Predicting Violent Conflict in Africa

# Predicting Violent Conflict in Africa
## Leveraging Open Geodata and Deep Learning for Spatio-Temporal Event Detection
### Darius A. Görgen
### B.Sc. Geography, University of Marburg
### 2021-05-05

---

# Content

Introduction
--
Methodology
--
Main Findings
--
Discussion
--
Conclusion
--
]

]

---

# Introduction

---

# Introduction

.pull-left[
- violent conflicts impede social and economical development and put environmental 
resources at risk

- we observe complex non-linear relationships with natural resources, e.g.
food production (see [Koren, 2018](https://doi.org/10.1093/ajae/aax106) and [Buhaug et al., 2015](https://doi.org/10.1088/1748-9326/10/12/125015))

- research activities on violent conflicts are found between inferential statistics
and prediction models (see [Ward et al, 2010](https://doi.org/10.1177/0022343309356491))

- environmental variables as predictors are barley found despite a public focus
on environmental change and its consequences

- on the African continent violent conflict seems to be on the rise during the
21st millennium
]

<br>
<br>
<br>
.pull-right[
![:scale 90%](assets/01-intro-conflicts-1.svg)
]

---

# Research question

---

# Research question

<br>

*Does a modern deep learning framework and the vast availability
of open geodata positively impact the task of spatio-temporal conflict detection?*

<br>

- **H1**: Environmental predictors increase the performance of deep learning models for the
conflict prediction task over models based solely on the conflict history and structural
variables.

- **H2**: Aggregating predictor and response variables on the basis of sub-basin watersheds
delivers better predictive performance than aggregating on sub-national administrative
districts.

---

# Methodology

---

# Methodology

## (1) Spatio-Temporal Aggregation

- spatio-temporal aggregation on two units of analysis (_adm_ and _bas_) and 4 spatial
buffers (0, 50, 100, 200 km)

- temporal split of training, validation, and testing data sets (2001 - 2016, 2017 - 2018, 2019)

- splitting into 3 predictor sets: Conflict History (CH), Structural Variables (SV), and Environmental Variables (EV)

> `$X_t^L= x_t^1,x_{t+1}^1...,x_{t-N+1}^1,...\ ..., x_t^L,...,x_{t-N+1}^L$`

]

.pull-right[
<img src = "assets/wf.svg", width = 70%, height = 70%, style = "position:absolute; top: 20%; left: 45%;"></img>
]

---

# Methodology

## (2) Response Variable & Bayesian Optimization

- binary encoding of response variable (3 conflict classes + 1 combined class)

- Bayesian optimization of CNN-LSTM Hyperparameters based on combined conflict class (**cb**)

- 100 random trials + 100 optimization trials for all combinations of *aggregation units* **per** *predictor sets*

> `$x_{t} = arg \;\underset{x}{max} \, u(x|D_{1:t-1})$`

]

.pull-right[
<img src = "assets/wf.svg", width = 70%, height = 70%, style = "position:absolute; top: 20%; left: 45%;"></img>
]

---

# Methodology

## (3) Model training

- neural networks are initiated with parameters from BO

- simple logistic regression on the most complex predictor set for baseline reference

- training is repeated 10 times for each combination of *aggregation units* **x** *predictor sets* **x** *outcome classes* = 280 models

]

.pull-right[
<img src = "assets/wf.svg", width = 70%, height = 70%, style = "position:absolute; top: 20%; left: 45%;"></img>
]

---

# Methodology

## (4) Testing

- performance is logged based on the testing set for each model

- performance is logged for each month in a 12 month prediction horizon

- Metrics: F2-score, AUC, AUPR, Precision, Sensitivity, Specificity

> `$F_\beta = (1+\beta^2)\frac{Precision*Sensitivity}{\beta^2*Precision+Sensitivity}$`

- ANOVA: 
  - aggregation unit (*adm*, *bas*)
  - conflict class (**cb**, **sb**, **ns**, **os**)
  - logistic regression + predictor sets (LR, CH, SV, EV)

]

.pull-right[
<img src = "assets/wf.svg", width = 70%, height = 70%, style = "position:absolute; top: 20%; left: 45%;"></img>
]

---

# Results

---

# Results

## Global Performance

---

# Results

## Temporal Performance

---

# Results

## Spatial Performance - *adm*

]

---

# Results

## Spatial Performance - *bas*

]

