Hydrology Packages for Modelling Rainfall Streamflow Time Series

[brant]

Hi. I'm wondering if there are any hydrologists out there that could provide me with some advice. I have two sets of time series data: daily precipitation and daily peak discharge data. The data spans from the year 2000 until 2022. I am trying to model the correlation between daily precipitation and change in daily peak discharge (from the previous day). I want to be able to see if the effect size of the relationship between daily precipitation and change in daily peak discharge from 2000 to 2013 differs from the effect size of the relationship between precipitation and change in daily peak discharge from 2018 to 2022.

My dataset is currently formatted as follows.

Date, PeakDischarge, Precipitation, DailyChange

I want to be able to compare the strength of the relationship between "Precipitation" and "DailyChange" during both time periods (2000 to 2013) and (2018 to 2022). Does anyone have any advice/guidance on this?

Thanks in advance. ), 2) that precipitation has a greater effect on peak discharge during the 2000-2013 time period, and 3) that precipitation has a reduced effect on peak discharge during the 2017-2022 time period. Does anyone have any advice/guidance on how to test this? Thanks in advance.

[technocrat]

@brant

I want to be able to show that

Be self-conscious of the dangers of confirmation bias that arise from framing questions this way. Even though you expect this to be the case, acquire the habit of suspended judgment, along these lines:

Given a long time series of daily observations of precipitation and peak discharge in a basin with a watershed [relevant characteristics], does precipitation as the treatment (or independent) variable for peak discharge as the response (or dependent) variable have predictive utility and, if so, does that utility exist over the entire series and sub-series equally?

Time series data present some considerations that do not arise with many other of the statistical tools that we use to assess the association between variables. This occurs from factors such autocorrelation and the heteroscedasticity of residuals in the usual linear regression situation. A good toolset is the {fpp3} package and associated free, online text.

I would start with taking one or two year's data and laying down a baseline suite of models using naive, mean, seasonally naive and random walk with drift. Chose a metric, such as root-mean-square-error and that will be the bogey to beat. Then model with tsrm, time series linear regresssion.