Workflow for Refining ClimateNA Temperature Predictions

Here we present a workflow and code for refining ClimateNA temperature predictions (Wang et al. 2016) using temperature data loggers and remote sensing accessed via Google Earth Engine. View temperature offsets and spatial data at https://bgcasey.users.earthengine.app/view/climateoffsets. View model summaries at https://bookdown.org/bgcasey/climateOffset_models/.

Ecological studies often rely on interpolated climate data to predict species distributions and identify climate change refugia. However, the scale of climate data does not always correspond to the scale of habitat conditions influencing organisms. ClimateNA, a freely available software package, addresses this by providing scale-free predictions of climate variables by interpolating gridded climate data and adjusting for elevation. While useful, ClimateNA predictions could be improved by incorporating other variables that influence micro-climatic variation. We developed methods to refine ClimateNA air temperature predictions using temperature data loggers and remote sensing data accessed via Google Earth Engine. Monthly temperature variables from 2005-2021 were calculated using near-surface temperatures gathered from 513 monitoring sites across Alberta, Canada. We used variables associated with terrain, vegetation structure, and atmospheric conditions in boosted regression trees to predict differences between ClimateNA temperature predictions and micro-climate conditions. We produced 30 m seasonal offset layers for mean, maximum, and minimum temperatures covering all of Alberta and British Columbia. Mean summer temperatures were on average -0.03°C (SD = 0.71) greater than ClimateNA predictions; maximum summer temperatures were on average 6.81°C (SD = 1.11) less than CimateNA predictions; and winter minimum temperatures were on average -0.81°C (SD = 0.98) greater than CimateNA predictions. Offset adjusted ClimateNA predictions should better reflect micro-climatic variation and improve the accuracy of species-habitat models.

---

• Import and clean temperature data
• Get XY coordinates of data loggers
• Temperature data quality control
• ClimateNA temperature
• Spatial covariates
• Boosted regression trees
• Generate offset raster layers
• References

Import and clean temperature data

First we gathered temperature data from temperature data loggers deployed across the province of Alberta.

Project code	Number of loggers	Time frame	Region	Reference
RIVR	88	2018-2020	Alberta	Estevo et al. (2022)
HILLS	152	2014-2016	Alberta	NA
WOOD	232	2005-2010	Alberta	Wood et al. (2017)
ALEX	41	2020	Alberta	NA

Sources of temperature data loggers.

The file 1_code/r_notebooks/1_ibutton_data_prepare.Rmd provides code and instructions for importing and cleaning iButton temperature data. We did the following for each data source:

1. Imported temperature data into the r project.
2. Identified and removed pre-deployment and post-retrieval temperature data.
3. Identified and fixed errors in date-time strings.
4. Identified and fixed errors in site names and made sure that each iButton deployment was associated with a unique identifier.
5. Calculated daily temperature summaries including:

   • Tmax (maximum temperature)

   • Tmin (minimum temperature)

   • Tavg (mean temperature)

6. Combined daily temperature data from all sources into a single data frame.

Get XY coordinates of data loggers

The file 1_code/r_notebooks/2_ibutton_data_xy.Rmd provides code and instructions for generating spatial dataframes and shapefiles from iButton coordinates. The file describes how to:

1. Get XY coordinates of iButton locations.
2. Convert a data frame with XY coordinates into a spatial data frame using the sf package in R (Pebesma 2022).
3. Export the spatial data frame as a shape file. The shape file of point locations was later uploaded as an asset to Google Earth Engine and used to extract environmental variables.
4. Create a map of the study area and iButton locations using the tmap package in R (Tennekes 2022).

Locations of temperature data loggers.

Temperature data quality control

The file 1_code/r_notebooks/3_ibutton_qualityControl.Rmd provides code and instructions for flagging outlier temperature data and identifing iButtons that were covered by snow during deployment. Once outliers were flagged and removed, we interpolated missing data and calculated monthly temperature metrics (Tmax, Tmin, and Tavg). The markdown file details the following steps:

1. Remove the first and last months of a temperature time series if they have less than 20 days of data.
2. Flag outlier temperatures. Calculate the difference between daily temperature metrics from iButton data and daily estimates from ERA5 (Hersbach et al. 2018). We flagged iButton data with temperature differences above Mean + 3*SD and below Mean - 3*SD.
3. Impute missing data with a maximum gap of 10 days using the na_interpolation function in the R package imputeTS (Moritz and Bartz-Beielstein 2017).
4. Flag iButtons that may have been buried by snow. An iButton was flagged as buried if it had a diurnal temperature range of <3 degrees for > 25 consecutive days (Wood et al. 2017).
5. Calculate monthly temperature summaries.
6. Exclude months with more than 10 missing or flagged days from the final data set.

