2011-7-29 jianqiang ren 1,2, zhongxin chen 1,2, huajun tang 1,2, fushui yu 1,2, qing huang 1,2...
TRANSCRIPT
2011-7-292011-7-29
Jianqiang RENJianqiang REN1,21,2, Zhongxin CHEN, Zhongxin CHEN1,21,2, Huajun TANG, Huajun TANG1,21,2, , Fushui YUFushui YU1,21,2, Qing HUANG, Qing HUANG1,21,2
不同长势冬小麦乳熟期冠层反射率2008 5 15年 月 日
0. 0
0. 1
0. 1
0. 2
0. 2
0. 3
0. 3
0. 4
0. 4
300 400 500 600 700 800 900 1000 1100
nm波长( )
相对
反射
率
N0 N1 N2 N3
Simulation of regional winter wheat yield Simulation of regional winter wheat yield by combining EPIC model and by combining EPIC model and
remotely sensed LAI based on remotely sensed LAI based on global optimization algorithmglobal optimization algorithm
1 Key Laboratory of Resources Remote-Sensing & Digital Agriculture, Ministry of Agriculture, China
2 Institute of Agricultural Resources & Regional Planning, Chinese Academy of Agricultural Sciences
1/24
Introduction1
Study area2
Method33
Data preparation44
Results and analysis5
Conclusions and future work6
Outline
2/24
1. Introduction
Crop yield information is critical to food security early warning
in a country or a region
Traditional crop yield forecasting methods
• agricultural statistical methods
• agricultural forecasting method
• climate model method
Main remote sensing models for crop yield estimation
• empirical model
• semi-empirical model
• crop growth mechanism model 3/24
1. Introduction
Combining RS data and crop growth model to simulate
crop growth and crop yield has been becoming important
research field
crop growth model: strong mechanism and time continuity
remote sensing: real-time features and spatial continuity
crop growth model + RS: strong mechanism + time/spatial
continuity
4/24
1. Introduction
The way of combining RS data with crop growth model
forcing strategy (Easy)
time series variable of crop model (such as LAI) retrieved from remote
sensing data was input into model directly
initialization/parametrization strategy (Complex)
responding parameters and initial values were optimum
• when the difference between simulated crop parameter and
related remote sensing data reached the minimum value (relative
complex)
• or when the difference between simulated reflectance and remote
sensing reflectance (most complex)
1. Introduction
The choice of optimization algorithm is critical to the
accuracy of simulation results, general methods include:
simulate anneal arithmetic
genetic algorithms
neural networks, etc
SCE-UA (Shuffled Complex Evolution method - University of Arizona)
developed by Q.Y. Duan at University of Arizona (Duan, 1993)
could improve accuracy and efficiency of crop growth
monitoring and yield forecasting (Zhao, 2005; Qin, 2006)
5/24
2. Study area
Study area E115.19 °– 116.53 °, N37.09 °– 38.36 °
includes 11 counties covering about 8815 km2
located in Hengshui City, Hebei Province, which is a
part of Huanghuaihai Plain in North China
Climate
temperate, semiarid, semi-humid and continental
monsoon climate.
Cropping system Winter wheat-summer maize (dominant double
cropping system )
Winter wheat : sowed (3rd 10-day of September----2nd 10-day of October)
mature (1st 10-day of June ----- 2nd 10-day of this month)
Ground survey plots: 75 in the year of 2004 and 2008
29 survey plots (in 2004) and 46 plots (in 2008) 6/24
3. Method
Flowchart of this research
Sensitivity analysis
Calibration of parameters
Elemental mapping unit (EMU)
Preparation of the average data
in each unit
When simulate
optimization object:
the simulated LAI
optimized parameters
planting date of crop, net N
fertilizer application rate and
planting density.
7/24
3. Method
3.1 Crop growth model EPIC (Environmental Policy Integrated Climate)
developed to assess the effect of soil erosion on soil productivity
by USDA in 1984.
Suitable to most of all crop simulation and needs daily climate
data as driver parameters (solar radiation, max. temperature, mini.
temperature and precipitation……)
Basic formula in EPIC model
ttt LAIRAIPAR 65.0exp15.0
AGYIELD HI B REG t BE IPAR t dt Where IPAR is intercepted photosynthetically active radiation; RA is solar radiation; BE is the crop parameter for converting energy to biomass; REG is the
value of the minimum crop stress factor; BAG is the aboveground biomass in T/Ha
for crop; HI is the harvest index
8/24
3.2. Global optimization algorithm SCE-UA (Duan, 1994)
3. Method
an efficient and global optimization algorithm not sensitive to parameter initialization value avoids optimization process relying on the prior knowledge the objective function as follows:
Where LAIsimi was simulated LAI; LAIobsi was remotely sensed LAI; n
was the number of EMU.
2
1
n
simi obsii
y LAI LAI
9/24
3. Method
3.3. Model parameters calibration
10/24
Parameters impacting the accuracy of simulated yield (Wu,2009) WA (potential radiation use efficiency)
HI (normal harvest index)
DMLA (maximum potential leaf area index)
DLAI (point in the growing season when leaf area begins to decline due to leaf senescence)
DLP1 (crop parameter control leaf area growth of the crop under non-stressed condition)
DLP2 (crop parameter control leaf area growth of the crop under non-stressed condition)
RLAD (leaf-area-index decline rate parameter)
WA and HI: most key parameters which affected the model
localization and the accuracy of simulated yield (Wu,2009).
