Long-term Trend of Global Land Precipitation: Uncertainties in Gauge-based Analyses

1
Long-term Trend of Global Land Precipitation
Uncertainties in Gauge-based Analyses

• Mingyue Chen1) , Pinging Xie2),
• John E. Janowiak2), Phillip A. Arkin3)

1) RS Information Systems, Inc. 2) Climate
Prediction Center, NCEP/NWS/NOAA 3) Earth Systems
Science Interdisciplinary Center, UMD
The 29th Annual Climate Diagnostics Prediction
Workshop, 2004
2
Background

• Long-term trends in temperature and precipitation
  have been examined using STATION OBSERVATIONS
  e.g. Karl et al. 1993 Lamb and Peppler 1991
• SPATIAL DISTRIBUTION of the long-term trend is
  needed for many applications such as model
  verifications
• Long-term trend in analysis field may be biased
  due to changes in gauge network

3
Objectives

• To describe the spatial distribution of
  long-term trend of precipitation using
  gauge-based analyses over land, and
• To explore ways to quantify uncertainties of the
  long-trend in the gauge based analyses due to
  changes of gauge networks

4
Data

• PREC/L
• The global monthly precipitation analysis over
  land from 1948-present
• Optimal interpolation (OI) of gauge observation
• 2.5o lat/lon
• Gauge observations
• Monthly precipitation collected in GHCN v2 of
  NCDC/NOAA
• Monthly precipitation collected in CAMS of CPC
• Over 17,000 stations
• From 1948 to the present

5
Linear Trend of Annual Mean Precipitation (PREC/L,
1948-2003)

• Increasing trend over the US, NW Australia,
• Decreasing trend over the equatorial Africa, E
  Australia,
• The similar patterns are observed in other
  published gauge
• based analyses, e.g. Dai et al. (1997), and New
  et al. (2000)

6
Spatial Distribution of Available gauges

• The spatial distribution of gauge network
  changes
• Good coverage in earlier years over most regions
• The US region has good coverage through the
  period

7
Time Series of the Total Number of Available
Gauges Used to Define the Gauge-Based Analysis

• The total number of available gauges changes
• The maximum during 1960s
• Decreased during later period

8
We conducted comparative studies to examine how
the magnitude of the gauge-based analyses vary
with

• Gauge network configuration and
• Interpolation algorithms

9
Detailed Examinations of the Gauge-Based
Analyses over the Sahel Region
10
Time series of reporting station number

• The number of gauge stations changes
• Subset stations with relatively high reporting
  rates

11
Experiment IComparisons of gauge-based analyses
using various gauge networks (1931-1980)

• Select a period with the best gauge availability
  over the region 1931 1980
• Construct analyses using observations at stations
  with 80 or higher reporting rates (the fixed
  network) and those available at 1921, 1931, ,
  1991, 2001 (the changing networks)
• Compare the trends calculated from the analyses
  based on different gauge networks
• Analyses are created using the OI and Shepard
  algorithms

12
Number of gauge stations on 0.5olat/lon grid

• The gauge coverage is reasonably well, but
• Less stations at the northern dry regions

13
Interpolation Algorithms

• OI (Optimal Interpolation of Gandin 1965)
• ?Interpolate the monthly anomalies
• ?Weighting statistically
• ?Add the interpolated anomalies to climatology
• Shepard (1965)
• ?Interpolate the monthly total
• ?Inverse-distance weighting
• ?Using 4-10 nearest stations

14
Areal mean of annual precipitation from
OI/Shepard over the Sahel region(1931-1980,
June-Sep.)

• Similar trends in the analyses with various gauge
  networks
• The RMSD is much less the magnitude of long-term
  trend
• OI interpolation is less affected by the gauge
  network
• than Shepard

15
Spatial distributions of annual mean, trend, RMSD
of trend(1931-1980, June-Sep.)

• Over most of the Sahel region the trend
  uncertainties due to
• change of gauge network is very limited
• The Shepard produce more small scale feature of
  trend pattern
• OI is less affected by the change of gauge
  network

16
Experiment IIComparisons of trends interpolated
using using various gauge networks for data
period 1948-2003

• Assume the trend calculated from the PREC/L
  gauge-based analysis for 1948 2003 is true
• Interpolate the trend using gauge networks for
  each year of the 56-year period
• Compare the 56 sets of interpolated trend
  distribution to get insight into the uncertainties

17
Spatial distribution of trend calculated with
gauge networks of different years(1948-2003)

• Trend distribution is smoothed
• The overall patterns
• of trend are similar even when networks are very
  sparse (e.g.2000)

18
Trend calculated with gauge networks of different
years over the Sahel region

• Overall, trends calculated using various gauge
  networks do not show big difference with that
  based on a dense network
• Differences in the calculated trend are larger
  when networks are sparser

19
Summary of Results for the Sahel Region

• The annual precipitation over the regions of
  Sahel have been decreasing during the periods of
  1931-1980 1948-2003
• The uncertainties exist due to the change of
  network through the period
• The magnitude of the uncertainties in trend is
  much less than that of the trend itself
• The OI algorithm produces gauge-based analysis
  with less alias in magnitude than the Shepard

20
Examinations over the Global Land Areas for
1948 2003
21
Spatial distributions of annual mean, trend, RMSE
of trend (1948-2003)
22
Summary and Future Work

• The spatial distribution of major trend of annual
  precipitation has been described from the
  long-term gauge based analysis
• The uncertainties due to the change of gauge
  network through the period has been explored
• The uncertainties are related to interpolation
  algorithms, the OI interpolation is better than
  the Shepard
• The trend are related to gauge network but the
  trend alias over the major trend regions is
  limited
• Future work is underway to further quantify the
  uncertainties, such as, significance test, etc.