CN104077475B - A kind of global land table based on many algorithm integrations evapotranspires estimating system and method - Google Patents

Abstract

The invention belongs to meteorological model field of measuring technique, it is related to a kind of global land table based on many algorithm integrations to evapotranspire estimating system and method.The method analyzes meteorological data and drives MODIS algoritic module, the PM remote sensing corrected evapotranspire algoritic module, the improved Priestly Taylor of algoritic module, Priestly Taylor that evapotranspire that evapotranspire to evapotranspire algoritic module, the algoritic module that evapotranspires based on empirical method again first with MODIS remotely-sensed datas and MERRA, and respective evapotranspiration is drawn respectively；Then ground is extracted from global flux website observation data by observation data extraction module according to website measured data sample quality to evapotranspire observation；Finally by Bayesian model method, the ground of the extraction evapotranspiration that observation and each algoritic module that evapotranspires calculate that evapotranspires is carried out integrated, form Bayesian model integration module.Compared with prior art, global land table of the invention evapotranspires estimating system and method has stronger stability, and applicable surface is broader, has broad application prospects.

Description

A kind of global land table based on many algorithm integrations evapotranspires estimating system and method

Technical field

The invention belongs to meteorological model field of measuring technique, it particularly relates to arrive a kind of based on the complete of many algorithm integrations Ball land table evapotranspires estimating system and method.

Background technology

Currently, global water resources increasingly plaque is weary, in order to rationally utilize and distribute water resource, in the urgent need to understanding difference in depth Evaporation water consumption situation under the conditions of vegetative coverage and Land_use change.Evapotranspiration dams evaporation capacity, transpiration amount, soil including vegetation Earth evaporation capacity and water surface evaporation, are the main components of regional water balance and energy balance, are not only followed in water circulation and energy There is extremely important effect during ring, and be the important tie of ecological process and hydrologic process.Land table evapotranspire be hydrosphere, Atmospheric thermodynamics and biosphere moisture and the main process parameter of energy exchange, are Terrestrial energy circulation, water circulation and carbon cycle In key variables.The surface evapotranspiration quantity data of specific region accurately and timely are obtained, in agricultural, the hydrology, forest and ecology etc. Field can play highly important effect.Therefore, a kind of measurement system and method are set up, realization is accurately and rapidly obtained Big region land table evapotranspires and estimates significant.

However, traditional assay method is not representative to Remote sensing situation interior on a large scale strong, and lay measuring point cost It is high, it is difficult to form practical observation grid.Chinese patent (CN 200810179250.0) discloses a kind of area based on remote sensing Evapotranspire quantity monitoring method in domain.The method is built using the visible ray/IRMSS thermal band and daily routine meteorological data of multidate The data processing chain of land surface evaporation monitoring on regional scale is found, it is achieved thereby that the maneuverability remote sensing prison of regional evapotranspiration Survey.Relative to the monitoring means evapotranspired is measured on traditional flux observation station, the multi-source multidimensional that remote sensing technology can provide earth's surface is more When phase information, with large area, macroscopic view, in real time and dynamic etc. advantage, be land table evapotranspire estimation open new approach.Remote sensing is estimated The method that calculation land table evapotranspires is a lot.Mainly there are the physical model based on Land surface energy budget, empirical statistics algorithm, Penman- Monteith algorithms, remote sensing deltic method, data assimilation method etc..These model methods are by Mo Ning-Ao Bu Hough phases Like theoretical key parameter (surface temperature, vegetation index etc.) and the determining that land table evapotranspires for setting up earth's surface meteorological element, remote-sensing inversion Amount contact is evapotranspired obtaining land table, but the evapotranspiration of the estimation between algorithms of different is widely different, in addition gap reach 2 times with On, overall accuracy is not high.And, there is very big uncertainty in the single algorithm that evapotranspires, minority is put down using simple many algorithms Mean Method, simple average value method it is uncertain larger, had a strong impact on the computational accuracy that land table evapotranspires.

The content of the invention

In order to solve the above-mentioned technical problem, evapotranspired estimation the invention provides a kind of global land table based on many algorithm integrations System and method.