Project	Site_StationKey	Month	Year	Tmax_Month	Tmin_Month	Tavg_Month
alex	10_B	8	2020	15.138	7.590	15.138
alex	10_B	9	2020	9.278	0.670	9.278
alex	10_B	10	2020	-0.945	-7.545	-0.945
alex	10_B	11	2020	-5.873	-12.753	-5.873
alex	10_B	12	2020	-6.405	-11.101	-6.405
alex	10_B	1	2021	-4.806	-7.047	-4.806

ClimateNA temperature

The file 1_code/r_notebooks/4_ibutton_climatesNA.Rmd provides instructions for:

1. Installing the ClimateNA desktop application and R package (Wang et al. 2016).
2. Extracting monthly ClimateNA temperature predictions for each iButton location.
3. Calculating the difference between monthly iButton and ClimateNA temperature metrics.

Project	Site_StationKey	Month	Year	Tmax_Month	Tmin_Month	Tavg_Month	Tmax_cNA	Tmin_cNA	Tavg_cNA	Tmax_diff	Tmin_diff	Tavg_diff
alex	10_B	8	2020	15.138	7.590	15.138	21.5	8.0	14.8	6.36	0.41	-0.34
alex	10_B	9	2020	9.278	0.670	9.278	16.4	3.2	9.8	7.12	2.53	0.52
alex	10_B	10	2020	-0.945	-7.545	-0.945	5.1	-4.7	0.2	6.05	2.84	1.15
alex	10_B	11	2020	-5.873	-12.753	-5.873	-1.3	-11.6	-6.5	4.57	1.15	-0.63
alex	10_B	12	2020	-6.405	-11.101	-6.405	-4.8	-14.6	-9.7	1.60	-3.50	-3.30
alex	10_B	1	2021	-4.806	-7.047	-4.806	-9.6	-19.9	-14.7	-4.79	-12.85	-9.89

Spatial covariates

Spatial covariates were extracted using Google Earth Engine’s online code editor at code.earthengine.google.com.

You cam download the Google Earth Engine scripts by entering git clone https://earthengine.googlesource.com/users/bgcasey/climate_downscaling into your working directory. Earth engine scripts can also be found as .js files in 1_code/GEE/. Just copy them them into the Google Earth online code editor.

The file 1_code/r_notebooks/5_covariates_gee_spatial.Rmd provides code for:

1. Cloning the GEE project’s git and accessing the GEE code.
2. Importing csv files of summarized covariates generated in Google Earth Engine.
3. Importing and processing raster layers for each covariate. All were generated in Google Earth Engine.
4. Setting up data for use in boosted regression trees.

We calculated and/or extracted the following metrics in Google Earth Engine:

