Regional Turbulent Water And Heat Fluxes From Airborne Eddy Covariance Measurements

The direct measurements of water and heat flux exchanges are the key to obtain accurate information on the exchange of matter and energy between the land-surface and the atmosphere.At present,the observations of turbulent water and heat flux exchanges between the land surface and atmosphere at a regional scale are the focuses and difficulties for the researches of land surface process.The ground-based eddy-covariance(EC)method provides reliable long-term continuous observations for these turbulent fluxes,but usually with a small spatial representative(i.e.footprint,typically several hundred meters of space around the tower).The scintillometer-based method could measures the turbulent heat fluxes at a large spatial range,but its observation density is low,and it is still difficult to provide a sufficient regional flux observation.The aircraft-based EC method could measures the turbulent flux at large spatial scale directly.In recent years,the airborne EC method has get more and more attention,and the reliability of its measurements has been widely recognized as well.However,relevant research of airborne EC measurements is few in China.This study focused on the technology and application methods of airborne eddy covariance turbulent flux measurements.Firstly,this thesis described the system composition,the basic principle of airborne EC technology and the methods of system calibration and data processing in details based on the sophisticated airborne flux measurement platform Sky Arrow 650 environment research aircraft(ERA).Airborne flux measurements were collected during 2008 in the Netherlands.Then,some deficiencies in the current airborne EC method are analyzed,and the solutions or methods to solve the deficiencies or improve the airborne EC methods are proposed.The key researches and conclusions in this thesis are as follows:The airborne turbulent flux measurements obtain the instantaneous turbulent flux information along the flight path.The choice of the window length for calculating the turbulent flux is one of the crucial factors determining the accuracy and the physical meaning of airborne EC measurements.Firstly,this study gave the optimum spatial averaging length(i.e.window length)for calculating accurate turbulent heat fluxes from airborne eddy covariance measurements under near neutral to unstable atmospheric stability conditions,to reduce the negative influences from mesoscale turbulence and to estimate local meaningful turbulent heat fluxes accurately.The turbulent fluxes based on the optimum window length could better represent the average of the turbulent fluxes of the underlying surface.Based on the collected extensive airborne EC measurement data,this thesis used the Ogive analysis method to find out the possible window length range under near neutral to unstable atmospheric stability conditions,and then used the possible window lengths to calculate the heat fluxes and their uncertainty,and based on which defined the final optimum window length.The results show that the choice of the optimum window lengths strongly depends on atmospheric stability conditions.Under near neutral conditions,local turbulence is mixed insufficiently and vulnerable to heterogeneous turbulence.A relatively short window length is needed to exclude the influence of mesoscale turbulence,and we found the optimum window length ranges from 2000 m to 2500 m.Under moderately unstable conditions,the typical scale of local turbulence is relative large,and the influence of mesoscale turbulence is relatively small.We found the optimum window length ranges from 3900 m to 5000 m.Under very unstable conditions,large convective eddies dominate the transmission of energy so that the window length needs to cover the large eddies with large energy transmission.We found the optimum window length ranges from 4500 m to 5000 m.In order to obtain an exact information on the representative underlying surface of the measured turbulent flux along the flight path,secondly,this study presented a simple method for footprint evaluation over a heterogeneous surface based on the characteristics of airborne EC measurements.The traditional footprint analysis of airborne flux measurements is based on the homogeneous surface assumption which ignore the spatial heterogeneity of surface flux contribution.This study used the latest two-dimensional parameterization Lagrangian footprint model.Firstly,we divided the flux calculated window into several sub-windows and calculated the sub-footprint of each sub-window.Through the sub-footprint manner,we could introduced the influence of heterogeneity into the footprint analysis.Then,we merged and normalized the sub-footprints within the flux calculated window to output the final footprint.In addition,this study