A kind of global land table based on many algorithm integrations of the present invention estimating system that evapotranspires includes：MODIS evapotranspires calculation Method module, the PM remote sensing corrected is evapotranspired algoritic module, and Priestly-Taylor evapotranspires algoritic module, improved Priestly- Taylor evapotranspires algoritic module, the algoritic module that evapotranspires based on empirical method, observes data extraction module, and Bayesian model is integrated Module；Analyzing meteorological data again first with MODIS remotely-sensed datas and MERRA drives MODIS to evapotranspire algoritic module, the PM that corrects Remote sensing evapotranspire algoritic module, the improved Priestly-Taylor of algoritic module, Priestly-Taylor that evapotranspire evapotranspires algorithm mould Block, the algoritic module that evapotranspires based on empirical method, draw respective evapotranspiration respectively；Then according to website measured data sample matter Amount is extracted ground and is evapotranspired observation by observation data extraction module from global flux website observation data；Finally by Bayes's mould Type method, the evapotranspiration that observation is evapotranspired on the ground of extraction and each algoritic module that evapotranspires is calculated is carried out integrated, forms pattra leaves This model integrated module.

A kind of global land table based on many algorithm integrations of the present invention evapotranspires evaluation method, the specific step of methods described Suddenly include：

1) MODIS is built to evapotranspire algoritic module：Canopy is divided into wet canopy and dry canopy, soil is divided into water saturation soil Earth and moisture unsaturation soil, evapotranspire algorithm according to MODIS, calculate and the impedance of design earth's surface, aerodynamic resistance and side Interlayer parameter；

2) the PM remote sensing corrected is built to evapotranspire algoritic module：Evapotranspired algorithm according to traditional Penman-Monteith, to temperature The defect for spending restriction factor linear change is corrected；

3) Priestly-Taylor is built to evapotranspire algoritic module：According to existing Priestly-Taylor models, utilize Ecosystem stress factors extend Priestly-Taylor coefficients, obtain Priestly-Taylor and evapotranspire algorithm；

4) improved Priestly-Taylor is built to evapotranspire algoritic module：Soil moisture is parameterized using apparent thermal inertia Factor of evaporation amendment soil moisture restriction factor, obtains improved Priestly-Taylor and evapotranspires algorithm；

5) algoritic module that evapotranspires based on empirical method is built：Rule of thumb Evapotranspiration by Using algorithm, design is evapotranspired input, defeated Go out and computing module；

6) carried from global flux website observation data by observation data extraction module according to website measured data sample quality Ground is taken to evapotranspire observation；

7) analyze meteorological data actuation step 1 again using MODIS remotely-sensed datas and MERRA) MODIS evapotranspire algorithm mould Block, step 2) the PM remote sensing corrected evapotranspires algoritic module, step 3) Priestly-Taylor evapotranspires algoritic module, step 4) change The Priestly-Taylor for entering evapotranspires algoritic module, step 5) algoritic module that evapotranspires based on empirical method, draw respectively each Evapotranspiration；

8) by Bayesian model method, by step 6) ground extracted evapotranspires observation and step 7) each calculation of evapotranspiring The evapotranspiration that method module is calculated carries out integrated, formation Bayesian model integration module.

The step 1) in the MODIS algorithms that evapotranspire be：

LE=LE_{wet_c}+LE_trans+LE_soil

In formula, LE evapotranspires for land table, LE_{wet_c}It is the evaporation that wet canopy evaporation process is produced, LE_transFor Vegetation canopy steams Rise the transpiration of process generation, LE_{wet_soil}It is the evaporation that saturated soil moisture evaporation is produced, LE_{soil_pot}For unsaturated soil is potential Evaporate the evaporation for producing, LE_soilIt is soil evaporation, Δ is the slope of saturation vapour pressure and temperature curve, and γ is wet and dry bulb constant, VPD is the saturation vapour pressure and actual vapor pressure difference of air, and RH is relative humidity, r_aIt is aerodynamic resistance, r_sFor earth's surface hinders It is anti-, F_cIt is vegetation coverage, F_wetIt is humidity area coverage, ρ is atmospheric density, C_pIt is the specific heat capacity of air, R_ncIt is Vegetation canopy Shared net radiation, rhrc is canopy corresponding part aerodynamic resistance, and rvc is Ground Heat transfer impedance, P_aFor big Air pressure, R_nsNet radiation shared by surface soil, r_totIt is whole aerodynamic resistance, r_asIt is partial air dynamics Impedance, β, λ and ε are coefficient；

The step 2) in temperature limiting factor correction method be：

In formula, m_TIt is the temperature limiting factor, T_aIt is air themperature, T_optIt is vegetation growth optimal temperature, is set to 25 DEG C；

The step 3) in the Priestly-Taylor algorithms that evapotranspire be：

In formula, f_wetIt is relative surface humidity, RH is relative humidity, f_wet=RH⁴, f_gIt is green Canopy cover degrees, f_TTo plant By the temperature limiting factor, f_MIt is vegetation moisture restriction factor, f_SMIt is soil moisture restriction factor, G is soil heat flux.

