exclude: true <style type="text/css"> code.r{ font-size: 16px; } pre { font-size: 16px !important; } </style> --- class: split-two <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/4.7.0/css/font-awesome.min.css"> .column.bg-main1[ .font2.comfortaa.coral[Prediction of Seasonal Agricultural Productivity Anomalies Derived from MODIS Data for the Cultivated Land of Chile]<br></br></br> .font1.comfortaa[Francisco Zambrano, Anton Vrieling, Andy Nelson, Muichele Meroni, and Tsegaye Tadesse <br><br><br>] <img src="slide_img/logo_hemera.png" width=20%><img src="slide_img/Logo-UMAYOR.png" width=10%><br> .font1.comfortaa[
<i class="fas fa-link faa-vertical animated " style=" color:white;"></i> hemera.umayor.cl
<br>] .font1.comfortaa[
<i class="fab fa-github faa-pulse animated " style=" color:white;"></i> frzambra
<br>
<i class="fab fa-twitter faa-pulse animated " style=" color:00acee;"></i> @frzambra
<br>
<i class="fas fa-envelope faa-pulse animated " style=" color:white;"></i> francisco.zambrano@umayor.cl
<br>] .font_large[2019 Joint Satellite Conference, Boston, MA, USA </br> <img src="slide_img/ams_CENT_tall_tag.png" width=10%><img src="slide_img/noaa.jpg" width=10%><img src="slide_img/EUMETSAT.jpg" width=25%><br> October 3, 2019] .font1.comfortaa[
<i class="fas fa-link faa-vertical animated " style=" color:white;"></i> https://bit.ly/2pA0mnr
] ] .column.bg-main3.center[ .vmiddle[<img src="slide_img/sequia_agricola_araucania.gif" width=95%>] ] <!-- --- --> <!-- class: bg-main1 --> <!-- # Why am I here? --> <!-- -- --> <!-- ### Teach you how to code in R ❌ --> <!-- -- --> <!-- ### Teach you how you *should* use R ✔️ --> <!-- -- --> <!-- ### Teach you how to make writing R code enjoyable ✔️ --> <!-- -- --> <!-- ### Teach you how to learn R ✔️ --> <!-- --- --> <!-- class: middle bg-main1 --> <!-- # How most academics learn R --> <!-- <img src="slide_img/throw_into_pool.gif" width=50%> --> <!-- --- --> <!-- class: middle bg-main1 --> <!-- # How .yellow[should] you use R? --> --- layout: true --- class: split-two with-border border-white fade-row2-col1 fade-row3-col1 fade-row4-col1 .column[ .split-four[ .row.bg-main1[.content.font2[ What is the .yellow[problem] and how we .yellow[addressed] it? ]] .row.bg-main4[.content.font2[ Deriving the proxy (.yellow[output])
<i class="fas fa-map-marker-alt faa-float animated " style=" color:yellow;"></i>
]] .row.bg-main4[.content.font2[ Predictor variables
<i class="fas fa-chart-line faa-float animated "></i>
]] .row.bg-main4[.content.font2[ Prediction models
<i class="fas fa-walking faa-float animated "></i>
]] ]] .column.bg-main1[.content.center.vmiddle[ <!-- <img src="slide_img/GRETA-TRUMP-740x430.jpg" width=80%> --> <img src="slide_img/2014-2018-global-temperatures-gif.gif" width=80%> ]] <!-- --- --> <!-- class: hide-row2-col1 hide-row3-col1 hide-row4-col1 --> <!-- <img src="slide_img/GRETA-TRUMP-740x430.jpg" width=80%> --> <!-- <img src="slide_img/2014-2018-global-temperatures-gif.gif" width=80%> --> <!-- --- --> <!-- class: hide-row3-col1 hide-row4-col1 --> <!-- <img src="slide_img/sf_hex.gif" width=40%> --> <!-- <img src="slide_img/r4ds.png" width=40%> --> <!-- <img src="slide_img/chripsck_overflow.png" width=40%> --> <!-- --- --> <!-- class: hide-row4-col1 --> <!-- <img src="slide_img/DeepLearn.png" width=80%> --> <!-- --- --> <!-- class: --> <!-- <img src="slide_img/travolta.gif" width=80%> --> <!-- --- --> <!-- class: fade-row2-col1 fade-row3-col1 fade-row4-col1 --> --- layout: false class: bg-main1 # Cultivated land of Chile .center[<img src="slide_img/study_area_chile.jpg" width=110%>] --- layout: false class: bg-main1 # .yellow[Agricultural drought] in Chile .font2.comfortaa[A persistent .yellow[rainfall deficit] has been affecting Central-South Chile .yellow[since 2007].] .center[<img src="slide_img/SPI12_zcNDVI_2000-2017.png" width=85%>] <!-- --- --> <!-- layout: false --> <!-- class: bg-main1 --> <!-- # .yellow[Sequía] en Chile --> <!-- ## Año 2019 --> <!-- .center[<img src="slide_img/declaEmergenciaAgricola.png" width=80%>] --> --- layout: false class: bg-main1 # How we can study agricultural drought at regional scale? .font2.comfortaa[ - Crop growth model
- Agricultural census (yield, production, )
- Using a .yellow[proxy] for agricultural productivity.
