how i can design a research project using remote sensing to predict yield of wheat
I believe that you can use remote sensing to well predict the yield of wheat but it depends on your research scale as different scales of research require different resolution data. If your study area is a pilot site of about thousands of hectares, you may use very high resolution data such as QuickBird, IKONOS , WorldView, GeoEye, SPOT/Pleiades or RapidEye, etc., if your research is sub-national to national scale, you may consider Landsat data, if it is regional or continental scale, you should use multi-resolution data such Landsat and MODIS data. You have to define you research scale.
The importance of your work is to build remote sensing-based yield models using vegetation indices (VIs) or LAI. For the time being, there is no universally applicable yield model due to difference in soil and climate conditions worldwide. So you have to develop models relevant to your research area. To do so, if you have not spectroradiometer, you can acquire a satellite image at peak greenness of wheat (before flowering), and then conduct some preprocessing such as atmospheric correction and multispectral transformation to derive LAI and VIs such as NDVI, EVI, SARVI and OSAVI. At the same time, you go to field to select about 36 (or as many as you can) plots (plot siez: 1m*1m) respectively distributed in poorly, moderately, well performed wheat area. While harvesting is conducted, you should note the yield of each of these plots. Then you can calibrate the yield of part of these plots, e.g., 24 plots representing different performance, with their corresponding VIs (NDVI, EVI, SARVI, OSAVI etc.) or LAI values you have produced either by linear or nonlinear models. After obtaining the models, you have to validate them using the remained 12 plots (those were not used yet for modeling) to see whether your model(s) can predict yield accurately at the known plots. You can also convert the unit of the developed models in g/m2 into t/ha.
After validation of your model(s), you should identify the wheat plantation in your whole research area by classification approach (usually supervised classification). You can mask out the non-wheat area in the VIs or LAI images, then apply models to the VIs or LAI images to obtain pixel-based yield image of your whole study area. If you want to know the total wheat production in your study area, you just need to carry out a statistical analysis but depending on which software you will use. You can also export the yield image as a tif image and then import it in ArcGIS using zonal statistics to get the total yield.
If your study area is a large region or continent, it will involve a multiscale approach to extend your models from local sites to regional scale using multiresolution data. If this is your case, you may contact me for further suggestion.
Have a good try.
Laston,
Multi-spectral images were used for the assessment of crop yield estimation for long time by using NDVI and LAI.
I think this link will give you the exact idea, merit and its comparison with real scenario.
http://blogs.worldbank.org/impactevaluations/measuring-yields-space
Probably one of the oldest applications of RS dating back to the early years (1980s) of satellite imagery. In broad terms: NDVI (green biomass proxy) as obtained with optical sensors, field calibration of the mid-resolution imagery either with clipped green biomass or ground reflectance measurement at the spectral bands of the imagery. The challenges are cloud free imagery at the relevant dates and field sizes in relation to surrounding vegetation (large fields in flat terrain like in the west of the USA or the Ukraine are easy), small fields in Tunesia proofed to be more complex. NDVI measurement over the growing season at coarse spatial, but fine temporal resolution imagery (e.g. MODIS) could be even better.
There is no full-prove system that I am aware of. The technology, if exist, will be too restrictive to be of any practical use.
I believe that you can use remote sensing to well predict the yield of wheat but it depends on your research scale as different scales of research require different resolution data. If your study area is a pilot site of about thousands of hectares, you may use very high resolution data such as QuickBird, IKONOS , WorldView, GeoEye, SPOT/Pleiades or RapidEye, etc., if your research is sub-national to national scale, you may consider Landsat data, if it is regional or continental scale, you should use multi-resolution data such Landsat and MODIS data. You have to define you research scale.
The importance of your work is to build remote sensing-based yield models using vegetation indices (VIs) or LAI. For the time being, there is no universally applicable yield model due to difference in soil and climate conditions worldwide. So you have to develop models relevant to your research area. To do so, if you have not spectroradiometer, you can acquire a satellite image at peak greenness of wheat (before flowering), and then conduct some preprocessing such as atmospheric correction and multispectral transformation to derive LAI and VIs such as NDVI, EVI, SARVI and OSAVI. At the same time, you go to field to select about 36 (or as many as you can) plots (plot siez: 1m*1m) respectively distributed in poorly, moderately, well performed wheat area. While harvesting is conducted, you should note the yield of each of these plots. Then you can calibrate the yield of part of these plots, e.g., 24 plots representing different performance, with their corresponding VIs (NDVI, EVI, SARVI, OSAVI etc.) or LAI values you have produced either by linear or nonlinear models. After obtaining the models, you have to validate them using the remained 12 plots (those were not used yet for modeling) to see whether your model(s) can predict yield accurately at the known plots. You can also convert the unit of the developed models in g/m2 into t/ha.
After validation of your model(s), you should identify the wheat plantation in your whole research area by classification approach (usually supervised classification). You can mask out the non-wheat area in the VIs or LAI images, then apply models to the VIs or LAI images to obtain pixel-based yield image of your whole study area. If you want to know the total wheat production in your study area, you just need to carry out a statistical analysis but depending on which software you will use. You can also export the yield image as a tif image and then import it in ArcGIS using zonal statistics to get the total yield.
If your study area is a large region or continent, it will involve a multiscale approach to extend your models from local sites to regional scale using multiresolution data. If this is your case, you may contact me for further suggestion.
Have a good try.
We have found a fairly acceptable and validated result by correlating cumulative crop water use or ET to yield using Landsat and surface energy balance equation.
use this link, blogs.worldbank.org/impactevaluations/measuring-yields-space
You will have to first tell apart the wheat crop from other crops or grasses if your study area is a large region or continent. Then you can make a yield model by region using remote sensing index such as NDVI,EVI or mixed remote sensing index with the meteorological data.
I suggest to look at the initiative/methodology of SUPARCO (Space Agency of Pakistan) for the crop yield estimation using RS
Dear, Mr. Laston
To be able to estimate the harvest of wheat seems to me almost the same as the estimated harvest in my country. You can use remote sensing technology combined with field research. Some things that you need to master is:
1. generative and vegetative phase of wheat; the generative phase is usually marked by the growth of leaves and ovaries. In remote sensing techniques can usually be seen the reflection of the pixel values of the object / value (the natural color is usually marked with green color), on Landsat imagery can use a combination of the band 321, or 432 band.
2. Phase vegetative phase of maturity or wheat grain filling. on remote sensing technology is usually represented by the color yellow.
3. You can also using NDVI algorithms to estimate the harvest of wheat where NDVI formulated using Landsat imagery: (Band3 - Band4 / Band3 + Band4), provided that:
a. NDVI value is low (even negative) in the early crops (replanting of wheat).
b. High NDVI value during the generative because of high chlorophyll contains
c. NDVI values lower at harvest Wheat (wheat growth to the NDVI values shaped parabolic curve)
4. By using crop cutting experiments 2,5x2,5 or equal to 3 x 3 pixels (1 ha) carried out measurements in the field, then the production of wheat can be formulated using linear regression (Ordinary Least Square) as follows:
Production (Wheat) / Ton / ha = a (NDVI) + b
a = constant
b = Constants
Ok....Good Luck
Sir,
May I know what do you mean by the constants a and b here? what are the values to these constants.
I consider developing the model with your region environment is very important, and the next thing is to train more actual yields and correlated images.
- Badges
- Science topic
- Similar topics
- Geoscience
- Remote Sensing