The step 4) in soil moisture restriction factor be：

In formula, DT is air day and night temperature, DT_maxIt is maximum day and night temperature, takes 40 DEG C；

The step 5) in be based on the algorithm that evapotranspires of empirical method：

LE=R_n(a₀+a₁NDVI+a₂T_a+a₃DTaR)

In formula, DTaR is day day and night temperature, and NDVI is normalized differential vegetation index, a₀, a₁, a₂And a₃It is empirical coefficient；

The step 8) in Bayesian model integration module computing formula be：

In formula, r is represented needs evapotranspiring for estimation, and the observation evapotranspired at a time is r_t, model set K, i.e., {f₁,f₂,…,f_k, be estimation r all single model set, p (r | f_k) it is single model f_kEstimate that the probability of the r that evapotranspires is close Degree function, p (f_k|r_t) it is single model f_kPosterior probability, single model f can be reacted_kWith the degree of observation data fit.

The step 8) in, the probability density function p of the estimated value that Bayesian model integration module is obtained (r | f₁,f₂,…, f_k) be exactly single model probability density function weighted average, the posterior probability sum of all single models is 1, i.e.,So, the posterior probability of each model is regarded as weight w_k；

Assuming that p (r | f_k) it is Gaussian Profile,It is the average value of the Gaussian Profile, σ² _KIt is variance, uses parameterThe Gaussian Profile related to g () expressions be：

p(r|f_k)=g (r | θ_k)

The estimate for obtaining is the conditional expectation of r, cost function is optimized and is obtained weight with expectation-maximum algorithm w_k。

The step 1) by vegetation coverage F_cTo distinguish the global land table available energy of vegetation and soil distribution, profit Determine F with absorbed photosynthetically active radiation ratio FPAR products are planted in MODIS remotely-sensed datas_c；

The step 2) the PM remote sensing corrected evapotranspires in algorithm, and the maximum of unified moisture restriction factor VPD is set to 2.79KPa；

The step 3) R that evapotranspire in algorithm of Priestly-Taylor_nsUse surface net radiation R_nWith leaf area index LAI Calculate, specially R_ns=R_nExp (- kLAI), k are empirical coefficient；

The step 4) improved Priestly-Taylor evapotranspire algorithm soil moisture restriction factor in, DT is to pass through Day day and night temperature is calculated or is calculated by surface temperature；

The step 5) in empirical coefficient a₀, a₁, a₂, and a₃It is to observe data as training number by the use of global flux website Obtained according to recurrence；

The step 6) observation data extraction module in, website measured data sample quality be by quality identification judge Obtained from；

The step 7) leaf area index, vegetation index are chosen from MODIS remotely-sensed datas, meteorology is analyzed again from MERRA Basic meteorologic parameter, including temperature, relative humidity are chosen in data, drives each algoritic module to calculate respective evapotranspiring Amount.

The global land table includes forest, shrub, farmland, meadow and wetland.

Compared with the conventional method, a kind of global land table integrated based on multi-source data and many algorithms of the present invention steams Dissipating the beneficial effect of estimating system and method is：By theory analysis, experiment test and practical application, ground is made full use of to survey Data, remotely-sensed data and many algorithms are combined, and think that Bayesian model is relied on, existing clear and definite physical basis, and with stabilization The advantages of property is good, simple, flexible, easily operated, the algorithm that MODIS therein can be evapotranspired, the PM remote sensing corrected evapotranspire algorithm, Priestly-Taylor evapotranspire algorithm, improved Priestly-Taylor evapotranspire algorithm and empirical method the algorithm that evapotranspires it is comprehensive Intersection into, there is stronger stability than the single algorithm that evapotranspires, applicable surface is broader, has broad application prospects.

Brief description of the drawings

Fig. 1 is that the global land table based on many algorithm integrations that the present invention is used evapotranspires the structural representation of estimating system；

Fig. 2 is that global land table of the present invention based on many algorithm integrations evapotranspires the flow chart of evaluation method；

Fig. 3 is global flux website measured value checking.

Specific embodiment

A kind of global land table based on many algorithm integrations of the present invention is evapotranspired with reference to specific embodiment is estimated System and method is calculated to be described further, but protection scope of the present invention is not limited to this.