---

# Results

## ANOVA

.pull-left[
<table class="table table" style="margin-left: auto; margin-right: auto; font-size: 12px; margin-left: auto; margin-right: auto;border-bottom: 0;">
<caption style="font-size: initial !important;">Results of the Welch-James ANOVA.</caption>
 <thead>
<tr>
<th style="empty-cells: hide;border-bottom:hidden;" colspan="1"></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-weight: bold; " colspan="2"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">cb</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-weight: bold; " colspan="2"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">sb</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-weight: bold; " colspan="2"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">ns</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-weight: bold; " colspan="2"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">os</div></th>
</tr>
  <tr>
   <th style="text-align:left;"> Term </th>
   <th style="text-align:left;"> p </th>
   <th style="text-align:left;"> Sig. </th>
   <th style="text-align:left;"> p </th>
   <th style="text-align:left;"> Sig. </th>
   <th style="text-align:left;"> p </th>
   <th style="text-align:left;"> Sig. </th>
   <th style="text-align:left;"> p </th>
   <th style="text-align:left;"> Sig. </th>
  </tr>
 </thead>
<tbody>
  <tr>
   <td style="text-align:left;"> unit </td>
   <td style="text-align:left;"> 0.00e+00 </td>
   <td style="text-align:left;"> *** </td>
   <td style="text-align:left;"> 1.77e-03 </td>
   <td style="text-align:left;"> ** </td>
   <td style="text-align:left;"> 3.65e-01 </td>
   <td style="text-align:left;">  </td>
   <td style="text-align:left;"> 2.41e-08 </td>
   <td style="text-align:left;"> *** </td>
  </tr>
  <tr>
   <td style="text-align:left;"> type </td>
   <td style="text-align:left;"> 0.00e+00 </td>
   <td style="text-align:left;"> *** </td>
   <td style="text-align:left;"> 0.00e+00 </td>
   <td style="text-align:left;"> *** </td>
   <td style="text-align:left;"> 0.00e+00 </td>
   <td style="text-align:left;"> *** </td>
   <td style="text-align:left;"> 0.00e+00 </td>
   <td style="text-align:left;"> *** </td>
  </tr>
  <tr>
   <td style="text-align:left;"> unit:type </td>
   <td style="text-align:left;"> 3.77e-05 </td>
   <td style="text-align:left;"> *** </td>
   <td style="text-align:left;"> 2.70e-03 </td>
   <td style="text-align:left;"> ** </td>
   <td style="text-align:left;"> 1.89e-02 </td>
   <td style="text-align:left;"> * </td>
   <td style="text-align:left;"> 6.70e-04 </td>
   <td style="text-align:left;"> *** </td>
  </tr>
</tbody>
<tfoot><tr><td style="padding: 0; " colspan="100%">
<span style="font-style: italic;">General:</span> <sup></sup> The significance level is indicated according to: *p &lt;= 0.05, **p &lt;= 0.01, ***p &lt;= 0.001</td></tr></tfoot>
</table>

- assumption of equal variance for Fisher's ANOVA is violated
- Welch's ANOVA does not assume equal variance
- interaction terms are significant for all outcome variables
<br>
&rightarrow; focus on interaction terms

]

.pull-right[
<img src = "assets/appendix-anova-qq-1.svg", width = 33%, height = 33%, style = "position:absolute; top: 20%; left: 60%;"></img>
<img src = "assets/appendix-anova-resid-1.svg", width = 33%, height = 33%, style = "position:absolute; top: 55%; left: 60%;"></img>
]

---

# Results

## ANOVA

- regarding **H1**:
  - for *adm* districts, no significant difference from CH to EV set is observed for
  any outcome variable

- for *bas* districts a significant difference from the CH to the EV theme is 
  observed for the **cb** outcome variable accounting for 7.2 points in F2-score

- for *bas* districts and **sb** and  **os** outcome variables, the difference is not
  significant (p-value ~ 0.1), the estimate is about 4.6 to 4.9 points in F2-score

- regarding **H2**:
  - *bas* districts consistently show significant higher estimates than *adm* districts
  except for the **ns** outcome variable in a range between 3.7 to 9.9 points in F2-score

---

# Discussion

---

# Discussion

- subtle balance between a model's precision and sensitivity

<br>

- trade-off between spatial detail and predictive performance

<br>

- unfamiliarity by non-experts with watershed boundaries

<br>

- conflict types and modes currently not fully included

<br>

- focus on predictors available in gridded data sets

<br>

- observed differences might be due to subtle changes in DL architectures

---

# Conclusion

---

# Conclusion

- DL and open geodata can contribute to the conflict prediction task

<br>

- environmental variables aggregated on watershed levels seem to increase
performance

<br>

- availability of data on both sides of the equation is increasing

<br>

- predictive analysis enhances scientific understanding and adds value
to ongoing conflict prevention efforts

