AU2021105183A4 - A Method for Estimating Near-surface Air Temperature From Remote Sensing Data Based on Machine Learning - Google Patents
A Method for Estimating Near-surface Air Temperature From Remote Sensing Data Based on Machine Learning Download PDFInfo
- Publication number
- AU2021105183A4 AU2021105183A4 AU2021105183A AU2021105183A AU2021105183A4 AU 2021105183 A4 AU2021105183 A4 AU 2021105183A4 AU 2021105183 A AU2021105183 A AU 2021105183A AU 2021105183 A AU2021105183 A AU 2021105183A AU 2021105183 A4 AU2021105183 A4 AU 2021105183A4
- Authority
- AU
- Australia
- Prior art keywords
- temperature
- modis
- water vapor
- air temperature
- surface air
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000010801 machine learning Methods 0.000 title claims abstract description 5
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Chemical compound O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000004088 simulation Methods 0.000 claims abstract description 25
- 238000013528 artificial neural network Methods 0.000 claims abstract description 24
- 238000012360 testing method Methods 0.000 claims abstract description 23
- 230000005855 radiation Effects 0.000 claims abstract description 9
- 230000008859 change Effects 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 11
- CURLTUGMZLYLDI-UHFFFAOYSA-N Carbon dioxide Chemical compound O=C=O CURLTUGMZLYLDI-UHFFFAOYSA-N 0.000 claims description 6
- 101000836284 Homo sapiens Solute carrier organic anion transporter family member 1B3 Proteins 0.000 claims description 3
- 102100027239 Solute carrier organic anion transporter family member 1B3 Human genes 0.000 claims description 3
- 229910002092 carbon dioxide Inorganic materials 0.000 claims description 3
- 239000001569 carbon dioxide Substances 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 2
- 239000005427 atmospheric aerosol Substances 0.000 claims description 2
- 238000012795 verification Methods 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 abstract description 12
- 230000005540 biological transmission Effects 0.000 abstract description 3
- 238000004364 calculation method Methods 0.000 abstract description 3
- 230000007613 environmental effect Effects 0.000 abstract description 3
- 230000001537 neural effect Effects 0.000 abstract description 3
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000002834 transmittance Methods 0.000 description 4
- 241000132092 Aster Species 0.000 description 3
- 239000000443 aerosol Substances 0.000 description 3
- 239000007789 gas Substances 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- CBENFWSGALASAD-UHFFFAOYSA-N Ozone Chemical compound [O-][O+]=O CBENFWSGALASAD-UHFFFAOYSA-N 0.000 description 2
- 230000009471 action Effects 0.000 description 2
- 229930002875 chlorophyll Natural products 0.000 description 2
- 235000019804 chlorophyll Nutrition 0.000 description 2
- ATNHDLDRLWWWCB-AENOIHSZSA-M chlorophyll a Chemical compound C1([C@@H](C(=O)OC)C(=O)C2=C3C)=C2N2C3=CC(C(CC)=C3C)=[N+]4C3=CC3=C(C=C)C(C)=C5N3[Mg-2]42[N+]2=C1[C@@H](CCC(=O)OC\C=C(/C)CCC[C@H](C)CCC[C@H](C)CCCC(C)C)[C@H](C)C2=C5 ATNHDLDRLWWWCB-AENOIHSZSA-M 0.000 description 2
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 2
- 239000013049 sediment Substances 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 description 1
- 241000196324 Embryophyta Species 0.000 description 1
- 241001464837 Viridiplantae Species 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229910002091 carbon monoxide Inorganic materials 0.000 description 1
- 230000001143 conditioned effect Effects 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000000053 physical method Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- CCEKAJIANROZEO-UHFFFAOYSA-N sulfluramid Chemical group CCNS(=O)(=O)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)F CCEKAJIANROZEO-UHFFFAOYSA-N 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W2201/00—Weather detection, monitoring or forecasting for establishing the amount of global warming
Landscapes
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Environmental & Geological Engineering (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Animal Husbandry (AREA)
- General Business, Economics & Management (AREA)
- Marine Sciences & Fisheries (AREA)
- Agronomy & Crop Science (AREA)
- Mining & Mineral Resources (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Human Resources & Organizations (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Economics (AREA)
- Atmospheric Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Ecology (AREA)
- Environmental Sciences (AREA)
- Radiation Pyrometers (AREA)
Abstract
The invention relates to a method for estimating near-surface air temperature from remote
sensing data based on machine learning, which can be applied to remote sensing application
departments such as meteorology, environmental monitoring, land management, agricultural
situation monitoring, and disaster monitoring. The method comprises three steps. In the first
step, the land surface temperature, emissivity and atmospheric water vapor content of each
pixel of MODIS data products are used as prior knowledge as input parameters of atmospheric
radiation transmission simulation software MODTRAN4, and forward simulation is carried out
in different regions and seasons for the 29th, 31st and 32nd thermal infrared bands of each
pixel of obtained remote sensing data MODIS, and a training and test database is established.