Other parameters: strongly influenced by crop varieties and difficult to
obtain in a large region.
3. Method
3.3. Model parameters calibration
11/24
3. Method
3.4. Model assimilation parameters
The accuracy of derived leaf area index had an important impact on crop final estimated yield.
We selected the simulated LAI as the optimized object
The parameters such as DMLA, DLAI, DLP1, DLP2, RLAD, crop planting date, plant density and amount of nitrogen fertilization have significant effects on the change of simulated LAI value (Clevers, 1996)
we selected the above parameters as optimization parameters for leaf area index simulation.
12/24
3. Method
3.5. Validation of results
Simulated crop yield
validated by the statistical crop yield data at county level;
Simulated crop management information
validated by the regional average information coming from each field survey plot because the custom field management was more stable in China.
statistical parameters
Root Mean Square Error (RMSE)
Coefficient of determination (R2)
Relative Error and Absolute Error
13/24
4. Data preparation
4.1. Basic data collection and process
Station climate data solar radiation, maximum temperature, minimum temperature,
precipitation, relative humidity and wind speed
interpolated at resolution of 250m using Kriging method
3rd 10-day of September, 2007 ---2nd 10-day of June, 2008
Soil map data (1:4,000,000)
soil depth, soil texture, bulk density, soil pH, organic carbon concentration
and calcium carbonate content of soil, etc
Field management data
planting date, harvesting date, fertilizer application rate, irrigation volume
and plant population, etc
14/24
4. Data preparation
4.2. Field observation
75 sampled plots
in the year of 2004 and 2008, more than 500m * 500m.
The number of sample sites was no less than 3 at each sample plot.
LAI measurement
manually at each growth stage. In each plot the average LAI of all
sampling sites was regarded as the final LAI value.
Yield
measured at ripening stage and the average yield of all sampling sites
was the final field-measured yield.
Field management information collection
planting date, plant density, net N fertilization application rate were
collected in each plot.
15/24
4. Data preparation
4.3. LAI retrieved from MODIS NDVI
Basic data
250m 16 day MODIS NDVI data downloaded from NASA website (273rd day,2007 to 161st day, 2008)
field measured LAI in each growth stage
Method
Neural Network method.
Validation
Relative error of simulated LAI was less than 5%
RMSE (0.29~1)
Result of remotely sensed LAI
(2008, 113 rd day)
16/24
4. Data preparation
4.4. Other auxiliary data
Crop map
provided by Key Laboratory of Resources Remote-Sensing &
Digital Agriculture, Ministry of Agriculture of China
Crop statistical yields at county level (2008)
provided by Agricultural Bureau of Hengshui City
17/24
5. Results and analysis
5.1. Result of simulated sowing date of winter wheat (2007)
Regional average simulated sowing date was the 290th day (Oct. 17, 2007)
Average field-investigated sowing date was the 289th day (Oct. 16, 2007).
Absolute error was only 1 day.
18/24
5. Results and analysis
5.2. Result of simulated plant density of winter wheat (2008)
Regional investigated plant density was 460.2 plants/m2
Average simulated plant density was 423.6 plants/m2
Average relative error of simulated plant density was -7.95%19/24
5. Results and analysis
5.3. Result of simulated N fertilization application rate (2008)
Mean simulated amount of net N fertilization was 270.34 kg/ha
Mean custom amount of ground survey was 296.70 kg/ha
Relative error of simulated net N fertilization application rate was -8.88%. 20/24
5. Results and analysis
5.4. Result of simulated winter wheat yield (2008)
5. 00
5. 20
5. 40
5. 60
5. 80
6. 00
6. 20
6. 40
6. 60
5. 00 5. 20 5. 40 5. 60 5. 80 6. 00 6. 20 6. 40 6. 60
Stat i st i c yi el d of wi nter wheat (t / ha)
Simu
late
d yi
eld
of w
inte
r wh
eat(
t/ha
)
y = 1. 0336x - 0. 0903
R2 = 0. 6436RMSE = 0. 208t / ha
Mean simulated yield was 5.94 t/ha
Relative error of simulated yield was 1.81%
RMSE of yield estimation was 0.208 t/ha 21/24
6. Conclusions and future work
(1) Comparing with the statistical data or the investigated data, we got
better simulated results which could meet the need of accuracy of
agricultural remote sensing monitoring.
(2) It was possible and feasible to estimate crop yield and simulate
regional crop growth and field management parameters through
integrating remotely sensed LAI into crop growth model.
(3) These above work had setup good foundation for further use of this
method to predict crop yield at larger region in China.
6.1 Conclusions
22/24
6. Conclusions and future work
(1)To expand application of the research method in a larger
region or the whole China
(2)To carry out research of grid cell size optimization in China
Considered the running efficiency and simulation accuracy, the optimal
grid cell size for provincial or national yield estimation should be studied
further.
(3) To carry out deeply research of other outer assimilation data
LAI was only considered as outer assimilation data, the NDVI, EVI or ET
etc would be used as outer assimilation data for further study.
6.2 Future work ( Discussion )
23/24