---

# Main references

.font90[
- Colaresi & Mahmood 2017. Do the robot: Lessons from machine learning to
improve conflict forecasting. Journal of Peace Research 54, 193–214. [https://doi.org/10.1177/0022343316682065](https://doi.org/10.1177/0022343316682065)
- Collier & Hoeffler 1998. On economic causes of civil war. Oxford Economic Papers 50, 563–573. [https://doi.org/10.1093/oep/50.4.563](https://doi.org/10.1093/oep/50.4.563)
- Fearon & Laitin 2003. Ethnicity, Insurgency, and Civil War. American 
Political Science Review 97, 75–90. [https://doi.org/10.1017/S0003055403000534](https://doi.org/10.1017/S0003055403000534)
- Halkia et al. 2020.
The Global Conflict Risk Index: A quantitative tool for policy support on conflict
prevention. Progress in Disaster Science 6, 100069. [https://doi.org/10.1016/j.pdisas.2020.100069](https://doi.org/10.1016/j.pdisas.2020.100069)
- Hegre et al. 2019. ViEWS: A political violence early-warning system. Journal of Peace Research 56,
155–174. [https://doi.org/10.1177/0022343319823860](https://doi.org/10.1177/0022343319823860)
- Homer-Dixon 1994. Environmental Scarcities and Violent Conflict: Evidence
from Cases. International Security 19, 5. [https://doi.org/10.2307/2539147](https://doi.org/10.2307/2539147)
- Kuzma et al. 2020. Leveraging Water Data in a Machine Learning-Based Model for Forecasting
Violent Conflict. Technical note. [WWW Document]. URL [https://www.wri.org/publication/leveraging-water-data](https://www.wri.org/publication/leveraging-water-data )
- Sachs & Warner 1995. Natural Resource Abundance and Economic Growth. 
National Bureau of Economic Research. [https://doi.org/10.3386/w5398](https://doi.org/10.3386/w5398)
- Ward et al 2010. The perils of policy by p-value:
Predicting civil conflicts. Journal of Peace Research 47, 363–375. [https://doi.org/10.1177/0022343309356491](https://doi.org/10.1177/0022343309356491)
- Yu et al. 2019. A Review of Recurrent Neural Networks: LSTM Cells and Network Architectures. 
Neural Computation 31, 1235–1270. [https://doi.org/10.1162/neco_a_01199](https://doi.org/10.1162/neco_a_01199)
]

---

# Extra Slides

---

# CNN

`$X^l_\beta = \sigma \Big( \sum\limits^L_{i=1} X^{l-1}_i \cdot k^l_{i\beta} + b^l_{i\beta} \Big)$`

]

---

# LSTM

.center[
<figure>
<img src = "assets/img/LSTM.png", width = 50%, height = 50%></img>
<a href="https://doi.org/10.1162/neco_a_01199">
<figcaption>Source: Yu et al. (2019)</figcaption>
</a>
</figure>

$$
`\begin{split}
f(t) = \sigma (W_{fh} h_{t-1} + W_{fx}x_t + b_f) \\
i(t) = \sigma (W_{ih} h_{t-1} + W_{ix}x_t + b_i) \\
\widetilde{c}(t) = tanh(W_{\widetilde{c}h} h_{t-1} + W_{\widetilde{c}x}x_t + b_{\widetilde{c}}) \\
c(t) = f(t) \cdot c_{t-1} + i(t) \cdot \widetilde{c}_t \\
o(t) = \sigma (W_{oh} h_{t-1} + W_{ox}x_t + b_0) \\
h_t = o_t \cdot tanh(c_t)
\end{split}`
$$
]

