Effects of abiotic factors on ecosystem health of Taihu Lake, China based on eco-exergy theory

A lake ecosystem is continuously exposed to environmental stressors with non-linear interrelationships between abiotic factors and aquatic organisms. Ecosystem health depicts the capacity of system to respond to external perturbations and still maintain structure and function. In this study, we explored the effects of abiotic factors on ecosystem health of Taihu Lake in 2013, China from a system-level perspective. Spatiotemporal heterogeneities of eco-exergy and specific eco-exergy served as thermodynamic indicators to represent ecosystem health in the lake. The results showed the plankton community appeared more energetic in May, and relatively healthy in Gonghu Bay with both higher eco-exergy and specific eco-exergy; a eutrophic state was likely discovered in Zhushan Bay with higher eco-exergy but lower specific eco-exergy. Gradient Boosting Machine (GBM) approach was used to explain the non-linear relationships between two indicators and abiotic factors. This analysis revealed water temperature, inorganic nutrients, and total suspended solids greatly contributed to the two indicators that increased. However, pH rise driven by inorganic carbon played an important role in undermining ecosystem health, particularly when pH was higher than 8.2. This implies that climate change with rising CO2 concentrations has the potential to aggravate eutrophication in Taihu Lake where high nutrient loads are maintained.


Water smapling and sample pretreamnt
We carried on field observations at 33 different sites across the Taihu Lake at monthly intervals during the year of 2013. At each location, 4L of water sample was collected at a depth of 0.5m underneath water surface using organic glass collector. It was used to determine water quality variables, including NH 4 -N, TN, NO 2 -N, NO 3 -N, TP, DTP, PO 4 -P and Chl-a.
For observations of phytoplankton, 1L of water sample was collected at a depth of 0.5m using glass collector. 15mL of Lugol's solution served as fixative solution was added into the sample for preservation in case sample deterioration.
For observations of zooplankton, 1L of water sample was collected at a depth of 0.5m using glass collector. For quantifying microzooplantkon, water sample was input into 1L plastic flask with the addition of 10mL of Lugol's solution served as fixative solution. For quantifying macrozooplanton such as cladocerans and copepods, the sample was treated by 5% (v/v) formaldehyde solution.
Sampling, transportation and preservation of water samples were referred to 1, 2 .
All samples were kept in vehicle mounted refrigerator that was set to constant temperature of 4 o C. Then, they are transported to laboratory within 4h.

Laboratory analysis of water samples
At each water sampling site, we used DO meter, pH microelectrode and secchi disk to measure DO concentration, water temperature, pH and water transparence of the lake. We also used GPS to obtain the latitude and longitude of each location, and the anemoscope was used to determine wind speed and wind direction on the spot. All other variables were determined in laboratory.

Water quality indicators
The water sample was deposited for 30min, then supernatant extracted using Siphon method was used for TN and TP determinations. After this, the sample was filtered by acetic acid -nitric acid synthetic fabric membrane (47mm， 0.45μm) and the filtrate was used for NH 4 -N, NO 2 -N, NO 3 -N, DTP, PO 4 -P. For Chl-a determination, the water sample was filtered by glass fiber membrane (47mm, 0.70μm), then filter membrane was put into 90% (v/v) acetone solution, and using solvent extraction it was kept for 4-12h. For TSS determination, the sample was filtered by oven-dried filter membrane (47mm, 0.45μm), then we dried the membrane again and weight it to calculate TSS mass. The detailed process, standard number and primary instrument during laboratory analysis were listed in Table S1.

Phytoplankton biomass
Microscopic counting method was used for phytoplankton determination. We took the well-shook water sample (≥100 ml) after treated by Lugol's solution, then carried on vacuum filtration using suction filter equipped with cellulose acetate membrane (45mm,1.2μm). The membrane with algae cells was put into beaker and then we added 5-8mL of purified water. After mixed, extracted solution was preserved.
Repeat the process for 3-5 times and the final volume of lotion was fixed to 30mL.
The solution was deposited and shook up, then we extracted 0.1mL and injected it into counting box (20×20mm 2 ). When covering the cover slip, it was warranted that there was no bubble inside the counting box, and no spill. The rapid detection count method was used by microscopic examination 10 × 40, 400X. The phytoplankton biomass expressed in mg/L was obtained by unit conversion in relation to number of phytoplankton cells per liter.

Zooplankton biomass
Microscopic counting method was used for zooplankton determination. For microzooplantkon determination, 1000mL of water sample was standing for 24h, and siphon pipe covered #25 plankton net was used to take up supernatant fraction. The remaining 10-30mL deposit was transferred into 50mL volumetric flask and then supernatant fraction was removed by siphon pipe and leave 10mL of deposit solution to be determined. concentrated into 10mL.
Generally, we collected macrozooplanton species, such as cladocerans and copepods, using #13 plankton net, and put it into 50mL volumetric flask. After standing for 24h, supernatant fraction was removed by siphon pipe and leave 10mL of deposit solution to be determined. When counting number, the deposit solution should be shook up, then using micro pipette we extracted 0.1mL and injected it into 0.1mL of counting box (20mm×20mm). The rapid detection count method was used by microscopic examination 10×40, 400X for counting protozoan, rotifer and nauplius.
For counting cladoceran and copepod, using micro pipette we extracted 1.0mL and injected it into 1.0mL of counting box (40mm×60mm). The microscopic examination 10×10, 100X was used. The zooplankton biomass expressed in mg/L was obtained by unit conversion in relation to number of individual zooplankton per liter.

Wet weight to C-biomass conversion
Approximately, 0.16 (the average of 0.22, 0.16, 0.11 and 0.16) was used to convert phytoplankton wet biomass to dry weight biomass in carbon unit 13 . For zooplankton species, we obtained that the ratio of dry weight to wet weight of biomass was approximately 0.19 14 , and carbon biomass roughly 32% of zooplankton dry weight 15 . Therefore, the unit conversion was 0.06 for converting wet weight to carbon biomass 16 . GBM is a decision-tree based approach, and thus highly non-linear without having to transform the independent variables. We used the package "gbm" in R-language, and then outputting a file of predictions. The model was a forest of tiny decision trees that come together to form a solution. The significant advantage of GBM was to provide a convenient and computationally efficient way to explore highly non-linear interactions between independent variables and the dependent variable. We implemented the linear fit of calculated values of ecological indicator of interest versus simulated results derived from GBM to evaluate model performance.

3.
We could also determine what independent variables were having the most impact on the solution space, from looking at the "influence" of each independent variable. The