Continental-scale decrease in net primary productivity in streams due to climate warming

An increase in stream temperature leads to a convergence of metabolic balance, overall decline in net ecosystem productivity, and higher CO2 emissions from streams, according to analyses of temperature sensitivity of stream metabolism across six biomes. Streams play a key role in the global carbon cycle. The balance between carbon intake through photosynthesis and carbon release via respiration influences carbon emissions from streams and depends on temperature. However, the lack of a comprehensive analysis of the temperature sensitivity of the metabolic balance in inland waters across latitudes and local climate conditions hinders an accurate projection of carbon emissions in a warmer future. Here, we use a model of diel dissolved oxygen dynamics, combined with high-frequency measurements of dissolved oxygen, light and temperature, to estimate the temperature sensitivities of gross primary production and ecosystem respiration in streams across six biomes, from the tropics to the arctic tundra. We find that the change in metabolic balance, that is, the ratio of gross primary production to ecosystem respiration, is a function of stream temperature and current metabolic balance. Applying this relationship to the global compilation of stream metabolism data, we find that a 1 °C increase in stream temperature leads to a convergence of metabolic balance and to a 23.6% overall decline in net ecosystem productivity across the streams studied. We suggest that if the relationship holds for similarly sized streams around the globe, the warming-induced shifts in metabolic balance will result in an increase of 0.0194 Pg carbon emitted from such streams every year.

Streams play a key role in the global carbon cycle. The balance between carbon intake 33 through photosynthesis and carbon release via respiration influences carbon emissions from 34 streams and depends on temperature. However, the lack of a comprehensive analysis of 35 the temperature sensitivity of the metabolic balance in inland waters across latitudes and 36 local climate conditions hinders an accurate projection of carbon emissions in a warmer fu-37 ture. Here, we use a model of diel dissolved oxygen dynamics, combined with high-frequency 38 measurements of dissolved oxygen, light, and temperature, to estimate the temperature sen-39 sitivities of gross primary production and ecosystem respiration in streams across six biomes, 40 from the tropics to the arctic tundra. We find that the change in metabolic balance, that 41 is, the ratio of gross primary production to ecosystem respiration, is a function of stream 42 temperature and current metabolic balance. Applying this relationship to the global com-43 pilation of stream metabolism data, we find that a 1 • C increase in stream temperature 44 leads to a convergence of metabolic balance, and to a 23.6% overall decline in net ecosys-45 tem productivity across the streams studied. We suggest that if the relationship holds for 46 similarly-sized streams around the globe, the warming-induced shifts in metabolic balance 47 will result in an increase of 0.0194 Pg carbon emitted from such streams every year. 48 Streams play a significant role in the transport, storage, and transformation of organic 49 carbon globally 1,2 . Recent estimates suggest that 0.8-1.8 petagrams (Pg) of carbon evade 50 from streams and rivers to the atmosphere annually 3,4 . This is comparable in size to the 51 net annual terrestrial-atmosphere and net ocean-atmosphere carbon exchange 5 . Stream 52 metabolism, which is governed by gross primary production (GPP) and ecosystem respiration 53 (ER), contributes substantially to the overall carbon flux out of streams. A recent study 54 estimated that stream metabolism is responsible for up to 28% of the total carbon flux 55 from streams to the atmosphere 6 , resulting in an estimated net flux of 0.12 Pg C per year 7 . 56 As GPP and ER are both temperature dependent processes, sustained climate warming 57 has the potential to profoundly alter the rates of carbon flux in and out of streams. Over 58 the past century, mean water temperature in US rivers and streams increased at a rate of 59 0.009-0.077 • C per year 8 , and stream temperatures are predicted to increase by 1-3 • C 60 with the doubling of atmospheric CO 2 concentration 9 . Consequently, understanding the 61 feedback between stream metabolism and global warming is crucial when considering global 62 or regional carbon cycles. 63 GPP and ER, the varied approaches employed to quantify them also have the potential to 97 influence the inferred ecosystem-level temperature dependence of GPP and ER. Incubations 98 of stream substrata at different temperatures 21,22 or mesocosm warming experiments 23 do 99 not include the entire focal ecosystem and may not encompass the processes key for deter-100 mining the temperature sensitivities of GPP and ER at the ecosystem level. Comparisons 101 among streams or within one stream over seasons 19,24-28 yield ecosystem-level estimates of 102 temperature sensitivities, but temperature independent differences among streams or seasons 103 due to hydrology 29 , geomorphology 22 , nutrient availability 30,31 , and light availability 26 can 104 easily confound the responses of GPP and ER to temperature. These confounding factors 105 render the estimated temperature dependence not purely a response to temperature, but 106 an integrated response to the suite of temperature dependent and independent differences 107 across streams or seasons.