---

# Network Structure

---

# Predictors

<table class="table" style="font-size: 10px; margin-left: auto; margin-right: auto;">
<caption style="font-size: initial !important;">Spatio-temporal properties of predictor variables.</caption>
 <thead>
  <tr>
   <th style="text-align:left;"> Name </th>
   <th style="text-align:center;"> Spatial Resolution </th>
   <th style="text-align:center;"> Temporal Resolution </th>
   <th style="text-align:center;"> Unit </th>
   <th style="text-align:left;"> Aggregation </th>
  </tr>
 </thead>
<tbody>
  <tr grouplength="1"><td colspan="5" style="border-bottom: 1px solid;"><strong>Baseline</strong></td></tr>
<tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Conflict history </td>
   <td style="text-align:center;"> - </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> binary </td>
   <td style="text-align:left;"> - </td>
  </tr>
  <tr grouplength="14"><td colspan="5" style="border-bottom: 1px solid;"><strong>Structural</strong></td></tr>
<tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Terrain Ruggedness Index (log) </td>
   <td style="text-align:center;"> 0.0008° </td>
   <td style="text-align:center;"> static </td>
   <td style="text-align:center;"> m </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Travel time (log) </td>
   <td style="text-align:center;"> 0.008° </td>
   <td style="text-align:center;"> static </td>
   <td style="text-align:center;"> minutes </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Livestock (log) </td>
   <td style="text-align:center;"> 0.08° </td>
   <td style="text-align:center;"> static </td>
   <td style="text-align:center;"> 2010 heads </td>
   <td style="text-align:left;"> sum </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Population (log) </td>
   <td style="text-align:center;"> 0.008° </td>
   <td style="text-align:center;"> yearly </td>
   <td style="text-align:center;"> persons </td>
   <td style="text-align:left;"> sum </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Youth bulge </td>
   <td style="text-align:center;"> 0.008° </td>
   <td style="text-align:center;"> yearly </td>
   <td style="text-align:center;"> % </td>
   <td style="text-align:left;"> sum </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Dependency ratio </td>
   <td style="text-align:center;"> 0.008° </td>
   <td style="text-align:center;"> yearly </td>
   <td style="text-align:center;"> % </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> GDP (log) </td>
   <td style="text-align:center;"> 0.08° </td>
   <td style="text-align:center;"> yearly </td>
   <td style="text-align:center;"> 2011 USD </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Cropland </td>
   <td style="text-align:center;"> 0.005° </td>
   <td style="text-align:center;"> yearly </td>
   <td style="text-align:center;"> % </td>
   <td style="text-align:left;"> sum </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Forest cover </td>
   <td style="text-align:center;"> 0.005° </td>
   <td style="text-align:center;"> yearly </td>
   <td style="text-align:center;"> % </td>
   <td style="text-align:left;"> sum </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Builtup area </td>
   <td style="text-align:center;"> 0.005° </td>
   <td style="text-align:center;"> yearly </td>
   <td style="text-align:center;"> % </td>
   <td style="text-align:left;"> sum </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Grassland </td>
   <td style="text-align:center;"> 0.005° </td>
   <td style="text-align:center;"> yearly </td>
   <td style="text-align:center;"> % </td>
   <td style="text-align:left;"> sum </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Shrubland </td>
   <td style="text-align:center;"> 0.005° </td>
   <td style="text-align:center;"> yearly </td>
   <td style="text-align:center;"> % </td>
   <td style="text-align:left;"> sum </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Barren land </td>
   <td style="text-align:center;"> 0.005° </td>
   <td style="text-align:center;"> yearly </td>
   <td style="text-align:center;"> % </td>
   <td style="text-align:left;"> sum </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Water bodies </td>
   <td style="text-align:center;"> 0.005° </td>
   <td style="text-align:center;"> yearly </td>
   <td style="text-align:center;"> % </td>
   <td style="text-align:left;"> sum </td>
  </tr>
</tbody>
</table>
]

.pull-right[
<table class="table" style="font-size: 10px; margin-left: auto; margin-right: auto;border-bottom: 0;">
<caption style="font-size: initial !important;">Spatio-temporal properties of predictor variables. (cont.)</caption>
 <thead>
  <tr>
   <th style="text-align:left;"> Name </th>
   <th style="text-align:center;"> Spatial Resolution </th>
   <th style="text-align:center;"> Temporal Resolution </th>
   <th style="text-align:center;"> Unit </th>
   <th style="text-align:left;"> Aggregation </th>
  </tr>
 </thead>
<tbody>
  <tr grouplength="14"><td colspan="5" style="border-bottom: 1px solid;"><strong>Environmental</strong></td></tr>
<tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Precipitation </td>
   <td style="text-align:center;"> 0.05° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> mm </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Precipitation anomaly </td>
   <td style="text-align:center;"> 0.05° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> mm </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> SPI </td>
   <td style="text-align:center;"> 0.05° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> - </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> SPEI </td>
   <td style="text-align:center;"> 0.05° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> - </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Land Surface Temperature </td>
   <td style="text-align:center;"> 0.05° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> K </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Evapotranspiration </td>
   <td style="text-align:center;"> 0.01° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> kg/m² </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Gross Primary Productivity </td>
   <td style="text-align:center;"> 0.01° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> kg C/m² </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Precipitatation agr. </td>
   <td style="text-align:center;"> 0.05° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> mm </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Precipitation anonmaly agr. </td>
   <td style="text-align:center;"> 0.05 </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> mm </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> SPI agr. </td>
   <td style="text-align:center;"> 0.05° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> - </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> SPEI agr. </td>
   <td style="text-align:center;"> 0.05° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> - </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Land Surface Temperature agr. </td>
   <td style="text-align:center;"> 0.005° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> K </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Evapotranspiration agr. </td>
   <td style="text-align:center;"> 0.005° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> kg/m² </td>
   <td style="text-align:left;"> mean </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> Gross Primary Productivity agr. </td>
   <td style="text-align:center;"> 0.005° </td>
   <td style="text-align:center;"> monthly </td>
   <td style="text-align:center;"> kg C/m² </td>
   <td style="text-align:left;"> mean </td>
  </tr>
</tbody>
<tfoot><tr><td style="padding: 0; " colspan="100%">
<span style="font-style: italic;">General:</span> <sup></sup> Variables denoted with agr. were calculated by a multiplicative interaction with a binary cropland mask.</td></tr></tfoot>
</table>
]

