PARTNERING FOR ADAPTATION

AND RESILIENCE AGUA

(PARA-AGUA) PROJECT
CONTRACT NO. 14S16492WK01, ORDER NO. OAA-TO-13-00037

DEVELOPING CLIMATE SCENARIOS

December 2014
1
This publication was produced for review by the United States Agency for International Development.
It was prepared by AECOM.

PARTNERING FOR ADAPTATION

AND RESILIENCE AGUA
(PARA-AGUA) PROJECT
CONTRACT NO. 14S16492WK01, ORDER NO. OAA-TO-13-00037
Submitted to:
AECOM International Development
Prepared by:
University Corporation for Atmospheric Research, acting on behalf of The National Center for
Atmospheric Research
David Yates and Hector Angarita
Subcontract Number:
Master Services Agreement No: 14S16492WK01
Deliverable: Report summarizing Activity 1, Activity 2.1, and Activity 3.1 and 3.2
Activity No. 1: Acquiring and Analyzing Global Climate Model Datasets for study basins and draft
basin scenarios for WEAP.
Activity No. 2.1: Development of GCM boundary forcing for current and future periods
Activity No. 2.2: Development of Current and Future regional climate datasets from WRF and
analysis of results.
Activity No 3.1 and 3.2: Archiving of historic and future boundary forcing data.
Project Task:
Task 3: Strengthening planning systems that optimize water use over the whole length of watersheds in
the context of climate change adaptation
Language:
Activity 1 is in Spanish, while Activity 2.1, 2.2, 3.1 and 3.2 are in English.
DISCLAIMER:
This document is made possible by the generous support of the American people through the U.S.
Agency for International Development (USAID). The contents of this document are the sole opinion of
AECOM and do not necessarily reflect the views of USAID or the U.S. Government.
2

Acronymns
AOGCM
Atmosphere-Ocean Global Climate Model
BCCA
Bias Correction-Constrructed Analog
BCSD
Bias Correcation-Spatial Disaggregation
CCSM
Community Climate System Model
CMIP-5 Climate Model Intercomparison Project- 5
DHI
Delft Hydraulics Institute

ECMWF

GHG
IDEAM
IPCC
AR5
KNN
LAC
LSM
MACA
MPAS
MPI-MR
NCAR
NOR-ESM
RCP
SEI
SENHAMI
SST
SWAT
USAID
WCRP
WEAP
WRF

European Center for Medium Range Forecast

Green House Gases

Instituto de Hidrologa, Meteorologa y Estudios Ambientales de Colombia
Intergovernmental Panel on Climate Change
Assessment Report 5
K-Nearest Neighbor
Latin America and the Caribbean
Land Surface Model
Multivariate Adaptive Constructed Analogs
Model for Prediction Across Scales
Max Plank Institute- Medium Resolution Model
National Center for Atmospheric Research
Norwegian Earth System Model
Representative Concentration Pathway
Stockholm Environment Institute
Servicio Nacional de Meteorologa e Hidrologa del Per
Sea Surface Temperature
Soil Water Assessment Model
United States Agency for International Development
World Climate Research Program
Water Evaluation and Planning System
Weather Research Forecast Model

Contenidos
Acronymns .................................................................................................................................................. 3
1 Presentacin .......................................................................................................................................... 8
2 Executive Summary ............................................................................................................................... 9
3 Actividad 1: Las proyecciones regionales sobre el clima para los Andes del norte del IPCC AR5 y la
adquisicin de datos y anlisis de los MCG para las cuencas de estudio y el desarrollo de la cuenca
escenarios climticos especficos para WEAP. .................................................................................... 11
3.1 Propuesta metodolgica para incluir variabilidad climtica y escenarios de cambio climtico en
el modelos WEAP ......................................................................................................................... 11
3.2 Otros estudios del efecto del cambio climtico con informacin disponible para Chira-Piura ..15
3.3 Metodologa ................................................................................................................................. 17
3.3.1 Fuentes de informacin- CMIP-5 ...................................................................................... 19
3.3.2 Registros de la red meteorolgica .................................................................................... 20
3.3.3 Comparacin del desempeo y priorizacin de modelos................................................. 21
3.3.4 Reduccin de escala - downscaling ................................................................................... 24
3.4

Resultados de las simulaciones y discusin. ............................................................................. 30

3.4.1 Resultados para la Cuenca Chinchin ............................................................................... 30
3.4.2 Resultados para la Cuenca Chira-Piura ............................................................................. 36

3.5

Resumen de la Actividad 1 ........................................................................................................ 49

3.6

Referencias para la Actividad 1 ................................................................................................. 49

4 Activity 2.1 (Development of GCM boundary forcing for current and future periods).......................51
5 Activity 2.2. (The WRF Regional Climate Simulations) ........................................................................ 55
5.1

WRF benchmark simulations with ERA-Interim ........................................................................ 56

5.2

WRF simulations with CCSM for current and future periods. ................................................... 58

Appendix: Description of Software ........................................................................................................... 62

Lista de figuras
Figura 1 Trayectorias representativas de concentracin RCP- en trminos de a. Forzamiento
Radiativo y b.Emisiones de CO2 de la industria y del consumo energtico. Tomada de
(Moss et al. 2010) .................................................................................................................................... 12
Figura 2

Niveles de integracin de los diferentes modelos climticos globales (GCMs) en sus

diferentes generaciones de desarrollo. Tomado de Yates et al. .................................................. 13

Figura 3

Proyecciones de temperatura presentadas en la comunicacin ms reciente sobre cambio

climtico del IDEAM. Se presentan series de temperatura media del aire observados (1976 a
2005), y futuros (2011 a 2070) del promedio de modelos para cada RCP. Tomado de (Ruiz
et al. 2013) ................................................................................................................................................. 15

Figura 4

Climate wizard.......................................................................................................................................... 16

Figura 5

reas adoptadas para evaluar el desempeo de los modelos GCM en la representacin de

los atributos diferentes atributos del clima de la cuencas Chinchin (alto) y Chira-Piura
(inferior). .................................................................................................................................................... 18

Figura 6

Caracterizacin de atributos del clima observado y modelado a diferentes escalas .............. 21

Figura 7

Indicadores utilizados para caracterizar atributos del GCM y del clima observado. Adaptado
de (Gutmann et al. 2013) ....................................................................................................................... 22

Figura 8

Seal del Nio- Ocenica (Oceanic Nio Index ONI)................................................................ 23

Figura 9

Nocin del espectro de potencia (periodigrama), ........................................................................... 23

Figura 10 Diferentes aproximaciones para la reduccin de escala. Adaptado de Maraun et al. (2010),
Rev of Geo. .................................................................................................................................................. 24
Figura 11 Ilustracin del algoritmo de K-NN-bootstraping............................................................................. 27
Figura 12 Ejemplo de una secuencia de clima generada mediante el mtodo K-NN. ............................... 28
Figura 13 Procedimiento de correccin de eventos extremos ...................................................................... 29
Figura 14 Presentacin de Resultados GCM para Cuenca Chinchin, Colombia ...................................... 30
Figura 15 Comparacin del patrn estacional (mensual) observado y simulado por los diferentes
GCM ........................................................................................................................................................... 32
Figura 16 Cambios proyectados en diferentes atributos de la precipitacin por los dos modelos de
mejor desempeo MPI-ESM-MR y CCSM4 ....................................................................................... 34
Figura 17 Presentacin de Resultados GCM para Cuenca Piura, Peru ......................................................... 36
Figura 18 Presentacin de precipitacin mes GCM Resultados para Cuenca Piura, Peru ....................... 40
Figura 19 Periodogram plots para Cuenca Chira-Piura, Peru ......................................................................... 42
Figura 20 Cambios proyectados en diferentes atributos de la precipitacin por los dos modelos de
mejor desempeo MPI-ESM-MR y CCSM4 para Cuenca Chira-Piura, Per. ............................ 44
Figure 21 Terrain height (m, color scale at bottom) and land/sea mask for CCSM4 (top) and 6-km
WRF (bottom). Actual coastlines and political boundaries shown in black. ............................. 52
5

Figure 22 Schematic of the bias-correction method for a generic variable. The CCSM4 perturbation
term is added to the 1981-2005 ERA-Interim mean term. ........................................................... 54
Figure 23 Schematic of the bias-correction method for a generic variable. The CCSM4 perturbation
term is added to the 1981-2005 ERA-Interim mean term. ........................................................... 54
Figure 24 The WRF domains for northern South America (Domain 1 at 18-km resolution); and the
higher resolution domain (Domain 2 at a 6-km resolution) over the northern Andean
Region of South America. ...................................................................................................................... 55
Figure 25 Figure of CCSM 4 output (Activity 2.1; GCM boundary forcing). Run 5 configuration was
used. ............................................................................................................................................................ 57
Figure 26 CMORPH monthly mean December precipitation for 7 years (left) and December 2010
(right). ......................................................................................................................................................... 58
Figure 27 CCSM4 estimate of the ENSO output for the contemporary period. 1981 was the La Nia
event, while 1997 and 2057 have been the El Nio events in which we have focused our
WRF modeling efforts. ........................................................................................................................... 59
Figure 28 Domain 2 (6-km) total annual rainfall (top left), topography (top right) and daily rainfall given
across a transect for a current La Nia, a current El Nio and a future El Nio. ................... 60

Lista de tablas
Tabla 1 Caractersticas generales del modelo de la cuencas Chira-Piura (Peru) y Chichina (Colombia)
actualmente implementado en WEAP ..................................................................................................... 15
Tabla 2 Otras fuentes de informacin sobre cambio climtico disponibles para la zona de estudio ...... 15
Tabla 3 Listado de modelos CMIP5 utilizados en este estudio ......................................................................... 20
Tabla 4a Fuentes de informacin hidrolgica utilizadas en este estudio para cuenca Chira-Piura, Per 20
Tabla 5 Algunas tcnicas estadsticas de reduccin de escala (statistical downscaling) .................................. 25
Tabla 6 Resumen de la comparacin de atributos modelados y observados en los diferentes GCM
analizados. ....................................................................................................................................................... 35
Tabla 7 Resumen de la comparacin de atributos modelados y observados en los diferentes GCM
analizados. ....................................................................................................................................................... 39

1 Presentacin
El reporte contiene la metodologa utilizada por el NCAR para analizar los datos y la variabilidad de los
modelos climticos globales, con la finalidad de desarrollar los escenarios de cambio climtico a
insertarse en el modelo WEAP, que se aplicar en las cuencas del ro Chinchin en Colombia y ChiraPiura en Per.
Se analizan las proyecciones diarias de los modelos de clima global (MCG), expuestos en el quinto
reporte (AR5) del Panel Intergubernamental sobre Cambio Climtico (IPCC). Se ha creado un archivo
de precipitacin y temperatura diaria (1TB), a partir de 10 MCG, y se han generado rutinas en el
lenguaje estadstico R para procesar y evaluar estos escenarios en las cuencas de inters. En general,
los MCG sobreestiman la precipitacin en los Andes del norte, para periodos contemporneos; y solo
tres de ellos simulan de alguna manera la estacionalidad, distribucin espacial y frecuencia de
precipitacin en la regin. Estos modelos son MPI-MR, CCSM4 y NOR-ESM. Todos los modelos
climticos indican calentamiento en el siglo 21 como resultado del incremento de las concentraciones
de gases de efecto invernadero.
A partir de estos archivos de precipitacin y temperatura diaria, se ha desarrollado un procedimiento
de regionalizacin estadstica (statistical downscaling) que utiliza informacin de los MCG
conjuntamente con datos de observacin de precipitacin y temperatura diaria para generar
proyecciones de clima futuro en las cuencas de Chira-Piura y Chinchin. El mtodo general se conoce
como vecino mas cercano K (K-NN), que es un esquema de muestreo estadstico que genera el
tiempo a nivel diario, preservando el clima estacional y sus dependencias espaciales y temporales en
distintas estaciones de una regin. El algoritmo est condicionado por las caractersticas de los datos
estadsticos del MCG, tales como condiciones clido-hmedas incrustadas en las proyecciones del
MCG. Este algoritmo se describe en el reporte y luego es usado para generar un ensamble de
escenarios climticos para ambas cuencas.
Un segundo conjunto de actividades est abocado a desarrollar proyecciones climticas regionales
basadas en el Modelo de Clima Regional (MCR) del NCAR, denominado WRF (Weather Research
Forecast model). El proceso de utilizar un MCR para proyectar el impacto del cambio climtico en el
clima regional usualmente se conoce como regionalizacin dinmica (dynamical-downscaling). Un MCR
necesita datos meteorolgicos de un MCG en sus fronteras, que permiten integrar el modelo hacia el
futuro. Para ello se han utilizado los resultados del modelo CCSM4, generados como parte del 5to
reporte del IPCC. El primer paso para usar WRF para la regionalizacin dinmica, se gener un
conjunto de archivos de frontera climtica intermedia con correccin en los sesgos (a set of biascorrected intermediate climate boundary files) utilizados como condiciones iniciales del modelo WRF.
CCSM4 es un modelo climtico global acoplado de fuente abierta y de dominio pblico, con datos
generados del IPCC y a los que pueden accederse desde la cuadrcula federada de la tierra
(https://www.earthsystemgrid.org/). CCSM4 est compuesto de cuatro modelos que simulan la
atmsfera, el ocano, la superficie terrestre y los glaciares marinos. A partir de estos archivos NCAR
corre simulaciones del modelo climtico regional y analiza los resultados.
Para la regin noroccidental de Sudamrica, hemos configurado el modelo WRF con dominio
doblemente anidado (doubly nested-domain), que incluye una malla de baja resolucin de 18 km que se
extiende en el ocano pacfico y que captura importantes caractersticas dinmicas de temperatura
superficial del mar. La malla de 18 km se extiende Amrica Central por el norte y supera la frontera
entre Per y Chile al sur. Una malla con resolucin final de 6 km se focaliz en una regin que incluye
Colombia y Per, para asegurarse que las cuencas de inters estn apropiadamente representadas en
las simulaciones dinmicas con WRF. Se configur el WRF para la malla basada en pruebas de varias
configuraciones parametradas usando patrones espaciales y cantidades de precipitacin como criterios
8

primarios. Esta configuracin de WRF se us despus para simular el clima presente y futuro,
seleccionando las anomalas ENSO, que considera la fase clida (El Nio) y la fase fra (La Nia) en las
condiciones actuales y futuras del clima. Estos resultados se presentan en el informe.