also gave the traditional footprint analysis method based on the homogeneous surface assumption.Results showed that the footprint considering the surface heterogeneity could identify the spatial variability of surface flux contribution induced by surface heterogeneity,However,the footprint based on the homogeneous surface assumption tends to ignore some flux contributions from special land types.For extending the observed flux information along the flight path to the entire study region,thirdly,this study gave a comprehensive methodology to upscale the turbulent heat fluxes measured by aircraft to the regional scale continuously covered land surface heat fluxes using the artificial neural network(ANN)model and footprint analysis based on homogeneous assuming.The ANN model was been used to learn the non-linear relationships between the observed turbulent heat fluxes and the land surface as well as atmospheric driving factors.Firstly,the airborne EC measurements in August 2008 were used as data foundation.Land surface temperature(LST)and enhanced vegetation index(EVI)products from MODIS were used as the main land surface parameters,and other surface parameters(e.g.albedo,DEM,etc.)that affected the turbulence flux were also taken into account.Those land surface parameters combining with various atmospheric parameters were used as potential input variables to select the optimal input variables through mutual information method.Then,the back-propagation neural network(BP-NN)model was used to infer a function relationship between the turbulent heat fluxes and the selected optimal input variables.After fully validation the reliability of the ANN models(sensible heat flux:R~2=0.8,RMSE=16.6 W/m~2;latent heat flux:R~2=0.75,RMSE=42.7 W/m~2),this study selected two case regions in the central and western of the Netherlands to construct the regional heat flux maps based on the trained ANN model.The spatial resolution of the land surface heat fluxes map is 250 m for the two selected regions,and the up-scaled results were compared with both the ground-based and airborne measurements.Results showed that the aggregated heat fluxes intensity from the heat fluxes maps were consistent with the turbulent heat fluxes observed by ground-based EC measurements.Compared with the airborne measurements,for the central region,the sensible heat flux is slightly underestimated by 12.3%,and the latent heat flux is slightly underestimated by 11.5%.For the western region,the sensible heat flux is underestimated by 15.2%,and the latent heat flux is overestimated by 7.6%.On the whole,the errors are within a reasonable range,which verifies the feasibility of the proposed method.In order to fully consider the affect of surface heterogeneity on the spatial expansion of airborne flux observations,fourthly,this study presented an upscaling method that using the footprint analysis method of considering surface heterogeneity and deep learning model to construct the regional heat fluxes maps.Because of the more complex non-linear relationships after considering the effect of heterogeneity,the traditional ANN model cannot model this complicated relationship well.Therefore,this study tried to use the popular deep learning method to model the complex land surface process which considers the influence of surface heterogeneity.In this study,a deep belief network(DBN)was used to pre-train the ANN model through a unsupervised way,and then used the BP-NN to fine-tune the model parameters through a supervised way.DBN-ANN models with better generalization performance were obtained(sensible heat flux:R~2=0.84,RMSE=15.1 W/m~2;latent heat flux:R~2=0.83,RMSE=34.3 W/m~2).The land surface heat fluxes maps of the two selected case regions with spatial resolution of 250 m were constructed based on the trained DBN-ANN models.Then the up-scaled results were compared with the ground-based measurements,airborne measurements as well as the previous up-scaled results.Results showed that the heat fluxes intensity of the heat fluxes maps were consistent with the turbulent heat fluxes observed by ground-based EC measurements as well.Thanks to the powerful non-linear learning ability of deep learning method,after considering the heterogeneity,the accuracy of the up-scaled regional land surface heat fluxes maps have been improved.Compared with the airborne measurements,for the central region,the sensible heat flux is underestimated by 6.8%,and the latent heat flux is underestimated by 14.6%.For the western region,the sensible heat flux is underestimated by 8.1%,and the latent heat flux is overestimated by 1.6%.It can be considered that although considering the influence of surface heterogeneity increases the complexity of the land surface process simulation,the deep learning method could learn this complexity in a certain extent and obtain more an accuracy result.