- Commonly by vegetation indices (VIs) derived from satellite images. - NDVI one o the most common used indices. ] --- class: bg-main1 # .yellow[Proxy] proposed by our study .font2.comfortaa[Cumulative NDVI within the growing season] .center[<img src="slide_img/proxy_zcNDVI.png" width=80%>] --- class: bg-main1 # Proposed prediction by our study .font2.comfortaa[ - What we want to predict .yellow[(output)]? the proxy of agricultural productivity .yellow[(zcNDVI)] - What variables we used for the prediction (.yellow[Inputs]): - the proxy itself months before the end of the growing season (EOS). - Standardized precipitation index (SPI) at 1-, 3-, 6-, 12- y 24-months . - Climatic oscillation indices (PDO and MEI) - .yellow[When?:] forecast lead times from .yellow[one- to six-month] before the end of the growing season (EOS) - .yellow[Where?:] at census units, considering those which has cultivated land. ] --- class: split-two with-border border-white fade-row1-col1 fade-row3-col1 fade-row4-col1 .column[ .split-four[ .row.bg-main1[.content.font2[ What is the .yellow[problem] and how we .yellow[addressed] it? ]] .row.bg-main4[.content.font2[ Deriving the proxy (.yellow[output])
<i class="fas fa-map-marker-alt faa-float animated " style=" color:yellow;"></i>
]] .row.bg-main4[.content.font2[ Predictor variables (.yellow[inputs]) ]] .row.bg-main4[.content.font2[ Prediction models
<i class="fas fa-chart-line faa-float animated "></i>
]] ]] .column.bg-main1[.content.center.vmiddle[ <img src="slide_img/modis.png" width=80%> <img src="slide_img/sf_hex.gif" width=40%> <img src="slide_img/R_SpatialAnalytics.png" width=40%> ]] --- layout: true class: split-two with-border border-white .column[ .split-four[ .row.bg-main1[.content.font2[ Were determined the census units having > 10% of cultivated land. ]] .row.bg-main4[.content.font2[ Per census unit the Start Of the growing Season (SOS) was estimated ]] .row.bg-main4[.content.font2[ Per census unit the End Of the growing Season (EOS) was estimated ]] .row.bg-main4[.content.font2[ Finally, for each census units was calculated the proxy `\((zcNDVI)\)`]] ]] .column.bg-main3[.content.center.vmiddle[ {{content}} ]] --- class: fade-row2-col1 fade-row3-col1 fade-row4-col1 <img src="slide_img/study_area_croplands.png" width=40%> --- class: fade-row1-col1 fade-row3-col1 fade-row4-col1 <img src="slide_img/growing_season_sos.png" width=40%> <img src="slide_img/phenologyScale.png" width=40%> --- class: fade-row1-col1 fade-row2-col1 fade-row4-col1 <img src="slide_img/growing_season_length_EOS.png" width=40%> <img src="slide_img/phenologyScale.png" width=40%> --- class: fade-row1-col1 fade-row2-col1 fade-row3-col1 <img src="slide_img/proxy_zcNDVI.png" width=100%> <img src="slide_img/ImageCube.png" width=40%> <img src="slide_img/tidy-2.png" width=40%> --- layout: false class: bg-main1 # Validation of the .yellow[proxy] <img src="slide_img/validation_proxy.jpg" width=60%> <img src="slide_img/Proxy-1.png" width=35%> --- class: split-two with-border border-white fade-row1-col1 fade-row2-col1 fade-row4-col1 .column[ .split-four[ .row.bg-main1[.content.font2[ What is the .yellow[problem] and how we .yellow[addressed] it? ]] .row.bg-main4[.content.font2[ Deriving the proxy (.yellow[output])
<i class="fas fa-map-marker-alt faa-float animated " style=" color:yellow;"></i>
]] .row.bg-main4[.content.font2[ Predictor variables (.yellow[inputs]) ]] .row.bg-main4[.content.font2[ Prediction models
<i class="fas fa-chart-line faa-float animated "></i>
]] ]] .column.bg-main1[.content.center.vmiddle[ <img src="slide_img/CHC_USAID_logos.png" width=80%> <img src="slide_img/modis.png" width=80%> <img src="slide_img/meiv2.timeseries.png" width=80%> <img src="slide_img/PDO_Phase.gif" width=80%> ]] --- layout: true class: split-two with-border border-white .column[ .split-four[ .row.bg-main1[.content.font2[ All the predictors (zcNDVI, SPIs, PDO, and MEI) were calculated from the end of the season... ]] .row.bg-main4[.content.font2[ ...for forecasting lead times of one-month... ]] .row.bg-main4[.content.font2[ ...two-month... ]] .row.bg-main4[.content.font2[ ...until six-month...]] ]] .column.bg-main3[.content.center.vmiddle[ {{content}} ]] --- class: fade-row2-col1 fade-row3-col1 fade-row4-col1 <img src="slide_img/pred_lead0.png" width=100%> --- class: fade-row1-col1 fade-row3-col1 fade-row4-col1 <img src="slide_img/pred_lead1.png" width=100%> --- class: fade-row1-col1 fade-row2-col1 fade-row4-col1 <img src="slide_img/pred_lead2.png" width=100%> --- class: fade-row1-col1 fade-row2-col1 fade-row3-col1 <img src="slide_img/pred_lead6.png" width=100%> --- layout: false class: split-two with-border border-white fade-row1-col1 fade-row2-col1 fade-row3-col1 .column[ .split-four[ .row.bg-main1[.content.font2[ What is the .yellow[problem] and how we .yellow[addressed] it? ]] .row.bg-main4[.content.font2[ Deriving the proxy (.yellow[output])