2 Executive Summary
This report describes the development of climate scenarios to aid in the development of integrated
water management activities in river basins in both Colombia and Peru, under the servicing contract
AID-EDH-I-00-08-00024. According to the requirements of the contract, the report contains the
methodology to analyze Global Climate Model data and their variability and develop climate change
scenarios for the WEAP model for the Chira-Piura basin in Peru and the Chinchin basin in Colombia
Climate change affects water resources, agriculture, and ecosystems in the Latin America and
Caribbean (LAC) region. Increasing temperatures and alterations of the hydrological cycle can affect
crop productivity and biodiversity and cause more frequent and extreme weather events, leading to
more intense flooding and drought. Andean glaciers and pramos, vital sources of fresh water for tens
of millions of people, are under severe threat. Policymakers, researchers, and water resources
managers across the LAC region need assistance in adapting to the impacts of climate change and help
reduce vulnerability, strengthen resilience and build adaptive capacity. As part of the Partnering for
Adaptation and Resilience Agua (PARA-Agua) Project, a regional project that is working directly with
scientists, decision-makers, and communities to strengthen watershed resilience to climate change
impacts. Under this project, NCAR is assisting AECOM, the Stockholm Environment Institute and their
partners to develop innovative climate adaptation approaches at the watershed level in Colombia and
Peru, and to facilitate regional replication of best practices through twinning partnerships. NCARs role
is to develop regional climate change projections that can be used in watershed and water management
models within the region.
In this report, we summarize our analysis of the Intergovernmental Panel on Climate Change (IPCC)
Assessment Report 5 (AR5) archive of daily Global Climate Model (GCM) projections. We have
created a 1 TB archive of daily precipitation and temperature data from 10 GCMs, and generated
scripts in the R statistical language to process and evaluate these scenarios over the watersheds of
interest. In general, the GCMs overestimate precipitation over the region of the Northern Andes for
the contemporary period. Of the ten climate models, we found that only three seem to represent the
seasonality, spatial distribution, and frequency of precipitation over the region with some level of skill.
These models were the Max Plank Institutes Medium Resolution model (MPI-MR), the NCAR
Community Climate System Model (CCSM4) and the Norwegian Earth System Model (NOR-ESM). All
the climate models indicate warming into the 21st century, as a result of increasing greenhouse gas
concentrations.
From this GCM archive of daily precipitation and temperature, we have developed a statistical
downscaling procedure that uses the information from the GCMs, along with observational data of
daily precipitation and temperature to generate projections of future climate over the watersheds of
the Chira-Piura Watershed in Peru and the Chinchin Watershed in Colombia. The general method is
known as K-Nearest Neighbor or K-NN, which is a statistical resampling scheme that generates daily
weather, preserving the seasonal climate and their spatial and temporal dependencies at multiple
stations in a given region. The algorithm is conditioned on the statistical characteristics of GCM data,
such as warmer-wetter conditions embedded within the GCM projection. The algorithm is described
in the report and then used to generate an ensemble of climate scenarios for both the Chira-Piura and
Chinchin Watersheds. The work of analyzing the GCM data and generating climate scenarios for
9

watershed management are in support of Activity 1 and Activity 4 in the NCAR sub-contract
agreement with AECOM.
A second set of NCAR activities (Activities 2 and 3) center around the development of regional
climate projections based on the NCARs Regional Climate Model (RCM), the Weather Research
Forecast model, commonly referred to as WRF. This process of using an RCM to project the impact of
climate change on regions climate is commonly referred to as dynamical-downscaling. Regional climate
models require on their boundary, meteorological data from a Global Climate Model to integrate the
model forward in time. For this study, we have used output from the NCAR Community Climate
Systems Model (CCSM4) that was generated as part of the IPCC 5th Assessment Report process. In a
first step to using WRF for dynamical downscaling, we generated a set of bias-corrected intermediate
climate boundary files that were used as the initial and boundary conditions for the WRF model.
The CCSM4 is a coupled global climate model (GCM) and is open source and in the public domain,
with datasets generated from these IPCC runs freely available from the Earth Federation Grid
(https://www.earthsystemgrid.org/). The CCSM4 is comprised of four component models that simulate
the atmosphere, ocean, land surface and sea-ice (Activity 3). With these files from CCSM4 generated
and archived, NCAR then ran the regional climate model simulations and have analyzed their results
(Activity 2.2).
For the region of the Northwestern South America, we have setup the WRF model configured as a
doubly nested-domain, that includes a coarse resolution domain of 18-km that extends well out into
the Pacific Ocean and which captures important, dynamical sea-surface temperature characteristics.
The 18-km domain also extends north into Central America, and South beyond the border of Chile. A
final resolution domain of 6-km was then focused over a region that includes Colombia and Peru, to
ensure that watersheds of interest are appropriately represented during the dynamical simulations with
WRF. We have established a WRF model configuration with WRF for our domain based on testing of
various parameterization configurations using spatial patterns and rainfall amounts as the primary
criteria. We have then used this WRF configuration to simulate the present-day and future climate of
select El Nio-Southern Oscillation (ENSO) anomalies, which considers the warm phase (El Nio) and
the cold phase (La Nia) under current and future climate conditions. These results are presented in
this report.
Activity 4 will use the results from the WRF model and integrate these data along with the K-NN
method, into a refined set of climate data suitable for hydrologic model at the watershed level,
specifically suitable for use in the Water Evaluation and Planning (WEAP) decisions support system.
These data will be made available to the water resource system models and will be included in the final
report.

10

3 Actividad 1: Las proyecciones regionales sobre el clima para los

Andes del norte del IPCC AR5 y la adquisicin de datos y anlisis
de los MCG para las cuencas de estudio y el desarrollo de la
cuenca escenarios climticos especficos para WEAP.
3.1 Propuesta metodolgica para incluir variabilidad climtica y escenarios de cambio
climtico en el modelos WEAP
Las evaluaciones de cambio climtico en los sistemas hidrolgicos fundamentalmente buscan establecer
como, a diferentes escalas espaciales y temporales, los cambios en los atributos del clima pueden
modificar la disponibilidad de agua en una cuenca. Por ejemplo, buscan determinar cmo los cambios
en las magnitudes esperadas o en los patrones estacionales de la temperatura del aire, precipitacin y
humedad relativa, modificarn los procesos de circulacin de agua que ocurren a escala de la cuenca: la
escorrenta, evaporacin, transpiracin, y almacenamiento de humedad en el suelo. A su vez, buscan
establecer como estos cambios pueden producir alteraciones en los sectores hidro-dependientes
(como la hidroenerga, los acueductos y la agricultura) o los ecosistemas que dependen de patrones
especficos de disponibilidad de agua en la cuenca.
Este tipo de estudios supone tres grandes de interrogantes:
1.

Cules son los atributos actuales del clima, entendidos como los atributos estadsticos del estado
del tiempo?, y a su vez, cuales son los cambios en dichos atributos antes los factores de cambio
asociados a factores antrpicos o naturales, por ejemplo, los cambios en la concentracin de
gases de efecto invernadero y aerosoles en la atmsfera, cambios en la radiacin solar, etc. Estos
atributos incluyen por ejemplo, las magnitudes medias, de las variables de estado del tiempo y su
variabilidad, los patrones estacionales (su variacin a lo largo de una ao), las tele-conexiones
macroclimticas (por ejemplo la relacin del fenmeno del nio), y los ciclos seculares, o de baja
frecuencia.

2.

Cul es la respuesta hidrolgica de una cuenca o un rea de una cuenca determinada al clima?, o
en otras palabras, el movimiento y almacenamiento del agua en los procesos hidrolgicos que
ocurren en la cuenca, como la escorrenta, la evapotranspiracin, el almacenamiento en el suelo,
la recarga/descarga de acuferos, etc. A su vez, cul ser la respuesta hidrolgica de una cuenca
ante cambios en el clima y otros factores antropognicos sobre los sistemas naturales y humanos?
Es decir, cules sern los patrones del cambio espacial y temporal en la disponibilidad de recursos
hdricos.

3.

Finalmente, Cules son las implicaciones de estos impactos a escala de cuencas sobre el sistema
socio-econmico y ambiental?

En conjunto, estos interrogantes suponen una gran dificultad y constituyen un reto al desarrollo
cientfico actual, pues requieren predecir el comportamiento de sistemas altamente complejos: el
sistema climtico, los sistemas hidrolgicos y sus factores de alteracin antrpicos y naturales, cada
uno compuesto de muchos elementos mutuamente interdependientes y con ciclos de
retroalimentacin no muy bien comprendidos. Ante este reto, existen avances que permiten, si bien
con niveles significativos de incertidumbre, vislumbrar a nivel exploratorio cuales son los futuros
posibles del sistema climtico y sus efectos sobre los sistemas hidrolgicos, de forma que contribuyan
como gua en el proceso de planificacin.
Con respecto al primer interrogante, un nmero importante de instituciones cientficas de diferentes
pases, han desarrollado desde la dcada de 1960s investigaciones orientadas a simular el sistema
11

climtico clima terrestre y predecir los cambios esperados por factores naturales y antrpicos. Los
factores de cambio analizados incluyen por ejemplo, el incremento en la concentracin atmosfrica de
gases de efecto invernadero y aerosoles, los cambios en la cobertura y el uso del suelo 1y los cambios
en el albedo por el derretimiento del hielo. Dichos esfuerzos han sido coordinados desde la dcada
1990 por el Programa Mundial de Investigaciones Climticas de las Naciones Unidas (WCRP), que
desarrolla cada 5 a 7 aos las proyecciones climticas globales a travs de su Proyecto de
Intercomparacin de Modelos Acoplados (CMIP, Coupled Modeling InterComparison Project).
En septiembre 2013 el WCRP public su quinto informe de evaluacin (CMIP5) sobre el estado del
arte del entendimiento cientfico del cambio climtico. Este informe presenta los resultados compilados
de la modelacin de clima realizados por 36 instituciones de diferentes pases, bajo un consenso sobre
los periodos de simulacin (1850-2100), y los escenarios de los factores de alteracin del sistema
climtico, denominados rutas representativas de concentracin, o RCP. (por sus siglas en ingls:
Representative concentration Paths), que se ilustran en la Figura 1. Cada RCP es definido como un
conjunto posible de cambios antrpicos y/o naturales en propiedades del sistema terrestre asociadas
principalmente con el balance radiativo del sistema climtico, tales como: Concentracin de aerosoles
atmosfricos (Volcnicos, Sulfatos, etc), Concentracin de gases de efecto invernadero, Polvo,
Cambios de cobertura terrestre, Ozono (troposfrico y estratosfrico), Carbono Orgnico, Sal Marina
y Radiacin solar. Por lo tanto, trayectorias de forzamiento radiativo no estn asociadas con los
escenarios socioeconmicos o de emisin nicos, sino son la combinacin de diferentes futuros
econmicos, tecnolgicos, demogrficos, polticos e institucionales (Ruiz et al. 2013).
Figura 1 Trayectorias representativas de concentracin RCP- en trminos de a. Forzamiento Radiativo y b.Emisiones de
CO2 de la industria y del consumo energtico. Tomada de (Moss et al. 2010)

Desde 1850, los niveles de CO2 han aumentado ms de un 40 por ciento, por encima del nivel preindustrial de
278 ppm en torno.

12

El proyecto CMIP5 pretende proveer una herramienta basada en mltiples modelos y escenarios,
para ilustrar el rango de los cambios posibles en el clima durante el siglo 21. Los modelos utilizados
en el CMIP5 varan desde aquellos que representan solamente la interaccin entre el ocano y la
atmsfera, hasta modelos del sistema climtico terrestre que representan adems las interacciones de
los ciclos biogeoqumicos, la vegetacin, capas de hielo, etc., como se muestra en la Figura 2. Sin
embargo, mediante los Escenarios RCP, la investigacin del sistema climtico global se realiza en la
actualidad sobre una base homognea que permite su posterior comparacin y sntesis.

