Dhaka Weather Forcasting API

Features

1. Endpoint to get on average coolest districts per hour basis based on next 7 days data.
2. Predict Dhaka's temperature for future date and time.
3. Refresh temperature data every 24 hours with corn job.
4. Extensive documentation of process and performance.
5. Easy installation and setup with docker.

Run the app

1. Clone this repository
2. Move to the code directory i.e.

   cd code

3. To build and run the app with docker run following docker compose command.

    docker compose up -build

This will download required docker images and other project dependencies. 4. Visit to http://localhost:8000/docs to access the API documentation and swagger.

About Endpoints

/coolest

This api returns the coolest districts for next 7 days in ascending order of average temperatures in a given hour. As query parameter this api endpoint takes two parameters.

Parameter	Description	Default Value	Range
hour	Hour of the day when the temperature will be considered to calculate coolness.	14	0 to 23
size	Length of the result set. How many districts will be returned.	10	1 to 64

Example curl command

curl --location 'localhost:8000/coolest?hour=14&size=10'

Response body

    [
        {
            "id": "54",
            "division_id": "7",
            "name": "Sylhet",
            "bn_name": "সিলেট",
            "lat": "24.8897956",
            "long": "91.8697894",
            "average_temperature": 26.67
        },
        ...
    ]

/predict

Predicts temperature of a future date. Date should be today + 6 days.

Parameter	Description
date	Date for which the temperature prediction should be made. Date can be formatted like "2024-06-07".

Example curl command

curl --location 'localhost:8000/predict?date=2024-06-10'

Response body

    {
        "date": "2024-06-10",
        "predicted_temperature": 31.32,
        "lower_bound": 30.39,
        "upper_bound": 32.3
    }

How It Works

Coolest Districts

Coolest district calculation processing happens in several steps:

Step-1: Fetch all districts data, we keep it a network call to avoid stale data in case data related to a districts changes, though it is a rare event.
Step-2: Make two lists, one containing latitudes and another longitudes of the districts.
Step-3: Fetch weather data for coming 7 days using openmeteo.
Step-4: Calculate average temperature for each hour of every districts.
Step-5: Save the calculated data for future use.
Step-6: Serve the pre-calculated data through /coolest api endpoint.

Temperature Prediction

We used prophet time series model to fine tune for our temperature prediction purpose. Temperature prediction happens in following steps:

Step-1: This function fetch weather data for Dhaka.
Step-2: Then use hourly data to get an average temperature of the day to create a pandas data-frame which then used to fit/fine-tune the prophet model.
Step-3: Fine-tuned model is saved in global model variable to be used later to predict weather.
Step-4: Use the saved model to predict temperature for future dates.

Performance

/coolest

As we save the collected and processed data in runtime memory, response time is well below constrained 500ms.
Following chart will bear truth to the above claim. We notice initial request takes more time and as time progress response time stabilize to 3ms.

hour	size	response_time
14	10	12
0	4	4
0	64	7
9	20	3
10	5	3

/predict

Models performs better when future date is nearer to the training date and slips away further as we go further the time. This can be improved with more training data. Here we are fitting the model with only 7 days of temperature data.

Date	Distance	Predicted
2024-06-10	10 days	27.99
2024-07-10	40 days	20.99
2025-07-10	375 days	-7.73