My company is providing an IOT service to monitor dissolved oxygen level in the sea for a fish farmer. Dissolved Oxygen is a very critical factor in fish farming. If DO is low, the fish will die. So my purpose in using TimeGPT is to predict what DO level will be like in the coming hours or days so the farmer can take proactive action (eg. turn on the pump to increase oxygen in the water). DO is influenced by many factors eg. weather, water temperature, ammonia, nitrites, Ph level). I have current and historical information of DO, weather, water temperature, etc. How can I incorporate these exogenous features into DO forecast ? Can TimeGPT help ? The problem is that I DO NOT HAVE FUTURE VALUES of those exogenous variables. I only have current and historical data. How do I do forecast ? From below tutorial, it is mentioned that future values of exogenous variables are needed. https://docs.nixtla.io/docs/tutorials-exogenous_variables Important If you want to use exogenous variables when forecasting with TimeGPT, you need to have the future values of those exogenous variables too.

It’s my understanding too that you need future values for exogenous variables in TimeGPT. I.e. it’s not like N-HITS or NBEATS-X which can handle exogenous variables only known for past values.

Hello! So, you're right, to use exogenous variables, you have to provide their future values to the model. In section 4 of the same tutorial, we show how to generate forecasts for your features. Here is the link: https://docs.nixtla.io/docs/tutorials-exogenous_variables I hope it helps!

Since this is a limitation of Nixtla, do you have any suggestions how to predict future values if we do not have future values of exogenous variables ?

Hello! This is not a limitation. You can simply predict your exogenous features with TimeGPT. Again, you can follow this tutorial, where it shows how to handle exogenous features and predict exogenous features when their future values are not readily available. https://docs.nixtla.io/docs/tutorials-exogenous_variables

thank you very much for your clarification.

@Marco If I already have future values, for would I be able to use those or do I have to generate new ones

If you already have future values for your given forecast horizon, no need to regenerate them!