The second step is to train and test the training and test data sets repeatedly by using neural
network. The third step is the inversion calculation of MODIS actual image data to obtain the
near-surface air temperature distribution in the target region, which can be used for weather
forecasting, environmental monitoring, agricultural monitoring and disaster monitoring, etc.
2/4
MODIS I B
Band:29, 21, 32,
2, 17T, 1B, 19
Inversion gonthm
of land surface
temperature and
emissivity
Land surface temperature Land surface emissivilty
ear-. ce MODTRAN Training and Dynanne
air learning neural
temperature simulation test database network
range t +
Atmosphfenc water vapor N ear -surface
content range and other air
atmospheric parameters temperature
Figure 3
Input
MODIS
data of T29
on-board Output
brightness T31 near-surface air
temperature NSAT
in 29th. 31st temperature
and 32nd m
bands and
atmospheric W
water vapor
content Input output
node Hidden node node
Figure 4
Description
2/4
MODIS I B Band:29, 21, 32, 2, 17T, 1B, 19
Inversion gonthm of land surface temperature and emissivity
Land surface temperature Land surface emissivilty
ear-. ce MODTRAN Training and Dynanne air learning neural temperature simulation test database network range t
+ Atmosphfenc water vapor N ear -surface content range and other air atmospheric parameters temperature
Figure 3
Input MODIS data of T29 on-board Output brightness T31 near-surface air temperature NSAT in 29th. 31st temperature and 32nd m bands and atmospheric W water vapor content Input output node Hidden node node
Figure 4
[A Method for Estimating Near-surface Air Temperature From Remote Sensing
Data Based on Machine Learning]
The invention relates to a method for estimating near-surface air temperature from remote
sensing data based on machine learning, which breaks through that limitation of traditional
ground observation. It can be applied to remote sensing departments such as meteorology,
agriculture, environmental monitoring and drought monitoring.
Near-surface air temperature refers to the temperature about 2 meters away from the
surface, which is a very important parameter in climate change research.
MODIS remote sensor was successfully launched with Earth observation satellite in 1999
and 2002, which opened up a new way for dynamic monitoring of global and regional
resources and environment. MODIS is a medium-resolution remote sensing system with
36 bands (see Figure 1), which can obtain global observation data once every 1 ~ 2 days.
Its flight is synchronized with the sun, and at least two images of day and night can be
obtained in the same area every day, and they are received free of charge, so it is very
suitable for dynamic monitoring of regional resources and environment at large and
medium scale. Eight of the 36 bands of MODIS are thermal infrared bands, as shown in
Table 1, so it is very suitable for analyzing the spatial difference of surface heat at regional
scale. At present, there are many inversion algorithms of land surface temperature for
MODIS remote sensing data, and the inversion accuracy of land surface temperature and emissivity has been guaranteed to some extent. At present, the method of estimating near surface air temperature by MODIS data has not been published.
Table 1 Technical parameters of MODIS remote sensor
[MODISLevel 1B Product User's Guide, ForLevel 1B Version 4.2.0 (Terra) and Version
4.2.1 (Aqua).]