Embodiment 1

Global land table based on many algorithm integrations estimating system that evapotranspires of present invention design includes：MODIS evapotranspires algorithm mould Block, the PM remote sensing corrected is evapotranspired algoritic module, and Priestly-Taylor evapotranspires algoritic module, improved Priestly-Taylor Evapotranspire algoritic module, the algoritic module that evapotranspires based on empirical method, observes data extraction module, Bayesian model integration module； Analyzing meteorological data again first with MODIS remotely-sensed datas and MERRA drives MODIS to evapotranspire algoritic module, the PM remote sensing corrected Evapotranspire algoritic module, the improved Priestly-Taylor of algoritic module, Priestly-Taylor that evapotranspire evapotranspires algoritic module, base In the algoritic module that evapotranspires of empirical method, respective evapotranspiration is drawn respectively；Then according to website measured data sample quality by Observation data extraction module is extracted ground and is evapotranspired observation from global flux website observation data；Finally by Bayesian model side Method, the evapotranspiration that observation is evapotranspired on the ground of extraction and each algoritic module that evapotranspires is calculated is carried out integrated, forms Bayes's mould Type integration module.

The evapotranspire specific steps of evaluation method of the corresponding global land table based on many algorithm integrations include：

1) MODIS is built to evapotranspire algoritic module：Canopy is divided into wet canopy and dry canopy, soil is divided into water saturation soil Earth and moisture unsaturation soil, evapotranspire algorithm according to MODIS, calculate and the impedance of design earth's surface, aerodynamic resistance and side Interlayer parameter；

2) the PM remote sensing corrected is built to evapotranspire algoritic module：Evapotranspired algorithm according to traditional Penman-Monteith, to temperature The defect for spending restriction factor linear change is corrected；

3) Priestly-Taylor is built to evapotranspire algoritic module：According to existing Priestly-Taylor models, utilize Ecosystem stress factors extend Priestly-Taylor coefficients, obtain Priestly-Taylor and evapotranspire algorithm；

4) improved Priestly-Taylor is built to evapotranspire algoritic module：Soil moisture is parameterized using apparent thermal inertia Factor of evaporation amendment soil moisture restriction factor, obtains improved Priestly-Taylor and evapotranspires algorithm；

5) algoritic module that evapotranspires based on empirical method is built：Rule of thumb Evapotranspiration by Using algorithm, design is evapotranspired input, defeated Go out and computing module；

6) carried from global flux website observation data by observation data extraction module according to website measured data sample quality Ground is taken to evapotranspire observation；

7) analyze meteorological data actuation step 1 again using MODIS remotely-sensed datas and MERRA) MODIS evapotranspire algorithm mould Block, step 2) the PM remote sensing corrected evapotranspires algoritic module, step 3) Priestly-Taylor evapotranspires algoritic module, step 4) change The Priestly-Taylor for entering evapotranspires algoritic module, step 5) algoritic module that evapotranspires based on empirical method, draw respectively each Evapotranspiration；

8) by Bayesian model method, by step 6) ground extracted evapotranspires observation and step 7) each calculation of evapotranspiring The evapotranspiration that method module is calculated carries out integrated, formation Bayesian model integration module.

The step 1) in the MODIS algorithms that evapotranspire be：

LE=LE_{wet_c}+LE_trans+LE_soil

In formula, LE evapotranspires for land table, LE_{wet_c}It is the evaporation that wet canopy evaporation process is produced, LE_transFor Vegetation canopy steams Rise the transpiration of process generation, LE_wet__so_ilIt is the evaporation that saturated soil moisture evaporation is produced, LE_so_il__po_tFor unsaturated soil is latent In the evaporation that evaporation is produced, LE_soilIt is soil evaporation, Δ is the slope of saturation vapour pressure and temperature curve, and γ is that wet and dry bulb is normal Number, VPD is the saturation vapour pressure and actual vapor pressure difference of air, and RH is relative humidity, r_aIt is aerodynamic resistance, r_sIt is ground Table impedance, F_cIt is vegetation coverage, F_wetIt is humidity area coverage, ρ is atmospheric density, C_pIt is the specific heat capacity of air, R_ncIt is vegetation Net radiation shared by canopy, rhrc is canopy corresponding part aerodynamic resistance, and rvc is Ground Heat transfer impedance, P_a It is atmospheric pressure, R_nsNet radiation shared by surface soil, r_totIt is whole aerodynamic resistance, r_asFor partial air is dynamic Mechanical impedance, β, λ and ε are coefficient；

The step 2) in temperature limiting factor correction method be：

In formula, m_TIt is the temperature limiting factor, T_aIt is air themperature, To_ptIt is vegetation growth optimal temperature, is set to 25 DEG C；

The step 3) in the Priestly-Taylor algorithms that evapotranspire be：

In formula, f_wetIt is relative surface humidity, RH is relative humidity, f_wet=RH⁴, f_gIt is green Canopy cover degrees, f_TTo plant By the temperature limiting factor, f_MIt is vegetation moisture restriction factor, f_SMIt is soil moisture restriction factor, G is soil heat flux.