Figura 2 Niveles de integracin de los diferentes modelos climticos globales (GCMs) en sus diferentes generaciones de
desarrollo. Tomado de Yates et al.

Los modelos del CMIP5 son la principal fuente de informacin disponible actualmente para predecir el
clima en horizontes de largo plazo, y pueden informar sobre el rango de futuros posibles del clima
terrestre frente a diferentes cambios naturales o antrpicos, y a su vez, informar los modelos
hidrolgicos utilizados para representar las condiciones locales en las cuencas y los efectos esperados
por los cambios en el clima. Sin embargo, la aplicacin de los resultados del CMIP5 en la estimacin del
efecto del cambio climtico global sobre sobre los sistemas hidrolgicos, presenta varias limitaciones.
En primer lugar el nivel de detalle espacial alcanzado por los modelos globales no es suficiente para
representar los atributos del clima local. Si bien un modelo de escala global simula los patrones de
circulacin utilizando una grilla de entre 1 a 2 millones de puntos, en la prctica la resolucin espacial
horizontal alcanzada vara entre 1 a 3 (aproximadamente 40 a 120 km en el ecuador, que equivalen
reas de entre 1600 a 14400 km2). Esto significa que los modelos solamente proporcionan informacin
promedio del clima de la atmsfera libre en grandes reas de la superficie, y no alcanzan a
representar atributos locales que dependen de las condiciones fsicas o biticas, por ejemplo, por
caractersticas orogrficas, patrones de circulacin local asociados a la precipitacin convectiva,
almacenamientos locales, etc.

13

Por esta razn, para apoyar la evaluacin de los impactos locales en los recursos hdricos existe la
necesidad de desarrollar proyecciones climticas a escala reducida. Las tcnicas de Reduccin de escala
Downscaling consisten en asociar las propiedades de la atmsfera libre obtenidas a partir de las
salidas de los GCM, a las condiciones meteorolgicas locales, ya sea mediante modelos matemticos de
base fsica (downscaling dinmico) o modelos estadsticos (downscaling estadstico). Tales mtodos
recurren a los registros meteorolgicos locales para establecer los modelos ptimos.
Otra limitacin en la aplicacin de los modelos climticos globales est asociada a la no linealidad de las
ecuaciones que describen los diferentes componentes del sistema climtico: la atmsfera, el ocano, la
biota, etc., y sus interacciones. Por lo tanto, si bien la dependencia de los procesos naturales en el
tiempo sugiere su previsibilidad, las complejas relaciones entre los componentes de los modelos,
incluso con cantidades pequeas de incertidumbre (por ejemplo, en las condiciones iniciales) dan lugar
a la imprevisibilidad de los procesos despus de un cierto horizonte temporal (Koutsoyiannis 2010).
Esta limitacin, generalmente se trata mediante enfoques de mltiples modelos, realizaciones y
escenarios, que en conjunto, buscan informar sobre el rango de futuros posibles esperados del clima,
esencialmente en trminos de propiedades de largo plazo, sin ahondar en la precisin de las
predicciones.
Con respecto al segundo interrogante, para comprender los posibles efectos de los cambios en el
clima en la disponibilidad de agua en las cuencas, es necesario utilizar modelos hidrolgicos, que
permitan representar adecuadamente los diversos procesos que ocurren en la cuenca. Por ejemplo,
como incrementos en la temperatura, cambios en las rachas de das secos o hmedos, o cambios en
los patrones estacionales de lluvia, etc., pueden conducir a cambios en los volmenes totales de
evapotranspiracin, escorrenta o almacenamientos en el suelo. A su vez, para comprender los cambios
en los patrones de circulacin de agua en las cuencas, no solo es necesario incluir aspectos
relacionados con el cambio en el clima y sus efectos directos en los procesos hidrolgicos, sino
tambin indirectos en trminos de los cambios en las condiciones de operacin de la infraestructura
como almacenamientos, trasvases y desviaciones de agua y la irrigacin. Los anlisis de cambio
climtico, por lo tanto, tambin buscan estimar las complejas interacciones entre el clima, las
actividades humanas hidro-dependientes, y la hidrologa de las cuencas. Existen diversos modelos
hidrolgicos que pueden capturar a la escala de las cuencas hidrogrficas los procesos antes
mencionados, entre desde modelos aglutinados como los propuestos por Domnguez (2005) hasta
modelos distribuidos y semi-distribuidos como SWAT , MikeSHE (DHI), WEAP (Yates et al. 2005;
Young et al. n.d.). Debido a su caracterizacin de los procesos hidrolgicos, muchos de estos modelos
requieren datos climticos de alta resolucin (por ejemplo, la precipitacin, la temperatura, el viento, la
radiacin solar, y as sucesivamente) a escalas de tiempo relativamente finas (diarias o semanales).
En las cuencas Chira-Piura (Per) y Chinchin (Colombia), est liderando el desarrollo de los sistemas
de modelamiento para estas cuencas utilizando el WEAP Decision Support Modelo. Estos modelos
integran la oferta y demanda de agua de las cuencas, y el uso de un conjunto de escenarios que
incluyen el crecimiento de la poblacin, el uso del suelo, las limitaciones ambientales, y el clima, para
estudiar los flujos de regmenes. Como parte de este proceso, este estudio busca proveer informacin
sobre escenarios de clima futuro, que sumados a los dems componentes que determinan la oferta y
demandas hdricas, permitan establecer el rango de futuros posibles de alteracin del rgimen
hidrolgico en las cuencas.

14

Tabla 1 Caractersticas generales del modelo de la cuencas Chira-Piura (Peru) y Chichina (Colombia)
actualmente implementado en WEAP
Modelo

Caractersticas
El dominio de modelacin es la totalidad de la cuenca Chira-Piura
(30.000 km), representado mediante 122 unidades hidrolgicas
de anlisis. El modelo opera a paso de tiempo mensual.

Cuenca Chira-Piura

El modelo actualmente incluye la infraestructura hidrolgico de

gran escala (4 embalses) y otras demandas hdricas de gran
magnitud (distritos de riego) y centros poblados
El dominio de modelacin es Ch (1.030 km), representado
mediante 31 unidades hidrolgicas de anlisis. El modelo opera a
paso de tiempo mensual.

Cuenca Chichina
El modelo actualmente incluye la infraestructura hidrolgico en la
cuenca (1 embalses).

3.2 Otros estudios del efecto del cambio climtico con informacin disponible para
Chira-Piura
Otras fuentes de informacin disponible a escala local se describen en la Tabla 2. Si bien estos estudios
proveen informacin de contexto sobre la cuenca, los requerimientos especficos de este estudio no
hacen viable su utilizacin.
Tabla 2 Otras fuentes de informacin sobre cambio climtico disponibles para la zona de estudio
FUENTE

REPORTADO

DESCRIPCIN

Worldclim

(Hijmans et al.
2013)

WorldClim es un conjunto de grillas de clima global generadas mediante

reduccin de escala estadstica de los resultados del IPCC5-CMIP5,
interpoladas para zonas terrestres mundiales (excluyendo la Antrtida) con
una resolucin espacial de 30 (~1 km de resolucin espacial). Las
variables del clima considerados son: la precipitacin media, y la
temperatura mnima y mxima. Los resultados estn disponibles a paso de
tiempo mensual para tres periodos de tiempo: Actual (representativo de
1950-2010), 2050 y 2070.
La reduccin de escala de los periodos futuros (2050 y 2070), se realiz
adoptando como referencia al data set WorldClim 1.4 construido mediante
la interpolacin de los registros del perodo 1950-2000 (mtodo SPLINEANUSPLIN). (Hijmans et al. 2005).

Comunica
cin
Nacional
sobre
el
Cambio
climtico

IDEAM. (Ruiz
et al. 2013)

En Colombia en trabajos previos de anlisis de cambio climtico han sido

fundamentalmente desarrollados por el IDEAM. En su comunicacin ms
reciente sobre la evaluacin de los efectos del Cambio Climtico en
Colombia basados en el informe IPCC V4-CMIP3, se presentan
proyecciones de escala reducida para diferentes regiones del pas,
utilizando mtodos estadsticos y dinmicos. Los resultados se encuentran
disponibles a escala de tiempo mensual, y proporcionan valores de
referencia indicativos sobre los cambios medios esperados en la
temperatura media del aire (del orden de 1.4C para el periodo 2011-2040 y
de 2.4C para el periodo 2041-2070) y de la precipitacin por su parte, con
reducciones proyectadas de hasta un 36%. La figura 4 ilustra algunas de
las proyecciones presentadas en este estudio.
Figura 3 Proyecciones de temperatura presentadas en la comunicacin ms

15

FUENTE

REPORTADO

DESCRIPCIN
reciente sobre cambio climtico del IDEAM. Se presentan series de temperatura
media del aire observados (1976 a 2005), y futuros (2011 a 2070) del promedio de
modelos para cada RCP. Tomado de (Ruiz et al. 2013)

Cambio
Climatico
en el Peru:
Impactos
Economic
os
Y
Sociales
Climate
Wizard

SENAMHI
(Avalos, 2009;

Anlisis de futuros cambios del clima en regiones alto andinas del Per. Las
evidencias presentadas en este trabajo sugieren una tendencia de
calentamiento en la regin alto andina del Per (incremento de das y
noches clidas). En relacin a lluvias, la tendencia general es de reduccin
en los totales anuales (incremento de das secos consecutivos).
http://www.senamhi.gob.pe/?p=0901&idNota=090326

Maurer et al
(2007)

ClimateWizard es un conjunto de grillas de clima global generadas

mediante reduccin de escala estadstica de los resultados del IPCC4CMIP3 con una resolucin espacial de 30 (~50 km de resolucin espacial).
Las variables del clima considerados son: la precipitacin media, y la
temperatura mnima y mxima. Los resultados estn disponibles para el
promedio mensual multianual de tres periodos: Actual (representativo de
1950-2000), 2050 y 2080.
Figura 4 Climate wizard

16

FUENTE

REPORTADO

DESCRIPCIN

3.3 Metodologa
El propsito de la metodologa propuesta es generar series de proyecciones climticas de las variables
de temperatura y precipitacin para la cuenca Chira-Piura en Per, y la cuenca Chinchin en Colombia,
a nivel de las unidades de anlisis del modelo WEAP, informadas por las predicciones del proyecto
CMIP5, que sumadas al clima actual, permitan establecer un rango de comportamiento hidrolgico
posible para los aos 2010 a 2080.
Especficamente, los objetivos trazados son:

Compilar informacin de estudios CMIP5 para modelos climticos globales cuyos resultados se
encuentren disponibles a resolucin diaria.

Establecer el desempeo actual de los modelos para diferentes secciones de la cuenca

Magdalena Cauca en trminos de diferentes atributos del clima a diferentes escalas temporales
e identificar los 2 modelos con mejor desempeo, que pueden informar la generacin de series
climticas para la cuenca.

Generar 2 escenarios de cambio climtico de escala reducida para la cuenca magdalena Cauca,
a escalas temporales diarias y mensuales, que corresponden a condiciones extremas en
trminos de humedad y temperatura, en el horizonte 2010 2050, que reflejen los cambios en
los patrones de estado del tiempo representados en las simulaciones del clima global.

17

Figura 5 reas adoptadas para evaluar el desempeo de los modelos GCM en la representacin de los atributos diferentes
atributos del clima de la cuencas Chinchin (alto) y Chira-Piura (inferior).

18

3.3.1

Fuentes de informacin- CMIP-5

La principal fuente de informacin disponible en la actualidad para establecer los escenarios de cambios
en el clima consiste en los resultados compilados por el proyecto CMIP5. Estos resultados
comprenden 61 modelos de circulacin global - o GCM - desarrollados por 36 instituciones
internacionales. Para cada uno de los modelos, son reportados datos a escala global, de diferentes
conjuntos de resultados, tales como:

Experimentos histricos, que reproducen los atributos del clima durante el periodo 1850 a
2005

Experimentos futuros, que comprenden el periodo 2005 2100, en cuatro escenarios de

cambios antrpicos y naturales en las propiedades del sistema climtico, denominados RCPs o
rutas representativas de concentracin: RCP2.6, RCP4.5, RCP6.0 y RCP8.5.

Experimentos Histricos Miscelneos: que buscan reproducir los atributos del clima
durante el periodo 1850 a 2005, en condiciones de forzamiento simplificado. Estas
modelaciones tienen como propsito proveer informacin sobre la importancia relativa de
diferentes forzamientos climticos en los atributos generales del clima (por ejemplo, los gases
de efecto invernadero, los aerosoles atmosfricos).

