Overview

The CADP forecasts temperature using ECMWF forecasts as a starting point. The ECMWF forecasts are gridded and known to have numerous deficiencies. By integrating local weather data with the gridded forecasts, we can improve forecast quality significantly.

For demonstration purposes, we use official weather station observations and show how much we can reduce the forecast error. The following analysis is for 60 days of data.

CHHATRAPATI SHIVAJI INTERNATIONAL, IN

Weather forecasts consist of multiple forecasts at different forecast horizons. In the following chart, we compare the error at each forecast horizon between the CADP forecast (black bars) and the base ECMWF forecast (gray bars). The better the forecast, the less error should exist (hence smaller is better).

3 hour forecast

The following charts show a detailed view of our 3 hour forecast. The first graph compares the CADP forecast (black) with the ECMWF forecast (gray) and the actual observations (orange).

The first 7 days of our forecast (black dashed line) are ignored in the error calculation as they are reserved for the “burn in” period of the model.

We plot the residuals to get a better understanding of what variation could not be captured by the model.

24 hour Forecast

48 hour Forecast

90 hour Forecast