<i class="fas fa-map-marker-alt faa-float animated " style=" color:yellow;"></i>
]] .row.bg-main4[.content.font2[ Predictor variables (.yellow[inputs]) ]] .row.bg-main4[.content.font2[ Prediction models
<i class="fas fa-chart-line faa-float animated "></i>
]] ]] .column.bg-main1[.content.center.vmiddle[ <img src="slide_img/sf_hex.gif" width=40%> <img src="slide_img/r4ds.png" width=40%> <img src="slide_img/R_SpatialAnalytics.png" width=40%> ]] --- layout: false class: bg-main1 # Deep Learning vs Linear regresion .font2.comfortaa[ - We used a modeling scheme of .yellow[LOOCV (leave-one-out cross-validation)] - Leaving one season out each time .center[<img src="slide_img/loocv.png" width=80%>] <!-- `$$RMSE_{cv} = \sqrt{\frac{\sum{(zcNDVI^S-\widehat{zcNDVI^S})^2}}{n}}$$` --> <!-- `$$R^2_{cv}$$` --> ] --- class: bg-main1 .font2.comfortaa[ # Deep Learning vs Regresión Lineal - Deep Learning: package .yellow[H2O] `(LeDell et al., 2014)` in
<i class="fab fa-r-project faa-bounce animated " style=" color:#165CAA;"></i>
- Hyperparameter optimization .yellow[Random Grid Search] - 15 predictors `X` 758 units `X` 17 seasons `X` 6 lead times = 1.159.740 data - Linear regresion: function .yellow[lm] in
<i class="fab fa-r-project faa-pulse animated " style=" color:#165CAA;"></i>
- 12 predictors `X` 758 units `X` 17 seasons `X` 6 lead times = 927.792 regresions - We selected the predictor that reach the minimum error `\((RMSE_{cv})\)` - Models evaluation: `\(RMSE_{cv}\)` y `\(R^2_{cv}\)` ] --- class: bg-main1 # Results .center[<img src="slide_img/boxplot_accuracies_r2.jpg" width=90%>] --- class: bg-main1 # Results .center[<img src="slide_img/boxplot_accuracies_rmse.jpg" width=90%>] --- class: bg-main1 # Results ## Linear regresion .center[<img src="slide_img/olr_map.jpg" width=90%>] --- class: bg-main1 # Results ## Deep Learning .center[<img src="slide_img/dl_map.jpg" width=90%>] --- layout: false class: bg-main1 # Final remarks .font2.comfortaa[ - Deep Learning (Black Box). - a simple linear regresion (.yellow[in this case]) equals Deep Learning - not enough data for Deep Learning `(LeDell, 2019)` - we should incorporate .yellow[soil moisture] as predictor - we .yellow[need] field data for validation ] --- class: bg-main1 center #Thanks! <img src="slide_img/thank-you.gif" width=90%> --- class: split-two <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/4.7.0/css/font-awesome.min.css"> .column.bg-main1[ .font2.comfortaa.coral[Prediction of Seasonal Agricultural Productivity Anomalies Derived from MODIS Data for the Cultivated Land of Chile]<br></br></br> .font1.comfortaa[Francisco Zambrano, Anton Vrieling, Andy Nelson, Muichele Meroni, and Tsegaye Tadesse <br><br><br>] <img src="slide_img/logo_hemera.png" width=20%><img src="slide_img/Logo-UMAYOR.png" width=10%><br> .font1.comfortaa[
<i class="fas fa-link faa-vertical animated " style=" color:white;"></i> hemera.umayor.cl
<br>] .font1.comfortaa[
<i class="fab fa-github faa-pulse animated " style=" color:white;"></i> frzambra
<br>
<i class="fab fa-twitter faa-pulse animated " style=" color:00acee;"></i> @frzambra
<br>
<i class="fas fa-envelope faa-pulse animated " style=" color:white;"></i> francisco.zambrano@umayor.cl
<br>] .font_large[2019 Joint Satellite Conference, Boston, MA, USA </br> <img src="slide_img/ams_CENT_tall_tag.png" width=10%><img src="slide_img/noaa.jpg" width=10%><img src="slide_img/EUMETSAT.jpg" width=25%><br> October 3, 2019] .font1.comfortaa[
<i class="fas fa-link faa-vertical animated " style=" color:white;"></i> https://bit.ly/2pA0mnr
] ] .column.bg-main3.center[ .vmiddle[<img src="slide_img/sequia_agricola_araucania.gif" width=95%>] ]