1. Focus of analysis – reason for using PRECIS RCM
PRECIS RCM experiment is part of a larger topic of PhD research into the “The nature and mechanisms of climate variability and change in East and Central Africa and their impact on Terrestrial Hydrology in Uganda”
Project Aims and Objectives;
Develop a soil-moisture balance model to represent basin hydrology (groundwater, surface flows, surface storage)
Evaluate ability of RCM to reproduce the magnitude as well as spatial and temporal distribution of African precipitation (daily time-step)
Assess impact of climate variability/change and tectonic setting on terrestrial hydrology
Model hydrological and climatological extreme events
The aim of using the PRECIS RCM was to be able to infill a currently sparse observational network, and therefore run future estimations of changes in river flow, groundwater and surface water storage. The work to date primarily focuses on a 2500 km² river catchment in north western Uganda. Therefore analysis of the RCM will be carried out on both a regional and basin scale.

2. The PRECIS regional climate model and its ability to represent observed precipitation on annual, monthly and daily timescales.
Experiment Design
Model domain – Lon 26-40E Lat 8N-8S
EW points 60 NS points 71
Boundary Data – ECMWF ERA40 reanalysis data ( )
Resolution 0.22
Diagnostics – Daily and climate meaning
Length 45 years
Output – PP format
Changes were made to the elevation of inland lakes, Lake Victoria – 1133m, Lake Albert – 619m, Lake Edward – 920m, Lake George – 912m, Lake Kyoga – 914m, Lake Trikana – 375m, Lake Tanganyika 773m.
Also being run for the same region is the baseline experiment - ( ) and a future scenario run using the three member ensemble of SRES A2 scenario experiments ( )
Region for which validation has been carried out Lon 28-37E Lat 4S-6N
Three model domains were initially tested as follows
Region 1 – UCLAI – LON 26-48E, LAT 8S-18N
Region 2 – UCLAJ – LON 26-40E, LAT 8S-8N
Region 3 – UCLAL – LON 22-42E, LAT 11S-15N
The three model experiments were run for the period and analysis of the data to observed CRU and UDEL precipitation was carried out on the monthly timescale. The bias, RMSE and spatial correlation were calculated between modelled and observed data. The above region was selected based on these results as it provided the most realistic representation of the observed datasets, in terms of spatial correlation, bias and RMSE. This region was subsequently used for all experiment model runs in the region.

3. PRECIS Climatology Vs Observations –(1960-90)
Methods of Comparison- All results were compared in terms of Spatial correlation, Root mean squared error (RMSE) and bias. Three observed datasets were used CRU, UDEL, and VASCLIMO, all grided to the 0.5 degree.
MEAN MONTHLY PRECIPITATION
PERCENTAGE BIAS
SCORR
RMSE
PRECIS
VASCLIMO
CRU
UDEL
PRECIS-CRU
PRECIS-VAS
PRECIS-UDEL
Jan
4.04
1.81
1.64
1.69
146.42
122.94
139.47
0.78
0.79
3.06
3.09
3.03
Feb
4.83
2.46
2.21
2.17
118.27
95.88
122.23
0.70
0.71
3.35
3.16
3.30
Mar
5.28
3.40
3.27
3.32
61.45
55.01
59.01
0.52
0.56
2.85
2.75
2.83
Apr
4.63
5.11
4.93
5.16
-6.12
-9.41
-10.40
0.09
0.08
2.52
2.56
May
3.58
4.41
3.94
4.06
-9.07
-18.86
-11.92
0.43
0.48
1.98
1.95
1.97
Jun
262
2.55
2.63
18.90
15.75
15.28
0.77
1.50
1.67
1.51
Jul
2.90
2.69
2.72
13.64
7.64
6.39
0.83
0.85
1.32
1.73
1.26
Aug
3.47
3.18
3.07
3.20
13.14
9.15
8.40
0.75
2.02
Sep
3.34
3.23
3.28
21.97
18.20
20.14
0.74
1.86
2.06
1.77
Oct
4.64
3.51
3.42
3.54
35.69
32.12
31.15
0.47
2.44
Nov
3.81
3.56
3.63
13.46
6.11
11.43
0.58
0.61
2.19
2.15
2.16
Dec
3.70
2.31
2.12
74.34
60.31
67.45
0.76
2.71
2.73
For analysis PRECIS data was regrided to the 0.5 degree resolution for comparison against CRU, VASCLIMO and UDEL data and to the 2.5 degree resolution for comparison with the NCEP dataset. Never, was the coarser resolution observational data regrided to a finer resolution than it was derived.
Spatial correlations have not been calculated between PRECIS and VASCLIMO at present due to the differences between the two data sets in representing Lake Victoria. For direct comparison Lake Victoria will need to blocked out the dataset as no values are interpolated over the lake in the VASCLIMO dataset.
PRECIS overestimates precipitation in all months except April, with the worst errors occurring between Dec-Mar
Worst bias between Dec-Mar reaching peaks of 146%, the magnitude of precipitation is reasonably represented in APR, MAY, JUL, AUG,NOV
The spatial representation of precipitation is best In the two dry seasons (DJF, JJA)
The worst spatial correlations are in the first rains (MAM) with almost zero correlation in April.