The step 4) in soil moisture restriction factor be：

In formula, DT is air day and night temperature, DT_maxIt is maximum day and night temperature, takes 40 DEG C；

The step 5) in be based on the algorithm that evapotranspires of empirical method：

LE=R_n(a₀+a₁NDVI+a₂T_a+a₃DTaR)

In formula, DTaR is day day and night temperature, and NDVI is normalized differential vegetation index, a₀, a₁, a₂And a₃It is empirical coefficient；

The step 8) in Bayesian model integration module computing formula be：

In formula, r is represented needs evapotranspiring for estimation, and the observation evapotranspired at a time is r_t, model set K, i.e., {f₁,f₂,…,f_k, be estimation r all single model set, p (r | f_k) it is single model f_kEstimate that the probability of the r that evapotranspires is close Degree function, p (f_k|r_t) it is single model f_kPosterior probability, single model f can be reacted_kWith the degree of observation data fit.

The step 8) in, the probability density function p of the estimated value that Bayesian model integration module is obtained (r | f₁,f₂,…, f_k) be exactly single model probability density function weighted average, the posterior probability sum of all single models is 1, i.e.,So, the posterior probability of each model is regarded as weight w_k；

Assuming that p (r | f_k) it is Gaussian Profile,It is the average value of the Gaussian Profile, σ² _KIt is variance, uses parameterThe Gaussian Profile related to g () expressions be：

p(r|f_k)=g (r | θ_k)

The estimate for obtaining is the conditional expectation of r, cost function is optimized and is obtained weight with expectation-maximum algorithm w_k。

The step 1) by vegetation coverage F_cTo distinguish the global land table available energy of vegetation and soil distribution, profit Determine F with absorbed photosynthetically active radiation ratio FPAR products are planted in MODIS remotely-sensed datas_c；

The step 2) the PM remote sensing corrected evapotranspires in algorithm, and the maximum of unified moisture restriction factor VPD is set to 2.79KPa；

The step 3) R that evapotranspire in algorithm of Priestly-Taylor_nsUse surface net radiation R_nWith leaf area index LAI Calculate, specially R_ns=R_nExp (- kLAI), k are empirical coefficient；

The step 4) improved Priestly-Taylor evapotranspire algorithm soil moisture restriction factor in, DT is to pass through Day day and night temperature is calculated or is calculated by surface temperature；

The step 5) in empirical coefficient a₀, a₁, a₂, and a₃It is to observe data as training number by the use of global flux website Obtained according to recurrence；

The step 6) observation data extraction module in, website measured data sample quality be by quality identification judge Obtained from；

The step 7) leaf area index, vegetation index are chosen from MODIS remotely-sensed datas, meteorology is analyzed again from MERRA Basic meteorologic parameter, including temperature, relative humidity are chosen in data, drives each algoritic module to calculate respective evapotranspiring Amount.

When a kind of global land table based on many algorithm integrations of the present invention evapotranspires estimating system and method is used cooperatively, its Operating procedure is specific as follows：

1st, evapotranspired algorithm (Mu et al.2011) according to MODIS, LAI, NDVI and the EVI using MODIS are (enhanced Vegetation index) product combine analyze MERRA meteorological datas, estimation area evapotranspiration again；

2nd, evapotranspired algorithm (Yuan et al.2010) according to the PM that corrects, produced using LAI, NDVI and EVI of MODIS Product are combined and analyze MERRA meteorological datas, estimation area evapotranspiration again；

3rd, evapotranspired algorithm (Fisher et al.2008) according to Priestly-Taylor, using LAI, NDVI of MODIS And EVI products are combined and analyze ERRA meteorological datas, estimation area evapotranspiration again；

4th, evapotranspired algorithm (Yao et al.2013) according to improved Priestly-Taylor, using the LAI of MODIS, NDVI and EVI products are combined and analyze ERRA meteorological datas, estimation area evapotranspiration again；

5th, evapotranspired algorithm (Wang et al.2008) according to improved empirical remote sensing, using LAI, NDVI of MODIS with And EVI products are combined and analyze ERRA meteorological datas, estimation area evapotranspiration again；

6th, ground is extracted by Global Link discharge observation station data quality to evapotranspire observation；

