🎯 Key Takeaways for quick navigation: 00:02 🌾 Introduction to Rice Mapping Process - Overview of using Sentinel 1 and 2 in Google Earth Engine (GEE) for rice mapping. - Explanation of the process, including monthly data acquisition and clustering. 02:04 🗺️ Selecting Region of Interest - Description of how to select the region of interest in GEE. - Importance of choosing a small area for smooth code execution. 04:02 📅 Filtering Sentinel 2 Images - Filtering Sentinel 2 images based on start and end dates. - Managing metadata like Cloudy pixels for image selection. 08:08 🌱 Calculating NDVI - Explanation of calculating NDVI (Normalized Difference Vegetation Index). - Renaming bands and adding the NDVI to image collection. 12:45 📊 Generating Monthly NDVI Images - Creating monthly NDVI images for the rice growing season. - Using a sequence to obtain a series of monthly images. 17:06 🧩 Combining Sentinel 1 and Sentinel 2 Data - Combining Sentinel 1 and Sentinel 2 data into one stacked image. - Preparing the data for clustering. 19:08 📊 Clustering for Unsupervised Classification - Clustering the combined data into 30 clusters. - Assigning values to clusters based on visual inspection. 25:21 🗺️ Assigning Rice and Non-Rice Labels - Reassigning values to clusters to identify rice and non-rice areas. - Mapping the clusters for rice mapping. 28:57 🌾 Comparing Results with Rice Maps - Comparing the generated rice map with a rice map from MODIS data. - Discussion on the accuracy of the generated rice map. Made with HARPA AI
Thank you very much for your wonderful demonstration. However, I encountered a problem: some clusters were not shown on the graph. for example, I used 30 clusters for my study area but the graph showed only 0, 1, 7, 13 and 16 clusters. May I have your advice to address this issue?
If you are asking for the MODIS rice map, this is from the article, www.sciencedirect.com/science/article/abs/pii/S0168192321002227. Request them to provide you the maps, but they are mainly for Nepal.
Hello, congratulations, interesting work. I wonder how clusters present different stages of rice? And at the same time, each pixel of this cluster can be taken as a 10x10 area, and how should we interpret the temporal part with the random colors?
As per my understanding, the clusters are formed by the temporal similarity as well. The pixels which spectral profile were similar all over the period / due to their closeness were clustered as a cluster. This would mean the rice that were transplanted together or grew together. Clusters are not directly representing the stages of rice. The spectral chart we are observing represents the stages of rice. The idea is, as shared in the video, the clusters showing some distinct spectral curve distinct to the rice should be the presence of the rice. The observations of more and more clusters will probably refine the rice growing area But large no. of clusters would rather make obsolete for the observations . Regarding 10 * 10m - what's being shown by the pixels can not directly represent the field. It would be the general representation. Hope this helps, or correct if it doesn't seem logical. Thankyou for the question.
thank you for explanation, in the end fo this code are displaying 3 maps: remapped_cluster, remapped_cluster10, and remapped_cluster20; these 3 maps are composite from startdate and enddate, this correct?. so how to display remapped_cluster20 in certain month? thank you sir
Yes, the three clusters are from the start date and end date, just changing the discrete values, and we are seeing them to get the idea of clusters. Clusters only of the certain one month is not possible there. And also not meaningful. Although, we can simply try to use the images only from the month and then start clustering. It also depends on the number of images within one month. Thankyou
100 is just the value originally used in the paper itself. Using 100 would just mean there would be more Sentinel 2 images in the region. It is okay if the area usually has the lower clouds, or the region is smaller such that it won't be greatly shadowed by the clouds. Other than this, there are no such specific reasons for this. Thankyou
thanks your reply, again sentinel1 we do not need any further processing like correction or calculating coefficients, you just download and use them directly? I am curiosity?