Spectral Spectral Ground Signal to Main application region bandwid resolution (m) noise ratio fields th Chlorophyll 620-670nm 50nm 128snr absorption of 250 vegetation Clouds and plants, 841-876nm 35nm 201snr landover land cover Differences 459~479nm 20nm 243snr between soil and vegetation 545-565nm 20nm 228snr Green plants
20nm 500 74snr Leaf/canopy 1230~1250nm difference Snow/cloud 1628~1652nm 20nm 275snr difference difference Land and cloud 2105-2135nm 50nm 110snr caracrsi characteristics 405~420nm l5nm 880snr 438~448nm l0nm 8380snr ocean color 483~493nm l0nm 1000 802snr plankton 526-536nm l0nm 754snr
546~556nm 10nm 750snr Ocean water color, sediment
662~672nm l0nm 910snr Sediment, atmosphere
673-683nm l0nm 1087snr Chlorophyll fluorescence
10nm 586snr Aerosol 743~753nm characteristics
862~877nm l5nm 516snr Aerosol/atmospheri c characteristics 890-920nm 30nm 167snr 931~941nm 10nm 57snr Cloud/atmosphere characteristics 915-956nm 50nm 250snr
180nm 0.05 Sea surface 3660~3840nm NEAT temperature
3929-3989nm 50nm 2.00 Forest fire/volcano NEAT 3929-3989nm 50nm 0.07 NEAT Cloud/surface 0.07 temperature 4020~4080nm 50nm NEAT
4433~4498nm 50nm 0.25 NEAT Atmospheric 0.25 humidity/cloud 4482~4549nm 50nm NEAT
1504NEA Cirrus clouds and 1360~1390nm 30nm Tarsl T aerosols
6535-6895nm 360nm 0.25 NEAT Atmospheric 0.25 humidity 7175~-7475nm 300nm NEAT
0.05 8400-8700nm 300nm NEAT Surface temperature
0.25 9580-9880nm 300nm NEAT Ozone
10780~11280 0.05 Cloud/surface nm NEAT temperature
11770~12270 0.05 Cloudtop nm500nm NEAT height/surface temperature 13185~13485 300nm 0.25 nm NEAT 13485~13785 300nm 0.25 nm NEAT ____________________________Cloud top height 13785~14085 300nm 0.25 nm NEAT 14085~14385 300nm 0.25 nm NEAT
The purpose of the present invention is to provide a method for estimating near-surface air
temperature from remote sensing data MODIS, so as to overcome the practical difficulties
that the existing near-surface air temperature is difficult to obtain by physical methods, and
the shortcomings that interpolation of meteorological stations is difficult to guarantee
accuracy, especially the shortcomings that timeliness is difficult to guarantee in remote
areas. The method can also further improve the estimation accuracy of near-surface small
scale air temperature and surface evapotranspiration.
To achieve the above purpose, the method for estimating the near-surface air temperature
from remote sensing data MODIS provided by the present invention is as follows.
In the first step, the simulation database of on-board radiance temperature in 29th, 31st and
32nd bands of MODIS remote sensor is established.
1-1) Select the atmospheric profile mode, atmospheric path, radiation mode and scattering
mode of the region where the obtained image is located as input parameters.
1-2) Read out the values of land surface temperature (LST), emissivity (Ei) and atmospheric
water vapor content (w) of each pixel of the corresponding MODIS product. For each pixel,
LST-2 KSLSTSLST+2 K, Ei-0.03<E<Eif0.03, and w-0.13w~w~w+0.13w are input into
MODTRAN4 as prior knowledge of each pixel for simulation and establishment of training
and test database.
1-3) Read in the land surface temperature and emissivity values corresponding to MODIS
data, and simulate the possible changes of land surface temperature and near-surface air
temperature according to the change range defined in 1-2).
1-4) Read in the initial value of atmospheric water vapor content, and simulate the change
of atmospheric water vapor content in the process according to the error change range
limited in 1-2).
1-5) Input the height of MODIS satellite sensor and default other parameters such as
atmospheric aerosol and carbon dioxide.
1-6) Perform simulation according to wavelength ranges of 29th, 31st and 32nd bands of
MODIS data, and output simulated on-board radiance of 29th, 31st and 32nd bands of
MODIS data.
1-7) Convert the on-board radiance obtained by each simulation into brightness
temperature, and establish a corresponding database for each pixel together with the land
surface temperature, emissivity and atmospheric water vapor content input by each
simulation.
In the second step, neural network training and test
2-1) Divide the simulation database in the first step into two groups. One group is a training
data set, one group is a test data set.
2-2) Carry out training by taking the on-board brightness temperature and atmospheric
water vapor content of MODIS 29th, 31st and 32nd bands in the training data set as the
input nodes of the neural network and the near-surface air temperature as the output node.
2-3) Input the on-board brightness temperature and atmospheric water vapor content of the
test data set into the trained neural network, and output the near-surface air temperature.
