Analysis Of Dryland Carbon Dynamics Under Climate Change

Terrestrial carbon cycle has been the active research subject in the areas of climate change,sustainable development,etc.Drylands cover 41%of the Earth's land surface and support more than 38%of the global population.Grasslands,shrublands,and savannas are the primary ecosystems in dryland regions.The dryland ecosystems are fragile and very sensitive to frequent extreme climate events such as drought,floods,and heatwaves.The desertification in dryland areas due to climate change and human activities has caused serious negative effects on social and economic development,especially for developing countries.Although dryland ecosystems are less productive than humid ecosystems,they play a critical role in regulating the trend and interannual variability of the global terrestrial carbon cycle.Previous studies have illustrated that under the influence of climate change,global drylands area is projected to expand substantially.However,it remains unclear how dryland expansion will influence future variability in dryland productivity.In this dissertation,first,in order to provide a continuous carbon flux data with good quality for long-term study,several machine learning algorithms are evaluated for filling the carbon flux gaps with different lengths,and a new framework is proposed to fill the long-length gaps in carbon fluxes;Second,the change of carbon cycle and its impact mechanism at a typical grassland site in semi-arid area of the Loess Plateau(SACOL)are examined;Third,with the carbon flux data obtained from the global flux network(FLUXNET),the characteristics of carbon fluxes of various dryland vegetation types and their responses to environmental variables are assessed;Finally,how the projected dryland expansion and degradation lead to shifts in regional and subtype contributions to global drylands GPP variability in response to future climate change are quantified.The main conclusions as follows:(1)Using the data obtained over a semi-arid sagebrush ecosystem,four machine learning(ML)algorithms including artificial neural network(ANN),k-nearest neighbors(KNN),random forest(RF),and support vector machine(SVM),are employed and evaluated for gap-filling CO₂ fluxes with different gap lengths.The performance decreases significantly with increasing of gap lengths.ANN and RF algorithm slightly outperform the other algorithms in filling gaps ranging from hours to days,while RF is more time efficient than ANN.We also find that gap-filling uncertainty is much larger than measurement error for the nighttime half-hourly data.To overcome these difficulties,instead of filling all gaps altogether,we propose a two-layer gap-filling framework based on RF.With this framework,the model performance improves significantly,especially for the nighttime data.The accumulated NEE annual uncertainty is about 15 g Cm^-2year^-1 on average for the proposed approach,which is smaller than the interannual variability during these four years.(2)Using the CO₂ flux and meteorological data obtained from the Semi-Arid Climate and Environment Observatory of Lanzhou University(SACOL)during 2007-2011,the inter-annual variations in carbon fluxes and the environmental controlling parameters are analyzed.The maximum net ecosystem exchange(NEE)of CO₂ fluxes at SACOL(-10.3(?)mol m^-2s^-1)is lower than humid region grasslands(-27(?)mol m^-2s^-1).The diurnal variation of NEE has obvious seasonal differences.The cumulative gross primary production(GPP)has the largest value in summer,followed by autumn,whereas the largest cumulative NEE is in autumn,and followed by summer.From 2007to 2010,the annual accumulation of NEE at SACOL station was less than zero(i.e.,carbon sink),while the precipitation in 2011 was very small and the overall performance was a carbon source.There are obvious seasonal differences in water use efficiency and carbon use efficiency at this site,which are significantly higher in the growing season than in the non-growing season.Water limits the photosynthesis capacity of vegetation,but the maximum photosynthesis capacity is generally 40 days later than the maximum precipitation,and vegetation has a lagged response to precipitation.Moreover,the precipitation of the previous year changes the moisture of the soil,which affects the productivity of the vegetation in the following year.(3)The characteristic of carbon fluxes and its respond to environmental parameters over various dryland vegetation types have been analyzed using the FLUXNET data sets.The ability of photosynthesis and respiration in Croplands is higher than that of other vegetation types,and the interannual variability is also greater than others.Shrubland has the lowest carbon absorption capacity.The respiration of different vegetation types increases exponentially with the increase of soil temperature.The correlation between respiration and soil temperature is more significant with the increase of soil moisture.The respiration of Cropland varies with temperature was 4-5times higher than other vegetation types.The photosynthesis of various vegetation types has a lagged response to changes in precipitation,generally around 40 days.The more accumulated precipitation within 40 days,the more beneficial it is to the growth of vegetation.The GPP increases significantly with the increase of ET,and the water use efficiency of cropland is the highest,while that of shrublands is relatively low.Analyzing the carbon use efficiency of different vegetation types suggest that the carbon use efficiency of cropland is also the highest,while the lowest is shrublands.(4)Global drylands demonstrated increasing trends and spatiotemporal variations in MODIS GPP from 2000 to 2014.The largest positive contributions to the increased dryland GPP trend occurred in North America,East Asia,whereas the largest contributions to the global drylands IAV occurred in Australia,South America,and Africa.Long-term warming and drying trends would accelerate dryland expansion,which would lead to shifts in regional contributions to the global drylands GPP and IAV and changes in subtype contributions across regions.Global drylands GPP will increase by 12%(25%)by the end of this century under RCP4.5(RCP8.5)relative to the 2000-2014 global drylands GPP baseline.The semiarid subtype continues to dominate global drylands GPP variability.Dryland expansion and climate-induced conversions among sub-humid,semi-arid,arid,and hyper-arid subtypes will lead to substantial changes in regional and subtype contributions to global drylands GPP variability.Our results highlight the vulnerability of dryland subtypes to more frequent and severe climate extremes and suggest that regional variations will require different mitigation strategies.