4. PRECIS Climatology Vs Observations –(1960-90)
Is model error within the constraints of error in observed datasets?
MEAN MONTHLY PRECIPITATION
PERCENTAGE BIAS
SCORR
RMSE
PRECIS
VASCLIMO
CRU
UDEL
VAS-UDEL
UDEL-CRU
CRU-VAS
Jan
4.04
1.81
1.64
1.69
6.90
2.82
-10.53
0.96
1.72
0.41
0.50
Feb
4.83
2.46
2.21
2.17
11.85
-1.81
-11.43
0.94
2.08
0.64
Mar
5.28
3.40
3.27
3.32
2.52
1.51
-4.15
0.90
2.36
0.73
Apr
4.63
5.11
4.93
5.16
-1.11
4.56
-3.63
0.81
3.59
0.92
1.06
May
3.58
4.41
3.94
4.06
7.88
3.13
-12.07
3.70
0.87
1.24
Jun
3.03
262
2.55
2.63
-0.40
3.04
-2.72
2.77
0.59
0.89
Jul
2.90
2.69
2.72
-1.18
6.38
-5.57
0.95
2.99
0.66
0.99
Aug
3.47
3.18
3.07
3.20
-0.69
4.19
-3.66
0.93
0.82
1.22
Sep
3.34
3.23
3.28
1.62
1.50
-3.20
3.26
1.02
Oct
4.64
3.51
3.42
3.54
-0.74
-2.70
0.91
3.36
0.71
1.01
Nov
3.81
3.56
3.63
4.77
1.79
-6.93
2.44
0.67
0.76
Dec
2.31
2.12
4.26
3.95
-8.75
2.13
0.57
Highest Bias in Jan, Feb, May – But always less than 11% significantly lower than between modelled and observed datasets (range 6-146%)
Lowest spatial correlations between observed data still in Apr and May
Relatively small variation in magnitude of observed precipitation (Low RMSE)
PRECIS does not fall within the constraints of error between observed datasets

5. PRECIS Climatology Vs Observations –(1960-90)
TEMPORAL ANALYSIS OF RAINFALL
Early representation of the peak rainfall in both seasons (~1 Month)
Poor representation of magnitude of precipitation in first dry season (DJF)
Reasonable representation of magnitude of precipitation Apr-Sep.
No trend in overestimation or underestimation of precipitation
Good agreement between all observed datasets

