The Gospel according to Jeff

For about a year now Jeff Watson and his team have been posting river level forecasts to the Horizons Regional Council website. This is a project undertaken on behalf of the Emergency Management Office and I'm happy to say has received generally supportive feedback from the community. Recently however a few agencies have been complaining that our long-term forecasting is inaccurate and their decision-making has been affected by this. Well, that's where the discussions have become quite interesting in Council and have ranged from stopping posting the forecasts to the website to wrapping it in 10 pages of legal disclaimer and description. I'm pleased that we'll be taking a middle of the line approach in the next few weeks to provide a short disclaimer and brief description of the limitations of the forecasts. It's worth pointing out here that there are only a handful of organisations around the world prepared to put river level forecasts out into the public arena, and we are the only regional council in New Zealand to move it out of the back room.

So, brace yourself for what will probably be my lengthiest post since I created the blog…

Firstly, the forecasting models draw on a number of data inputs, the forecast rainfall, the actual rainfall and river levels, and then run these numbers through a virtual catchment to try and predict how high the river might get at a few critical spots (well that sounds fairly straight forward). Let's pick on the Whanganui model to try and explain some of the limitations of this forecasting.

Forecast rainfall… we automatically receive the forecast rainfall from the MetService every eight hours. We get this once global forecast models have run to provide input into their local climate models and it literally takes this long to happen. We have recently found out that MetService actually run seven different models and then pick the best to base their forecasts and warnings on (that makes sense). The river level forecast however draws on the same model every time, regardless of whether its being used by the MetService to develop their forecasts - we're seeing how we can pick up the best three forecasts that are chosen by the forecasters instead of the same model every time. Next (and this is where we get a bit 'Dr Who') there is the space-time-quantity continuum… does the rain that was forecast, up to 48 hours out, arrive where they said, when they said and in the intensities they said? You only need to be out by one degree of track to move the rain from one sub-catchment to the next and the model will be wrong. Equally, the forecast quantity presents a challenge - any more or any less than was forecast and the model will be wrong. In fact, the quantity might be right but it all arrives in two hours instead of 12 hours (you guessed it, the model will be wrong). Finally, if the front stalls or moves faster than expected the model will be wrong. So… you might as well poke a sticky wet finger in your ear and look at the sky (which is what some of the team did before we got the model).

Next, the actual rainfall… the Whanganui catchment is a large, hilly, sparcely populated and difficult to access part of the region. The number of sites we actually collect data from is very limited, to the point where it can be raining in a couple of sub-catchments and we won't even know. We are looking to improve the model by drawing on rain gauges outside of the catchment in Taranaki and Waikato and then interpolating the information across the catchments as well as adding an additional two rainfall sites into the catchment later this year. Of course if any of the equipment fails then this will make the model particularly vulnerable. The team keep on top of this as much as possible but sometimes our equipment gets stolen, hit by lightning, eaten by pests, filled with spiders or even used as target practice. Of course, of the rain that actually falls, some will evaporate, some will be absorbed and some will make it to the river and all three of these will be different for every event depending on what has happened before the rain arrived. The flood forecasting model attempts to deal with these effects.

So now we're interested in the water that actually makes it to the river… assuming it's arrived when, where and in the quantities forecast and wasn't evaporated or absorbed! The Whanganui catchment is made up of multiple sub-catchments, it may be raining in some but not others and each of these sub-catchments are feeding the river to varying degrees. Finally, the river height and flow will be recorded at a river level site (and possibly other parameters for other activities in Council). Now that we have a hard number on the river level it should be fairly straight forward to forecast how high the river will get in the town - right… we know the average travel time on the river and we know how much water is coming… but don't forget that more sub-catchments will be feeding the river before it gets to the town so we need to add these in too. Another little catch is this thing called a rating curve. To calculate the volume of water in the river, a cross-section of the river is taken and co-related to the river level site so… if the river is say 6m deep, the volume of water passing over the cross-section is say 1000 cubic metres/second (or cumecs). But what if the profile of the cross-section has changed (and Mother Nature does that to us), what if the river deepened at that spot and the volume was actually 2000 cumecs - the height could be the same on the gauge… this could have serious consequences down stream.

The Pipiriki site is a significant site for the Whanganui River. The site provides plenty of lead time for the town and is the basis for the Horizons and Wanganui District Council joint Flood Action Plan. There are two sites down river of this that enable us to confirm margins of error in the model. During possible events, staff in our emergency operations centre manually evaluate the data when it gets to this point so if the model is under-forecasting we should pick it up in plenty of time. If the model is over-forecasting… better safe than sorry.

When all said and done, forecasting how deep a complicated river system like the Whanganui will get in a few days time is a little more complicated than your average punter might think - but that won't stop us trying. No other agency is in a position to tell the local communities this level of information. Since the model has been running, we've had a couple of small events go through. Even though the model is designed to be more accurate in larger events, these small events have allowed us to re-calibrate the model (this should be in place soon). You can also see that improvements in forecasting data are being sought along with actual data from new and neighbouring sites. The main thing to remember is that the further out the forecast, the less accurate it will be (we can accurately tell you what actually happened though). While work continues to improve existing models, new models are being produced… we're expecting the Manawatu forecasting model to get plenty of hits when its up and running later this year.