7th, Bayesian model method is set up, the estimation result of separate sources is merged, obtain optimal estimation result. R represents the variable for needing estimation, and the observation of the variable at a time is r_t, model set K, i.e. { f₁,f₂,…,f_k, To estimate the set of all single model of r.According to breakdown law, it is necessary to be estimated the probability density function of the r that evapotranspires of calculation It is expressed as follows by the probability density function of each model：

Wherein, p (r | f_k) it is single model f_kEstimate the probability density function of the r that evapotranspires, p (f_k|r_t) it is single model f_k's Posterior probability, single model f can be reacted_kWith the degree of observation data fit.The posterior probability sum of all single models It is 1.So, the posterior probability of each model is regarded as weight w_k, this weight is exactly the parameter for needing estimation.It is expressed as：

Therefore, the estimated value for obtaining probability density function p (r | f₁,f₂,…,f_k) it is exactly single model probability density letter Several weighted averages, weighted value is the posterior probability of each function.Assuming that p (r | f_k) it is Gaussian Profile, f_kAIt is the Gaussian Profile Average value, σ² _KIt is variance, uses parameter θ_K={ f_kA, σ² _KThe Gaussian Profile related to g () expressions, it is as follows：

p(r|f_k)=g (r | θ_k)

The estimated value table for obtaining is shown as the conditional expectation of r, as follows：

It will be apparent that f_kAThe estimated value of single model, then the key of problem be then how the posterior probability of appraising model w_k, so as to get r probability density functions be more nearly with observation.According to bayesian theory, it is known that observe data surely, work as likelihood When function formula (4) value is maximum, the estimated value that bayes method is obtained for true value maximum probability (Duan and Phillips, 2010).Calculate for convenience, the logarithmic function for tackling the likelihood formula is optimized：

Wherein, ∑_(s,T)Represent in all spatial point s and the observation sum of all time point T, r_s,tRepresent sometime point T On the s of space, cost function is optimized with expectation-maximum algorithm (expectation-maximization, EM) algorithm (Raftery et al.2005), single model f_kAEstimated value and corresponding observation r_s,tEach is calculated as training data The posterior probability of model.

Both 8th, measurement evapotranspiration and the actual evapotranspiration of selection point position carry out correlation analysis, obtain studying in area Relation；

9th, evaluation model evapotranspires estimation situation in the world in research area.

Checking data set is FLUXNET200, whole world flux website observation data.Global FLUXNET station datas include 200 websites, distribution Asia, Europe, North America, South America, Africa and Oceania.Across the torrid zone, subtropical zone, temperate zone, cold With various climate zones, the various vegetation cover types such as desert, oasis, grassland, forest, savart Na, farmland and wetland are covered, it is full Foot observation data at least one year, disclosure satisfy that the estimation requirement evapotranspired under various vegetation patterns and weather SNNP.With complete Ball land table is research area, and the flux observation data in the research area are that 2000-2009 years section is collected, corresponding MODIS numbers According to analyze data is also 2000-2009 years section again with MERRA.In order to the evapotranspiration for verifying model estimates effect, calculate Obtain evapotranspiration amount carries out correlation analysis with the observation data at Global Link discharge observation station.According to each observation station of global FLUXNET The evapotranspiration of measurement and the evapotranspiration of the integrated estimation of bayesian algorithm set up scatter diagram, its result such as Fig. 3, mean error is 2.3W/m², coefficient correlation square (R²) it is 0.64, mean square deviation is 24.2W/m².Result shows, model estimate value and field observation Data have correlation higher.Visible by analyzing, Bayesian model Integrated Algorithm has very much to estimation global land table evapotranspiration Effect.

Claims (2)

1. a kind of global land table based on many algorithm integrations evapotranspires evaluation method, it is characterised in that the specific steps of methods described Including：

1) MODIS is built to evapotranspire algoritic module：Canopy is divided into wet canopy and dry canopy, soil be divided into water saturation soil and Moisture unsaturation soil, evapotranspires algorithm according to MODIS, calculates and the impedance of design earth's surface, aerodynamic resistance and boundary layer Parameter；

2) the PM remote sensing corrected is built to evapotranspire algoritic module：Evapotranspired algorithm according to traditional Penman-Monteith, temperature is limited The defect of factor linear change processed is corrected；

3) Priestly-Taylor is built to evapotranspire algoritic module：According to existing Priestly-Taylor models, using ecology System stress factors extend Priestly-Taylor coefficients, obtain Priestly-Taylor and evapotranspire algorithm；