---

# Bayesian Optimization

<table class="table" style="font-size: 7px; margin-left: auto; margin-right: auto;border-bottom: 0;">
<thead>
<tr>
<th style="empty-cells: hide;border-bottom:hidden;" colspan="1"></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-weight: bold; " colspan="2"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">Conflict History (CH)</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-weight: bold; " colspan="2"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">Structural Variables (SV)</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-weight: bold; " colspan="2"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">Environmental Variables (EV)</div></th>
</tr>
<tr>
<th style="empty-cells: hide;border-bottom:hidden;" colspan="1"></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-style: italic; " colspan="1"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">adm</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-style: italic; " colspan="1"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">bas</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-style: italic; " colspan="1"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">adm</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-style: italic; " colspan="1"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">bas</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-style: italic; " colspan="1"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">adm</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-style: italic; " colspan="1"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">bas</div></th>
</tr>
</thead>
<tbody>
  <tr>
   <td style="text-align:left;"> `$double\_cnn$` </td>
   <td style="text-align:center;"> Yes/Yes/Yes/Yes </td>
   <td style="text-align:center;"> Yes/Yes/Yes/Yes </td>
   <td style="text-align:center;"> Yes/Yes/Yes/Yes </td>
   <td style="text-align:center;"> No/No/No/No </td>
   <td style="text-align:center;"> Yes/Yes/Yes/Yes </td>
   <td style="text-align:center;"> Yes/Yes/Yes/Yes </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$a_{cnn}$` </td>
   <td style="text-align:center;"> softsign/hard_sigmoid/softsign/softmax </td>
   <td style="text-align:center;"> softsign/hard_sigmoid/sigmoid/softsign </td>
   <td style="text-align:center;"> softmax/softmax/hard_sigmoid/softmax </td>
   <td style="text-align:center;"> sigmoid/softplus/softplus/softmax </td>
   <td style="text-align:center;"> hard_sigmoid/softmax/hard_sigmoid/hard_sigmoid </td>
   <td style="text-align:center;"> softmax/softmax/hard_sigmoid/softmax </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$k_{cnn}$` </td>
   <td style="text-align:center;"> 106/125/45/78 </td>
   <td style="text-align:center;"> 86/99/39/128 </td>
   <td style="text-align:center;"> 95/70/63/42 </td>
   <td style="text-align:center;"> 41/102/67/43 </td>
   <td style="text-align:center;"> 19/41/94/99 </td>
   <td style="text-align:center;"> 92/76/66/42 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$\beta_{cnn}$` </td>
   <td style="text-align:center;"> 5/10/22/8 </td>
   <td style="text-align:center;"> 19/24/24/6 </td>
   <td style="text-align:center;"> 10/9/9/16 </td>
   <td style="text-align:center;"> 22/12/14/5 </td>
   <td style="text-align:center;"> 12/6/13/14 </td>
   <td style="text-align:center;"> 8/10/10/17 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$pool_1$` </td>
   <td style="text-align:center;"> `$max/max/avg./max$` </td>
   <td style="text-align:center;"> `$avg./max/max/max$` </td>
   <td style="text-align:center;"> `$max/avg./avg./max$` </td>
   <td style="text-align:center;"> `$avg./max/avg./avg.$` </td>
   <td style="text-align:center;"> `$max/avg./max/avg.$` </td>
   <td style="text-align:center;"> `$max/avg./avg./max$` </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$pool\_size$` </td>
   <td style="text-align:center;"> 15/23/3/14 </td>
   <td style="text-align:center;"> 24/19/10/21 </td>
   <td style="text-align:center;"> 12/18/17/18 </td>
   <td style="text-align:center;"> 16/6/10/9 </td>
   <td style="text-align:center;"> 12/16/9/22 </td>
   <td style="text-align:center;"> 12/18/17/16 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$pool_2$` </td>
   <td style="text-align:center;"> `$max/max/max/avg.$` </td>
   <td style="text-align:center;"> `$max/max/avg./avg.$` </td>
   <td style="text-align:center;"> `$max/max/max/avg.$` </td>
   <td style="text-align:center;"> `$max/max/avg./max$` </td>
   <td style="text-align:center;"> `$max/max/max/avg.$` </td>
   <td style="text-align:center;"> `$max/max/max/avg.$` </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$lstm\_layers$` </td>
   <td style="text-align:center;"> 2/2/2/2 </td>
   <td style="text-align:center;"> 3/3/3/3 </td>
   <td style="text-align:center;"> 2/2/2/2 </td>
   <td style="text-align:center;"> 2/2/2/2 </td>
   <td style="text-align:center;"> 3/3/3/3 </td>
   <td style="text-align:center;"> 2/2/2/2 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$n_1$` </td>
   <td style="text-align:center;"> 114/83/114/106 </td>
   <td style="text-align:center;"> 97/12/12/78 </td>
   <td style="text-align:center;"> 109/128/43/23 </td>
