This weather app will give you the most accurate forecasts

There are plenty of weather apps out there, but not all will keep you warm and dry.

Major weather services can range in their ability to predict rain and temperatures in a certain place by more than 25 percentage points, so picking the right app could make all the difference.

CNBC looked at how eight services predicted temperatures and rain at nearly 800 weather observations stations across the country, based on data from ForecastAdvisor (click there to enter in your own zip code). At the very top overall were Weather Underground and the Weather Channel, two brands owned by the same company, IBM.

When it comes to predicting whether it will rain one to three days ahead of time in 2015, Weather Underground and The Weather Channel have the highest average accuracy across the country at almost 84 percent. (The two apps sometimes give different results based on how a user's location is interpreted, but the underlying source is identical, the company said.)

The two apps came out as the most accurate options for predicting rain in a little over 60 percent of weather stations, which means that one of the other apps we looked at is still best in the other 40 percent. So the best app depends mostly on where you are: If you're in Miami, MeteoGroup is best overall. If you're in Los Angeles and worried about rain, stick with Dark Sky.

The Weather Channel and Weather Underground are again at the top of the stack nationwide for forecasting high temperatures, but AccuWeather far surpasses all rivals in its ability to predict low temperatures to within three degrees. That's in line with another analysis that found that AccuWeather is the most accurate temperature forecaster globally.

The business of predicting the weather can be far more challenging in some parts of the country than others. Areas near the coasts or other bodies of water are often easier to forecast than inland areas.

"There is wide variation in the ability of forecast models to make accurate predictions in different areas," said Eric Floehr, the founder of ForecastAdvisor. "The Plains states are notoriously difficult to predict because you're constantly having fronts battle it out."

Places like Florida, California and Alaska are easier to forecast with high accuracy. In California, apps predicted rain (or more likely, no rain) in some areas with more than 97 percent accuracy in 2015. Likewise, temperature forecasts are strongest in Florida. Overall, the hardest states to predict the weather in are North Dakota and Montana, where apps are only right about 64 percent of the time.

On average, the accuracy differences between various services are higher for temperature than precipitation. Almost all apps can predict rain with a high degree of confidence in California, but the gap is much higher between apps along the Gulf Coast, in the South and in Plains states.

To be sure, the straightforward metrics at ForecastAdvisor are just one of many ways to measure a weather app.

Floehr, who also runs an advanced service for measuring forecast accuracy called ForecastWatch, said that the basic statistics he publishes on ForecastAdvisor tend to generally be in line with more complex assessments. Still, it's hard to say conclusively that any one app is better than others.

"The technical work of collecting data from multiple providers over a comprehensive set of locations is a big task. So we use ForecastAdvisor for that," Samu Karanko, chief analyst at Finnish forecaster Foreca, said in an email. "Forecast verification is a science of its own and you need to study it for years to learn about all the pitfalls."

Data sourcing can make a difference too: MeteoGroup noted that the data in the ForecastAdvisor analysis comes from its website, while its WeatherPro app uses a new (and more accurate) API.

Additionally, what is perceived as accurate can change depending on how you use a service. That's why some weather services have been known to adjust their predictions to better serve their customers.

One famous example is the "wet bias" — when a forecaster reports a higher probability of rain than his or her models suggest to help keep people from getting wet if it does rain. The logic is that if consumers see that rain is unlikely and get wet, they'll be a lot angrier than if they see that rain is more likely and it doesn't rain.

"Sometimes a less accurate forecast is perceived as being a better forecast," Floehr said. "You could be more accurate, but you would have been more people unpleasantly surprised when they didn't have an umbrella when it rained. Different industries and users have different needs for forecasts and forecast accuracy."

Another example is scattered rain showers — if your app says scattered showers and it rains on four friend half a mile away but not you, was it wrong? The traditional way to deal with that uncertainty is by using probabilities, said Karanko of Foreca. While ForecastAdvisor doesn't track that data, Floehr has looked at rain probability accuracy for ForecastWatch.