CGMS Anhui & Yield estimation with RS CGMS Anhui & Yield estimation with RS
Our work We participate the following work package WP 21 Ground data collection WP 24 CGMS pilot in CHINA WP 41 Official yield data collection WP 44 Wheat Yield estimation based on remote sensing for HUAIBEI Plain WP 7 Networking and Sustainable partnership
Our work Organizing a CGMS workshop in China, 2011
Part 1 CGMS-Anhui
Level 1 Weather Station
Level 1 Weather Station
Level 1 Update the METDATA The data from meteorological department (Archive data, from 1990 to 2012) The data from NOAA GSOD, now we can download the real-time data from the NOAA GSOD FTP everyday.
Level 1 Interpolation Grid Weather The batch model give us a easy way to interpolation weather
Level 1 Grid Weather (Average daily temperature, 31/12/2012 )
Level 2 Crop simulation- using the batch model 1.Calculate the crop yield 2. Aggregation the grid yield
Level 3 Yield Forecast 1.Aggregation the Nuts yield 2. Prepare for forecast
Level 3 CGMS Statistical Tool
Level 3 CGMS Statistical Tool But only use the potential yield storage to estimate the crop yield, the result is not very good
Resent works and further works Update the CGMS dataset Prepared some NDVI, fAPAR and DMP data, plan to add these data into CST Integrate the CGMS Anhui (further work)
Part2 Prediction of Wheat Yields Using Multiple Linear Regression Models in the Huaibei Plain of China Beier Zhang (AIER - China ) Qinhan Dong (VITO - Belgium)
Contents Study area Phenology Trends of yields Data sets and methods Results of prediction Validation Discussions
Study area Huaibei Plain (include 6 prefectures) Area:64154 km 2 Arable area: km 2 Main soil type :Cambosols & Vertisols Main crop type: Winter wheat & Maize
Phenology SowingEmergenceTiller Wintering period Turning green JointingHeadingMaturityHarvest 10/1210/1912/112/202/103/104/225/156/1 Wheat: October to next year June Maize or soybeans: June to October
Trends of yields the trend must be considered There are significant yearly trend of wheat yield in every prefectures from 2000 to 2011, so the trend must be considered in the prediction
Data sets I. Biophysical variables based on RS: using SPOT-VGT Ten-daily series : every dekad from 1999 to 2012 Variables: Smoothed k-NDVI Building data sets of RS: The cumulative NDVI for all possible combinations (at least 2, at most 9, because the one phenological stage is less than 3 month) of consecutive dekads within the wheat growing period (2 nd dekad of Oct to 3 rd dekad of Jun).
Data sets For example YearO2O3……O2O3O3N1……O2N1O3N2……
Data sets III. Meteorology data sets Variables: include rainfall, temperature and duration of sunshine, from 1999 to 2012 Interpolation method: CGMS Level-1 give us the values of every grid (25km x 25km) in the study area. Calculate average values in every prefecture Building data sets of Meteorology data sets: The average rainfall, temperature and solar radiation of every phonelogical stage of wheat in every prefecture.
Methods Detrend method. We use two different methods 1.Add year as a variables into the model. 2.Separate the trend yield from real yield, and build the regression model with ΣNDVI and residual error First predicting the yield using regression to obtain the inter-annual trend (P T ) Calculate the residual error (official yield - P T ) Using ΣNDVI and meteorology data to predicting residual error(P R ) P T +P R
Methods Precision validation Leave-one-out (leave one year data out; regression model building using the rest of data to predict the left year; corellating the official yield with the predicted ones)
Results Prefecture Models R2R2 Absolute Error ConstantyearΣNDVIMeteorology Bengbu *year+2.099*O2N *F2F Bozhou *year+2.417*N Fuyang *year+4.606*O3N *Sg Huaibei *year+0.621*J3M *Sj Huainan *year+0.287*Sg+0.296*Em Suzhou *year+0.527*J2M *Es Regression models Using year , k-NDVI , and Meteorology Data
Results-detrend PrefectureTrend ModelR2R2 Bengbu *year0.667 Bozhou *year0.912 Fuyang *year0.741 Huaibei *year0.838 Huainan *year0.751 Suzhou *year0.754 Trend
Results-detrend Prefecture Models R2R2 ConstantΣNDVIMeteorology Bengbu *M *Sj+0.201*Em+0.293*Ee *Sm Bozhou *Sm+0.101*Em+0.298*Ew Fuyang *O3N *Sg0.681 Huaibei *J3M *Sm-0.152*Se0.585 Huainan *N *Sg+0.283*Em0.736 Suzhou *O *M *Ee0.592 Regression models after detrend
Validation Using Jack-knife method, comparing absolute error of different methods Prefecture Year & k-NDVI &Meteorology k-NDVI & Meteorology after Detrend Bengbu Bozhou Fuyang Huaibei Huainan Suzhou Mean Error
Validation After detrend Bengbu
Validation After detrend Bozhou
Validation After detrend Fuyang
Validation After detrend Huaibei
Validation After detrend Huainan
Validation After detrend Suzhou
Validation An example,If we want to estimate the yield of Building the trend model using the data from 2000 to 2011 Calculating residual error. Building the model using the above variables. Then Calculating the yield.
Validation The result of year PrefectureReal yield Estimate yieldabsolute error Bengbu Bozhou Fuyang Huaibei Huainan Suzhou Mean Error 0.31
Discussions The method We think the method using k-NDVI & Meteorology after detrend is better This method consider the fact of yield trend, RS and Meteorology. The average error of six prefecture in Huaibei Plain is about ton per ha, this is a quite good result.
Discussions Suggestion for further study Add these data into CST Add a new dataset from CGMS Level2 Do some field work, get the real crop yield about the field level, then build the model of this level. This work I think can adjust our method and make the result more accurately.
Thanks Grazie Merci Bedankt شكرا 谢谢