I need to create a polygon or buffer along an irregular shaped coastline in a 2D array of gridded sea temperature data.
Show older comments
I have a 578x235 grid of ocean temperature data in which the land pixels are NaN. I would like to create a 1 pixel buffer around the coastline so that all pixels adjacent to an existing NaN also become NaN. I'm unsure on the best way to do this. I thought I could create a logical array of NaNs and not NaNs but am not sure on the best way to extract a polygon and create the buffer after that step.
5 Comments
Umar
on 23 Jul 2024
Hi Angus,
Your aim is to create a 578x235 grid of ocean temperature data with land pixels represented as NaN values. Additionally, you need to implement a 1-pixel buffer around the coastline, ensuring that all pixels adjacent to land pixels (NaNs) also become NaNs. In order to accomplish this task, I will start by generating a 578x235 grid of random ocean temperature data using the rand function, set the first 100 rows of the grid as land pixels by assigning NaN values to them. Then, create a A logical array nan_mask to identify NaN and non-NaN values in the ocean_data. Now, to implement the 1-pixel buffer around the coastline, I will iterate through each pixel in the grid (excluding borders). For each land pixel, I will check if any of its adjacent pixels are also land pixels using the nan_mask. If any adjacent pixel is land, I will set the current pixel as land as well in the buffered_data. Finally, I will display the ocean temperature data grid with the coastline buffer using imagesc and add a colorbar for reference. Here is the snippet code as an example to illustrate it.
% Create a 578x235 grid of ocean temperature data with land pixels as NaN
ocean_data = rand(578, 235); % Generate random ocean temperature data
ocean_data(1:100, :) = NaN; % Set the first 100 rows as land (NaN)
% Create a logical array to identify NaN and non-NaN values
nan_mask = isnan(ocean_data);
% Implement a 1-pixel buffer around the coastline
buffered_data = ocean_data;
for i = 2:size(ocean_data, 1)-1
for j = 2:size(ocean_data, 2)-1
if ocean_data(i, j) == 0 % Check if the pixel is land
if any(any(nan_mask(i-1:i+1, j-1:j+1))) % Check if any adjacent pixel is land
buffered_data(i, j) = NaN; % Set the pixel as land
end
end
end
end
% Display the ocean temperature data grid with the coastline buffer
imagesc(buffered_data);
colorbar;
title('Ocean Temperature Data with Coastline Buffer');
Please see attached plot.
I hope this answers your question.
Angus
on 23 Jul 2024
Thank you very much for responding. I tried your suggestion but their appeared to be no change from ocean_data to buffered_data. For more context this is my data plotted without a buffer, NaN (land) is blue. 
Umar
on 23 Jul 2024
Hi Angus,
Could you please share your code, so I can take a look.
Angus
on 23 Jul 2024
Edited: Walter Roberson
on 25 Jul 2024
Umar
on 23 Jul 2024
Edited: Walter Roberson
on 25 Jul 2024
Hi Angus,
To achieve the desired 1-pixel buffer around the coastline, I had to update your code. Your code currently iterates through each pixel and checks if the pixel value is 0. If the pixel is 0 and any of its neighboring pixels are NaN, it sets the pixel value to NaN. However, I needed to extend this logic to create a buffer around NaN values. Here is the updated code that creates a 1-pixel buffer around the coastline:
nan_mask = isnan(meanSST);
buffered_data = meanSST;
for i = 2:size(meanSST, 1) - 1
for j = 2:size(meanSST, 2) - 1
if meanSST(i, j) == 0
if any(any(nan_mask(i-1:i+1, j-1:j+1)))
buffered_data(i, j) = NaN;
% Set neighboring pixels to NaN
buffered_data(i-1, j) = NaN;
buffered_data(i+1, j) = NaN;
buffered_data(i, j-1) = NaN;
buffered_data(i, j+1) = NaN;
end
end
end
end
So, In this updated code, it still checks if the current pixel is 0 and any of its neighbors are NaN. If the condition is met, we set the current pixel to NaN and also set its neighboring pixels (top, bottom, left, right) to NaN. Also, it makes sure This ensures that a 1-pixel buffer around the coastline is created, turning all adjacent pixels to NaN.
By incorporating these modifications, the code should now effectively create a 1-pixel buffer around the coastline by setting adjacent pixels to NaN based on the presence of NaN values in the neighborhood.
Now, to address Extracting a Polygon and Creating the Buffer
To extract a polygon representing the coastline and create a buffer around it, you can use the bwboundaries function to find the boundaries of the NaN regions in the buffered_data. For more information on this function, please refer to
Here's an example of how you can achieve this:
% Find boundaries of NaN regions
B = bwboundaries(isnan(buffered_data));
% Plot the original data with coastline buffer
subplot(2, 1, 1);
imagesc(lon, lat, buffered_data.');
hold on;
for k = 1:length(B)
boundary = B{k};
plot(boundary(:, 2), boundary(:, 1), 'r', 'LineWidth', 2);
end
colorbar;
title('Coastline with 1-Pixel Buffer');
% Plot the original data without buffer
subplot(2, 1, 2);
imagesc(lon, lat, meanSST.');
colorbar;
title('Original Coastline Data');
By incorporating the above modifications, you can update the code to create a 1-pixel buffer around the coastline and visualize the coastline with the buffer applied. Remember to adjust the code according to your specific data and requirements for accurate results.