   <td style="text-align:center;"> 23/59/38/92 </td>
   <td style="text-align:center;"> 88/28/85/122 </td>
   <td style="text-align:center;"> 107/127/38/18 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$d_1$` </td>
   <td style="text-align:center;"> 0.22/0.08/0.14/0.49 </td>
   <td style="text-align:center;"> 0.23/0.2/0.38/0.25 </td>
   <td style="text-align:center;"> 0.02/0.09/0.07/0.06 </td>
   <td style="text-align:center;"> 0.22/0.14/0.02/0.2 </td>
   <td style="text-align:center;"> 0.2/0.1/0.09/0.31 </td>
   <td style="text-align:center;"> 0.02/0.07/0.05/0.06 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$n_2$` </td>
   <td style="text-align:center;"> 79/53/62/123 </td>
   <td style="text-align:center;"> 128/34/84/65 </td>
   <td style="text-align:center;"> 83/82/128/21 </td>
   <td style="text-align:center;"> 37/114/83/56 </td>
   <td style="text-align:center;"> 111/19/13/104 </td>
   <td style="text-align:center;"> 76/80/128/21 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$d_2$` </td>
   <td style="text-align:center;"> 0.05/0.28/0.37/0.08 </td>
   <td style="text-align:center;"> 0.5/0.24/0.05/0.16 </td>
   <td style="text-align:center;"> 0.34/0.03/0/0.29 </td>
   <td style="text-align:center;"> 0.26/0.12/0.46/0.47 </td>
   <td style="text-align:center;"> 0.23/0.04/0.09/0.49 </td>
   <td style="text-align:center;"> 0.34/0.02/0.01/0.31 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$n_3$` </td>
   <td style="text-align:center;"> -/-/-/- </td>
   <td style="text-align:center;"> 85/92/96/57 </td>
   <td style="text-align:center;"> -/-/-/- </td>
   <td style="text-align:center;"> -/-/-/- </td>
   <td style="text-align:center;"> 54/109/106/13 </td>
   <td style="text-align:center;"> -/-/-/- </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$d_3$` </td>
   <td style="text-align:center;"> -/-/-/- </td>
   <td style="text-align:center;"> 0.437/0.231/0.255/0.172 </td>
   <td style="text-align:center;"> -/-/-/- </td>
   <td style="text-align:center;"> -/-/-/- </td>
   <td style="text-align:center;"> 0.202/0.477/0.284/0.285 </td>
   <td style="text-align:center;"> -/-/-/- </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$a_{dense}$` </td>
   <td style="text-align:center;"> `$softplus$` </td>
   <td style="text-align:center;"> `$elu$` </td>
   <td style="text-align:center;"> `$relu$` </td>
   <td style="text-align:center;"> `$softplus$` </td>
   <td style="text-align:center;"> `$selu$` </td>
   <td style="text-align:center;"> `$relu$` </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$n_{dense}$` </td>
   <td style="text-align:center;"> 21 </td>
   <td style="text-align:center;"> 97 </td>
   <td style="text-align:center;"> 32 </td>
   <td style="text-align:center;"> 38 </td>
   <td style="text-align:center;"> 95 </td>
   <td style="text-align:center;"> 29 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$a_{out}$` </td>
   <td style="text-align:center;"> `$sigmoid$` </td>
   <td style="text-align:center;"> `$sigmoid$` </td>
   <td style="text-align:center;"> `$sigmoid$` </td>
   <td style="text-align:center;"> `$sigmoid$` </td>
   <td style="text-align:center;"> `$sigmoid$` </td>
   <td style="text-align:center;"> `$hard\_sigmoid$` </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$\pi$` </td>
   <td style="text-align:center;"> 0.4529 </td>
   <td style="text-align:center;"> 0.6534 </td>
   <td style="text-align:center;"> 0.2404 </td>
   <td style="text-align:center;"> 0.3856 </td>
   <td style="text-align:center;"> 0.5697 </td>
   <td style="text-align:center;"> 0.275 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$\alpha$` </td>
   <td style="text-align:center;"> 0.9245 </td>
   <td style="text-align:center;"> 0.7928 </td>
   <td style="text-align:center;"> 0.8056 </td>
   <td style="text-align:center;"> 0.9087 </td>
   <td style="text-align:center;"> 0.722 </td>
   <td style="text-align:center;"> 0.8544 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$\gamma$` </td>
   <td style="text-align:center;"> 6.2896 </td>
   <td style="text-align:center;"> 5.8844 </td>
   <td style="text-align:center;"> 7.8011 </td>
   <td style="text-align:center;"> 1.1974 </td>
   <td style="text-align:center;"> 4.8053 </td>
   <td style="text-align:center;"> 8.1588 </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$opti$` </td>
   <td style="text-align:center;"> `$adagrad$` </td>
   <td style="text-align:center;"> `$adam$` </td>
   <td style="text-align:center;"> `$adamax$` </td>
   <td style="text-align:center;"> `$adagrad$` </td>
   <td style="text-align:center;"> `$adamax$` </td>
   <td style="text-align:center;"> `$adamax$` </td>
  </tr>
  <tr>
   <td style="text-align:left;"> `$lr$` </td>
   <td style="text-align:center;"> 0.0244 </td>
   <td style="text-align:center;"> 0.0116 </td>
   <td style="text-align:center;"> 0.0246 </td>
   <td style="text-align:center;"> 0.0258 </td>
   <td style="text-align:center;"> 0.0117 </td>
   <td style="text-align:center;"> 0.0226 </td>
  </tr>
</tbody>
<tfoot><tr><td style="padding: 0; " colspan="100%">
<span style="font-style: italic;">General:</span> <sup></sup> Multiple values indicate the results for the input branch of buffer size 0/50/100/200 km, respecitvley.</td></tr></tfoot>
</table>