6. PRECIS Climatology Vs Observations –(1960-90)
Dec PRECIS
Jan PRECIS
Feb PRECIS
Dec CRU
Jan CRU
Feb CRU

7. PRECIS Climatology Vs Observations –(1960-90)
Mar PRECIS
Apr PRECIS
May PRECIS
Mar CRU
Apr CRU
May CRU

8. PRECIS Climatology Vs Observations –(1960-90)
Jun PRECIS
Jul PRECIS
Aug PRECIS
Jul CRU
Aug CRU
Jun CRU

9. PRECIS Climatology Vs Observations –(1960-90)
Sep PRECIS
Oct PRECIS
Nov PRECIS
Nov CRU
Sep CRU
Oct CRU

10. Does PRECIS provide us with more knowledge than NCEP data?
MEAN MONTHLY PRECIPITATION
BIAS
PERCENTAGE BIAS
SCORR
RMSE
PRECIS
NCEP
CRU
NCEP-CRU
PRECIS-CRU
VAS-UDEL
UDEL-CRU
Jan
4.04
3.82
1.64
2.18
2.40
132.8
146.42
0.57
0.79
2.47
3.03
Feb
4.82
2.21
1.83
2.61
83.14
118.27
0.73
0.71
1.98
3.30
Mar
5.28
5.32
3.27
2.06
2.01
62.92
61.45
0.39
0.56
1.29
2.83
Apr
4.63
5.88
4.93
0.96
-0.30
19.39
-6.12
0.15
0.08
2.09
2.52
May
3.58
4.75
3.94
0.82
-0.36
20.77
-9.07
0.52
0.48
2.51
1.97
Jun
3.95
2.55
1.41
55.13
18.90
0.60
0.78
3.16
1.51
Jul
2.90
3.80
1.25
0.35
49.04
13.64
0.75
0.85
3.45
1.26
Aug
3.47
4.12
3.07
1.05
0.40
34.21
13.24
0.80
1.50
Sep
4.72
3.23
1.48
45.92
21.97
0.66
0.74
2.70
1.77
Oct
4.64
5.63
3.42
1.22
64.48
35.69
0.49
0.47
2.33
Nov
5.23
3.56
1.67
46.98
13.46
0.70
0.61
1.42
2.16
Dec
3.70
4.21
2.12
1.58
98.17
74.34
0.65
2.10
2.73
PRECIS data was regridded to a 2.5 degree resolution for this analysis.
NCEP does a better job in estimating the magnitude of precipitation in Jan and Feb, but overestimates mean precipitation to a greater extent than PRECIS for the rest of the year.
PRECIS has a lower percentage bias in all months apart from JAN, FEB.
PRECIS exhibits a higher spatial correlation in most months with the only significant differences in April and November, where the spatial correlation is poorer than NCEP.
April NCEP data has a very low spatial correlation similar to PRECIS
Wet seasons have a poorer spatial correlation than dry seasons.

11. Monthly and annual timescales?
A comparison of PRECIS and observational data was also carried out on monthly and annual timescales. The analysis was carried out for a four year period
The results of the analysis on monthly timescales very much confirmed what was illustrated in the climatology. The major findings were …
Difficulties representing the two rainy seasons
Worst correlations in the first rains MAM
May always the worst correlation
PRECIS fails to accurately represent either the spatial distribution or magnitude of precipitation for the large precipitation events/seasons

12. Daily Data – Station data vs Modelled
Comparisons of daily data will be carried out for the entire region using MIRA satellite data for the period , and at a basin scale using gridded station data for the period
The gridded station dataset was derived from 20 rain gauges in and around the Rukungiri catchment (see map on next slide). The raw datasets for the period had all there gaps in-filled based on the precipitation characteristics of the datasets to provide 15 years of complete data. These were gridded to a 0.25 degree resolution using an inverse distance weighting method.
The outline of the Rukungiri basin is shown on the map and the six PRECIS grid cells corresponding to the domain on which analysis was performed is identified by the green block.
Analysis of the number of rain days was also calculated in comparison to both the gridded station data and the CRU dataset. Analysis with the CRU dataset was carried out in the four CRU grid cells that contribute to the catchment area.