2-4) Compare the near-surface air temperature output in 2-3 with the corresponding near
surface air temperature.
In the third step, Inversion of near-surface air temperature
3-1) Read the 29th, 31st and 32nd bands of MODIS remote sensing image data and
atmospheric water vapor content data.
3-2) Convert the on-board brightness of the 29th, 31st and 32nd bands of MODIS data into
on-board brightness temperature (T29, T31 and T32) and extract the corresponding
atmospheric water vapor content W.
3-3) Input T29, T31, T32 and W in 3-2 into the neural network trained in the second step,
and output the near-surface air temperature (NSAT).
3-4) Carry out relevant verification and application analysis according to the corresponding
surface of the image.
According to the method, in step 1-3 of the first step, the change range of near-surface air
temperature is NSAT<LST+15 K, and the step change range in the simulation process is 2
According to the method, in step 1-4 of the first step, the initial value of atmospheric water
vapor content is the read atmospheric water vapor content (w), the limited range is w
0.13wswsw+.13w, and the step change range in the simulation process is 0.2 g/cm2
. According to the method, in step 1-5 of the first step, the input MODIS satellite sensor
height is 705 KM.
According to the method, in step 2-4 of the second step, the standard error of near-surface
air temperature is greater than 2 K, so both layers of hidden nodes are increased by 5, and
training and test are continued by repeating 2-2 until the standard error of near-surface air
temperature is less than 2 K.
The method has the beneficial effects that MODIS land surface temperature, emissivity
and atmospheric water vapor content are used as prior knowledge, the emissivity between
adjacent thermal infrared bands has a linear relationship, and there is a relationship between
transmittance and atmospheric water vapor content, and the potential information can be
well utilized by simulating with an atmospheric radiation transmission model, so that
unknowns are effectively reduced and the problem of insufficient equations in ill
conditioned inversion is solved. The inversion accuracy and calculation time are improved,
and the shortcoming of insufficient information for estimating near-surface air temperature
directly from satellite data in the past is overcomed. The method provides effective means
and technical support for climate change research, weather forecast, evapotranspiration, agricultural monitoring and disaster monitoring. Its operational practicability is simpler than that of traditional ground meteorological observation stations, and its surface accuracy is higher. In fact, the surface meteorological observation station is also an important supplementary source of data for further improving the accuracy of this method, and the combination of the two will greatly provide the regional estimation accuracy of near surface air temperature.
The invention will be further explained with reference to drawings and embodiments.
Figure 1 MODIS remote sensor.
Figure 2 A main flow diagram of the present invention
Figure 3 A schematic flow chart of establishing a simulation database of on-board radiance
temperature in 29th, 31st and 32nd bands of MODIS remote sensor
Figure 4 A schematic diagram of a multilayer neural network structure adopted by the
present invention
Figure 5 A schematic diagram of the neural network training and test flow of the present
invention
Figure 6 A schematic flow chart of estimating near-surface air temperature according to
the present invention
Figure 7 A comparison diagram of the ground surface measured data and inversion
results obtained by using the present invention
The inversion of land surface temperature and emissivity is based on the heat radiation
transfer equation, and the general expression is shown in Formula (1) [Mao Kebiao,
Research on the inversion method of land surface temperature based on MODIS data,
master's thesis, Nanjing University, May 2004];
Bi(Ti) = Ei()Ti()Bi(T,) + [1 - ri(6')][1 - Ei(8)]Ti(8) Bi(Tia)+ [1 - Ti(O)]Bi(Tia)
In the formula, T, represents the land surface temperature, Ti represents the on-board
brightness temperature obtained by the channel i at the height of the sensor, Ti(O)
represents the atmospheric transmittance of the channel i in the observation direction 0,
' represents the downward brightness temperature radiation direction of the atmosphere
and Ei(0) represents the surface emissivity of the channel i in the observation direction
0. Bi(Ti) is the radiation intensity received by the sensor, Bi(T,) is the radiation intensity
of the land surface, Tia represents the effective atmospheric average action temperature of
the channel i. The effective atmospheric average action temperature (Tia) varies with
wavelength. It is mainly determined by atmospheric water vapor content and near-surface
air temperature[Mao Kebiao, Huajun Tang, Xiufeng Wang, Qingbo Zhou, Daolong Wang,
Near-Surface Air Temperature Estimation From ASTER Data Using Neural Network,
InternationalJournalofRemote Sensing, 2008, 29(20): 6021-6028.].