Accepted Answer
More Answers (1)
John D'Errico
on 23 Jul 2024
Edited: John D'Errico
on 23 Jul 2024
Easy enough. Almost trivial, in fact. Of course, making a buffer zone of width exactly one pixel is alays going to be slightly problematic. But This will be quite good.
Firtst, I'll create a region, filled with NaNs. This one will be pretty simple.
A = ones(1000,1000);
[I,J] = meshgrid(1:1000);
A(sqrt((I - 500).^2 + (J - 500).^2) < 100) = NaN;
H = pcolor(isnan(A));
H.LineStyle = 'none';
axis equal
Ok, so a yellow, circular country. We can find the pixels that touch a NaN easily enough, using conv2, and xor.
bordermask = xor(isnan(conv2(A,ones(3),'same')),isnan(A));
spy(bordermask)
The definition of touching a NaN I used there is if a pixel is horizontally, vertically, or diagonally touching a NaN, then it is flagged in the mask.
If you wanted only horizontal or vertical connections, then the convolution kernel would be [0 1 0;1 1 1;0 1 0] instead of all ones.
5 Comments
Umar
on 23 Jul 2024
Edited: Walter Roberson
on 25 Jul 2024
Hi John,
I executed Angus code on my mobile Matlab, it displayed the following error, Unrecognized function or variable 'meanSST'. Any suggestions or ideas, it sounds trivial to solve the problem.
nan_mask = isnan(meanSST);
buffered_data=meanSST;
for i=2:size(meanSST,1)-1
for j=2:size(meanSST,2)-1
if meanSST(i,j) == 0
if any(any(nan_mask(i-1:i+1, j-1:j+1)))
buffered_data(i,j)=NaN;
end
end
end
end
subplot(2,1,1);
imagesc(lon,lat,buffered_data.');
colorbar;
subplot(2,1,2);
imagesc(lon,lat,meanSST.');
colorbar;
John D'Errico
on 23 Jul 2024
meanSST is apparently the name of the image array. If you don't define the array, how can anything work?
Umar
on 24 Jul 2024
Hi John, I know that, it is quite obvious. This is not what I meant to ask, I wanted you to solve Angus problem using your example to solve his issue mentioned above in the post comments“ I would like to create a 1 pixel buffer around the coastline so that all pixels adjacent to an existing NaN also become NaN. I'm unsure on the best way to do this. I thought I could create a logical array of NaNs and not NaNs but am not sure on the best way to extract a polygon and create the buffer after that step.” Since you volunteered to help out here.
John D'Errico
on 25 Jul 2024
Edited: John D'Errico
on 25 Jul 2024
@Umar I don't see the actual data posted, so I won't try to do it with that data, but the problem is still pretty easy to solve in my eyes. I'll make something up. I suppose if I knew how to use the mapping toolbox, that would give me some data.
fun = @(x,y) x.^2 + + x.*y + 2*y.^2 + sin(x+y).*cos(x-2*y)*8;
H = fcontour(fun,[-10 10 -10 10]);
H.LevelList = 50;
[x,y] = meshgrid(linspace(-10,10,100));
A = fun(x,y);
A(A < 50) = NaN;
As such, I've created an irregular hole, filled with NaNs. Do you agree?
surf(x,y,A,LineStyle = 'none')
Now, apply my scheme using conv2.
kernel = [0 1 0;1 1 1;0 1 0];
bordermask = xor(isnan(conv2(A,kernel,'same')),isnan(A));
H = fcontour(fun,[-10 10 -10 10]);
H.LevelList = 50;
H.LineColor = 'b';
hold on
plot(x(bordermask),y(bordermask),'.r')
As you can see, it created a 1 pixel border, falling entirely outside the region of NaNs. You can tell since I drew the contour on top, and as you should see, the blue contour falls entirely inside the red pixelated perimeter.
There are some points where that border is a little bumpy, but this is partly due to the coarseness of the mesh I used. As well, some of that is due to the definition of touching a NaN. But as you should understand, that is not my problem. The pixels that touch a NaN will not form a perfectly smooth set, and no matter how you generate them, they will have the same characteristics. Pixels are pixels. They will never be "smooth".
But I do want you to understand that NO loops were needed or used. Not much more than one line of code, and it will be blazingly efficient.
Thanks for your contribution. After reading your posted comments, I have to say that your code snippet does demonstrate the handling of NaNs in MATLAB contour plots. Your approach involving creating a hole filled with NaNs and then applying a scheme using conv2 to generate a border around the NaN region efficiently creates a 1-pixel border outside the NaN area and is evidenced by the contour plot. However, I do understand that the slight bumpiness in the border is attributed to mesh coarseness and the nature of NaN definitions, rather than yours implementation. Notably, the solution does not require loops and is computationally efficient, achieving the desired outcome effectively. Again, thanks for your help.
Categories
Find more on Map Display in Help Center and File Exchange