En este estudio, fueron utilizados los resultados de 10 GCMs, cuyo listado se presenta en la Tabla 3.
Los datos utilizados corresponden a las salidas globales de los modelos climticos, para los
19

experimentos Histrico (1850 a 2005) y RCP8.5, a paso de tiempo diario, para las variables
Precipitacin y Temperatura media.
Tabla 3 Listado de modelos CMIP5 utilizados en este estudio
CENTRO DE
MODELACIN
CCCma

INSTITUCIN

MODELO

Canadian Centre for Climate Modelling and Analysis

CNRM-CERFACS
CSIRO-QCCCE

CanESM2

Centre National de Recherches Meteorologiques / Centre Europeen de

CNRM-CM5
Recherche et Formation Avancees en Calcul Scientifique
Commonwealth Scientific and Industrial Research Organisation in
CSIRO-Mk3.6.0
collaboration with the Queensland Climate Change Centre of Excellence

MOHC (additional
Met Office Hadley Centre (additional HadGEM2-ES realizations contributed
realizations
by
HadGEM2
by Instituto Nacional de Pesquisas Espaciais)
INPE)
MPI-M

Max Planck Institute for Meteorology (MPI-M)

MPI-ESM-MR

MRI

Meteorological Research Institute

MRI-CGCM3

NCAR

National Center for Atmospheric Research

CCSM4

NCC

Norwegian Climate Centre

NorESM1-M

NIMR/KMA
NSF-DOE-NCAR

3.3.2

National Institute of Meteorological Research/Korea Meteorological

HadGEM2
Administration
National Science Foundation, Department of Energy, National Center for
CESM1(CAM5)
Atmospheric Research

Registros de la red meteorolgica

1. Cuenca Chira-Piura - La informacin hidrolgica y meteorolgica en la cuenca Chira-Piura

Cauca corresponde a los registros de la red de monitoreo hidrolgico y meteorolgico del
SENAMHI. La disponibilidad de estaciones y registros vara ampliamente en funcin de la
variable que se est analizando: 32 estaciones de precipitacin y 10 estaciones de temperatura.
2. Cuenca Chinchin En la Tabla 4 se presenta un resumen de las fuentes de informacin hidrolgica utilizadas el modelo
estudio.
Tabla 4a Fuentes de informacin hidrolgica utilizadas en este estudio para cuenca Chira-Piura, Per
INFORMACIN

AOS

Series de precipitacin (32 Estaciones)

1972-1991

Series de Temperatura (10 Estaciones)

1972-1991

FUENTE
SENAHMI

TIPO DE DATOS
Series de tiempo
(resolucin diaria)

Tabla 3b Fuentes de informacin hidrolgica utilizadas en este estudio para Cuenca Chinchin, Colombia.
INFORMACIN
Series de precipitacin (23 Estaciones)

AOS

FUENTE

1981-2010
IDEAM

Series de Temperatura (6 Estaciones)

1981-2010

20

TIPO DE DATOS
Series de tiempo
(resolucin diaria)

3.3.3

Comparacin del desempeo y priorizacin de modelos

La bondad de un modelo climtico debe verse en trminos de su capacidad de representar atributos

del clima, por definicin estadsticos, que ocurren a diferentes escalas temporales: desde elementos
asociados a pasos de tiempo diario hasta oscilaciones de aos o dcadas. En conjunto, estos atributos
caracterizan el clima y su variabilidad. La Figura 6 ilustra esta nocin. Los resultados de un GCM a
pasos de tiempo diario (en un dominio espacial relativamente amplio - entre 10000 a 40000 km2),
pueden compararse respecto al promedio regional observado a diferentes escalas de agregacin
temporal: Anuales, ej., las magnitudes medias anuales y sus desviaciones, Intra-anual, ej., la
estacionalidad de las magnitudes mensuales y los eventos extremos, la duracin de las rachas de das
secos, hmedos, etc.), e inter-anual, asociadas a oscilaciones seculares o teleconexiones macroclimticas
Es este estudio, se implementa un conjunto de 8 indicadores (cuantitativos y cualitativos) con los que
se busca medir el desempeo de diferentes modelos en la representacin de dichos atributos. A su
vez, estos indicadores permiten identificar y priorizar los modelos que mejor representan el clima
local.
Figura 6 Caracterizacin de atributos del clima observado y modelado a diferentes escalas

21

Figura 7 Indicadores utilizados para caracterizar atributos del GCM y del clima observado. Adaptado de (Gutmann et al.
2013)
INDICADOR
Sesgo (Bias)
Longitud de
rachas secas
(Dry_spell_le
nght)
Media diaria
(Daily mean)
Magnitud de
eventos
extremos
Fraccin de
das con lluvia
(P > 0.1 mm)
Wet fraction
Longitud
media de
rachas
humedas
(Wet spell
length)
Estacionalida
d mensual

Oscilaciones
de baja
frecuencia y
teleconexine
s
macroclimtic
as

FORMU
LA

[%]

DESCRIPCIN

: Media de la seal del GCM

: Media regional de la seal observada

95

P95: Percentil 95 (probabilidad de excedencia 0.05)

[%]

: Total de das secos en el periodo analizado (P 0.1 mm)

: Nmero de periodos sin lluvia en el periodo analizado
: Seal (GCM o Observada)
N: Nmero de das del periodo

: Total de das hmedos en el periodo analizado (P > 0.1 mm)

N: Nmero de das del periodo

: Total de das secos en el periodo analizado (P 0.1 mm)

: Nmero de periodos con lluvia en el periodo analizado

Reproduce o no el patrn estacional (un mximo, dos mximos)

: Periodograma (Vase Figura 8 Seal del Nio- Ocenica (Oceanic Nio

Index ONI)

Figura 9)
: Seal nio ocenica (media anual). Figura 8
X :Seal media anual del GCM
X :Seal media anual del observada

()

22

Figura 8 Seal del Nio- Ocenica (Oceanic Nio Index ONI)

Figura 9 Nocin del espectro de potencia (periodigrama),

23

3.3.4

Reduccin de escala - downscaling

El problema de la reduccin de escala consiste en identificar como utilizar las seales de clima global
generadas por un modelo GCM, para inferir los cambios en los atributos de las variables de estado
del clima local , a diferentes escalas temporales. En la Figura 10 y la Tabla 5, se ilustran diferentes
aproximaciones existentes para este fin, desde las ms sencillas, como los anlisis de sensibilidad
basados en incrementos constantes (
-Delta), hasta los mtodos dinmicos. Los mtodos estadsticos
por su parte, ocupan una posicin intermedia y comprenden mtodos paramtricos y no paramtricos.
Los primeros se basan en modelos, por ejemplo regresiones lineales de tipo = + , que
utilizan como predictores atributos de las series de un GCM () para predecir los atributos del clima
local ().
Por ejemplo, las tcnicas de Anlogos Construidos como BCCA y MACA, utilizan combinaciones
lineales de variables de estado presentes y pasados del GCM (Ver Tippett & DelSole (2013) para una
descripcin detallada) que son estadsticamente ptimas para estimar las caractersticas del clima local.
Por su parte, los mtodos no paramtricos, como los mtodos de reordenamiento o bootstrapping
(Gangopadhyay et al. 2005; Yates et al. 2003) construyen series sintticas como secuencias de
observaciones histricas, utilizando reglas probabilsticas que buscan preservan algunos atributos
presentes en las series histricas, como la estacionalidad, la variabilidad, la correlacin espacial, etc, y
que a su vez, se modifican en funcin de los cambios predichos por la seal del modelo climtico.
Figura 10 Diferentes aproximaciones para la reduccin de escala. Adaptado de Maraun et al. (2010), Rev of Geo.

24

Tabla 5 Algunas tcnicas estadsticas de reduccin de escala (statistical downscaling)

Metodo

Referencia

Descripcin

Delta

Incrementos constantes:

Temperatura: aditivos: e.g. +1C

Precipitacin: multiplicativos: e.g. 1.05% Pr

hstorica

BCCA
(Bias
Corrected Maurer, E. P., Hidalgo, H. G., Das, T.,
Constructed Analogs)
Dettinger, M. D., and Cayan, D. R.: The utility
of daily large-scale climate data in the
assessment of climate change impacts on
daily stream flow in California, Hydrol. Earth
Syst. Sci., 14, 11251138, doi:10.5194/hess14-1125-2010, 2010.

Preprocesamiento
(PP):
Eliminacin
tendencias y mapeo de cuantiles
Downscaling (DS): Anlogos construidos
Post Procesamiento (PostP): Ninguno

BCSD
(Bias
Corrected Wood, A. W., Maurer, E. P., Kumar, A., and
Statistical Disaggregation)
Lettenmaier, D. P.: Long-range experimental
hydrologic forecasting for the eastern United
States, J. Geophys Res., 107, 4429,
doi:10.1029/2001JD000659, 2002. Wood, A.
W., Leung, L. R., Sridhar, V ., and
Lettenmaier, D. P.: Hydrologic implications of
dynamical and statistical approaches to
downscaling climate model outputs, Climatic
Change, 62, 189 216, 2004.

PP: Eliminacin de tendencias y mapeo de

cuantiles para el llenado de vacos
DS: Interpolacin
PostP: Ninguno

ARRM

de

Stoner, A. M. K., Hayhoe, K., Yang, X. and PP: Ninguno

Wuebbles, D. J. (2012), An asynchronous DS: Interpolacin
regional regression model for statistical POstP: BC (Regression asincrona de cuantiles)
downscaling of daily climate variables. Int. J.
Climatol.. doi: 10.1002/joc.3603

MACA (Multivariate Adaptive Abatzoglou J.T. and Brown T.J. "A

Constructed Analogs)
comparison
of
statistical
downscaling
methods suited for wildfire applications "
International Journal of Climatology (2011)
doi: 10.1002/joc.2313
Bayesian
Regression

Neural

PP: Eliminacin de tendencias y mapeo de

cuantiles para el llenado de vacos
DS: Analogos construidos
PostP: BC y reemplazo de tendencia

Net

JVSD
(Joint
Variable Zhang, F. & Georgakakos, A. P. Joint variable
Statistical Downscaling)
spatial downscaling. Climatic Change 111,
(2011).
BCSA

Hwang, S. and Graham, W. D.: Development

and comparative evaluation of a stochastic
analog method to downscale daily GCM
precipitation, Hydrol. Earth Syst. Sci.
Discuss., 10, 2141-2181, doi:10.5194/hessd10-2141-2013, 2013.

HMM/NHMM (Homegeneous Robertson, Andrew W., Sergey Kirshner,

and Non-Homegeous Hidden Padhraic Smyth, 2004: Downscaling of Daily
Markov Method)
Rainfall Occurrence over Northeast Brazil
Using a Hidden Markov Model. J. Climate, 17,
44074424.doi:
http://dx.doi.org/10.1175/JCLI-3216.1
Tomado de: https://earthsystemcog.org/projects/downscalingmetadata/methods.

Para aplicaciones de modelacin de recursos hdricos, los mtodos no paramtricos ofrecen algunas
ventajas, entre otras:

Capturan atributos climticos de diferentes escalas temporales, tales como: la duracin de las
rachas de das secos y hmedos, que son de particular inters en la modelacin de sistemas
hidrolgicos, la estacionalidad, y las oscilaciones de baja frecuencia
25

Permiten conservar la correlacin espacial de las observaciones

Pueden construirse conjuntos arbitrariamente grandes de series de clima posible, e identificar

en dichos conjuntos secuencias asociadas a narrativas especficas en un contexto de
planificacin, por ejemplo, aquellas que contengan eventos, o rachas de eventos ms secos o
ms hmedas.

En este estudio se implementa el mtodo K-NN bootstraping (por sus siglas en ingls: K Nearest
Neighbor), para generar secuencias diarias de estado del tiempo (precipitacin y temperatura)
informadas a partir de la salida especfica de un GCM. Este proceso se repite para diferentes GCM
priorizados (por ejemplo CCSM4, MPI), para crear un conjunto de proyecciones a escala local, con las
que es posible examinar el rango de los cambios potenciales en los atributos del clima.
El mtodo KNN genera series aleatorias que preservan la probabilidad conjunta de los estados de
humedad consecutivos del clima regional a paso de tiempo diario, definidos como:
Seco (Dry), Precipitacin media diaria regional menor a min(0.3, P0.05),

Humedo (Wet): Precipitacin media regional entre 0.3 mm y P95

Estremadamente hmedo (Extremely wet): Precipitacin media regional > P95

La regla de probabilidad es derivada del anlisis de la seal observada y del GCM, tal y como se
muestra en el siguiente ejemplo:
Future (e.g. T = 2010-2040)

GCM, e.g.
CCSM4

Local climate

Present (T = 1970-2010)
J observed (Tpresent )

dry(t+1)

dry(t)
e-wet(t)

0.20
0.10
0.20

J model(Tpresent )

dry(t+1)

dry(t)

0.15
0.10
0.27

wet(t)

wet(t)

GCM
Predicted
shifts

e-wet(t)

wet(t+1) ewet(t+1)

0.10
0.20
0.12

0.05
0.02
0.01

model

(future,present )
dry(t)
wet(t)
e-wet(t)

e-wet(t)

J model (Tfuture)

dry(t+1)

dry(t)

0.20
0.10
0.20

wet(t)

0.03
0.05
0.01

wet(t)
e-wet(t)

dry(t+1)

0.05
0.00
-0.07

dry(t+1)

0.25
0.10
0.13

dry(t)

wet(t+1) ewet(t+1)

0.10
0.25
0.05

J shifted =
J observed + J

wet(t+1)

0.00
-0.05
0.07

wet(t+1) ewet(t+1)

0.10
0.15
0.19

0.05
0.02
0.01

wet(t+1) ewet(t+1)

0.10
0.20
0.12

0.03
0.05
0.01

ewet(t+1)

0.00
0.00
0.00

Donde:

(): Probabilidad observada para el periodo climtico presente T = [1970-2010]

(): Probabilidad del GCM, para el periodo climtico presente T = [1970-2010]

(): Probabilidad del GCM, para un periodo climtico especfico, por ejemplo, T = [20102040]
() = () + [ () ()]: Probabilidad para periodo proyectado
26

En la Figura 11 se presenta un ejemplo del algoritmo K-NN.

i.