If we say about the data from the earth engine, all required processing have been done - readily available for our analysis. So we can directly use it. If in other cases, if we are already more aware about what we need based on our purpose OR if we did not like the algorithms used in the earth engine, we can certainly try in the SNAP (which is the other freely available remote sensing software). But, more confusingly, sentinel-1 also has other type of data that is SLC type, along with GRD which we see in the earth engine, which might certainly requires other steps, which is complex than what we need. SAR data are complex, and the steps of the corrections or processing or any steps are the complex topic to learn or developing our own steps of algorithms might be other topics. There sure are people/resources from whom we can learn more. Anyway, what provided from earth engine is sufficient for us, I am sure. Thankyou
It might depend on what fusion mean. The video can be considered an example of using Sentinel 1 and 2 where two different data are handled by the classification. - We can try this in the SNAP tool box as well. step.esa.int/docs/tutorials/S1TBX%20Synergetic%20use%20of%20S1%20(SAR)%20and%20S2%20(optical)%20data%20Tutorial.pdf There are certain ways/equations that might have been developed integrating vegetation indices with radar indices. I cannot say about that more.
depends, when there is no training data, it will certainly be used. this can be helpful if we tried comparing with the supervised methods. but I cannot say more.
🎯 Key Takeaways for quick navigation:
00:02 🌾 Introduction to Rice Mapping Process
- Overview of using Sentinel 1 and 2 in Google Earth Engine (GEE) for rice mapping.
- Explanation of the process, including monthly data acquisition and clustering.
02:04 🗺️ Selecting Region of Interest
- Description of how to select the region of interest in GEE.
- Importance of choosing a small area for smooth code execution.
04:02 📅 Filtering Sentinel 2 Images
- Filtering Sentinel 2 images based on start and end dates.
- Managing metadata like Cloudy pixels for image selection.
08:08 🌱 Calculating NDVI
- Explanation of calculating NDVI (Normalized Difference Vegetation Index).
- Renaming bands and adding the NDVI to image collection.
12:45 📊 Generating Monthly NDVI Images
- Creating monthly NDVI images for the rice growing season.
- Using a sequence to obtain a series of monthly images.
17:06 🧩 Combining Sentinel 1 and Sentinel 2 Data
- Combining Sentinel 1 and Sentinel 2 data into one stacked image.
- Preparing the data for clustering.
19:08 📊 Clustering for Unsupervised Classification
- Clustering the combined data into 30 clusters.
- Assigning values to clusters based on visual inspection.
25:21 🗺️ Assigning Rice and Non-Rice Labels
- Reassigning values to clusters to identify rice and non-rice areas.
- Mapping the clusters for rice mapping.
28:57 🌾 Comparing Results with Rice Maps
- Comparing the generated rice map with a rice map from MODIS data.
- Discussion on the accuracy of the generated rice map.
Made with HARPA AI
thank you for your video!
Thank you very much, nice
Hi, thanks so much for such a great tutorial! Do you think we can apply it to map other crops such as corn or avocado in Latam?
Yes, absolutely. We need to fully understand the spectral trends of those crops first.
Thank you Brother ❤
thankyou ❤
Thank you very much for your wonderful demonstration. However, I encountered a problem: some clusters were not shown on the graph. for example, I used 30 clusters for my study area but the graph showed only 0, 1, 7, 13 and 16 clusters. May I have your advice to address this issue?
try reducing the study area or number of clusters, or time duration of the images,
Good Explanation, Thanks to make such tutorial, Would you please share MODIS Rice Map Download link?
If you are asking for the MODIS rice map, this is from the article, www.sciencedirect.com/science/article/abs/pii/S0168192321002227. Request them to provide you the maps, but they are mainly for Nepal.
¿Hola, cómo estás? Gracias por compartir, estoy siguiendo todos tus videos tutoriales. Saludos desde México.
hola, gracias por ver el video. me alegra ver el comentario.
Great Tutorial Brother! I want to ask, How do u know about the quantity of the pixel, u said 70 right?
do you mean the clusters, we specify this at the beginning,
70? I might not have said that? Can you point the time where I said that?
Yes, i just got the issue of kappa accuration asses, on that DOI the total pixel are 740 right? I am wondering where that number of pixel came from
Hello, congratulations, interesting work. I wonder how clusters present different stages of rice? And at the same time, each pixel of this cluster can be taken as a 10x10 area, and how should we interpret the temporal part with the random colors?