Tia = Ai + BiT 0 (Formula 2)
In the formula, To is the near-surface air temperature about 2 meters high, Ai is constant,
Bi is channel and i is coefficient. The near-surface air temperature is also affected by the land surface temperature. At a given place, the relationship between near-surface air temperature and land surface temperature is similar to Formula 2, but this relationship is not very stable, and it changes with time and place. In equation 1, there are three unknowns
(emissivity, land surface temperature and near-surface air temperature), which is a typical
ill-conditioned problem. If other conditions are not constructed, the equation has no
solution. In addition, the transmittance (T(O)) of each thermal infrared band is also
unknown. It is a function (Formula 3) of atmospheric water vapor content and other gases.
Ti(0) = f(W, 0) (Formula 3)
W is the atmospheric water vapor content, 0 means other gases (carbon dioxide, nitric
oxide, ozone, methane, carbon monoxide, etc.). These gases are stable relative to the
atmospheric water vapor content, and their influence can be obtained by standard
atmospheric profile simulation. The transmittance of thermal infrared band is very sensitive
to water vapor. The split window algorithm uses the different sensitivities of two thermal
infrared bands to water vapor to eliminate the influence of water vapor, and then gets the
land surface temperature by inversion calculation. For different ground objects in different
bands, emissivity is almost a constant. Mao et al. (2008) [ Mao, K., Shi J., Tang H., Li Z.L.,
Wang X. and Chen K., A Neural Network Technique for Separating Land Surface
Emissivity and Temperature from ASTER Imagery, IEEE Transactionson Geoscience and
Remote Sensing, 2008, 46(1): 200-208.]proposed to overcome the ill-conditioned problem
by using the local linear relationship between emissivity of adjacent bands to reduce
unknowns. The equation can be described as Formula 4.
Ei(0) = Ci + DiEj(0) (Formula 4)
Ei(0) and Ej(0) are the emissivity of different bands (i, j) at the observation angle 0. C,
is a constant, Di is the coefficient of the channel i. For the same land type, the emissivity
of different bands can be expressed by the emissivity of one band, thus reducing the
emissivity of different bands to one. Because it is difficult to describe all bands accurately
with several functions, this potential information is not fully utilized[Mao, K., Shi J., Tang
H., Li Z.L., Wang X. and Chen K., A Neural Network Technique for Separating Land
Surface Emissivity and Temperature from ASTER Imagery, IEEE Transactions on
Geoscience and Remote Sensing, 2008, 46(1): 200-208.]. In addition, the linear
simplification of nonlinear functions (such as Planck function) will also produce errors.
In order to overcome the disadvantage that the traditional inversion algorithm needs to
spend a lot of time to calculate, the present invention uses neural network (NN) without
knowing the relationship between input and output parameters, and can simulate the
training data set through the atmospheric radiation transmission model (MODTRAN4),
and determines the relationship between input and input data through analog data training.
The implementation (method) of this embodiment mainly includes three steps, as shown in
Figure 2.
The first step is to read the land surface temperature, emissivity and atmospheric water
vapor content of each pixel from MODIS land surface temperature and emissivity products
and atmospheric water vapor content products, and take them as the input parameters of
MODTRAN4. For example, the land surface temperature of a pixel is 300K, the emissivity
of 29th, 31st and 32nd bands is 0.95, 0.96 and 0.98 respectively, and the atmospheric water
vapor content is 1.2 g/cm2 . Taking these product values as known input parameters of
MODTRAN, the near-surface air temperature varies from 300 K to 315 K, and the mid
latitude atmospheric profile is adopted. The simulation process is shown in Figure 3.