---

# ANOVA

<table class="table" style="font-size: 8px; margin-left: auto; margin-right: auto;border-bottom: 0;">
<caption style="font-size: initial !important;">Results of the Games-Howell test for difference in mean values.</caption>
 <thead>
<tr>
<th style="empty-cells: hide;border-bottom:hidden;" colspan="1"></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-weight: bold; " colspan="2"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">cb</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-weight: bold; " colspan="2"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">sb</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-weight: bold; " colspan="2"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">ns</div></th>
<th style="border-bottom:hidden;padding-bottom:0; padding-left:3px;padding-right:3px;text-align: center; font-weight: bold; " colspan="2"><div style="border-bottom: 1px solid #ddd; padding-bottom: 5px; ">os</div></th>
</tr>
  <tr>
   <th style="text-align:left;"> Contrast </th>
   <th style="text-align:left;"> Est. </th>
   <th style="text-align:left;"> p </th>
   <th style="text-align:left;"> Est. </th>
   <th style="text-align:left;"> p </th>
   <th style="text-align:left;"> Est. </th>
   <th style="text-align:left;"> p </th>
   <th style="text-align:left;"> Est. </th>
   <th style="text-align:left;"> p </th>
  </tr>
 </thead>
<tbody>
  <tr grouplength="4"><td colspan="9" style="border-bottom: 1px solid;"><strong>EV:LR</strong></td></tr>
<tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> adm:adm </td>
   <td style="text-align:left;"> 0.29299 </td>
   <td style="text-align:left;"> 2.4e-10*** </td>
   <td style="text-align:left;"> 0.40543 </td>
   <td style="text-align:left;"> 4.6e-10*** </td>
   <td style="text-align:left;"> 0.25543 </td>
   <td style="text-align:left;"> 7.1e-08*** </td>
   <td style="text-align:left;"> 0.28137 </td>
   <td style="text-align:left;"> 1.4e-10*** </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> adm:bas </td>
   <td style="text-align:left;"> 0.22865 </td>
   <td style="text-align:left;"> 3.1e-10*** </td>
   <td style="text-align:left;"> 0.34574 </td>
   <td style="text-align:left;"> 7.9e-10*** </td>
   <td style="text-align:left;"> 0.22216 </td>
   <td style="text-align:left;"> 3.0e-07*** </td>
   <td style="text-align:left;"> 0.22065 </td>
   <td style="text-align:left;"> 0.0e+00*** </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> bas:adm </td>
   <td style="text-align:left;"> 0.36458 </td>
   <td style="text-align:left;"> 9.9e-12*** </td>
   <td style="text-align:left;"> 0.42837 </td>
   <td style="text-align:left;"> 3.4e-10*** </td>
   <td style="text-align:left;"> 0.30397 </td>
   <td style="text-align:left;"> 3.7e-10*** </td>
   <td style="text-align:left;"> 0.32709 </td>
   <td style="text-align:left;"> 2.2e-10*** </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> bas:bas </td>
   <td style="text-align:left;"> 0.30024 </td>
   <td style="text-align:left;"> 9.4e-11*** </td>
   <td style="text-align:left;"> 0.36869 </td>
   <td style="text-align:left;"> 3.9e-10*** </td>
   <td style="text-align:left;"> 0.27070 </td>
   <td style="text-align:left;"> 3.1e-10*** </td>
   <td style="text-align:left;"> 0.26636 </td>
   <td style="text-align:left;"> 5.3e-11*** </td>
  </tr>
  <tr grouplength="4"><td colspan="9" style="border-bottom: 1px solid;"><strong>EV:CH</strong></td></tr>
<tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> adm:adm </td>
   <td style="text-align:left;"> 0.02770 </td>
   <td style="text-align:left;"> 1.2e-01 </td>
   <td style="text-align:left;"> 0.04354 </td>
   <td style="text-align:left;"> 2.1e-01 </td>
   <td style="text-align:left;"> -0.02863 </td>
   <td style="text-align:left;"> 7.7e-01 </td>
   <td style="text-align:left;"> 0.02135 </td>
   <td style="text-align:left;"> 2.1e-01 </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> adm:bas </td>
   <td style="text-align:left;"> -0.04619 </td>
   <td style="text-align:left;"> 4.9e-06*** </td>
