We are developing a model which can predict the presence of blanket bog where it is found in England.
A number of random forest models have been fitted for Cumbria, and one model has been chosen to predict blanket bog in Cumbria.
This is initially based on Cumbria
M.rf.3.final Random Forest
1106 samples 15 predictor 2 classes: 'absent', 'present'
No pre-processing Resampling: Bootstrapped (25 reps) Summary of sample sizes: 1106, 1106, 1106, 1106, 1106, 1106, ... Resampling results across tuning parameters:
mtry Accuracy Kappa
2 0.9349493 0.8686078
8 0.9341944 0.8671090
15 0.9301934 0.8589484
Accuracy was used to select the optimal model using the largest value. The final value used for the model was mtry = 2.
Coefficients:
"elev" "aspect" "slope" "outflow" "inflow"
"surf" "gdd" "gsl" "rain_ann" "rain_daily"
"raindays_10mm" "raindays_1mm" "temp_mean" "temp_min" "temp_max"
See notebook/bb_run_models.3.nb.html for more info.
need to update this up to model 7a
Following Model 7a, trying the approach out on a different area: Yorkshire Dales and Nidderdale. Main lessons from Model 7a:
The predictors used in this model are as follows:
All climate data was derived from UKCP09: Met Office gridded land surface climate observations - long term averages at 5km resolution (Met Office, 2017). We have used the 1960 to 1990 long term averages (averages for more recent time periods are also available) on the assumption that blanket bog presence is most likely to correlate with historic climate than recent climatic changes. However this needs to be tested against other ranges of available climate data and it other variables which determine extent of blanket bog.
Full methodology and lots of further information is available from Met Office 2017, and in Perry et al, 2005. The monthly long-term averages were aggregated to seasonal and annual data using the same methodology as the Met Office used for its 25km seasonal averages:
For the days of frost and days of rain variables the seasonal and annual averages are the total of the individual monthly averages. For the remaining variables the seasonal and annual averages are the mean of the monthly averages (allowing for differences in month length). To facilitate combining the baseline data with the UKCP09 climate projections, the 25 km baseline averages for rainfall have been expressed in units of millimetres per day (rather than total millimetres, as for the 5 km data sets).
Each season is comprised of three calendar months, as follows:
The following datasets are used as predictors in the model:
| Dataset | Units |
|---|---|
| Growing degree days annual average | days |
| Growing season length annual average | days |
| Total rainfall annual average | mm |
| Mean daily rainfall annual average | mm |
| Days of rain above 10mm annual average | days |
| Days of rain above 1mm annual average | days |
| Mean annual maximum temperature | deg C |
| Mean annual temperature | deg C |
| Mean annual minimum temperature | deg C |
TBC
The model will be trained on EXPLAIN TRAINING DATA
explain what models are used, what models were tried and rationale for choice.
This project was carried out in R (R Core Team, 2016) and is a mixture of R script (.R) and Rmarkdown Notebook (.rmd). The code in this repository is intended to be run in the order below. However any data output from one script that is to be passed forward to a later script has been saved with save() either as .rda file or as a raster stack into a .tiff file. The later script then loads the files. This is particularly useful for the .Rmd files which can all be independently 'knitted'. Currently, no scripts are 'sourced' into another script.
| Script filename | Type | Purpose |
|---|---|---|
| ukcp09DataImport.R | R script | imports climate data as ESRI ASCII files, converts to raster and calculates seasonal and annual averages |
| bb_data_prep.Rmd | Rmarkdown | data preparation. Imports topo and hydro .tiff files and converts to raster, calculates slope and aspect rasters, resamples climate rasters |
| bb_input_data.Rmd | Rmarkdown | Creates input dataset for models by extracting location info from peat depth file and extracting predictor variables from rasters |
| bb_model_selection.Rmd | Rmarkdown | Runs a number of models and assesses performance |
| bb_model_run.Rmd | Rmarkdown | Runs final model and creates final outputs |
| bb_predict.Rmd | Rmarkdown | Uses model to predict blanket bog with a raster stack of predictors |
Met Office (2017): UKCP09: Met Office gridded land surface climate observations - long term averages at 5km resolution. Centre for Environmental Data Analysis, accessed on 01/10/2017. http://catalogue.ceda.ac.uk/uuid/620f6ed379d543098be1126769111007
Perry, Matthew, and Daniel Hollis. "The generation of monthly gridded datasets for a range of climatic variables over the United Kingdom." International Journal of Climatology 25.8 (2005): 1041-1054. https://www.metoffice.gov.uk/binaries/content/assets/mohippo/pdf/p/8/monthly_gridded_datasets_uk.pdf
R Core Team (2016). "R: A language and environment for statistical computing. R Foundation for Statistical Computing"", Vienna, Austria. URL http://www.R-project.org/.