108
Given the complexity of ecosystem-level temperature sensitivities and the challenges as-109 sociated with quantifying them, it is not surprising that various patterns have been reported.  In studies that simultaneously examined the temperature dependence of GPP and ER in 114 streams, a shift toward heterotrophy with warming has been observed in some instances 23,27 , 115 but a recent synthesis based on geothermal streams concluded that warming increases GPP 116 and ER to the same extent and results in no net change in metabolic balance 32 . To date, 117 simultaneous quantification of the temperature dependence of GPP and ER have been con-118 strained to mesocosm incubations or geothermal streams. Thus, there is still uncertainty 119 about whether streams will become more heterotrophic (decreasing GPP/ER and NEP) or 120 more autotrophic (increasing GPP/ER and NEP) at the continental scale in response to 121 continued warming. Simultaneously quantifying the ecosystem-level temperature sensitiv-122 ities of GPP and ER in streams across broad bio-climatic regions is key to resolving such 123 uncertainty.

124
Here, we estimate the temperature sensitivities of GPP and ER in streams from six dis-125 tinct biomes. We utilize the response of DO concentration to diel temperature variation 126 and dynamic models of DO concentration to infer the temperature dependence of GPP and 127 4 spatial or seasonal temperature gradients, we avoid the implicit assumption that differences 129 in stream metabolism along the temperature gradient are mainly attributed to temperature 130 differences, and thus minimize the influence of factors that covary spatially or seasonally with 131 temperature 34 . Moreover, this dynamic modeling approach allows us to estimate tempera-132 ture dependence of GPP and ER for each stream, and thus characterize stream to stream Station, Alaska, USA (ARC)). In each biome, we measured DO concentration, photosyn-147 thetically active radiation, and water temperature at a 5 or 10 minute interval for 1-2 weeks 148 in multiple stream reaches throughout a watershed. We modeled the response of DO con-149 centration to diel temperature variation to estimate ecosystem-level activation energies of 150 GPP and ER. Specifically, we modeled the dynamics of DO concentration as: Here, P max (mg O 2 L −1 min −1 ) is the maximum primary production rate, α (mg O 2 L −1 158 s m −2 µE −1 min −1 ) is the slope of the light response curve of primary production at low of metabolic balance in streams is the ratio of GPP to ER, which, for our formulation of the 187 instantaneous rates of GPP and ER, is: The formulation of GPP/ER has the form of an Arrhenius equation, and thus, E ap − E ar 189 is the apparent activation energy of GPP/ER and determines how instantaneous metabolic 190 balance changes with temperature. A positive E ap −E ar means that GPP/ER will increase as 191 temperature increases and a negative E ap −E ar means GPP/ER will decrease as temperature perature increase, shown as a decrease in the inter-site variability of GPP/ER ( Fig. 4(a)).

241
Second, the convergence in metabolic balance is asymmetric. The magnitude of decrease in 242 GPP/ER in streams with high temperatures and high daily GPP/ER was larger than the 243 magnitude of increase in GPP/ER in streams with low temperatures and low daily GPP/ER.

244
Such asymmetry suggests that warming will influence the metabolic balance of streams with 245 high temperature and daily GPP/ER more substantially, which translates to such streams 246 8 becoming stronger carbon sources (i.e. lower GPP/ER).
We quantified warming-induced changes in NEP, the difference between GPP and ER, 249 based on the simulated warming experiment. We estimated that a 1 • C increase in tem-250 perature will increase GPP from 0.89 to 1.12 (g O 2 m −2 day −1 ), and ER from 3.45 to 4.27  Fig. 4(c)), in response 256 to a 1 • C increase in temperature. Although our prediction of shifting toward more net 257 heterotrophy in response to warming is consistent with predictions based on metabolic the-258 ory 10 , it differs importantly in that it results from the asymmetric convergence of metabolic 259 balance, not a universal shift towards heterotrophy for all streams (Fig. 4(d)).

260
The predictions for how GPP/ER and NEP will change with warming do not come

316
We used posterior means of the parameters for further statistical analyses. 317 We made two special considerations when estimating parameters. First, low diel vari-318 ability in temperature in some streams prevented us from estimating E ap and E ar with 319 confidence. Thus, we only used E ap and E ar estimates with 95% highest posterior density 320 intervals narrower than 500 KJ mol −1 for further statistical analyses. This is to ensure that 321 the estimated E ap and E ar are mainly determined by the data, not by the uniform priors.
where GP P/ER current and GP P/ER warming are daily GPP/ER currently and under the    Pg C year −1 .