Using empirical and mechanistic models to predict crop suitability and productivity in climate change research

Using empirical and mechanistic models to predict crop suitability and productivity in climate change research

Anton Eitzinger [email protected]. Laderach, C. Navarro, B. Rodriguez

Decision and Policy Analysis DAPA, CIAT Nairobi, June 13th 2013

Why crop modeling in climate change?… assessing the impact of climate change on productivity and climate-suitability of crops and production systems … and understand the limiting factors

… using well-established empirical and mechanistic models such as Ecocrop, Maxent, DSSAT, …..that allow for the incorporation of spatial data and fine-tuned biophysical data

How?

Stations by variable:

• 47,554 precipitation • 24,542

tmean • 14,835

tmax y tmin

What is WorldClim?

Sources:•GHCN•FAOCLIM•WMO•CIAT•R-Hydronet•Redes nacionales

- 3 0 .1

3 0 .5

M e a n a n n u a lt e m p e r a t u r e ( º C )

0

1 2 0 8 4

A n n u a l p r e c i p i t a t i o n ( m m )

B

PREC• Generate interpolated climate surfaces using ANUSPLIN-SPLINA with weather station data

• Cross validating (25 iterations

• uncertainty

TMP

Uncertainty of climate data and models

BValidation of climate surface (25 iterations)

BCompare original worldclim with interpolated

GCMs are the only way we can predict the future

climate

Using the past to learn for the future

The ModelsGCM “Global Climate Model”

The Delta Method

Options – Statistical Methods

• Use anomalies and discard baselines in GCMs– Climate baseline: WorldClim– Used in the majority of studies– Takes original GCM timeseries– Calculates averages over a baseline and

future periods (i.e. 2020s, 2050s)– Compute anomalies– Spline interpolation of anomalies– Sum anomalies to WorldClim

Climate data• For current climate (baseline)

we used historical climate data from WorldClim www.worldclim.org

• Future climate: global climate models (GCMs) from IPCC (AR5) – SRES A2, A1B, ..

• Downscaling to provide higher-resolution (2.5 arc-minutes ~ 5 kilometer)

http://ccafs-climate.org

EcoCropThe database was developed 1992 by the Land and Water Development Division of FAO (AGLL) as a tool to identify plant species for given environments and uses, and as an information system contributing to a Land Use Planning concept.

In October 2000 Ecocrop went on-line under its own URL www.ecocrop.fao.org. The database now held information on more than 2000 species.

In 2001 Hijmans developed the basic mechanistic model (also named EcoCrop) to calculate crop suitability index using FAO Ecocrop database in DIVA GIS.

In 2011, CIAT (Ramirez-Villegas et al.) further developed the model, providing calibration and evaluation procedures.

open

Suitability modeling with EcocropEcoCrop, originally by Hijman et al. (2001), was further developed, providing calibration and evaluation procedures (Ramirez-Villegas et al. 2011).

It evaluates on monthly basis if there are adequate climatic conditions within a growing season for temperature and precipitation…

…and calculates the climatic suitability of the resulting interaction between rainfall and temperature…

How does it work?

• database held information on more than 2000 species

What happens when Ecocrop model runs?1

2

3

4

5

67

8

9

10

11

12

12 potentialgrowing seasons

1 kilometer grid cells(climate environments)

The suitability of a location (grid cell) for a crop is evaluated for each of the 12 potential growing seasons.

Growing season

0 24 100 80

𝑇𝑘𝑖𝑙𝑙=4+Tkill ( initial )

𝑇𝑘𝑖𝑙𝑙𝑇𝑚𝑖𝑛

𝑇𝑜𝑝𝑚𝑖𝑛

𝑇𝑚𝑎𝑥

𝑇𝑜𝑝𝑚𝑎𝑥

𝑇 (𝑋 )−𝑇𝑚𝑖𝑛

𝑇𝑜𝑝𝑚𝑖𝑛−𝑇𝑚𝑖𝑛

=𝑇 𝑠𝑢𝑖𝑡 1−𝑇 (𝑋 )−𝑇 𝑜𝑝𝑚𝑎𝑥

𝑇𝑚𝑎𝑥−𝑇 𝑜𝑝𝑚𝑎𝑥

=𝑇 𝑠𝑢𝑖𝑡

𝑇 𝑠𝑢𝑖𝑡=0

𝑇 𝑠𝑢𝑖𝑡=100

For temperature suitabilityKtmp: absolute temperature that will kill the plant Tmin: minimum average temperature at which the plant will grow Topmin: minimum average temperature at which the plant will grow optimally Topmax: maximum average temperature at which the plant will grow optimally Tmax: maximum average temperature at which the plant will cease to growFor rainfall suitabilityRmin: minimum rainfall (mm) during the growing season Ropmin: optimal minimum rainfall (mm) during the growing season Ropmax: optimal maximum rainfall (mm) during the growing season Rmax: maximum rainfall (mm) during the growing season Length of the growing seasonGmin: minimun days of growing seasonGmax: maximum days of growing season