The second step is to use the neural network software. Neural network is different from
traditional methods, because it does not need to know the inversion algorithm (rules)
accurately. Because neural network can extract information from complex and imprecise
data, neural network can be used to extract pattern prediction [Hornik K. M., Stinchcombe
M., and White H., Multilayer feedforward networks are universal approximators, Neural
Network, 1989, 4(5): 359-366], as shown in Figure 4. In this embodiment, a dynamic
learning neural network (DL) is used to train and test the database established in the first
step. The dynamic neural network uses Kalman filter to increase the convergence speed
during training and improve the ability to understand nonlinear problems. The weight of
each node of the neural network are initialized to random numbers between (-1, 1). The
Kalman filtering process is an iterative process of root mean square estimation. Each update
of network weights is based on the previous weight learning of new input data sets, and the
weight updates of output nodes are independent of each other. Because the dynamic
learning neural network based on Kalman filter only needs two iterations to reach the
required root mean square threshold, and the inversion result is stable, the root mean square
error is usually set to 10e-3 , and the number of iterations is 2. See the introduction of for
more information[Tzeng Y. C., Chen K. S., Kao W. L. , and Fung A. K., A Dynamic
learning nerual network for remote sensing applications, IEEE Trans. Geosci. Remote
Sensing, 1994, 32(5): 1096-1102.].
The whole simulation and training process is shown in Figure 5. The simulation data is
divided into two parts: training data set 89 and test data set 51. After repeated training and test, the inversion results of near-surface air temperature are shown as TO' in Table 1. It can be seen from Table 1 that the inversion results are very good, and the average error and standard error are about 0.8 K and 0.9 K respectively. The main reason for the improvement of accuracy is that land surface temperature, emissivity and atmospheric water vapor content are priori knowledge. The inversion errors considering land surface temperature and emissivity are about 1 K and 0.015 respectively[Wan, Z., Y. Zhang, Y. Q. Zhang, and Z. L. Li , Validation of the land surface temperature products retrieved from Moderate
Resolution Imaging Spectroradiometer data, Remote Sens. Environ., 2002, 83: 163
180.], The error of atmospheric water vapor content is 13%[ Kaufman Y. J., Gao B.C.,
Remote sensing of water vapor in the near-IR from EOS/MODIS. IEEE Trans. Geosci.
Rem. Sens., 1992, 30, 871-884.]. Here, we consider that the inversion error of land surface
temperature is 2 K and the emissivity error is 0.03. For each pixel, LST-2
K<LST<LST+2 K, Ei -0.03< E<< Ei +0.03, and w-0.13ww~w+0.13w are input into
MODTRAN4 as prior knowledge of each pixel for simulation and establishment of training
and test data. For example, the land surface temperature of a pixel is 300 K, the emissivity
of 29th, 31st and 32nd bands are 0.95, 0.96 and 0.98 respectively, and the atmospheric
water vapor content is 1.2 g/cm 2 . Land surface temperature of (297-303K), emissivity of
29th, 31st and 32nd bands of 0.92-0.97, 0.93-0.99 and0.96-1, and atmospheric water vapor
content of 1-1.5 g/cm 2 are input into MODTRAN4 as prior knowledge. The training data
set is 836 and the test data set is 392. After repeated training and test, the near-surface air
temperature is shown as To" in Table 1. The average error and standard deviation are 1.5 K
and 1.8 K, respectively, which can meet the requirements in current application.
Table 1 Inversion Error Table
Hidden TO' Hidden TO" node node r SD r SD
5-5 0.981 1.1 10-10 0.964 2.5
10-10 0.981 1.1 20-20 0.968 2.4
15-15 0.982 1.1 30-30 0.969 2.2
20-20 0.986 one 40-40 0.978 2 25-25 0.989 0.9 50-50 0.982 1.8 30-30 0.988 one 60-60 0.975 2.1 35-35 0.982 1.1 70-70 0.975 2.1
40-40 0.981 1.2 80-80 0.974 2.2 R: correlation coefficient; SD: standard deviation. TO' and TO" are near-surface air
temperatures.
The third step is to use the neural network trained in the second module to actually invert
the remote sensing image data MODIS. In order to compare with the actual
meteorological observation station data, we selected MODIS data of two meteorological
stations in Xiaotangshan and Hailar (2004), extracted the on-board brightness
temperature, land surface temperature, emissivity and atmospheric water vapor content
corresponding to MODIS 29th, 31st and 32nd bands of corresponding single points
according to longitude and latitude, and established training databases respectively (the
process is as shown in Step 1, Figure 3), and then carried out actual specific training and
test process as shown in Figure 6. The comparison result is shown in Figure 7, and the
average error ( n )ITO about rToIis 1.6 K.