4) improved Priestly-Taylor is built to evapotranspire algoritic module：Soil water evaporation is parameterized using apparent thermal inertia Factor amendment soil moisture restriction factor, obtains improved Priestly-Taylor and evapotranspires algorithm；

5) algoritic module that evapotranspires based on empirical method is built：Rule of thumb Evapotranspiration by Using algorithm, design evapotranspire input, export and Computing module；

6) extract ground from global flux website observation data by observation data extraction module according to website measured data sample quality Face is evapotranspired observation；

7) analyze meteorological data actuation step 1 again using MODIS remotely-sensed datas and MERRA) MODIS evapotranspire algoritic module, step The rapid PM remote sensing 2) corrected is evapotranspired algoritic module, step 3) Priestly-Taylor evapotranspires algoritic module, step 4) improved Priestly-Taylor evapotranspires algoritic module, step 5) algoritic module that evapotranspires based on empirical method, respective steaming is drawn respectively Dissipate amount；

8) by Bayesian model method, by step 6) ground extracted evapotranspires observation and step 7) each algorithm mould that evapotranspires The evapotranspiration that block is calculated carries out integrated, formation Bayesian model integration module；

The step 1) in the MODIS algorithms that evapotranspire be：

LE=LE_{wet_c}+LE_trans+LE_soil

${\mathrm{LE}}_{wet\_c}=\frac{\[\mathrm{\Δ}\×R_{nc}+\mathrm{\ρ}\×C_p\×VPD\×F_c/rhrc\]\×F_{wet}}{\mathrm{\Δ}+\frac{P_a\×C_p\×rvc}{\mathrm{\λ}\×\mathrm{\ϵ}\×rhrc}}$

${\mathrm{LE}}_{trans}=\frac{\[\mathrm{\Δ}\×R_{nc}+\mathrm{\ρ}\×C_p\×VPD\×F_c/r_a\]\×(1-F_{wet})}{\mathrm{\Δ}+\mathrm{\γ}\×(1+r_s/r_a)}$

${\mathrm{LE}}_{wet\_soil}=\frac{\[\mathrm{\Δ}\×R_{ns}+\mathrm{\ρ}\×C_p\×VPD\×(1-F_c)/r_{as}\]\×F_{wet}}{\mathrm{\Δ}+\mathrm{\γ}\×r_{tot}/r_{as}}$

${\mathrm{LE}}_{soil\_pot}=\frac{\[\mathrm{\Δ}\×R_{ns}+\mathrm{\ρ}\×C_p\×VPD\×(1-F_c)/r_{as}\]\×(1-F_{wet})}{\mathrm{\Δ}+\mathrm{\γ}\×r_{tot}/r_{as}}$

${\mathrm{LE}}_{soil}={\mathrm{LE}}_{wet\_soil}+{\mathrm{LE}}_{soil\_pot}\×{\left(\frac{RH}{100}\right)}^{VPD/\mathrm{\β}}$

In formula, LE evapotranspires for land table, LE_{wet_c}It is the evaporation that wet canopy evaporation process is produced, LE_transIt is Vegetation canopy transpiration process The transpiration of generation, LE_{wet_soil}It is the evaporation that saturated soil moisture evaporation is produced, LE_{soil_pot}For unsaturated soil potential evaporation is produced Raw evaporation, LE_soilIt is soil evaporation, Δ is the slope of saturation vapour pressure and temperature curve, and γ is wet and dry bulb constant, and VPD is The saturation vapour pressure of air and actual vapor pressure difference, RH is relative humidity, r_aIt is aerodynamic resistance, r_sIt is earth's surface impedance, F_c It is vegetation coverage, F_wetIt is humidity area coverage, ρ is atmospheric density, C_pIt is the specific heat capacity of air, R_ncShared by Vegetation canopy Net radiation, rhrc be canopy corresponding part aerodynamic resistance, rvc be Ground Heat transfer impedance, P_aIt is atmospheric pressure, R_nsNet radiation shared by surface soil, r_totIt is whole aerodynamic resistance, r_asIt is partial air dynamics impedance, β, λ and ε are coefficient；

The step 2) in temperature limiting factor correction method be：

$m_T=\exp \[-{\left(\frac{T_a-T_{opt}}{T_{opt}}\right)}^2\]$

In formula, m_TIt is the temperature limiting factor, T_aIt is air themperature, T_optIt is vegetation growth optimal temperature, is set to 25 DEG C；