Las secuencias de estado del tiempo son construidas mediante una caminata aleatoria de
registros histricos, seleccionados del clima regional en una ventana de 2*W-das
alrededor del da juliano n que se busca predecir.

ii.

El mtodo inicializa la simulacin seleccionando aleatoriamente una fecha en la ventana

contenida entre [n-W,n+W].

iii.

A partir del estado de humedad registrado en la fecha seleccionada, utilizando la matriz de

probabilidad conjunta, se establece el estado de humedad del periodo siguiente,

iv.

se identifican cuales valores observados en la ventana cumplen la condicin de estados

humedad consecutivos establecida en el paso anterior (por ejemplo hmedo-seco).

v.

El valor siguiente se la simulacin se determina aleatoriamente entre estos puntos, o

puntos candidatos. La probabilidad de ocurrencia de los estados consecutivos que cumplen
la condicin seleccionada, se describe mediante la siguiente funcin de distribucin de
probabilidad:
() =

Donde es la distancia entre el estado de humedad presente, y el estado de humedad de los puntos
candidatos:
= (() ())

Una vez seleccionado el valor del da siguiente, se repite la secuencia desde el paso iii. El resultado del
proceso es una secuencia de fechas (Vase Figura 12), que puede ser utilizada para reconstruir series
sintticas de estado del tiempo de las diferentes meteorolgicas (TA, HR) en las estaciones locales en
las que se cuenta con observaciones.
Figura 11 Ilustracin del algoritmo de K-NN-bootstraping

i.

ii.

iii.

iv.

27

v.

vi.

Las series sintticas generadas por el mtodo K-NN no incluyen seal de cambio de temperatura
debido a que se basan en las magnitudes del periodo histrico. En el mtodo la seal de incremento de
temperatura se estima como:
[]
() = ()
Donde:

(): Media mvil de W das de la seal de temperatura en el modelo para el dia t (Por
ejemplo 01-Ene 2015)

[]: Media mvil de W das de la seal media de temperatura en el modelo para el

periodo de referencia (Por ejemplo 1970-2010)

Figura 12 Ejemplo de una secuencia de clima generada mediante el mtodo K-NN.

28

Las secuencias de precipitacin son posteriormente evaluadas en trminos de tendencias en las

magnitudes de eventos extremos. Estas magnitudes son corregidas mediante un mapeo de cuantiles
sobre la funcin de distribucin de extremos corregida, como se muestra en la Figura 13.
Figura 13 Procedimiento de correccin de eventos extremos

29

3.4 Resultados de las simulaciones y discusin.

3.4.1

Resultados para la Cuenca Chinchin

Comparacin de las magnitudes de diferentes atributos del clima observados y simulados por
diferentes GCM para el periodo 1970-2000. a. Sesgo relativo, b. Precipitacin media diaria, c. Duracin
media de rachas secas, d. duracin media de rachas hmedas, e. Fraccin de das con lluvia y f. P5.
Percentil de probabilidad de excedencia 5%. A partir de este anlisis, parece que el de los modelos de
Noruega NOR-ESM-NCAR CCSM4, el Max Plank Institutos MPI-RM, y son de mejor rendimiento.
Aunque como prueba por las figuras de abajo, para algunos ndices, algunos modelos funcionan mejor
que otros. Por ejemplo, la longitud de racha seca en el modelo NOR-ESM es mucho ms largo que los
otros modelos, mientras que para otras mtricas, que preformes bastante bien.
Figura 14 Presentacin de Resultados GCM para Cuenca Chinchin, Colombia

30

31

Figura 15 Comparacin del patrn estacional (mensual) observado y simulado por los diferentes GCM

32

33

Figura 16 Cambios proyectados en diferentes atributos de la precipitacin por los dos modelos de mejor
desempeo MPI-ESM-MR y CCSM4

34

Tabla 6 Resumen de la comparacin de atributos modelados y observados en los diferentes GCM analizados.
Modelo

CCSM4
GCM
obs
CNRM-CM5
GCM
obs
CSIRO-Mk3-6-0
GCM
obs
HadGEM2-CC
GCM
obs
MPI-ESM-MR
GCM
obs
MRI-CGCM3
GCM
obs
MRI-ESM1
GCM
obs
NorESM1-M
GCM
obs

Sesgo

Duracin media de
rachas secas (das)

Precipitacin
media diaria

Percentil 95
(mm)

Fraccin de
das con lluvias

Duracin
media de
rachas
hmedas
(das)

22%

3.75
2.45

6.64
5.45

25.60
19.16

75%
74%

11.36
7.01

59%

1.89
2.36

9.16
5.76

21.77
19.64

97%
76%

51.83
7.55

165%

2.49
2.36

15.30
5.76

32.36
19.64

92%
76%

28.38
7.55

161%

1.81
2.36

15.02
5.76

35.07
19.64

97%
76%

56.41
7.55

-4%

3.29
2.36

5.55
5.76

17.71
19.64

65%
76%

6.13
7.55

90%

3.04
2.36

10.98
5.76

36.44
19.64

78%
76%

10.95
7.55

86%

2.99
2.36

10.74
5.76

36.11
19.64

78%
76%

10.44
7.55

7%

8.90
2.36

6.16
5.76

18.42
19.64

69%
76%

19.51
7.55

Series WEAP compatibles de tiempo diarias se han generado para cada uno de los escenarios de MCG.
Escenarios similares de datos se generaron para la cuenca del Piura y se colocan en una caja de reenvo
para el acceso de personal SEI. Estos datos consisten en series de tiempo todos los das de
precipitacin y temperatura en un formato similar al que se muestra a continuacin:
;fecha,Las Brisas (4150),La Esperanza (3280)
1/1/1981,5,5
1/2/1981,3,3
....
12/30/2050,0,3,0,1
12/31/2050,0.3,0,4

35

3.4.2

Resultados para la Cuenca Chira-Piura

Se presentan los resultados de los modelos de circulacin general ms de la cuenca del Piura, lo que
demuestra una vez ms que el modelo MPI-RM aparece tener la mayor habilidad para simular el clima
de la regin.
Precipitation stats:
Figura 17 Presentacin de Resultados GCM para Cuenca Piura, Peru

36

37

38

Tabla 7 Resumen de la comparacin de atributos modelados y observados en los diferentes GCM analizados.

Modelo

Sesgo

Precipitacin
media diaria

Percentile 95
(mm)

Fraccin de
das con
lluvias

Duracin
media de
rachas secas
(das)

Duracin
media de
rachas
hmedas
(das)

159%

5.47

19.28

72%

3.37

8.52

2.11

10.04

39%

8.50

5.40

4.45

17.41

62%

3.54

5.75

1.80

8.66

37%

8.03

4.66

5.49

17.05

78%

3.06

10.84

1.80

8.66

37%

8.03

4.66

6.06

12.00

68%

3.76

8.03

8.89

34%

9.35

4.85

9.26

28.53

59%

10.63

15.54

2.29

11.12

40%

6.13

4.08

9.40

27.46

89%

2.95

23.68

2.29

11.12

40%

6.13

4.08

3.31

15.21

41%

5.50

3.83

2.29

11.12

40%

6.13

4.08

7.70

28.33

64%

5.33

9.62

2.02

9.96

37%

8.60

5.10

8.17

29.30

64%

5.51

9.98

2.02

9.96

37%

8.60

5.10

7.21

21.94

72%

5.18

13.31

1.75

8.58

34%

9.38

4.80

CanESM2
GCM
obs
CCSM4
GCM

147%

obs
CESM1-CAM5
GCM

205%

obs
CNRM-CM5
GCM

135%

obs
CSIRO-Mk3-6-0
GCM

304%

obs
HadGEM2-CC
GCM

310%

obs
MPI-ESM-MR
GCM

44%

obs
MRI-CGCM3
GCM

281%

obs
MRI-ESM1
GCM

304%

obs
NorESM1-M
GCM
obs

312%

39

Figura 18 Presentacin de precipitacin mes GCM Resultados para Cuenca Piura, Peru

40

41

Figura 19 Periodogram plots para Cuenca Chira-Piura, Peru

CanESM2

CCSM4

CESM-CAM5

CNRM-CM5

CSIRO MK3

HagGEM2 CC

MPI-ESM-MR

42

MRI-CGCM3

MRI-ESM1

NorESM1

43

Figura 20 Cambios proyectados en diferentes atributos de la precipitacin por los dos modelos de mejor
desempeo MPI-ESM-MR y CCSM4 para Cuenca Chira-Piura, Per.

44

45

46

47

48

3.5 Resumen de la Actividad 1

Este informe presenta series meteorolgicas de estado de tiempo (diarias) informadas por cambio
climtico cuencas Chira-Piura en Per y Chinchin en Colombia, para escenarios de cambio climtico
de temperatura media y precipitacin, usando los resultados ms recientes del proyecto CMIP5 Escenario RCP8.5.
Los escenarios se generaron usando el mtodo K-NN, para que corrige el sesgo y espacialmente
desagregados los datos, para secuencias de resolucin diaria desde 2015 hasta 2089, tomando como
referencia las observaciones entre 1970-2010 y los modelos MPI-ESM-MR (Max Plank Institute) y
Community Climate Systems Model (CCSM) cuyas proyecciones climticas se encontr reflejan la
mayora de los patrones de estado del tiempo locales. Si bien se evalu una cantidad importante se
modelos, el escenario seleccionado se bas en un enfoque de secuencias informadas por el modelo de
mejor desempeo, debido a que se encontr que hay modelos que presentan sesgos considerables en
la estimacin de los diferentes atributos del clima, y que en consecuencia, se consideran no
representativos de los factores generadores del clima local.
Las series sintticas fueron generadas para las estaciones locales (1259 estaciones de precipitacin y
184 estaciones de temperatura). Los resultados de este anlisis aportan y complementan a los estudios
existentes sobre cambio climtico en la cuenca, especficamente en la representacin a escala reducida
de atributos de la variabilidad climtica a pasos de tiempo diarios y semanales, que hasta el momento
no se encuentran disponibles en los estudios previos implementados por el IDEAM y otras
instituciones (WorldClim y ClimateWizard). Si bien estos resultados aportan mayor detalle en relacin
con las escalas temporales a los estudios previos, a niveles de agregacin mayores (anuales y
mensuales) los resultados obtenidos corroboran la informacin obtenida por otros estudios, por
ejemplo, los incrementos esperados en la temperatura a mitad a 2050 para el escenario RCP 8.5 (+0.9
0.4C).

3.6 Referencias para la Actividad 1

Domnguez, E., 2005. Pronstico probabilstico de afluencias para la evaluacin de riesgos en
embalses. Avances en Recursos Hidrulicos, 12, p.25.
Gangopadhyay, S., Clark, M. & Rajagopalan, B., 2005. Statistical downscaling using K nearest neighbors. Water Resources Research, 41(2), p.n/an/a. Available at:
http://doi.wiley.com/10.1029/2004WR003444 [Accessed June 20, 2014].
Gutmann, E. et al., 2013. Statistical Downscaling Inter Comparison, Available at:
http://www.cesm.ucar.edu/events/ws.2013/presentations/SDWG/gutmann.pdf.
Hijmans, R.J. et al., 2005. Very high resolution interpolated climate surfaces for global land
areas. International Journal of Climatology, 25(15), pp.19651978. Available at:
http://doi.wiley.com/10.1002/joc.1276 [Accessed May 25, 2014].
Hijmans, R.J., Cameron, S. & Parra, J., 2013. WorldClim - Global Climate Data. Available at:
http://www.worldclim.org/about.