Claims (5)
1. A method for estimating near-surface air temperature from remote sensing data based
on machine learning, characterized in that:comprising the following steps:
Firstly, the simulation database of on-board radiance temperature in 29th, 31st and 32nd
bands of MODIS remote sensor is established;
1-1) select the atmospheric profile mode, atmospheric path, radiation mode and scattering
mode of the region where the obtained image is located as input parameters ;
1-2) read out the values of land surface temperature (LST), emissivity (Ei) and atmospheric
water vapor content (w) of each pixel of the corresponding MODIS product; for each pixel,
LST-2 K<LST<LST+2 K, Ei-0.03Ej<Eif0.03, and w-0.13w~w~w+0.13w are input into
MODTRAN4 as prior knowledge of each pixel for simulation and establishment of training
and test database;
1-3) read in the land surface temperature and emissivity values corresponding to MODIS
data, and simulate the possible changes of land surface temperature and near-surface air
temperature according to the change range defined in 1-2);
1-4) read in the initial value of atmospheric water vapor content, and simulate the change
of atmospheric water vapor content in the process according to the error change range
limited in 1-2);
1-5) input the height of MODIS satellite sensor and default other parameters such as
atmospheric aerosol and carbon dioxide;
1-6) perform simulation according to wavelength ranges of 29th, 31st and 32nd bands of
MODIS data, and output simulated on-board radiance of 29th, 31st and 32nd bands of
MODIS data;
1-7) convert the on-board radiance obtained by each simulation into brightness
temperature, and establish a corresponding database for each pixel together with the land
surface temperature, emissivity and atmospheric water vapor content input by each
simulation;
the second step, neural network training and test
2-1) divide the simulation database in the first step into two groups. One group is a training
data set, one group is a test data set.
2-2) carry out training by taking the on-board brightness temperature and atmospheric
water vapor content of MODIS 29th, 31st and 32nd bands in the training data set as the
input nodes of the neural network and the near-surface air temperature as the output node;
2-3) input the on-board brightness temperature and atmospheric water vapor content of the
test data set into the trained neural network, and output the near-surface air temperature;
2-4) compare the near-surface air temperature output in 2-3 with the corresponding near
surface air temperature;
step 3: inversion of near-surface air temperature
3-1) read the 29th, 31st and 32nd bands of MODIS remote sensing image data and
atmospheric water vapor content data;
3-2) convert the on-board brightness of the 29th, 31st and 32nd bands of MODIS data into
on-board brightness temperature (T29, T31 and T32) and extract the corresponding
atmospheric water vapor content W;
3-3) input T29, T31, T32 and W in 3-2 into the neural network trained in the second step,
and output the near-surface air temperature (NSAT);
3-4) carry out relevant verification and application analysis according to the corresponding
surface of the image.
2.The method according to claim 1,characterized in that: in step 1-3 of the first step, the
change range of near-surface air temperature is NSAT<LST+15 K, and the step change
range in the simulation process is 2 K.
3.The method according to claim 1,characterized in that: in step 1-4 of the first step, the
initial value of atmospheric water vapor content is the read atmospheric water vapor
content (w), the limited range is w-0.13wswsw+.13w, and the step change range in the
simulation process is 0.2 g/cm2 .
4.The method according to claim 1,characterized in that:in step 1-5 of the first step, the
input MODIS satellite sensor height is 705 KM.
