1. What is the NIWA35?
The NIWA35 is a model that uses the latest advancements in artificial intelligence to take a coarse resolution forecast and increase its resolution. This results in a more skillful forecast at longer lead times. NIWA35 is one-of-a-kind. It’s the first climate outlook in New Zealand to be updated on a daily basis, allowing for a real-time assessment of the risk for weather hazards and extremes, like droughts and floods, over the next five weeks.
2. What is sub-seasonal forecasting?
“Sub-seasonal” forecasting is any forecast that occurs during the time between the “weather” and the “seasons”; i.e. out to 10-50 days. Sub-seasonal forecasting is not the same as a weather forecast; it attempts to bridge the gap between weather time scales and seasonal time scales, where model forecasts are traditionally more skillful. Research indicates that the most effective way to use sub-seasonal (often abbreviated S2S, short for “sub-seasonal to seasonal”), is to look for themes, trends, and anomalies.
3. What is an anomaly?
An anomaly is the difference between what is forecast and what is normal for that time of year. The current standard climate normal period is the 30 years from 1991-2020.
4. What is the “model initiation”?
This is the first time step of the model forecast. Sometimes called the “forecast reference time”.
5. The model forecast started three days ago! Is this okay?
Yes, this is expected. The data required to make predictions and run the model are only fully available over two days from the model initiation. But that’s ok; NIWA35 predicts out to 35 days and its greatest utility are forecasts weeks in the future.
6. How accurate is the model (the NIWA35)?
When we look at some case studies in the recent past, there are indications that the NIWA35 model is skillful out to 3 weeks, and in some cases during very dry periods out to the full 5 weeks. However, S2S models sometimes take extra time to pick up on extreme rainfall events, such as tropical cyclones or atmospheric rivers, whose path may not become evident until a week out. Generally, the forecast confidence is highest out to week 2, and the forecast skill drops after that. As a part of this project with MPI, NIWA is undertaking a comprehensive validation of the NIWA35 and the results will be published sometime in 2023 or 2024.
7. Can I get the data for the NIWA35 forecasts?
The underlying data of the NIWA35 is not yet available, but at the end of this project it can be made available subject to usage, terms, and conditions.
8. What do the percentiles in the forecasts mean?
The NIWA35 produces a range different predictions for rainfall and drought because it is an ensemble model. Because of this, forecasts can be displayed as percentages, which can represent the likelihood of outcomes. For instance, the 25th percentile represents the forecast value at which 75% of forecast values are above, the 50th percentile represents the median (half of the forecast values are above, half are below), and the 75th percentile represents the forecast value at which 25% of forecast values are above. The 25th to 75th percentile or sometimes the 10th to 90th percentile are the ranges in which the bulk of the forecast values lie, making it useful for dealing with the uncertainties of sub-seasonal forecasts. The weekly outlooks are based on the ensemble mean - an average of all the outcomes.

Forecasting products: rainfall

35-day rainfall outlook

The median (50th percentile), lower quartile (25th percentile), and upper quartile (75th percentile) of forecast rainfall anomaly over the next 35 days as a whole. The median is the mid-range scenario, the lower quartile is an indication of a lower-bound drier outcome, and the upper quartile is an indication of a higher bound wetter outcome. When the 50th percentile is similar to either the 25th or 75th percentile, confidence is higher in the predicted outcome.

Territorial Authorities seasonal accumulations and outlook

An overview of observed rainfall (red line) over the last two months and the forecast rainfall range over the next 35 days (blue plume) compared to normal (teal curve). This is also compared to the last three calendar years (dashed lines).

Weekly outlook

Rainfall as a difference from normal over the next five weeks, shown at a five kilometer resolution. Forecast confidence is typically highest in weeks one and two.

Using these forecasts

The NIWA35 is a sub-seasonal forecast, which is not the same as a weather forecast; it attempts to bridge the gap between weather time scales and seasonal time scales, where model forecasts are traditionally more skillful. Research indicates that the most effective way to use sub-seasonal (often abbreviated S2S, short for “sub-seasonal to seasonal”), is to look for themes, trends, and anomalies.

Forecasting products: NZDI

The New Zealand Drought Index (NZDI) combines rainfall, soil moisture and evapotranspiration to give an estimate of “meteorological” drought.

35 day outlook

An overview of the forecast wettest (10th percentile) and driest (90th percentile) conditions that could be expected, as well as the middle scenario (50th percentile) of the NZDI condition within the next 35 days. These plots provide information on the extent and severity of possible dryness, but not on the timing. Please refer to the “Weekly outlook” for more information on timing.

Weekly outlook

The most likely severest NZDI category during each week over the next five weeks, shown at a five kilometer resolution. Forecast confidence is typically highest in weeks one & two.

Using these forecasts

The NIWA35 is a sub-seasonal forecast, which is not the same as a weather forecast; it attempts to bridge the gap between weather time scales and seasonal time scales, where model forecasts are traditionally more skillful. Research indicates that the most effective way to use sub-seasonal (often abbreviated S2S, short for “sub-seasonal to seasonal”), is to look for themes, trends, and anomalies.

Forecasting products: Seasonal forecast

Seasonal outlooks of unusually dry (lower tercile, orange colours) or unusually wet (upper tercile, green colours) are updated once a month around the 11th. These outlooks are created by combining data from 8 different international forecast centres, as available via the Copernicus Climate Change Service (C3S).

The seasonal outlooks provide a high-level overview of areas that may experience unusually dry or wet conditions and the chances for such conditions. The level of detail available at this longer time horizon isn’t as high as what is available in a traditional weather forecast or sub-seasonal outlook.

The seasonal information is updated once a month whereas the sub-seasonal information is updated once a day. It is best practice to check the sub-seasonal outlook tab for shorter-term trends. Over the course of a month, drivers in the climate system can change and influence the weather that Aotearoa/New Zealand experiences. These changes may not be captured until the next update of the seasonal forecast information.

Issues, concerns or feedback?

This trial is all about making the products useful for you. As meteorologists & data scientists we can take a crack, but collaboration creates excellence. For concerns and comments, contact meteorologist Ben Noll.

More information

Interested in learning more about the methodology behind the NIWA35 model? You can find that on the NIWA website at our High-resolution drought forecasting page.
Want to get a bit further into the weeds? You can check out the scientific publication of the methodology behind the NIWA35.