The step 3) in the Priestly-Taylor algorithms that evapotranspire be：

${\mathrm{LE}}_{trans}=(1-f_{wet})f_g{f}_T{f}_M\mathrm{\α}\frac{\mathrm{\Δ}}{\mathrm{\Δ}+\mathrm{\γ}}{R}_{nc}$

${\mathrm{LE}}_{soil}=\[f_{wet}+f_{SM}(1-f_{wet})\]\mathrm{\α}\frac{\mathrm{\Δ}}{\mathrm{\Δ}+\mathrm{\γ}}(R_{ns}-G)$

${\mathrm{LE}}_{wet\_c}=f_{wet}\mathrm{\α}\frac{\mathrm{\Δ}}{\mathrm{\Δ}+\mathrm{\γ}}{R}_{nc}$

In formula, f_wetIt is relative surface humidity, RH is relative humidity, f_wet=RH⁴, f_gIt is green Canopy cover degrees, f_TIt is vegetation temperature Degree restriction factor, f_MIt is vegetation moisture restriction factor, f_SMIt is soil moisture restriction factor, G is soil heat flux；

The step 4) in soil moisture restriction factor be：

$f_{SM}={\left(\frac{1}{DT}\right)}^{DT/{\mathrm{DT}}_{\max }}$

In formula, DT is air day and night temperature, DT_maxIt is maximum day and night temperature, takes 40 DEG C；

The step 5) in be based on the algorithm that evapotranspires of empirical method：

LE=R_n(a₀+a₁NDVI+a₂T_a+a₃DTaR)

In formula, DTaR is day day and night temperature, and NDVI is normalized differential vegetation index, a₀, a₁, a₂And a₃It is empirical coefficient；

The step 8) in Bayesian model integration module computing formula be：

$p\left(r\right|f_1,f_2,...,f_K)=\underset{k=1}{\overset{K}{\Σ}}p\left(r\right|f_k)p\left(f_k\right|r_t)$

In formula, r is represented needs evapotranspiring for estimation, and the observation evapotranspired at a time is r_t, model set K, i.e. { f₁, f₂,…,f_k, be estimation r all single model set, p (r | f_k) it is single model f_kEstimate the probability density letter of the r that evapotranspires Number, p (f_k|r_t) it is single model f_kPosterior probability, single model f can be reacted_kWith the degree of observation data fit；

The step 8) in, the probability density function p of the estimated value that Bayesian model integration module is obtained (r | f₁,f₂,…,f_k) It is exactly the weighted average of single model probability density function, the posterior probability sum of all single models is 1, i.e.,So, the posterior probability of each model is regarded as weight w_k；

Assuming that p (r | f_k) it is Gaussian Profile,It is the average value of the Gaussian Profile, σ² _KIt is variance, uses parameterAnd g () represents that the Gaussian Profile of correlation is：

p(r|f_k)=g (r | θ_k)

$p\left(r\right|f_1,f_2,...f_K)=\underset{k=1}{\overset{K}{\Σ}}{w}_{k}g\left(r\right|{\mathrm{\θ}}_k)$

The estimate for obtaining is the conditional expectation of r, cost function is optimized and is obtained weight w with expectation-maximum algorithm_k；

The step 1) by vegetation coverage F_cTo distinguish the global land table available energy of vegetation and soil distribution, utilize Absorbed photosynthetically active radiation ratio FPAR products are planted in MODIS remotely-sensed datas and determines F_c；

The step 2) the PM remote sensing corrected evapotranspires in algorithm, and the maximum of unified moisture restriction factor VPD is set to 2.79KPa；

The step 3) R that evapotranspire in algorithm of Priestly-Taylor_nsUse surface net radiation R_nCalculated with leaf area index LAI, Specially R_ns=R_nExp (- kLAI), k are empirical coefficient；

The step 4) improved Priestly-Taylor evapotranspire algorithm soil moisture restriction factor in, DT is by daytime day Night differential thermal calculation is obtained or is calculated by surface temperature；

The step 5) in empirical coefficient a₀, a₁, a₂, and a₃It is to observe data by the use of global flux website to be returned as training data Obtain；

The step 6) observation data extraction module in, website measured data sample quality be by quality identification judge and obtain Arrive；

The step 7) leaf area index, vegetation index are chosen from MODIS remotely-sensed datas, meteorological data is analyzed again from MERRA It is middle to choose basic meteorologic parameter, including temperature, relative humidity, drive each algoritic module to calculate respective evapotranspiration.

2. a kind of global land table based on many algorithm integrations according to claim 1 evapotranspires evaluation method, it is characterised in that The global land table includes forest, shrub, farmland, meadow and wetland.