49

Koutsoyiannis, D., 2010. HESS Opinions A random walk on water. Hydrology and Earth
System Sciences, 14(3), pp.585601. Available at: http://www.hydrol-earth-systsci.net/14/585/2010/.
Moss, R.H. et al., 2010. The next generation of scenarios for climate change research and
assessment. Nature, 463.
Ruiz, F. et al., 2013. Informe de Escenarios de Cambio Climtico para Temperatura y
Precipitacin en Colombia, Bogot.
Tippett, M.K. & DelSole, T., 2013. Constructed Analogs and Linear Regression. Monthly
Weather Review, 141(7), pp.25192525. Available at:
http://journals.ametsoc.org/doi/abs/10.1175/MWR-D-12-00223.1 [Accessed June 26,
2014].
Yates, D. et al., 2003. A technique for generating regional climate scenarios using a nearestneighbor algorithm. Water Resources Research, 39(7), p.n/an/a. Available at:
http://doi.wiley.com/10.1029/2002WR001769 [Accessed June 5, 2014].
Yates, D. et al., 2005. WEAP 21. A demand, priority and preference driven water planning
model. Part 2: Aiding Freshwater Ecosystem Service Evaluation. Water International,
30(4), pp.501512.
Young, C.A. et al., MODELING THE HYDROLOGY OF CLIMATE CHANGE IN CALIFORNIA
S SIERRA NEVADA FOR SUBWATERSHED SCALE ADAPTATION 1. , pp.115.

50

4 Activity 2.1 (Development of GCM boundary forcing for current

and future periods)
To quantify 21st century climate change over the Northern Andean region, we performed simulations
using version 3.5.1 of the Weather Research and Forecasting Model (WRF, Skamarock et al. 2008).
WRF is a fully compressible conservative-form nonhydrostatic atmospheric model with demonstrated
ability for resolving small-scale phenomena and clouds (Skamarock and Klemp 2008). Here, WRF is
employed to dynamically downscale climate fields from NCARs comparatively coarse-scale gridded
global climate model (GCM), the Community Climate System Model- Version 4 (CCSM4) which covers
the global domain. The WRF simulations are comparatively fine-scale, covering the regional domain of
the Northern Andes, which is relevant for assessing climate change impacts at regional-to-local scales.
The goal of this regional climate modeling activity was to develop projections of regional climate at
fine spatial and temporal scale, that reflect the large-scale features and temporal trends from
Atmosphere Ocean Global Climate Model (AOGCM or GCM for short) simulations (AR5), but also
the historical patterns of climate variables at the regional and local scale (Fowler et al., 2007; Wood et
al., 2004). To achieve this, a regional climate model (RCM) was deployed that dynamically downscaled
the climate of the Northern Andes using GCM data for lateral boundary conditions. Boundary
conditions are time-dependent lateral meteorological data that drive the high-resolution RCM. This
driving data is derived from GCMs (or analyses of observations) and can include GHG and aerosol
data. The RCM simulations should better represent the topographic gradients across the domain,
including the influence of the Andean mountain region, which potentially increases and re-distributes
precipitation due to enhanced lifting.
WRF benchmark simulations were performed over a historical period to best estimate the true state
and dynamics of the atmosphere. The benchmark simulations in this study derive their initial and
boundary conditions from the European Centre for Medium-Range Weather Forecasting (ECMWF)
Interim Reanalysis (ERA-Interim; Dee et al. 2011). ERA-Interim is considered to be the most accurate
atmospheric reanalysis available at the present time (e.g., Lorenz and Kunstmann 2010).
We then used data from the using the bias-corrected CCSM4 output, for the contemporary period (5years at 18 and 6 km) and the RCP8.5 future period (selective years of ENSO states), which is
described below. The purpose of performing the GCM-driven WRF climate simulations for the
historical period is to:
1. Generate a dataset that can be used to validate the GCM-driven WRF simulations against the
reanalysis-driven WRF benchmark simulations described above for some common historical
period, and
2. Provide a baseline dataset against which future GCM-driven WRF climate simulations can be
assessed.
The purpose of performing the GCM-driven WRF climate simulations for the future period is to
provide a projection for the future state of the atmosphere in some latter portion of the 21st century.
The WRF climate simulations in this study derive their initial and boundary conditions from version 4
of the Community Climate System Model (CCSM4; Gent et al. 2009). CCSM4 is a subset of Version 1
of the Community Earth System Model (CESM1; Hurrell et al. 2013). CCSM4 is a coupled global
climate model (GCM) comprised of four component models that simulate the atmosphere, ocean, land
surface and sea-ice. The CCSM4 simulations used to generate the present dataset were performed in
support of the Coupled Model Intercomparison Experiment Phase 5 (CMIP5; Taylor et al., 2012) and
the Fifth Assessment Report of the Intergovernmental on Climate Change (IPCC 2013). CCSM4 ranks
at the top of all CMIP5 GCMs in its ability to simulate observed temperature and rainfall globally
51

(Knutti et al. 2013). Model fields were obtained from the National Center for Atmospheric Research
on their Globally Accessible Storage Environment (GLADE), and are also available from the Earth
System Grid - Program for Climate Model Diagnosis and Intercomparison (ESG-PCMDI) gateway at
Lawrence Livermore National Laboratory (http://pcmdi3.llnl.gov/esgcet/home.htm). The CMIP5 model
scenarios used to construct the present bias-corrected dataset include a historical simulation and three
future projections. The historical simulation was forced by observed natural and anthropogenic
atmospheric composition changes spanning 1861-2005.
Figure 21 compares the terrain from the CCSM4 GCM to that from the WRF 18-km (top) and 6-km
(bottom) domains (described below). It is clear that CCSM4, with a spatial resolution of 0.9 degrees
latitude x 1.25 degrees longitude (approximately 100 km), cannot adequately resolve the topography of
the Andes and other important orography of the region compared to WRF, demonstrating the
necessity of performing the WRF dynamical downscaling simulations in order to provide a dataset that
is appropriate for assessing climate change in the region.
CCSM3

WRF

Figure 21. Terrain height (m, color scale at bottom) and land/sea mask for CCSM4 (top) and 6-km WRF
(bottom). Actual coastlines and political boundaries shown in black.

52

The future projections are the Representative Concentration Pathway (RCP; Moss et al. 2010) 4.5, 6.5
and 8.5 scenarios, which span 2006-2100. RCP4.5 is a low-to-moderate emissions scenario with GHG
radiative forcing reaching 4.5 W m-2 near 2100. It represents a trajectory that may be plausible if, for
instance, GHG emissions pricing were introduced in order to limit radiative forcing (Thompson et al.
2011). RCP6.0 is a moderate GHG emissions scenario that is similar to RCP4.5 in that a variety of
strategies for reducing GHGs would be applied to eventually stabilize radiative forcing near the end of
the 21st century (Masui et al. 2011). RCP8.5 is a high-emissions scenario with greenhouse-gas (GHG)
radiative forcing reaching 8.5 W m-2 near 2100. It represents a plausible trajectory if little is done to
curb greenhouse gas emissions (Riahi et al. 2011). While an ensemble of CCSM4 simulations was
performed for each scenario in order to characterize model-based uncertainty, only output from
Ensemble Member #6 -- also known as the "Mother of All Runs (MOAR)" -- was used to construct the
present files, because that is the only member that has available at 6-hourly intervals the full threedimensional fields required to force WRF or MPAS. The specific CCSM4 Ensemble Member #6
simulations used were: b40.20th.track1.1deg.012 (20th Century), b40.rcp4_5.1deg.006 (RCP4.5),
b40.rcp6_0.1deg.006
(RCP6.0)
and
b40.rcp8_5.1deg.007
(RCP8.5)
(additional
details
at: http://www.cesm.ucar.edu/experiments/cesm1.0/).
Like all GCMs, CCSM4 contains regional-scale biases due to its relatively coarse spatial resolution and
a limited representation of many complex physical processes. Such biases can adversely affect the
dynamical downscaling process and contribute to uncertainty. To remedy these biases, it is common to
bias correct the climate model output before using it to drive regional-scale models like WRF (e.g.,
Rasmussen et al. 2011, Xu and Yang 2012, Done et al. 2013). The present Intermediate files were
constructed with a recently-developed bias correction method that uses a global atmospheric
reanalysis to correct for the mean bias in the CCSM4 3-dimensional temperature, geopotential height,
wind, and humidity fields, as well as the surface pressure, sea level pressure, sea surface temperature
(SST), skin temperature, and soil temperature and moisture fields (Bruyere et al. 2013). Although the
bias in the mean state is corrected with the reanalysis fields for each variable, the methodology still
allows synoptic-scale and climate-scale variability to change in the future as simulated by CCSM4. The
reanalysis used for the bias correction is the European Centre for Medium-Range Weather Forecasting
(ECMWF) Interim Reanalysis (ERA-Interim; Dee et al. 2011), obtained from NCAR's Research Data
Archive for the surface and pressure levels (ECMWF 2009, 2012). ERA-Interim is considered one of
the most accurate atmospheric reanalysis available at the present time according to a variety of
measures (e.g., Lorenz and Kunstmann 2010).
Bias-corrected Intermediate files that can be used as the initial and boundary conditions for the WRF
were constructed with simulations from Version 4 of the Community Climate System Model (CCSM4;
Gent et al. 2011). A description of the bias correction procedure is provided in the Appendix. The
bias-corrected CCSM4 output was produced by summing the 1981-2005 average 6-hourly annual cycle
(the Reynolds-average mean term) from Era-Interim ("ERAINT") and a 6-hourly perturbation term (the
Reynolds-average eddy term) from CCSM4:

CCSM
= CCSM + CCSM
ERAINT
= ERAINT + ERAINT
CCSM
ERAINT + CCSM
=
R
Where the overbar terms are the mean climatology, primed terms are perturbations from the
climatology, and CCSM R is the revised (bias-corrected) CCSM4 output at 6-hourly intervals, which is
subsequently used as the initial and boundary conditions for the WRF climate simulations. A schematic
of the bias-correction method is shown in Figure 22.
53

Using the bias-correction technique, 6-hourly Intermediate files were generated for the
historical period and the three RCP scenarios. The files have the following naming
conventions:
1. Historical Period (1951-2005): CCSM4_CMIP5_MOAR_BC_20THC:YYYY-MM-DD_HH
2. RCP4.5 Scenario (2006-2100): CCSM4_CMIP5_MOAR_BC_RCP45:YYYY-MM-DD_HH
3. RCP6.0 Scenario (2006-2100): CCSM4_CMIP5_MOAR_BC_RCP60:YYYY-MM-DD_HH
4. RCP8.5 Scenario (2006-2100): CCSM4_CMIP5_MOAR_BC_RCP85:YYYY-MM-DD_HH
where YYYY is the present year, MM is the present month, DD is the present day and HH is
the present hour.

Figure 22. Schematic of the bias-correction method for a generic variable. The CCSM4 perturbation term is
added to the 1981-2005 ERA-Interim mean term.

Figure 23 shows the precipitation anomaly over the Northern Andean region for the CCSM4 RCP 8.5
experiment. The take-home message of this plot is that the scenario that we are using generally shows
a slightly increasing rainfall trend over the region while the temperature anomaly is approximately
3.5oC for the RCP 8.5 projection (not shown).
Figure 23. Schematic of the bias-correction method for a generic variable. The CCSM4 perturbation term is
added to the 1981-2005 ERA-Interim mean term.

54

5 Activity 2.2. (The WRF Regional Climate Simulations)

The WRF computational domains are shown in Figure 24, with the outer domain on a grid spacing of
18-km resolution (D1), which covers much of northern South America and the Eastern Pacific
Ocean. Nested inside the 18-km domain is a 6-km domain (D2) covering the region that includes
Colombia and Peru. The WRF simulations feature 40 vertical levels from the surface to 10 hPa (about
30 km above the surface). The WRF simulations are reinitialized every eight days, and each eight-day
period is preceded by a 12-hour period that allows the WRF hydrological fields to spin up, and which is
subsequently discarded. Throughout the simulations, four-dimensional data assimilation (FDDA,
Stauffer and Seaman 1994) -- i.e., grid nudging is employed on the 18-km domain to keep the
model solution from diverging from the large-scale global boundary conditions, which are described in
detail below.
Physical parameterization schemes, which simulate the sub-grid scale processes in WRF empirically,
include the Lin microphysics scheme, the RRTM longwave radiation scheme, Dudhia shortwave
scheme, MM5 surface layer scheme, Noah land surface model, YSU PBL scheme, and the Grell-Devenyi
convective scheme (for both the 18-km and 6-km domains). These parameterizations are chosen
because they yielded optimal WRF performance over Northern Andean regions when compared to
satellite remote precipitation for the case study periods. The global domains providing the initial and
lateral boundary conditions for the WRF dynamical downscaling simulations come from two sources
depending on whether they are benchmark simulations or climate simulations, which are described
next.
Figure 24. The WRF domains for northern South America (Domain 1 at 18-km resolution); and the higher
resolution domain (Domain 2 at a 6-km resolution) over the northern Andean Region of South America.