P

P

P

P

• Growing season: xx days (average of Gmin/Gmax)

• Temperature suitability (between 0 – 100%)

• Rainfall suitability (between 0 – 100%)

• Total suitability = TempSUIT * RainSUIT

If the average minimum temperature in one of these months is 4C or less above Ktmp, it is assumed that, on average, KTMP will be reached on one day of the month, and the crop will die. The temperature suitability of that month is thus 0%. If this is not the case, the temperature suitability is evaluated for that month using the other temperature parameters. The overall temperature suitability of a grid cell for a crop, for any growing season, is the lowest suitability score for any of the consecutive number of months needed to complete the growing season

The evaluation for rainfall is similar as for temperature, except that there is no “killing” rainfall and there is one evaluation for the total growing period (the number of months defined by Gmin and Gmax) and not for each month. The output is the highest suitability score (percentage) for a growing season starting in any month of the year.

(climate) Suitability modelling

A1B / 2030current

current A1B / 2030

(climate) Suitability modelling

Change in climate-suitability“assumptions on regional level”

losses gains

Change in climate-suitability Lossesgains

• Maximum entropy methods are very general ways to predict probability distributions given constraints on their moments

• Predict species’ distributions based on environmental covariates

What is Entropy Maximization?

• You can think of Maxent as having two parts: a constraint• component and an entropy component

• The output is a probability distribution that sums to 1• For species distributions this gives the relative probability of observing

the species in each cell• Cells with environmental variables close to the means of the presence

locations have high probabilities

MaxEnt model

B

21

Input: Crop evidence (GPS points)19 bioclimatic variables of current (worldclim) & future climateOutput:Probability of distribution of coffee (0 to 1)

MaxEnt model

Bioclimatic variables for suitability modeling

• Bio1 = Annual mean temperature• Bio2 = Mean diurnal range (Mean of monthly (max temp - min temp))• Bio3 = Isothermality (Bio2/Bio7) (* 100)• Bio4 = Temperature seasonality (standard deviation *100)• Bio5 = Maximum temperature of warmest month• Bio6 = Minimum temperature of coldest month• Bio7 = Temperature Annual Range (Bio5 – Bi06)• Bio8 = Mean Temperature of Wettest Quarter• Bio9 = Mean Temperature of Driest Quarter• Bio10 = Mean Temperature of Warmest Quarter• Bio11 = Mean Temperature of Coldest Quarter• Bio12 = Annual Precipitation• Bio13 = Precipitation of Wettest Month• Bio14 = Precipitation of Driest Month• Bio15 = Precipitation Seasonality (Coefficient of Variation)• Bio16 = Precipitation of Wettest Quarter• Bio17 = Precipitation of Driest Quarter• Bio18 = Precipitation of Warmest Quarter• Bio19 = Precipitation of Coldest Quarter

derived from monthly temperature & precipitation

Coffee suitability - Maxent Results Nicaragua

B

Results

Variable AdjustedR2

R2 due to variable

% of totalvariability

Present mean

Change by 2050s

Locations with decreasing suitability (n=89.8 % of all observations)BIO 14 – Precipitación del mes más seco 0.0817 0.0817 24.8 24.49 mm -3.27 mm

BIO 04 – Estacionalidad de temperatura 0.1776 0.0959 29.1 0.83 0.166BIO 12 – Precipitación anual 0.2057 0.0281 8.5 2462.35 mm -24.31 mmBIO 11 - Temperatura media del cuarto más frío 0.2633 0.0576 17.5 20.11 ºC 1.86 ºC

BIO 19 - Precipitación del cuarto más frío 0.2993 0.0155 4.7 169.13 mm -7.08 mm

BIO 05 - Temperatura máxima del mes más cálido 0.3198 0.0102 3.1 28.45 ºC 2.30 ºC

BIO 13 - Precipitación del mes más húmedo 0.2838 0.0205 6.2 450.27 mm 10.72 mm

Otros - - 6.2

Coffee suitability - Maxent Results Nicaragua

B

a Average of Q1 of GCMs

b Average of GMSs

c Average of Q3 of GCMs

d Measure of agreement of models

e standard deviation of GCMs

b

c

e

Uncertainty of model output (Maxent) using 19 GCMs SRES A2 – timeserie 2040 – 2069 (2050)

Decision Support System for Agro technology Transfer (DSSAT)

+

• For 2 DSSAT-varieties (IB0006 ICTA-Ostua, IB0020 BAT1289

– “INTA Fuerte Sequia”, “INTA Rojo”, and “Tío Canela 75” originating from Nicaragua– “ICTA Ostua” and “ICTA Ligero” originating from Guatemala– “BAT 304” originating from Costa Rica– “SER 16”, SEN 56”, “NCB 226”, and “SXB 412” originating from CIAT, Colombia.