   <td style="text-align:left;"> 0.04032 </td>
   <td style="text-align:left;"> 8.4e-01 </td>
   <td style="text-align:left;"> 0.06320 </td>
   <td style="text-align:left;"> 6.6e-01 </td>
   <td style="text-align:left;"> 0.00803 </td>
   <td style="text-align:left;"> 1.0e+00 </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> bas:adm </td>
   <td style="text-align:left;"> 0.09929 </td>
   <td style="text-align:left;"> 7.8e-06*** </td>
   <td style="text-align:left;"> 0.06649 </td>
   <td style="text-align:left;"> 1.6e-02* </td>
   <td style="text-align:left;"> 0.01991 </td>
   <td style="text-align:left;"> 8.6e-01 </td>
   <td style="text-align:left;"> 0.06706 </td>
   <td style="text-align:left;"> 1.6e-05*** </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> bas:bas </td>
   <td style="text-align:left;"> 0.02540 </td>
   <td style="text-align:left;"> 3.0e-03** </td>
   <td style="text-align:left;"> 0.06327 </td>
   <td style="text-align:left;"> 4.1e-01 </td>
   <td style="text-align:left;"> 0.11175 </td>
   <td style="text-align:left;"> 1.1e-01 </td>
   <td style="text-align:left;"> 0.05374 </td>
   <td style="text-align:left;"> 2.5e-02* </td>
  </tr>
  <tr grouplength="4"><td colspan="9" style="border-bottom: 1px solid;"><strong>EV:SV</strong></td></tr>
<tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> adm:adm </td>
   <td style="text-align:left;"> 0.00130 </td>
   <td style="text-align:left;"> 1.0e+00 </td>
   <td style="text-align:left;"> 0.01441 </td>
   <td style="text-align:left;"> 8.7e-01 </td>
   <td style="text-align:left;"> 0.04002 </td>
   <td style="text-align:left;"> 8.6e-01 </td>
   <td style="text-align:left;"> 0.01428 </td>
   <td style="text-align:left;"> 7.6e-01 </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> adm:bas </td>
   <td style="text-align:left;"> -0.03179 </td>
   <td style="text-align:left;"> 5.5e-04*** </td>
   <td style="text-align:left;"> -0.02382 </td>
   <td style="text-align:left;"> 4.8e-01 </td>
   <td style="text-align:left;"> -0.01401 </td>
   <td style="text-align:left;"> 9.6e-01 </td>
   <td style="text-align:left;"> -0.01023 </td>
   <td style="text-align:left;"> 9.8e-01 </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> bas:adm </td>
   <td style="text-align:left;"> 0.07289 </td>
   <td style="text-align:left;"> 2.0e-11*** </td>
   <td style="text-align:left;"> 0.03736 </td>
   <td style="text-align:left;"> 1.9e-02* </td>
   <td style="text-align:left;"> 0.08856 </td>
   <td style="text-align:left;"> 1.1e-01 </td>
   <td style="text-align:left;"> 0.06000 </td>
   <td style="text-align:left;"> 3.4e-04*** </td>
  </tr>
  <tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> bas:bas </td>
   <td style="text-align:left;"> 0.03980 </td>
   <td style="text-align:left;"> 5.9e-05*** </td>
   <td style="text-align:left;"> -0.00087 </td>
   <td style="text-align:left;"> 1.0e+00 </td>
   <td style="text-align:left;"> 0.03453 </td>
   <td style="text-align:left;"> 1.0e-02* </td>
   <td style="text-align:left;"> 0.03549 </td>
   <td style="text-align:left;"> 1.1e-01 </td>
  </tr>
  <tr grouplength="1"><td colspan="9" style="border-bottom: 1px solid;"><strong>EV:EV</strong></td></tr>
<tr>
   <td style="text-align:left;padding-left: 2em;" indentlevel="1"> bas:adm </td>
   <td style="text-align:left;"> 0.07159 </td>
   <td style="text-align:left;"> 3.4e-11*** </td>
   <td style="text-align:left;"> 0.02295 </td>
   <td style="text-align:left;"> 4.0e-01 </td>
   <td style="text-align:left;"> 0.04854 </td>
   <td style="text-align:left;"> 4.0e-02* </td>
   <td style="text-align:left;"> 0.04571 </td>
   <td style="text-align:left;"> 8.7e-06*** </td>
  </tr>
</tbody>
<tfoot><tr><td style="padding: 0; " colspan="100%">
<span style="font-style: italic;">General:</span> <sup></sup> Contrasts are indicated by ":" reading as the left-hand side compared to the right-hand side. Est. indicates the difference in mean, p indicates the p value with significance level according to: *p &lt;= 0.05, **p &lt;= 0.01, ***p &lt;= 0.001</td></tr></tfoot>
</table>