55

5.1 WRF benchmark simulations with ERA-Interim

The ERA-Interim fields employed here for the benchmark simulations have a ~0.7 grid spacing on 38
vertical levels. Sea surface temperature (SST) data at the lower oceanic boundaries of these benchmark
simulations are from version 2 of the National Oceanic and Atmospheric Administration (NOAA)
Optimum Interpolation (OISST) 0.25 degree product (Reynolds et al. 2007). A first and critical step in
configuring the WRF model to simulate both the current and future climate of the region, is to
evaluate which set of model physics gives the best representation of the regional climate. We did this
by testing various configurations of the model setup, which are summarized in Figure 25. This figure
shows a 6-day accumulated precipitation over the 6-km domain of the Northern Andes for a strong La
Nia year, for the period 12 to 17 December, 2010 for different configurations of the WRF model
(Runs 1 to 8) compared to a satellite-based observational product-CMORPH (Joyce 2004).
Figure 25 shows the various configurations of the model in terms of WRF model configuration,
including the convective parameterization (KF is Kain-Fritsch, GD is the Grell-Devenyi scheme),
microphysics (Lin, Thompson), the radiation and planetary boundary layer schemes including (YSU
PBL). These various runs were used to conclude that the best performing set of model configurations.
Run #5 seemed to exhibit the best performance in terms of a reasonable comparison to the CMORPH
rainfall estimates, as the values look reasonable compared to other gauge-based precipitation
climatologies. Importantly, run Number 5 used convective parameterization (CP), while when CP was
turned off and convection was explicitly resolved (Run 6), the rainfall patterns were much more
concentrated as compared with Run 5, which was more distributed in nature. Also, note that in the
southwestern portion of Colombia, the spatial patterns of rainfall is much different, with the influence
of the Andean topography demonstrated. Note the low precipitation band stretching from the
southwest to the northeast of the domain.
The December 2010 precipitation pattern is similar to 7-year climatology, but with more intense
precipitation north and south of isthmus and greater localized amounts over the higher terrain and the
Amazon region (Fig 26). Keep in mind CMORPH data are averaged over 30 km boxes while WRF
model inner domain has a 6 km grid resolution. WRF modeled values are generally comparable to
CMORPH, with maximum rates generally 2-5 time the CMORPH, which is consistent with the higher
resolution of the atmospheric model. The CMORPH will smooth out rainfall, and so we should
observe considerably higher rainfall rates as simulated by WRF. Interestingly, using convective
parametrization on the 6km grid results in precipitation patterns that more closely match CMORPH,
with CP runs generally having lower max precipitation rates, with precipitation from CP runs generally
exhibiting more smoothly varying spatial patterns, which are more coherent and look better over the
Amazon region. We concluded that Run 5 was subjectively determined be the best performing
configuration for this case study in terms of accumulated precipitation. Note the lower precipitation
values over the higher terrain of the Andean plateau; which is a difficult attribute to model, with some
of the WRF runs not able to capture this (Fig 25 and Fig 26).

56

Figure 25. Figure of CCSM 4 output (Activity 2.1; GCM boundary forcing). Run 5 configuration was used.
Run 1
Kain-Frisch CP
LIN and YSU

Run 2
KF CP off, Lin,
and YSU

Run 3
GD CP On, Lin,
YSU

Run 4
GD CP off, Lin,
YSU

CMORPH (OBS)
(280 mm)

*Run 5
GD CP ON,
Thomson, YSU

Run 6
GD CP Off,
Thompson, YSU

Run 7
GD On, Goddard
MP, YSU

Run 8
GD Off, Goddard
MP, YSU

CMORPH (OBS)
(280 mm)

57

Figure 26. CMORPH monthly mean December precipitation for 7 years (left) and December 2010 (right).

Dec 2003-2010 CMORPH Climo (max760 mm)

Dec 2010 CMORPH (max =1100 mm)

5.2 WRF simulations with CCSM for current and future periods.
Due to computing constraints (simply not enough computing time or resources available to run long
transient simulations at high resolution), our focus for running current and future climate simulations
was to focus on physical process understanding, rather that long runs of the WRF model. What we
hope to understand is the nature of the climate over the region of interest (northern Andes), both
now and in the future. For this reason, we decided to focus on the use of the El Nio-Southern
Oscillation (ENSO) cycle and its different phases.
Our simulation with WRF were then based on the selection of different phases of the ENSO signal that
included: 1) extreme La Nia, 2) moderate La Nia, 3) neutral ENSO events, 4) a moderate El Nio
and 5) an extreme El Nio. Note that the historical ENSO events using the NCAR CCSM4 runs do
not coincide with observed historical events because the internal variability in CCSM4 is not
temporally consistent with the observed climate system.
To define the years for which to simulate, our ENSO definition was defined using Region 3.4 following
Steinhoff et al. (2014) for the geographic region defined over the area, 5S-5N, 170-120W, and we then
computed an SST Anomaly as: SSTa = (SST(month) SST(monthly_mean)). The periods that were
used included, the present day 1905-2005 and the future 2005-2100 and we then computed a Mean
SST for DJF. A major ENSO event was defined as those less than 25th percentile or greater than 75th
percentile which is a mean anomaly close to +/-1.5 K SST variation.
Figure 27 shows the DJF SST anomaly for the ENSO3.4 region from the CCSM4 results, where the
anomalies were computed by subtracting ~100 year monthly mean from the monthly value and
computing the average anomaly for the 3 month period DJF for each winter. The Red Dots and the
anomalies greater than 1.5oC (El Nio Winter DJF); Blue Dots are the anomalies less than -1.5oC (La
Nia Winter for DJF) and the green line is the detrended DFJ SST anomalies. Of note is the absence
58

of La Nia events in the latter half of the 21st century. There appear to be fewer, but longer lived El
Nio events in future climate which may require multi-year simulations to capture entire El Nio
period. Note that for La Nia, strong events become less frequently toward the end of present day
and occur only prior to 2060 in the future and their duration appears shorter in the future as well. We
use these results to select specific events in the future for understanding the impact of regional climate
change on precipitation and temperature.
Figure 27. CCSM4 estimate of the ENSO output for the contemporary period. 1981 was the La Nia event,
while 1997 and 2057 have been the El Nio events in which we have focused our WRF modeling efforts.

13 El Nio and 7 La Nia Events

10 El Nio and 8 La Nia Events.

Figure 28 shows WRF runs run to data for current and future conditions. We have run a major La
Nia event (1981 in the CCSM4 run), a strong El Nio event (1998 in the CCSM4 run) and a future El
Nio (2057).

59

La Nio 2057

El Nio 1997

La Nia 1981

Figure 28. Domain 2 (6-km) total annual rainfall (top left), topography (top right) and daily rainfall given
across a transect for a current La Nia, a current El Nio and a future El Nio.

Figure 26 shows includes a transect across the plains of Peru into the high Andean interior of daily
rainfall, depicting the low values on the plains and the higher values to east. The transect and mean
annual map highlights the complex spatial patterns of precipitation across the region near the Piura
Watershed of Peru. These data show the daily precipitation increasing from west to east, and show
60

that the total rainfall for the La Nia event is smaller than in the La Nio event. These data will be used
in a statistical manner to condition the K-NN statistical model presented under Activity 1 and Activity
3, to generate new future climate projections that can be used in the hydrological modeling activities
with WEAP. These data will be generated in January of 2015, once the final WRF simulations are
finished.

61

Appendix: Description of Software

These software scripts and codes were written for the Para-Agua, to enable the development of what
are known as intermediate files for the WRF model. These intermediate files are specially formatted
so that WRF can make use of them for the regional climate simulations. The codes are summarized
below:

a. Conversion of CESM to Intermediate Format

This module is called "CCSM4_TO_WRFI_CMIP5_V3", and is contained in a directory
of the same name. Within the directory, the following NCL script reads in CCSM4 data
from the CMIP5 archive and writes out the required fields in WRF Intermediate Format:
convert_ccsm_hybrid_nc_to_pressure_wrfint_3d.ncl
The CESM1 data are stored on NCARs Globally Accessible Data Environmenta
centralized file service known as GLADE, for easy access. These data are stored in
/glade/p/vetssg/data/CMIP5/output1/NCAR/CCSM4/. You can get an idea of where the
exact directories of interest are by looking in the driver script described in the section
on how to run the software below. Note that only one CCSM simulation -- Member #6,
or 6i1p1 -- has been archived in a manner that the full, 6-hourly 3D data required to
drive WRF is available. Therefore, you can drive WRF with the data from the 6i1p1
stream, for historical (from 1951-2005) and rcp45, rcp60, and rcp85 (from 2006-2100).
From the NCL script, the following fields are written out at 6-hourly intervals, to
Intermediate files called CCSM4_CMIP5_MOAR_CASE:YYYY-MM-DD_HH, where
"CASE" is either 20THC, RCP45, RCP60 or RCP85 and YYYY, MM, DD and HH have
their usual time conventions.
How to Run CCSM4_TO_WRFI_CMIP5_V3 Software
[Note: This software has only been tested on NCAR's Yellowstone/Geyser
supercomputing platform and instructions below are based on this architecture]
1. Compile the fortran routine that is called by NCL via a wrapper:
type "./prepare_software.csh" . If successful, the following library file will appear in the
"./SRC" directory: write_intermediate.so
Note that the Fortran code is designed to by compiled using the gnu-based compilers
that are packaged with ncl. This shouldn't require any additional modules to be loaded
on your part. Specifically, it is required that the intermediate files be written out in bigendian format. Gnu allows a special big-endian flag to be specified in the open
62

statement for the wrf intermediate file within the Fortran routine. We did not have any
success using other compilers.
2. Make sure you load all of the modules you may need:
type "module load ncl"
type "module load cdo"
3. Go to the SEAICE directory and run the get_seaice.csh script for your years of
interest (submit the script to the geyser queue using submit_job_to_queue.csh).
Unfortunately, this step is necessary because, as the time these files were created,
there was only monthly average sea ice fraction data available on GLADE, and we
need at least daily varying sea ice fields in order to have consistent data for our lower
boundaries. The only way to get the sea ice is to download it from the HPSS tape
storage, which is what this script does.
4. The next step is to simply run the driver script, run_process_ccsm_to_wrfi.csh
4a.: Modify the driver script to specify the years you want to create intermediate files
for. The intermediate files are created in 1-year chunks.
Modify the following line to specify the year or years (separate multiple years by
spaces), e.g.:
#pick a year
foreach yyyy (1960 1961)
4b. Run the script. It takes about 30 min to finish 1 year of data and write out all of the
intermediate files. A year of data is 40 Gb. Output files are 6-hourly and named per the
convention described above. Note to make sure the following file is in your directory,
which allows the SST and SEAICE fields to be interpolated from the POP grid to the
CCSM grid: map_gx1v6_to_fv0.9x1.25_aave_da_090309.nc (it should be there, so
just double check).
type "./run_process_ccsm_to_wrfi.csh" .
b. Conversion of Era-Interim to Intermediate Format
This module is called "ERAI_TO_WRFI_CMIP5_V3". The purpose of this software
package is to convert the surface and pressure-level ERA-Interim fields to 1) the same
6-hourly horizontal and vertical domain as the CCSM4 data that was processed in the
step above and 2) to Intermediate format. The purpose of this step is to facilitate the
bias-correction step, which is described in the following section. The ERA-Interim data
used are stored as dataset ds627.0 (6-hourly) and ds627.1 (monthly) on GLADE
63

courtesy
of
NCAR's
Research
Data
Archive
(see
the
script
"run_process_erai_to_wrfi.csh" for details on exact locations of the ERA-Interim data).
The procedure to run this software package is nearly identical to that described above
for "CCSM4_TO_WRFI_CMIP5_V3", so details are not provided here to avoid
repetition.
The
ncl
script
that
reformats
ERA-Interim
("convert_era_grib_to_ccsm_pressure_wrfint_3d.ncl") uses the gaussian-to-fixed
global grid functions that are available in NCL in order to do the horizontal grid
transformation. No vertical interpolation is necessary as all 26 of the vertical pressure
levels that are needed to match the CCSM4 vertical levels are already available. All
fields are available from the ERA-Interim output at 6-hourly intervals except for
"TAVGSFC" which is intentionally derived from the monthly mean ERA-Interim skin
temperature in order to maintain stable inland lake surface temperatures.
c. Bias-Correction
This module is written primarily in fortran and performs the Bruyere et al. (2013) bias
correction by reading in the CCSM4 and ERA-Interim intermediate files that were
created using the two software packages described above. If researchers from IDEAM
or SENHAM are interested in acquiring this dataset, please contact David Yates at
yates@ucar.edu.