As per my understanding, the clusters are formed by the temporal similarity as well. The pixels which spectral profile were similar all over the period / due to their closeness were clustered as a cluster. This would mean the rice that were transplanted together or grew together.
Clusters are not directly representing the stages of rice. The spectral chart we are observing represents the stages of rice. The idea is, as shared in the video, the clusters showing some distinct spectral curve distinct to the rice should be the presence of the rice.
The observations of more and more clusters will probably refine the rice growing area But large no. of clusters would rather make obsolete for the observations . Regarding 10 * 10m - what's being shown by the pixels can not directly represent the field. It would be the general representation.
Hope this helps,
or correct if it doesn't seem logical.
Thankyou for the question.
thank you so much...
sir how can i download the clustered ndvi images from GEE
try exporting like other image
thank you for explanation, in the end fo this code are displaying 3 maps: remapped_cluster, remapped_cluster10, and remapped_cluster20; these 3 maps are composite from startdate and enddate, this correct?.
so how to display remapped_cluster20 in certain month?
thank you sir
Yes, the three clusters are from the start date and end date, just changing the discrete values, and we are seeing them to get the idea of clusters.
Clusters only of the certain one month is not possible there. And also not meaningful. Although, we can simply try to use the images only from the month and then start clustering. It also depends on the number of images within one month.
Thankyou
@@ksabmagar7 great, i just whant to know the changing of this cluster10 or cluster20 mont by mont, so we can monitor the stage of rice.
where can I copy the code?
Please check the description. There is a link which lands you to the code.
What is the image he imported? At the beginning
data from sentinel 1 and 2.
hi, thanks your sharing, I have a question, why you set sentinel2 cloudy is less than 100, you want to make up this by sentinel1 or other reasons?
100 is just the value originally used in the paper itself.
Using 100 would just mean there would be more Sentinel 2 images in the region. It is okay if the area usually has the lower clouds, or the region is smaller such that it won't be greatly shadowed by the clouds.
Other than this, there are no such specific reasons for this. Thankyou
Thanks a lot Sir
I am from Iraq, I need your code to apply it in our region
Your help will be highly appreciated
Hello sir, please check the description for the codes.
Thank you Brother , I have a question,cluster number zero in the chart to which attributes we can their value
we have to look at how the line graph shows for the cluster number zero, we cannot be specific for sure.
Very nice
thankyou sir.
Hello, do you provide private advice? And if so, any contact email?
There are no such things from my side.
thanks your reply, again sentinel1 we do not need any further processing like correction or calculating coefficients, you just download and use them directly? I am curiosity?
If we say about the data from the earth engine, all required processing have been done - readily available for our analysis. So we can directly use it.
If in other cases, if we are already more aware about what we need based on our purpose OR if we did not like the algorithms used in the earth engine, we can certainly try in the SNAP (which is the other freely available remote sensing software).
But, more confusingly, sentinel-1 also has other type of data that is SLC type, along with GRD which we see in the earth engine, which might certainly requires other steps, which is complex than what we need.
SAR data are complex, and the steps of the corrections or processing or any steps are the complex topic to learn or developing our own steps of algorithms might be other topics. There sure are people/resources from whom we can learn more.
Anyway, what provided from earth engine is sufficient for us, I am sure.
Thankyou
@@ksabmagar7 thx
@@sophiez7952 thankyou :)
how to output the result？
You can export them using the export functions.
how to fuse sentinel 1 and 2
It might depend on what fusion mean. The video can be considered an example of using Sentinel 1 and 2 where two different data are handled by the classification.
- We can try this in the SNAP tool box as well. step.esa.int/docs/tutorials/S1TBX%20Synergetic%20use%20of%20S1%20(SAR)%20and%20S2%20(optical)%20data%20Tutorial.pdf
There are certain ways/equations that might have been developed integrating vegetation indices with radar indices. I cannot say about that more.
Clustering approach. Is it good bro?
depends, when there is no training data, it will certainly be used. this can be helpful if we tried comparing with the supervised methods. but I cannot say more.