Variable	Description	Resolution (m)	Source
elevation	NA	23.19	Canada (2015)
slope	NA	23.19	Canada (2015)
aspect	NA	23.19	Canada (2015)
northness	cosign of aspect in radians	23.19	Canada (2015)
TPI	Topographic Position Index, TPI neigborhoods range from 50-1000 meters.	23.19	Canada (2015)
HLI	Heat Load Index	23.19	Canada (2015)
TWI	Topographic Wetness Index: ln(α/tanβ)) where α=cumulative upslope drainage area and β is slope (Sørensen and Seibert 2007)	92.77	Yamazaki et al. (2019)
NDVI	Normalized Difference Vegetation Index: (NIR - R) / (NIR + R)	30.00	Survey (2018)
EVI	Enhanced Vegetation Index: G * ((NIR - R) / (NIR + C1 * R – C2 * B + L))	30.00	Survey (2018)
BSI	Bare Soil Index: ((R+SWIR) – (NIR+B)) / ((R+SWIR) + (NIR+B))	30.00	Survey (2018)
SAVI	Soil Adjusted Vegetation Index: ((NIR - R) / (NIR + R + L)) * (1 + L)	30.00	Survey (2018)
NDMI	Normalized Difference Moisture Index: (NIR - SWIR) / (NIR + SWIR)	30.00	Survey (2018)
SI	Shadow index: (1 - B) * (1 - G) * (1 - R)	30.00	Survey (2018)
LAI	Leaf Area Index: 3.618 *(2.5 * ((NIR - R) / (NIR + 6 * R - 7.5 * B + 1)))-0.118	30.00	Survey (2018)
NDSI	Normalized Difference Snow Index: (G - SWIR) / (G- SWIR)	30.00	Survey (2018)
snow	percent of snow cover	30.00	Survey (2018)
discrete_classification	Land cover classification	100.00	Buchhorn et al. (2020)
forest_type	Forest type for all pixels with tree cover > 1 %	100.00	Buchhorn et al. (2020)
tree-coverfraction	Percent of forest vegetation cover	100.00	Buchhorn et al. (2020)
cloud_fraction	Fractional cloud cover	11132.00	Heidinger et al. (2014)
cloud_probabilty	probability of cloud cover	11132.00	Heidinger et al. (2014)
aet	Actual evapotranspiration	4638.30	Abatzoglou et al. (2018)
def	Climate water deficit	4638.30	Abatzoglou et al. (2018)
pdsi	Palmer Drought Severity Index	4638.30	Abatzoglou et al. (2018)
pet	Reference evapotranspiration (ASCE Penman-Montieth)	4638.30	Abatzoglou et al. (2018)
pr	Precipitation accumulation	4638.30	Abatzoglou et al. (2018)
ro	Runof	4638.30	Abatzoglou et al. (2018)
soil	Soil moisture	4638.30	Abatzoglou et al. (2018)
srad	Downward surface shortwave radiation	4638.30	Abatzoglou et al. (2018)
swe	Snow water equivalent	4638.30	Abatzoglou et al. (2018)
vap	Vapor pressure	4638.30	Abatzoglou et al. (2018)
vpd	Vapor pressure deficit	4638.30	Abatzoglou et al. (2018)
vs	Wind-speed at 10m	4638.30	Abatzoglou et al. (2018)
CHILI	Continuous Heat-Insolation Load Index	90.00	Theobald et al. (2015)
canopy_height	model trained with GEDI LiDAR to retrieve canopy height from Sentinel-2 images	10.00	Lang et al. (2022)
canopy_standard_deviation	model trained with GEDI LiDAR to retrieve standard deviation of canopy height from Sentinel-2 images	10.00	Lang et al. (2022)

Spatial covariates evaluated and used in our analysis.

Boosted regression trees

We used the spatial variables extracted via Google Earth Engine in boosted regression trees to predict differences between ClimateNA and iButton temperatures. The file 1_code/r_notebooks/6_boosted_regression_trees.Rmd provides the code and methods used. We built boosted regression trees using the dismo and gbm R packages. We repeated the following steps for Tmax, Tmin, and Tavg for the summer, fall, winter, and spring seasons:

1. Tune BRT parameters using a grid of parameter options and the gbm R package (Kuhn, Johnson, et al. 2013; Greenwell et al. 2022).

2. Build models using the tuned parameters in the gbm.step function from the dismo package (Hijmans et al. 2021)

3. Drop variables that don’t improve model performance using dismo::gbm.simplify.

4. Rerun models with the reduced set of predictor variables.

Generate offset raster layers

Generate offset rasters from the final models using dismo::predict. See 1_code/r_notebooks/7_make_predictive_offset_raster.Rmd for code.

To export the offset layers to a Google Drive:

1. paste the following into the Google Earth Engine code editor.

//load temperature offsets
var temperature_offsets = ee.Image("projects/ee-bgcasey-climate/assets/temperature_offsets");

// // Optional: define a study area. Upload a shape file as a new asset and add it to the code below.
// var aoi= ee.FeatureCollection("projects/ee-bgcasey-climate/assets/alberta_bc");

Export.image.toDrive({ 
  image: temperature_offsets,
  description: 'temperature_offsets', // the desired file names
  folder: 'temperature_offsets', //specify a google drive folder
  crs:'EPSG:3348', // set the CRS
  scale: 30, // set the scale of the exported images
  // region: aoi, //set the study area if defined
  maxPixels: 6000000000
});

2. if you want to, define a study area by uploading a shape file as an asset and adding the asset to the code.

3. Click run.

   

4. Open the tasks tab and click RUN next to the temperature_offsets task.

   

5. Click RUN in the image export pop-up window.

   

References

Abatzoglou, John T, Solomon Z Dobrowski, Sean A Parks, and Katherine C Hegewisch. 2018. “TerraClimate, a High-Resolution Global Dataset of Monthly Climate and Climatic Water Balance from 1958–2015.” Scientific Data 5 (1): 1–12.