5.The method according to claim 1,characterized in that: in step 2-4 of the second step, the
standard error of near-surface air temperature is greater than 2 K, so both layers of hidden
nodes are increased by 5, and training and test are continued by repeating 2-2 until the
standard error of near-surface air temperature is less than 2 K.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| AU2021105183A AU2021105183A4 (en) | 2021-08-11 | 2021-08-11 | A Method for Estimating Near-surface Air Temperature From Remote Sensing Data Based on Machine Learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| AU2021105183A AU2021105183A4 (en) | 2021-08-11 | 2021-08-11 | A Method for Estimating Near-surface Air Temperature From Remote Sensing Data Based on Machine Learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| AU2021105183A4 true AU2021105183A4 (en) | 2021-11-18 |
Family
ID=78513857
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| AU2021105183A Ceased AU2021105183A4 (en) | 2021-08-11 | 2021-08-11 | A Method for Estimating Near-surface Air Temperature From Remote Sensing Data Based on Machine Learning |
Country Status (1)
| Country | Link |
|---|---|
| AU (1) | AU2021105183A4 (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114414090A (en) * | 2021-12-14 | 2022-04-29 | 厦门大学 | Surface temperature prediction method and system based on remote sensing image and multilayer sensing |
| CN116030354A (en) * | 2023-03-29 | 2023-04-28 | 东华理工大学南昌校区 | Geological disaster analysis method and system based on remote sensing data fusion |
| CN117539855A (en) * | 2023-11-16 | 2024-02-09 | 中国科学院东北地理与农业生态研究所 | Method for extracting lake surface temperature by using MODIS |
-
2021
- 2021-08-11 AU AU2021105183A patent/AU2021105183A4/en not_active Ceased
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114414090A (en) * | 2021-12-14 | 2022-04-29 | 厦门大学 | Surface temperature prediction method and system based on remote sensing image and multilayer sensing |
| CN116030354A (en) * | 2023-03-29 | 2023-04-28 | 东华理工大学南昌校区 | Geological disaster analysis method and system based on remote sensing data fusion |
| CN117539855A (en) * | 2023-11-16 | 2024-02-09 | 中国科学院东北地理与农业生态研究所 | Method for extracting lake surface temperature by using MODIS |
| CN117539855B (en) * | 2023-11-16 | 2024-06-11 | 中国科学院东北地理与农业生态研究所 | Method for extracting lake surface temperature by using MODIS |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2021105183A4 (en) | A Method for Estimating Near-surface Air Temperature From Remote Sensing Data Based on Machine Learning | |
| Jiang et al. | A deep learning algorithm to estimate hourly global solar radiation from geostationary satellite data | |
| CN101634711B (en) | Method for estimating temperature of near-surface air from MODIS data | |
| Palmate et al. | Climate change impact on forest cover and vegetation in Betwa Basin, India | |
| Freitas et al. | Quantifying the uncertainty of land surface temperature retrievals from SEVIRI/Meteosat | |
| Chen et al. | A simple retrieval method of land surface temperature from AMSR-E passive microwave data—A case study over Southern China during the strong snow disaster of 2008 | |
| AU2021105287A4 (en) | Method for Simultaneously Retrieving Surface Temperature and Emissivity from Remote Sensing Data Based on Deep Learning | |
| CN111024618A (en) | Water quality health monitoring method and device based on remote sensing image and storage medium | |
| CN101320475B (en) | Operating range estimation method of infrared imaging system under complex background condition | |
| CN101295022A (en) | Method for ground surface temperature and emissivity inversion by remote sensing data ASTER | |
| CN110595968B (en) | PM2.5 concentration estimation method based on geostationary orbit satellite | |
| Sun et al. | Land surface temperature estimation from the next generation of Geostationary Operational Environmental Satellites: GOES M–Q | |
| AU2021106083A4 (en) | Method for retrieving atmospheric water vapor content based on deep learning and remote sensing data | |
| CN101936877A (en) | Method for inverting atmospheric water vapor content from MODIS (Moderate Resolution Imaging Spectroradiometer) data | |
| Mao et al. | Retrieval of land-surface temperature from AMSR2 data using a deep dynamic learning neural network | |
| Lim et al. | Tropical cyclone forecasts impact assessment from the assimilation of hourly visible, shortwave, and clear-air water vapor atmospheric motion vectors in HWRF | |
| CN101655564A (en) | Method for inversing surface temperature and emissivity from MODIS data | |
| CN114005048A (en) | Multi-temporal data-based land cover change and thermal environment influence research method | |
| CN113447137A (en) | Surface temperature inversion method for unmanned aerial vehicle broadband thermal imager | |
| He et al. | Direct estimation of land surface albedo from simultaneous MISR data | |
| AU2021105233A4 (en) | Method of Retrieving Surface Temperature from Passive Microwave Remote Sensing Data AMSR E | |
| CN114357885B (en) | Photosynthetic effective radiation scattering proportion prediction method fusing multisource data | |
| CN113779863B (en) | Ground surface temperature downscaling method based on data mining | |
| CN105403201A (en) | Pixel decomposition-based remote sensing image atmosphere path radiance acquiring method | |
| Bagis et al. | An adaptive spatiotemporal agricultural cropland temperature prediction system based on ground and satellite measurements |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| AMFD | Amendment of filing date |
Effective date: 20210811 |
|
| FGI | Letters patent sealed or granted (innovation patent) | ||
| MK22 | Patent ceased section 143a(d), or expired - non payment of renewal fee or expiry |