the sections. Similarly, the dam 4 region is used to model the scenario in which dam 4 is removed, and the dam 6 region is used to model the scenario in which dam 6 is removed. The abundance was counted by snorkeling two times each year (summer and early winter) in these regions beginning in 1995, but complete estimates for all zones under consideration were available only from 2009 to 2014.

            In the phase 1 study, we defined a spring discharge index to be the average daily discharge from May to September to measure of the impact of typhoons each year. However, this index did not capture several additional factors that influence population dynamics. In this study, we devise a typhoon score that accounts for these additional factors and use this score in place of the spring discharge index.

We assign each typhoon a pathway score that takes into account the continuity of the typhoon (a measure of the degree it has broken into smaller pieces), whether it is concurrent with a monsoon, the proximity to the Wuling basin, and its radius. We also assign a frequency score to each typhoon that measures the length of time between consecutive typhoons. This score is important because typhoons generally have a larger impact on the salmon population when they occur in closer temporal proximity. Then, we assign each individual typhoon a score, defined as the product of the pathway score and the frequency score. Finally, we define an annual typhoon score by summing the individual typhoon scores for all typhoons that occurred during that calendar year and dividing by 200 for scaling purposes.

            The annual typhoon scores for 1990-2014 were computed based on observed data from the Typhoon Database (http://rdc28.cwb.gov.tw/) and are used in models that consider time periods within these years. For models that project to future years, a typhoon score is selected randomly each year, based on the historical mean, the correlation between observed consecutive years, and the standard deviation of the error between observed scores and the mean of the observed scores. The typhoon scores and the discharge indices had similar directional trends for 12 years, but there are 6 instances in which they change in opposite directions between consecutive years. There are also differences in the magnitude of change between the two scores over consecutive years.

            We included an option in the phase 1 study to account for anticipated increases in typhoon intensity associated with climate change. The same type of climate change effect is used in the present study, but it is applied to the typhoon score rather than to the discharge index. We define P to be the percent growth in the typhoon score over the period 2014-2035. We first forecast the typhoon scores based on historical data and gradually increase these values linearly over time so that the 2035 typhoon score is increased by percent P. Based on recent trends, we apply the increasing factor to only 2/3 of the typhoon scores, selected randomly.

We include this climate change effect by varying P from 0 to 3.4 in increments of 0.1. As part of our analysis, we consider a likely climate change scenario based on a study that anticipates the quantity of precipitation in Taiwan to increase by 10% per decade (IPCC, 2014). We determined that  P = 1.4 represents a likely climate change scenario based on these expectations.

            The following parameters are taken from the phase 1 study and are the same across all stream regions: a Von Bertalanffy length at age curve, the probability of maturity as a function of length, and the fecundity (number of eggs) of a female as a function of length. The adult survival rates are calibrated as functions of the annual typhoon score. The number of juvenile recruits is  a function of the number of eggs and of the annual typhoon score. For Kaoshan Stream, these parameters were created separately for pre-removal conditions (data from 1995-2000) and post-removal conditions (data from 2009-2014). For all other regions, these parameters were calibrated for pre-removal conditions only (data from 2009-2014) because the dams are still intact in these regions.

            We observed that in Kaoshan Stream, there was very little change between the adult survival rates as a function of the annual typhoon score after the dams were removed. There was a noticeable increase in the number of recruits after dam removal, although the benefit decreases as the typhoon score increases. Based on these observations, we take the adult survival rates from the same curve that was calculated under pre-removal conditions for the dam 3, 4, and 6 regions. To estimate number of juvenile recruits under post-removal conditions in these regions, the number of recruits from pre-removal conditions is increased by the percent growth observed in Kaoshan Stream. We also consider alternate scenarios in which the percent growth in the number of recruits in these regions is smaller than that in Kaoshan by applying a scaling factor S (from 0 to 1 in increments of 0.1).

            For each model under consideration, we performed 1,000 simulations using Matlab 2014a and used the mean population size as the prediction for each year. For Kaoshan Stream, we performed the pre-removal model over 1995-2000 and the post-removal model over 2009-2014 and compared the predicted and observed values. We used these results to determine if replacing the discharge index with the typhoon score improved the accuracy of these models. We ran the pre-removal model from 1995 to 2014 to predict population trends assuming the dams had not been removed from Kaoshan Stream. We projected the post-removal model from 2014-2035 to predict future population trends assuming the dams have been removed.

            For the dam 3, 4, and 6 regions, we ran the model over 2009-2014 and compared the model predictions to observed data. Then, we projected each model from 2014 to 2035 under different combinations of S and P values to model different scenarios of dam removal combined with climate change. The typhoon score for 2014 was taken from the observed data, and our forecasting method was used for the remaining years. For scenarios in which both S and P vary, we performed multiple linear regression on the projected 2035 salmon abundance as a function of S, P, and SP to analyze the effect of dam removal, climate change, and their combined effect. In each dam removal scenario, we computed the predicted increase in the 2035 abundance as a result of dam removal.

                        We summarize the results in regards to the four objectives:

(1)  To  determine if the model refinements improve the accuracy of the predictions

            For both the pre- and post-removal models in Kaoshan Stream, the model explained a higher percentage of the observed data when we replaced the spring discharge index with the typhoon score, indicating that the typhoon score improved the accuracy of the model. However, we note that the model predictions for Kaoshan Stream were fairly similar between the phase 1 and 2 studies. Both models predicted that the population would reach zero by 2006 if the dams had not been removed. Also, both models predicted that the populations would stabilize near 400 individuals by 2018 with the dams removed, although phase 2 predicts a slightly lower value. The model also explains a high percentage of the data for the dam 3, 4, and 6 regions, although comparisons cannot be made since these regions were not considered in phase 1.

