What is the purpose of the web app?
What is the aim of the app?
The aim of this web app is to provide a portal to allow cane growers to insert their information based upon location using the google map.
Why is the information needed?
This information will be used to verify the results and improve the models. Without collecting real world data in this way it is impossible to know if the results are accurate.
What is real world data?
Real world data is data collected from sources that are associated with outcomes. In this case data is being collected from the cane growers.
How was the analysis done?
Using Sentinel 2 satellite images a harvest was deemed to have occurred if the green value went below the threshold of 0.32 during the harvest months (June to November inclusive).
How was the green value calculated?
The green value was calculated for each pixel (10 square meters) using RBG colour: GREEN / (RED + BLUE + GREEN)
Is 0.32 the correct value for the threshold?
Looking at the images 0.32 appears to be a good choice as the threshold, however this must be verified by real world data.
Is using the threshold for the green value the best way to determine if a harvest has occurred?
Maybe but other solutions could be possible. For instance a drop in the green value would also be expected. The data was analysed for the first drop in the green value during the harvest period however the results did not show the same clarity as the green threshold model.
Could the drop in green value model be improved?
Yes, the results showed that a threshold also needs to be used for this model. In addition fields that have been newly planted need to be removed before running the analysis.
Why wasn't this a problem for the threshold model?
Initially it was not a problem as many of these pixels did not have a value that went below 0.32 during the harvest period - 192154 or 5.5% in 2017 but only 52,381 or 1.5% in 2018. However on further analysis it may be better to remove the newly planted field pixels before checking if a pixel has dropped below the threshold.
Can newly planted fields be separated from ratoon fields using the satellite images?
To assess if a field was a new planting or a subsequent ratoon crop the minimum green value for the year was determined. If this occurred between January and April then a field was fallow at this time and subsequently planted with sugar cane that would not be harvested until the year after. If the minimum occurred between June and November then a harvest had occurred.
In 2017, 26.4% pixels did have a minimum value prior to May, with 13.8% in 2018. These values are higher than was calculated above for the 0.32 green value threshold. This suggests that the 0.32 threshold may be too low.
Is June to November the correct harvest period?
Although most sugar cane is harvested between July and November, June was included within the harvest time as there were reports of harvesting in June 2018.
Why are the harvest results given monthly?
The satellite images were mostly 10 to 14 days apart however often cloud and shadow prevented the reading of the RGB for a pixel. This makes the harvest date inaccurate and can cause a field to have several colors in the images.
How can you tell if the sugar cane has been burnt before harvesting?
To determine if the sugar cane has been burnt prior to harvesting the normalised burn ratio (NBR) was used:
NBR = (NIR-SWIR)/(NIR+SWIR)
This was calculated for each date and then the post harvest NBR was deducted from the pre harvest NBR. An increase in the NBR indicates a burnt area.
What is NIR and SWIR?
NIR is near infrared and SWIR is short wavelength infrared. Band 4 of the satellite images provide a NIR reading. As both band 11 and 12 provided SWIR data, the analysis was performed on both these sets of tiles. Band 12 proved to be a more sensitive than Band 11. Verification is needed to confirm which is the better band to use.
Is there a threshold for the NBR?
This reference for the NBR reports a threshold increase in the NBR value above 0.1 to be an indicator of a burn. However this was based upon fires in the United States. Use within sugar cane areas needs validation.
How can satellite images show quality of sugar cane as it is growing?
Healthy vegetation reflects more near-infrared (NIR) and green light compared to other wavelengths. But it absorbs more red and blue light.
This is measured using the Normalized Difference Vegetation Index, which calculates the difference between near-infrared (which vegetation strongly reflects) and red light (which vegetation absorbs) using this equation: NDVI = (NIR-Red)/(NIR+Red)
How do you get the NIR and Red values?
Red is in Band 4 and NIR is provided by Band 8 in the Sentinel 2 images.
This article explains the process.