read email from Ashley about things being counted differently for 2024 (and maybe also 2025)

Also, the SFM site in our Gate creek study was added to our fire project in 2023 (BayLee named it SFGate or something), but we do have data for macros and many other parameters going back to 2021 (Gate Creek Study) if you want to include that in overall dataset summaries. The NF Gate site they added is not at the same location as our NF sites from that study though.

HBI = Hilsenhoff Biotic Index– estimates the overall tolerance of the community in a sampled area, weighted by the relative abundance of each taxonomic group (family, genus, etc.). Organisms are assigned a tolerance number from 0 to 10 pertaining to that group’s known sensitivity to organic pollutants; 0 being most sensitive, 10 being most tolerant.

CG=Collector-Gatherer; CF=Collector-Filterer; PH=Piercer-Herbivore; PR=Predator; SH=Shredder; SC=Scraper; UN=Unknown CG+CF+PH+PR+SH+SC=1

Things to test: Could earlier years have stronger correlations between fire metrics and sensitivity metrics?? Aka are things recovering over time. What is: %DomTaxa??

Continuous Unbound: Taxa richness, #EPT Taxa, HBI, BecksBioticIndex, GenusRichness, BSTI, ShannonDiversity, Density, SimpsonsDiversity, Abundance, Cray+Moll % %EPT, %DomTaxa, %Chironomidae, %Clinger, %Tolerant, %intolerant(but normal), %Sensitive (but normal), %Tanytarsini, %Baetidae, %Diptera, %Hydropsychidae, %CF, %CG(normal), %MH, %PH, %PR (normal), %SC (normal), %SH,

Is %EPT fundamentally a proportion bounded between 0 and 1? Yes. That is why a beta GLMM is preferred—not because the histogram isn’t bell-shaped, but because the beta distribution naturally models bounded proportions and accounts for the relationship between the mean and variance.

## Warning in check_dep_version(dep_pkg = "TMB"): package version mismatch: 
## glmmTMB was built with TMB package version 1.9.21
## Current TMB package version is 1.9.17
## Please re-install glmmTMB from source or restore original 'TMB' package (see '?reinstalling' for more information)

Sediment Indices: BTSI FSBI BSTI (Benthic Sediment Tolerance Index) and FSBI (Fine Sediment Biotic Index)= often literature- or expert-based FSBI scores = statistically calibrated to sediment gradients

FSBI=∑(ni × si)/ ∑ni ni = number of individuals, si= fine sediment tolerance score for taxon. BSTI=∑(ai × ti)/ ∑ai ai = abundance, ti= sediment tolerance score for taxon.

We do have a stronger and stronger trend with BSTI (it is wrong in this file as BTSI) in years 3 and 4. The BSTI was specifically developed for Oregon (Hubler et al. 2016) the FSBI was developed for PNW (but most data from WA). “macroinvertebrate assemblages in more extensively burned watersheds shifted toward taxa more tolerant of fine sediments (higher BTSI) by the 3rd year post-wildfire”- Joe’s interpretation.

Merge with site details.

Raw data means Figures (for all metrics, not just macros)

Beta GLMM- because data are a % (don’t want predictions to be negative etc.). Values have to be between 0-1 for this model.

This model asks: Does EPT change through time? Does EPT differ among burn severity classes? Does the rate of change through time depend on burn severity? The third question is the interaction.Ecologically, this corresponds to asking:Do streams with different burn severities recover differently over time?

First chunk is just a single variable– following chunk loops over all % variables.

Checking of model fit: “Model assumptions were assessed using simulated residuals from the DHARMa package. Residual diagnostics showed no evidence of non-uniformity, overdispersion, or influential outliers, indicating that the beta GLMM provided an adequate fit to the data.” Chunk is just single example- EPT only

Interpreting results from the statistical model (^above). We want to know if recovery if different in low and high severity (burned severity is ordered correctly to do so). We are mostly interested in the effect of Year:Low and Year:High.

Hand made loop

## Running: EPT

## Running: DomTaxa
## Warning in newton(lsp = lsp, X = G$X, y = G$y, Eb = G$Eb, UrS = G$UrS, L = G$L,
## : Fitting terminated with step failure - check results carefully

## Running: Chironomidae

## Running: Clinger

## Running: Tolerant

## Running: Intolerant

## Running: Sensitive

## Running: Tanytarsini

## Running: Baetidae

## Running: Diptera

## Running: Hydropsychidae

## Running: CF

## Running: CG

## Running: MH

## Running: PR

## Running: SC

## Running: SH

LMER FOR continuous unbound data- linear mixed effects model. need to check for normal residuals, etc.

## Running: TaxaRichness

## Running: #EPT_Taxa

## Running: HBI

## Running: ShannonDiversity

## Running: Density

## Running: BecksBioticIndex

## Running: SimpsonsDiversity

## Running: Abundance

## Running: GenusRichness

## Running: BSTI

Effect Size Figure

##  [1] "%EPT"            "%DomTaxa"        "%Chironomidae"   "%Clinger"       
##  [5] "%Tolerant"       "%Intolerant"     "%Sensitive"      "%Tanytarsini"   
##  [9] "%Baetidae"       "%Diptera"        "%Hydropsychidae" "%CF"            
## [13] "%CG"             "%MH"             "%PR"             "%SC"            
## [17] "%SH"
##  [1] "TaxaRichness"      "#EPT_Taxa"         "HBI"              
##  [4] "ShannonDiversity"  "Density"           "BecksBioticIndex" 
##  [7] "SimpsonsDiversity" "Abundance"         "GenusRichness"    
## [10] "BSTI"

FFGs!!!!!!!!!! try Dirichlet regression

GAMM- for non linear situations. Need to think about if expected responses should be linear or not. Some could vary over years if perhaps very dependent on flow- other no, large wood counts, canopy, and dunno about macro indices. Could see if the raw values persite and as burn groups jump around a lot or not.

Metric tests! Make a correlation matrix to focus on the site detail metrics versus all of the indices.