Notes on the use of the CESM Intermediate files in WPS and WRF
1. This entire software package is intended to replace the "ungrib.exe" program in
WPS, because ungrib.exe cannot handle netcdf input, which is what the CESM data
are. Just as ungrib.exe is meant to create WRF intermediate files from grib input, this
package creates WRF Intermediate files from netCDF input. You will still be required to
run all other steps of WPS and WRF. A typical workflow would be:
Run geogrid.exe after specifying all of your namelist.wps parameters. You should
seriously consider modifying the namelist.wps (and possibly GEOGRID.TBL) to allow
for the inland lake surface type, which in turn will use the "TAVGSFC" variable to
initialize lake surface temperatures. Otherwise, make sure your lake surface
temperatures look reasonable. For instructions on how to do this, see
<http://www.mmm.ucar.edu/wrf/users/docs/user_guide_V3/users_guide_chap3.htm#_
Alternative_Initialization_of>. Note that if you choose to go with the inland lakes option,
you DO NOT have to run tavg_sfc.exe per the instructions at this hyperlink -- This has
already been done in the NCL script and TAVGSFC is already available to you within
the CCSM3D:XXXX-XX-XX_XX files (therefore, you also DO NOT need to add
"TAVGSFC" to the 'constants' section of namelist.wps as instructed at the link above).
64

-- Run the software in this directory *instead of ungrib.exe* to create your intermediate
files.
--Run metgrid.exe, making sure that you have properly specified the names of the
intermediate files in namelist.wps. You may also want to modify METGRID.TBL to
optimize interpolation of SSTs. See Below.
-- Run real.exe from the WRF directory once you have successfully created your
met_em files. Note that you may have to modify "NUM_LAND_CAT" in the wrf
namelist to reflect the new "lake" land surface type in your dataset. You can find the
value of NUM_LAND_CAT from doing an ncdump -h of any met_em file.
-- Run wrf.exe
2. The CCSM data has NO LEAP YEARS. In order to deal with this, if you will be
running simulations that span any leap years, you have to add DNO_LEAP_CALENDAR to "ARCH_LOCAL" in the configure.wps file before compiling
WPS (note that some versions of the instructions say to add this flag to "CPPFLAGS"
rather than "ARCH_LOCAL" -- you may have some trial-and-error to figure it out).
3. Downscaling GCM data has a different set of choices regarding simulation
strategies than, for example, using WRF for forecasting or hindcasts, since those
simulations involve driving WRF with "real" meteorological data rather than GCM data.
For starters, some users prefer not to do cold-starts (re-initializations) of WRF every
few days as is common for downscaling "real" data. Instead, they might run an entire
20-year period continuously, without any cold starts (in which case they use
intermittently-written "RESTART" files to get around wall clock constraints). One
advantage of running continuous simulations is that they allow the model soil state to
spin up via the land surface model (e.g., the Noah LSM that is coupled to WRF). It
typically takes about a year for soil fields to spin up in the LSM. Another advantage of
continuous simulations is that they don't require as much effort for preprocessing,
because there are fewer cold-starts to heed; since these are climate model runs, the
typical reasons for doing frequent cold starts (e.g., using high fidelity initial conditions
to constrain model accuracy with respect to the large-scale driving fields) are not
generally first-order considerations. However, you should keep a few things in mind if
you choose to do long-term (i.e., greater than a month-long) simulations:
4. Make sure you regularly update fields such as SST using the "sst_update" flag in
your WRF namelist. Otherwise, you might be using January SSTs in July if you
initialized a run in January of year X. This software provides 6-hourly TAVGSFC fields
65

so that the inland lakes can be updated along with SSTs (note that although
TAVGSFC is written out at 6-hourly intervals for convenience and consistency with
other fields, TAVGSFC is actually the monthly average skin temperature and therefore
only changes on the first day of each month; this approach prevents spuriously large
diurnal or day-to-day fluctuations of the lake surface temperatures that can otherwise
occur.
5. Think about whether or not spectral or grid nudging (fdda) would be appropriate.
This depends on your motives, although one theory is that it may be better not to
nudge during long GCM-downscaling runs because the GCMs themselves probably
don't have a great representation of the large-scale atmospheric forcing, and thus it
may be better to let WRF drift toward its "climatology" inside the model domain.
However, it could be advantageous to do fdda if WRF is drifting too much toward
unrealistic values over the course of your long simulations. Or, your objective may be
to have your simulations be more heavily constrained by the large-scale forcing. In any
case, if you do decide to nudge, you should only nudge the upper-most levels (i.e.,
above ~500 hPa) if you use grid nudging, or use spectral nudging in a manner that you
only nudge the large waves. You want to avoid dampening the energy near the surface
in WRF -- that is the whole reason for downscaling. Several recent publications
suggest that grid nudging, when done properly (only large scale forcing), can improve
simulations of extremes in WRF, at least when driving WRF with reanalysis data (see
Otte et al., 2012, J. Climate http://dx.doi.org/10.1175/JCLI-D-12-00048.1, or Glisan et
al., 2013, J. Climate, http://dx.doi.org/10.1175/JCLI-D-12-00318.1).
6. You may want to run a "spin-up" year prior to your period-of-interest and then throw
it out. This will allow full spin up of the soil state.
7. Fractional sea ice from CCSM has been included in the WRF Intermediate files for
your use. As mentioned below, interpolation of sea ice from the comparatively coarse
resolution of the CCSM domain to your WRF domain can be tricky. Therefore, be sure
to check your sea ice fields in the wrflowbdy_d0X file once you have interpolated
everything onto the WRF domain, to make sure they look realistic, and especially to
make sure there aren't issues near coastlines, where masking differences between
CCSM and WRF can be problematic. If you choose not to use fractional sea ice, make
sure to at least set " seaice_threshold = 271.35 " (-1.8 C) in the WRF namelist, which
will diagnose whether sea ice exists based on whether the SST is less than 271.35.
8. You may have to experiment with some of the interpolation options in
METGRID.TBL to refine your subsequent ingest of the WRF intermediate files into
metgrid.exe, in order to make sure you get fields in your met_em files that look right. In
66

particular, masked data -- SST and SEAICE -- interpolation can be tricky near the
coastlines when using the comparatively low resolution data from CCSM.
Here are some options (modifications to METGRID.TBL) that seem to work well:
========================================
name=SST
interp_option=sixteen_pt+wt_average_16pt+search
masked=land
interp_mask = LANDSEA(1)
fill_missing=-1.E+30
flag_in_output=FLAG_SST
missing_value=-1.E+30
========================================
========================================
name=SEAICE
interp_option=four_pt+average_4pt
interp_mask=LANDSEA(1)
masked=land
missing_value=-1.E30
fill_missing=0.
========================================
***Note that the "search" interpolation option for SST will mean that any *inland* lake
will take on the SST value from the nearest ocean grid point...this is usually very, very
inaccurate. Therefore, if you use this option, make sure to use the inland lakes option
("TAVGSFC") discussed elsewhere in this document.
9. In your namelist.wps file, the following is the correct way to specify the "&metgrid"
section (note that everything you should need to run the model is in the
CCWM4_CMIP5_MOAR_CASE file, unless you want to use some other fields of your
own):
&metgrid
fg_name= ' CCSM4_CMIP5_MOAR_CASE' ,
constants_name = ,
io_form_metgrid = 2,
/
67

Or, if you are working with the bias-corrected CCSM output:

&metgrid
fg_name= ' CCSM4_CMIP5_MOAR_BC_CASE' ,
constants_name = ,
io_form_metgrid = 2,
/
As noted above you may also want to set " geog_data_res = 'modis_lakes+30s', " or "
geog_data_res = 'usgs_lakes+30s', " in the &geogrid section of namelist.wps so that
you get the inland lakes surface type. This is a must-do action if you plan to use the
TAVGSFC field to diagnose inland lake temperatures.
10. Some of the programs in the WPS/util directory may be handy for checking your
intermediate files: rd_intermediate.ext (for seeing what's in them) and plotfmt.exe (for
plotting out intermediate data for a quick look)
Data Access via GLOBUS
Regional climate model data have been made publically available on the GLOBUS
data server. If there is interest in acquiring the RCM data from GLOBUS, first a user
must create a Globus account, which enables one to use either a web interface or
command line interface to transfer files between endpoints (http://globus.org). Globus
also offers a feature called Globus Connect Personal, which enables you to move files
easily to and from your laptop or desktop computer and other endpoints. The figure
below shows the GLOBUS link to the data. Users should download the GLOBUS
personal connect, create a user account, and then search for the endpoint,
dnyates#ParaAgua in the Transfer File Endpoint.

68

References
Bruyre C.L., J.M. Done, G.J. Holland, and S. Fredrick, 2013: Bias Corrections of
Global Models for Regional Climate Simulations of High-Impact Weather. Climate
Dynamics, doi:10.1007/s00382-013-2011-6.
Bruyre CL, Done JM, Holland GJ, Fredrick S. 2013. Bias corrections of global models for
regional climate simulations of high-impact weather. Clim. Dyn., doi:10.1007/s00382-0132011-6.
Dee, D.P., with 35 co-authors, 2011: The ERA-Interim reanalysis: configuration and
performance of the data assimilation system. Quart. J. R. Meteorol. Soc., 137, 553597 (DOI: 10.1002/qj.828).
Done JM, Holland GJ, Bruyre CL, Leung LR, Suzuki-Parker A. 2013. Modeling high-impact
weather and climate: lessons from a tropical cyclone perspective. Climatic Change,
doi:10.1007/s10584-013-0954-6.
ECMWF, 2009, updated monthly. ERA-Interim Project. Research Data Archive at the
National Center for Atmospheric Research, Computational and Information Systems
Laboratory. http://rda.ucar.edu/datasets/ds627.0/. Accessed 2014.
ECMWF, 2012. ERA-Interim Project, Monthly Means. Research Data Archive at the
National Center for Atmospheric Research, Computational and Information Systems
Laboratory. http://rda.ucar.edu/datasets/ds627.1/. Accessed 2014.
Gent PR, Danabasoglu G, Donner LJ, Holland MM, Hunke EC, Jayne SR, Lawrence DM, Neale
RB, Rasch PJ, Vertenstein M, Worley PH, Yang Z-L, Zhang M. 2011. The Community
Climate System Model version 4. J. Climate, 24: 4973-4991, doi: 10.1175/2011JCLI4083.1.
69

Hurrell, J.W., et al, 2013: The Community Earth System Model: A Framework for
Collaborative Research. Bull. Amer. Meteor. Soc., 94, 13391360.
IPCC (Intergovernmental Panel on Climate Change). 2013. Climate Change 2013: The Physical
Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change (Stocker TF, Qin D, Plattner G-K, Tignor M,
Allen SK, Boschufng J, Nauels A, Xia Y, Bex V, Midgley PM, eds). Cambridge, UK and New
York, NY:Cambridge University Press.
Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie, 2004: CMORPH: A method that produces
global precipitation estimates from passive microwave and infrared data at high spatial and
temporal resolution.. J. Hydromet., 5, 487-503.
Knutti, R., D. Masson, and A. Gettelman, 2013: Climate model genealogy: Generation CMIP5
and how we got there. Geophys. Res. Lett., 40, 11941199, doi:10.1002/grl.50256.
Lorenz C., and H. Kunstmann, 2012: The Hydrological Cycle in Three State-of-the-Art
Reanalyses: Intercomparison and Performance Analysis. J. Hydrometeor, 13, 13971420.
Masui T, Matsumoto K, Hijioka Y, Kinoshita T, Nozawa T, Ishiwatari S, et al. 2011. An
emission pathway for stabilization at 6 Wm2 radiative forcing. Clim Change 109: 59-76.
Moss, R.H., and co-authors. 2010: The next generation of scenarios for climate change
research and assessment. Nature, 463, 747-756 doi:10.1038/nature08823.
Rasmussen R, and Coauthors. 2011. High-Resolution Coupled Climate Runoff Simulations of
Seasonal Snowfall over Colorado: A Process Study of Current and Warmer Climate. J.
Climate, 24, 30153048.
Riahi, K., S. Rao, V. Krey, C. Cho, V. Chirkov, G. Fischer, G. Kindermann, and N. Nakicenovic,
2011: RCP8.5 A scenario of comparatively high greenhouse gas emissions. Climatic
Change, 109, 3357.
Skamarock WC, Klemp JB. 2008 A time-split non-hydrostatic atmospheric model for weather
research and forecasting applications. J Comput Phys, 227: 3465-3485
Skamarock, W.C., J.B. Klemp, M.G. Duda, L. Fowler, S-H. Park, and T.D. Ringler,
2012: A Multi-scale Nonhydrostatic Atmospheric Model Using Centroidal Voronoi
Tesselations and C-Grid Staggering. Monthly Weather Review, 240, 3090-3105,
doi:10.1175/MWR-D-11-00215.1.
Taylor K.E., R.J. Stouffer, and G.A. Meehl, 2012: An Overview of CMIP5 and the Experiment
Design. Bull. Amer. Meteorol. Soc., 93: 485498 doi:10.1175/BAMS-D-11-00094.1.
Thomson, A. M., K. V. Calvin, S. J. Smith,G. P. Kyle, A. Volke, P. Patel, S. Delgado-Arias, B.
Bond-Lamberty, M. A. Wise, L. E. Clarke, and J. A. Edmonds, 2011: RCP4.5: a pathway for
stabilization of radiative forcing by 2100. Climatic Change, 109, 77-94.

70

Xu Z, Yang Z-L. 2012. An improved dynamical downscaling method with GCM bias
corrections and its validation with 30 years of climate simulations. J. Climate 25: 6271-6286,
doi:10.1175/JCLI-D-12-00005.1.

71

U.S. Agency for International Development

PARA-Agua Project
Calle Miguel Dasso 134-Ofic. 702
San 72
Isidro
Lima, Peru