Buchhorn, Marcel, Myroslava Lesiv, Nandin-Erdene Tsendbazar, Martin Herold, Luc Bertels, and Bruno Smets. 2020. “Copernicus Global Land Cover Layers—Collection 2.” Remote Sensing 12 (6): 1044.

Canada, Natural Resources. 2015. “Canadian Digital Elevation Model, 1945–2011.” Natural Resources Canada Ottawa, ON, Canada.

Estevo, Cesar A., Diana Stralberg, Scott E. Nielsen, and Erin Bayne. 2022. “Topographic and Vegetation Drivers of Thermal Heterogeneity Along the Boreal–Grassland Transition Zone in Western Canada: Implications for Climate Change Refugia.” Ecology and Evolution 12 (6): e9008. https://doi.org/https://doi.org/10.1002/ece3.9008.

Greenwell, Brandon, Bradley Boehmke, Jay Cunningham, and GBM Developers. 2022. Gbm: Generalized Boosted Regression Models. https://github.com/gbm-developers/gbm.

Heidinger, AK, Michael J Foster, Andi Walther, and Xuepeng Zhao. 2014. “NOAA Climate Data Record (CDR) of Cloud Properties from AVHRR Pathfinder Atmospheres-Extended (PATMOS-x), Version 5.3.” NOAA National Centers for Environmental Information: NOAA CDR Program.

Hersbach, Hans, Bill Bell, Paul Berrisford, Gionata Biavati, András Horányi, Joaquín Muñoz Sabater, Julien Nicolas, et al. 2018. “ERA5 Hourly Data on Single Levels from 1979 to Present.” Copernicus Climate Change Service (C3s) Climate Data Store (Cds) 10 (10.24381).

Hijmans, Robert J., Steven Phillips, John Leathwick, and Jane Elith. 2021. Dismo: Species Distribution Modeling. https://rspatial.org/raster/sdm/.

Kuhn, Max, Kjell Johnson, et al. 2013. Applied Predictive Modeling. Vol. 26. Springer.

Lang, Nico, Walter Jetz, Konrad Schindler, and Jan Dirk Wegner. 2022. “A High-Resolution Canopy Height Model of the Earth.” arXiv Preprint arXiv:2204.08322.

Moritz, Steffen, and Thomas Bartz-Beielstein. 2017. “imputeTS: Time Series Missing Value Imputation in R.” The R Journal 9 (1): 207–18. https://doi.org/10.32614/RJ-2017-009.

Pebesma, Edzer. 2022. Sf: Simple Features for r. https://CRAN.R-project.org/package=sf.

Sørensen, Rasmus, and Jan Seibert. 2007. “Effects of DEM Resolution on the Calculation of Topographical Indices: TWI and Its Components.” Journal of Hydrology 347 (1-2): 79–89.

Survey, U. S. Geological. 2018. “Landsat 4-7 Surface Reflectance (Ledaps) Product Guide.” Sioux Falls, SD, USA: Department of the Interior, U.S. Geological Survey.

Tennekes, Martijn. 2022. Tmap: Thematic Maps. https://github.com/r-tmap/tmap.

Theobald, David M, Dylan Harrison-Atlas, William B Monahan, and Christine M Albano. 2015. “Ecologically-Relevant Maps of Landforms and Physiographic Diversity for Climate Adaptation Planning.” PloS One 10 (12): e0143619.

Wang, Tongli, Andreas Hamann, Dave Spittlehouse, and Carlos Carroll. 2016. “Locally Downscaled and Spatially Customizable Climate Data for Historical and Future Periods for North America.” PloS One 11 (6): e0156720.

Wood, Wendy H, Shawn J Marshall, Shannon E Fargey, and Terri L Whitehead. 2017. “Daily Temperature Data from the Foothills Climate Array Mesonet, Canadian Rocky Mountains, 2005-2010.” PANGAEA. https://doi.org/10.1594/PANGAEA.880611.

Yamazaki, Dai, Daiki Ikeshima, Jeison Sosa, Paul D Bates, George H Allen, and Tamlin M Pavelsky. 2019. “MERIT Hydro: A High-Resolution Global Hydrography Map Based on Latest Topography Dataset.” Water Resources Research 55 (6): 5053–73.