**The Problem With Modern-Forecasting**

This was originally posted in a weather forum that I am, so please excuse specific references. Thanks. :)

**The Problem With Modern-Forecasting**

I don't know if this is a subject that will interest many of you, but recently, it has for me. I decided to back and look through my introductory forecasting books during the past few days, and I found some really disturbing things that I find lacking in today's meteorological world. Albeit I am not a professional in the least, and could barely be considered an amateur forecaster.

One thing that I find really great about many in this group is that we don't have consistent "modelology". Most people here are aware that models do not constitute the completeness of what it means to forecast; in fact, if anything, models are just guides; not forecasts. The best that a model can do is SIMULATE how the assumed variables will change through a specific period of time [and remember, simulations can be FAR from reality], not forecast. There is a big difference. However, I still feel as if something is missing.

With the many advances in science and technology over the past decade or so, society has really begun to advance at an accelerating rate; however, I believe we are finally beginning to realize the worst fears of the dystopian sci-fi novels; progress ISN'T always good. And I believe something similar could be said of meteorology. Progress in model and computer development has progressed so much that it has almost removed the human-element from forecasting. At some NWS offices, they are now being told to go with the 24-hour MOS data instead of questioning it; at other offices, model-forecasts are taken at face value for medium-range forecasts. Among many meteorologists, looking at short-range model forecasts for mesoscale events has replaced real analysis [even though it would be almost always advisable to ignore any mesoscale-model within 12-hours of the event, which is the time period in which human-forecast skill exponentially increases over the short-term model forecast] of surface-maps and satellite imagery. Essentially, a lot of modern-meteorology has become an "interpret and regurgitate" mentality; i.e. "we can't have any possible idea of what the next seven-days will do, therefore, regurgitate what the medium-range model showed and hope to apologize when the system changes". I will be the first to admit that I have done this; and I know many of us have done it on here. But don't you realize that the human-element is completely gone? Even when we look at various models and compare various features, and then take a collective "total" so to speak of each forecast [or decide whether or not to throw out outliers], do not ensembles already do this? Isn't there a component of humanity that is needed here? I mean, we are the only self-aware beings on this planet [as far as we know], and we are the only organisms in this solar system attempting to understand our universe, shouldn't the human still be playing the higher part that a computer doesn't and can't ever play? And that directs me to the article I was going to post: http://www.flame.org/~cdoswell/forecasting/human_role/future_forecasters.html I urge anyone interested to read this article, because he first addresses issues in the NWS itself, and then goes on to critique private forecasting use of models. He then goes on to argue why human forecasters will always be essential.

What we are doing now is not forecasting; whether or not we have individuals who can forecast in an excellent way [which we do; I know many who can], its something that is affecting all meteorologists. The surge in technological capabilities has lead to an even greater reliance on computers in themselves to do the forecasting for us, and essentially, we leave without any understanding of what's going on.

This brings up another article that I wanted to post; a speech by Dr. Len Snellman on this problem from National Weather Digest in 1991. http://www.nwas.org/members/snellman.php

Of course, I think there are other problems with meteorology today, including observational limitations, knowledge of our atmosphere, technology, etc. But I also think that this is a hidden problem that we sometimes ignore. Not to say that all meteorologists do this [by no means!], its just that many meteorologists fall into this trap.

What I think is necessary for forecasting is this:

I mean, I think we do have a real problem; because the forecast process should be sort-of like any scientist. We must apply the scientific method in a way that is unique to meteorology. I wouldn't even doubt that using a practical way from mechanics or other field would work. What is needed, however, is one that is structured to include human reasoning, cognitive abilities and intuition into the forecast, rather than just interpreting model-charts and adding some insight every now and then.

Since the scientific method [in general] begins with a question, I believe that the forecasting process should begin with a question of "what is going on". In other words, all forecasts should begin with analysis. Analyzing includes looking at the atmosphere, whether through surface observations, satellite-imagery, radar, and even processed-data fields from super short-range models.

In the next step of the scientific method, we develop a possible answer to our problem/solution. This would be analogous to diagnosis in meteorology. We must first synthesize our analysis into something that can lead to other questions and/or be a possible explanation for what is going on through what we observed in the analysis. If we haven't done a sufficient analysis, we can't do a sufficient diagnosis. We must be able to explain what we see on the weather-map in terms of our own experience, knowledge, conceptual models, etc. This will be more difficult for the amateur with less atmospheric knowledge than the professional with experience and much more atmospheric knowledge.

In the next-step of the scientific method, a scientist will develop a test/experiment for the hypothesis. This is essentially a continued step of what is required of diagnosis in meteorology. We continually monitor what we are analyzing, and now we synthesize model-output with the observations to try to develop a coherent picture of what exactly is going on; since the atmosphere is a continuous fluid and in constant motion, we only get a brief snapshot of what is going on through our analysis. That is why diagnosis is necessary, to get the "dT/dt" [change in the temperature with respect to time] part of analyzing a surface temperature map or the "dV/dt" [change in the wind vector with respect to time] part of analyzing surface observations, and thus we can then at least approximate the change that will occur in a specific period of time. That is essentially what forecasting is all about; predicting what "x" is going to be by estimating the current "dx/dt" [with "x" standing for any scalar variable, or a vector].

This is going to be a lot harder with longer-range forecasts; especially after the 1-4 day period. This is when we have to rely upon ensemble and operational model forecasts for determining what will happen. But that doesn't give us the right to rely completely on models; that's why we have global satellite imagery! That's why we have conceptual models and correlations/experience! I believe that is good to use as well with our model-data. But like I said, in this period, I think its unavoidable that we use models if we are planning on forecasting for the 6+ day time-frame.

Of course, I don't have anything against models, and I would definitely have to say that without models, modern-forecasting would not have made near the advances that we have made. But on the other hand, misuse of model forecasts doesn't make us better forecasters.