(2) To predict population trends if the dams are not removed, including the effect of climate change

            When climate change is not included in the pre-removal model, the Kaoshan population is predicted to decline to near zero by 2006. The observed population in Kaoshan Stream declined to 24 in 2001 but improved after the dam removal, fluctuating between 100 and 1,000 after 2001. The population after dam removal is larger than the model predicted it would have been if the dams had not been removed, suggesting that dam removal may have reversed the trend toward zero.

            The population in the dam 3 region is predicted to decline from the observed value 1291 in 2014 to near zero by 2029 if the dam is not removed. The populations in the dam 4 and 6 regions are predicted to increase from 2014 to 2015 and then to gradually decline. The predicted 2035 abundance is larger than the observed 2014 values in both of these regions. However, the decline over 2015-2035 in both regions is cause for concern, especially if the 2015 abundance is less than predicted.

            When we vary the climate change parameter in the pre-removal model, there is little effect on the dam 3 region because the population is predicted to approach zero even without climate change. The predicted 2035 abundance in Kaoshan and in the dam 4 and 6 regions initially declines quickly in response to P, but the sensitivity decreases for larger values of P. In these regions, the predicted 2035 abundance is near zero only at unexpectedly large values of P. The effect of climate change in our likely scenario (P = 1.4) reduces the 2035 abundance by 233, 147, and 130 in Kaoshan Stream, the dam 4 region, and the dam 6 region, respectively. However, lower levels of climate change would reduce the 2035 abundance significantly due to the fast initial response to P.

(3) To determine the impact of each dam removal on salmon abundance, including the combined effect with climate change

            Each region is predicted to benefit in some way from dam removal, although these benefits are reduced by climate change. All regions are predicted to experience growth in the 2035 abundance as a result of dam removal. The populations in Kaoshan Stream and the dam 3 region were predicted to approach zero if the dams had not been removed, but are predicted to stabilize at positive values after dam removal if the climate change level is sufficiently small. If the dams are not removed, the populations in the dam 4 and 6 regions are predicted to decline (but not to approach zero by 2035) if the climate change level is sufficiently small. Dam removal could stabilize the population in the dam 4 region at lower levels of climate change. However, the population in the dam 6 region did not stabilize in any of the scenarios under consideration.

            The multiple linear regression equations suggest that dam removal will augment the negative impact of climate change, with the most significant impact on the dam 3 region. In other words, an equivalent level of climate change results in a larger decline in the 2035 abundance in the scenario of effective dam removal compared to the scenario of less effective dam removal for all three regions. This is due in part to having larger 2035 abundance in the effective dam removal scenario, which allows more opportunity for decline.

            The model predicts that removing dam 3 would have the largest potential growth in the 2035 abundance in comparison with the scenario of no dam removal. However, the dam 3 region performs better than the other two regions only for relatively large values of S combined with small values of P. Although the dam 3 region experiences the largest growth at large values of S, it also has the largest negative response to P when S is large.The model also predicts that removing dam 4 results in larger population growth than removing dam 6 at virtually all combinations of S and P.

            The model predicts that without the climate change effect, the dam 3 region will suffer the largest population decline if its dam is not removed, being the only region with the population declining to near zero over the next 20 years. Consequently, this region seems to have the greatest need for conservation. However, interpretation of the results requires consideration of the variety of combinations of the effectiveness of dam removal in increasing the number of recruits and the possible effects of climate change. If the climate change factor is sufficiently large, the dam 4 and 6 regions are also predicted to decline to fairly close to zero within 20 years if the dams are not removed (< 15 individuals at P = 3.4). The model predicts that the dam 3 region has the largest potential growth in the 2035 abundance as a result of dam removal, which is reasonable considering that removing dam 3 would affect a larger region than removing either dam 4 or 6. However, the conditions for reaching near its full potential are more limited for the dam 3 region, requiring both a highly effective response to dam removal and a small effect from climate change. If dam removal results in only a small increase in the number of recruits and/or the climate change level is sufficiently large, the dam 4 and 6 regions will reach closer to their full potential, while the benefit in the dam 3 region would be very small. Management decisions should consider the range of possible scenarios for dam removal and climate change, including the best and worst case scenarios, along with the likelihood of these scenarios.

            This study provides useful tools for research in related fields that consider impacts from severe weather, climate change, and/or dam removal. Our typhoon score proved to be an effective measure for the impact of typhoons on salmon survival rates and number of recruits. Variations could be made to define similar measures in habitats where different species are affected by severe weather events. In addition, our model demonstrates a technique for incorporating effects of climate change into the typhoon score. This study also develops a technique for making predictions for the impact of dam removal in streams where dams have not yet been removed, based on observations from a similar habitat where pre- and post-removal data is available.



IPCC (2014) Summary for policymakers. In: Climate change 2014: Impacts, adaptation, and vulnerability. Part A: Global and sectoral Aspects. Contribution of working group II to the fifth assessment report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp 1-32

Good pieces need to be seen.


Laurie Battle 仈羅利

Laurie Battle 仈羅利

Laurie Battle is a professor of mathematics at Montana Tech specializing in mathematical modeling. She is a Fulbright Senior Scholar at National Chung Hsin University in Taichung, where she researches endangered Formosan landlocked salmon and teaches graduate courses on different modeling techniques.

Explore more

Research & Reflections

fulbright taiwan online journal