• Sowing on:– Primera (Beginning of June)– Postrera (Beginning of September)

• After recollecting data during 2011

results will be used

in a post-project-analysis

to calibrate 2 initial DSSAT varieties

run it again for trial sites and find

spatial and temporal analogues

Accompanying field trials in 5 countries to calibrate DSSAT

Planting date: Between 15th of April and 30th of June1

Variety 1: IB0006 ICTA-Ostua Variety 2: IB0020 BAT1289

Soil 1: IB00000005 (generic medium silty loam) Soil 2: IB00000008 (generic medium sandy loam)

Fertilizer 1: 64 kg / ha 12-30-0 6 to 10 days after germination and 64 kg / ha Urea (46% N) at 22 to 25 days after germination. Fertilizer 2: 128 kg/ha 18-46-0 Fertilizer application on sowing and 64 kg/ha UREA at 22 to 30 days after germination.

Weather data input:

Current climateAverage of 99 MarkSim daily outputs

Future climateEnsemble of 19GCM & 99 MarkSim outputs for 2020 & 2050

Runs: 17,800 points x 3 climates x 99 MarkSim-samples x 8 trials

DSSAT “Tortillas on the Roaster” in Central America

Results: yield change for year 2020 (Primera) – 8 trials

Trial 3 – high performance / high impactVariety 1: ICTA-OstuaSoil 1: generic medium silty loamFertilizer 2: 128 kg/ha 18-46-0 Fertilizer application on sowing and 64 kg/ha UREA at 22 to 30 days after germination

Trial 7 – medium high performance / less impactVariety 1: ICTA-OstuaSoil 2: generic medium sandy loamFertilizer 2: 128 kg/ha 18-46-0 Fertilizer application on sowing and 64 kg/ha UREA at 22 to 30 days after germination

Statistical negative and positive outliers of predicted yield change by 2020

31

Areas where the production systems of crops can be adaptedAdaptation-Spots Focus on adaptation of production system

Areas where crop is no longer an optionHot-SpotsFocus on livelihood diversification

New areas where crop production can be establishedPressure-SpotsMigration of agriculture – Risk of deforestation!

Identifying Impact-Hot-Spots and select sites for socio-economic analysis

32

• Beans as most important income (sell 70% of harvest)• Climate variability (intense rain, drought), missing labor & credits,

high input costs, … forces them to changes• Increasing livestock displace crops into hillside areas• Half of farmer rent their land• Distance to market is far• Mostly no road access in rainy season• They buy inputs/sell produce from/to farm-stores

(they call them: Coyotes)

Result: Sample-site 1 - Texistepeque (Las Mesas), Santa Ana ,El Salvador

Message 2: Adaptation Strategies must be fine-tuned at each site!

Las MesasAltitude: 667 m (about 2188 feet) Hot-spot -141 kg/ha

For 2020:• 35 mm less rain (current 1605mm)• mean temperature increase 1.1°CFor 2050:• 73mm less rain ( -5%)• mean temperature increase 2.3°C• hottest day up to 35°C (+ 2.6°C)• coolest night up to 17.7°C (+

1.8°C)

Hot-spot

33Message 3: There can be winners if they adapt quickly!

Result: Sample-site 2 – Valle de Jamastran, Danlí, Honduras Adaptation-spot

JamastranAltitude: 783 m (about 2568 feet) Adaptation-spot -115 kg/ha

• Active communities with already advanced agronomic management of maize-bean crops

• Favorable soil conditions and management• Long-term technical assistance / training• Irrigation schemes (e.g. 50 mz of 17 bean producers)• Diversification options (vegetables, livestock)• Market channels through processing industries• Advanced infrastructure (electricity, roads)• Need to optimize water use efficiency • Credit problems

For 2020:• 41 mm less rain (current 1094 mm)• mean temperature increase 1.1°CFor 2050:• 80 mm less rain ( -7%)• mean temperature increase 2.4°C• hottest day up to 34.2°C (+ 2.6°C)• coolest night up to 17°C (+ 2.1°C)

Decision support system modelling (for benchmark sites)

Agronomic managementExpert & farmer survey

Integrated crop-soil modeling

160 LDSF sample sites

Baseline domains

Impact2030 A1b

Experimental [n] cultivars[n] fertilizer application

[n] seasons

Application domains

Analysis of biophysical systems and simulating crop yield in relation to management factors. Combine these models with field observations that allow adjustment of the models in the course of the growing season .

Future24 GCM

A1B (IPCC)

CurrentworldClim

Validation with available station data

Daily weather generatorMarkSIM

Weather station data

(daily)

Climate data

yield

soil management

• Downscaling is inevitable.• Continuous improvements are

being done• Strong focus on uncertainty

analysis and improvement of baseline data

• We need multiple approaches to improve the information base on climate change scenariosDevelopment of RCMs (multiple: PRECIS not enough)Downscaling empirical, methods HybridsWe tested different methodologies

Conclusions climate data

Conclusions crop models• Ecocrop, when there is a lack on crop

information, for global or regional assessment

• Maxent, perennial crops with presence only data (coordinates) available

• DSSAT, only for few crops (beans, maize, …), high data input demand and calibrated field experiments are necessary

• We need to communicate uncertainty of model predictions

Empiricalmodels

Mechanisticmodels