13. Daily Data – Station data vs Modelled
GRID 1
GRID 4
GRID 2
GRID 5
GRID 3
GRID 6
= CATCHMENT
= RAIN GAUGE
= PRECIS GRID SQ
= CRU GRID SQ

14. Daily Data – Station data vs Modelled – PROBABILTY DISTRIBUTION
FREQUENCY
RANGE
STAT 1
PRECIS 1
629
1002 5
3539
1462 10
1028
1568 15
230
835 20 43
351 25 6
102 30 48 35 27 40 17 45 8 50 55 7 60 4 65 70 75 80 3 85 2 90 1 95
100
The probability distribution for grid cell 1 ( ) indicates….
An overestimation by PRECIS in the number of rain days of zero rainfall
A massive underestimation by PRECIS of the number of rain days between 5-10 mm
An overestimation in the number of rain days of between mm
PRECIS underestimates small rainfall events and overestimates large rainfall events, this is common for all six grid cells.
The location of GRID Cell one can be clearly identified on the previous slide.

15. Daily Data – Station data vs Modelled – NUMBER OF RAIN DAYS
GRID CELL
PRECIS
STATION 1
3290
3498 2
2979
3391 3
3346 4
3263
3237 5
2904
3360 6
2793
3276
Analysis of the gridded station and PRECIS data indicates...
PRECIS consistently underestimates the number of precipitation days in all grid cells, by approximately 10% the only exception is grid cell 4.
PRECIS also significantly overestimates the amount of precipitation in all grid cells. There is an approximately 80% overestimation of precipitation for grid cells 1-5 with slightly reduced errors for grid cell 6.
There is almost no correlation between the gridded station data and PRECIS on a daily timestep, with all correlations below 0.2. This is likely to increase on monthly, seasonal and annual timescales, however the larger errors in magnitude of precipitation will always limit the agreement.
TOTAL PRECIPITATION
GRID CELL
PRECIS
STATION 1 2 3 4 5 6
Results for the NUMBER of rain days and the correlation were carried out for the period not 1980 as with the other analysis
GRID CELL
CORRELATION 1 2 3 4 5 6

16. 1960-90 Climatology for the Rukungiri region
Region of analysis – LON E LAT N.
Mean PRECIS
Mean CRU
BIAS
RMSE
SCORR
Jan
157.64
64.53
93.11
95.64
0.79
Feb
166.58
75.37
91.28
93.99
Mar
158.19
112.80
45.39
48.33
0.42
Apr
105.85
136.04
-30.19
34.65
May
56.91
90.94
-34.04
35.54
0.30
Jun
34.44
41.68
-7.24
8.23
0.91
Jul
39.71
31.60
8.11
8.85
0.92
Aug
89.89
72.03
17.87
18.74
0.87
Sep
132.65
111.58
21.07
22.62
0.47
Oct
158.44
117.64
40.80
41.57
0.16
Nov
137.46
130.29
7.17
14.33
0.09
Dec
140.05
87.35
52.70
55.78
0.70
The bias in Jan, Feb remains high but is significantly lower than for the entire region
Spatial correlation increased in MAM but significantly reduced in SON
Best spatial correlations remain in the dry season
Still a poor representation of the onset of rains
RUKUNGIRI REGION IS THE FOUR GRID CELLS THAT ENCOMPASS THE RUKUNGIRI CATCHMENT

17. MIRA satellite data
Comparisons of MIRA satellite data at the 0.1 degree resolution and PRECIS model output were carried out on the regional scale at the daily time-step, using RMSE, BIAS and spatial correlation. For the period
The set of descriptive statistics for the RMSE and BIAS between the two data sets indicates a relatively consistent error between the two datasets indicated by a low standard deviation.
BIAS
RMSE
MEAN
1.73
4.53
MEDIAN
1.79
4.49
MODE
1.92
2.23
S.DEV
0.19
0.57
RANGE
1.44
4.57
MIN
0.48
2.01
MAX
6.58
The spatial correlations were calculated for a small sample of this data set in The results indicate very low spatial correlations between the two datasets often less than 0.3 with maximum correlations in the order of 0.5.
Some of this error may come from the relatively poor spatial resolution of PRECIS in comparison to the MIRA data and also the stringent test of analysis on the daily time-step. The relatively low bias, however, suggests that the magnitude of precipitation agrees fairly well between the two datasets.