如何基于1980-2018年MERRA2 .nc文件绘制CLDHGH变量时间序列图

Got it, let's tackle this step by step. You've already nailed plotting a single day's CLDHGH time series, so scaling this up to the full 1980–2018 period just needs a few key additions: generating the complete date range, looping through each file, aggregating your data, and building a proper time axis for plotting.

First, we need to create all the yyyymmdd date strings that match your filename pattern. In MATLAB, you can do this with datetime utilities:

% Define your start and end dates
start_date = datetime(1980, 1, 1);
end_date = datetime(2018, 3, 30);

% Generate every day in the range
all_dates = start_date:days(1):end_date;

% Convert dates to the yyyymmdd string format used in your filenames
date_strings = datestr(all_dates, 'yyyymmdd');

Next, we’ll iterate over each date, load the corresponding .nc file, apply your 0-value cleanup, and store the data. Note: If you’re working with the full dataset (over 330k hourly timesteps), memory might be an issue—so I’ll include two approaches:

Approach A: Full hourly data (if memory allows) #

This stores every hour’s 65x80 grid data for the entire period:

% Base path and file extension for your .nc files
file_base = 'D:\cloudnc\MERRA2_100.tavg1_2d_rad_Nx.';
file_ext = '.SUB.nc';

% Initialize an empty array to hold all CLDHGH data
cldhigh_full = [];

for i = 1:length(date_strings)
    % Build the full filename
    current_file = [file_base, date_strings(i), file_ext];
    
    % Skip missing files to avoid errors
    if ~exist(current_file, 'file')
        warning('File %s not found—skipping this date.', current_file);
        continue;
    end
    
    % Read the CLDHGH variable
    cldhigh = ncread(current_file, 'CLDHGH');
    
    % Apply your existing 0-value handling (replace ... with your actual logic)
    cldhigh(cldhigh == 0) = NaN; % Example: replace 0 with NaN
    
    % Append the day's data to the full dataset
    cldhigh_full = cat(3, cldhigh_full, cldhigh);
end

Approach B: Memory-efficient (compute hourly means on the fly) #

If storing the full 3D array is too much, calculate regional hourly means as you go to save memory:

hourly_mean = [];

for i = 1:length(date_strings)
    current_file = [file_base, date_strings(i), file_ext];
    
    if ~exist(current_file, 'file')
        warning('File %s not found—adding NaNs for missing hours.', current_file);
        hourly_mean = [hourly_mean, NaN(1,24)];
        continue;
    end
    
    cldhigh = ncread(current_file, 'CLDHGH');
    cldhigh(cldhigh == 0) = NaN;
    
    % Calculate mean across the 65x80 grid for each hour
    daily_hourly_means = squeeze(mean(mean(cldhigh, 1), 2));
    hourly_mean = [hourly_mean, daily_hourly_means];
end

To plot your data with meaningful dates, generate a time vector that matches every hour in your dataset:

full_time_axis = start_date:hours(1):end_date + hours(23);

Choose the plot that fits your needs:

Option 1: Regional hourly mean time series #

If you used Approach B:

figure;
plot(full_time_axis, hourly_mean, 'LineWidth', 0.5);
xlabel('Time');
ylabel('Mean CLDHGH');
title('Hourly Mean CLDHGH (1980-01-01 to 2018-03-30)');
datetick('x', 'yyyy-mm', 'keepticks'); % Format x-axis dates for readability
grid on;

Option 2: Single grid point time series #

If you used Approach A and want to plot a specific grid cell (e.g., row 20, column 30):

% Extract data for the target grid point
grid_point_data = cldhigh_full(20, 30, :);

figure;
plot(full_time_axis, grid_point_data);
xlabel('Time');
ylabel('CLDHGH at Grid (20, 30)');
title('Hourly CLDHGH at Grid Point (20, 30) (1980-01-01 to 2018-03-30)');
datetick('x', 'yyyy-mm', 'keepticks');
grid on;

• Parallelize for speed: If you have MATLAB’s Parallel Computing Toolbox, replace for with parfor to process multiple files at once.
• Log missing files: Add a line to write missing dates to a text file (e.g., fprintf(log_file, '%s\n', date_strings(i));) for later troubleshooting.
• Chunk processing: If even the full hourly mean array is too big, process the data in monthly chunks instead of all at once.

内容的提问来源于stack exchange，提问作者Vaishali Jain